JP7550991B2 - System, method, and program for estimating faulty part and replacement part - Google Patents

Description

The present disclosure relates to a system, method, and program for estimating faulty parts and replacement parts.

In many cases, equipment repairs require the replacement of parts. When the equipment to be repaired is installed in a customer's factory, laboratory, etc., the customer contacts a call center to inquire about the equipment's malfunction status, and the call center staff often infers the cause of the equipment's malfunction and replacement parts from the details of the inquiry. However, inferring the cause of the malfunction and replacement parts requires not only advanced specialized knowledge but also a wealth of experience. For this reason, depending on the level of expertise of the call center staff, there is a possibility that the estimation of the cause of the malfunction and the replacement parts may differ. Therefore, for example, Patent Document 1 describes a repair and replacement part instruction system that can estimate candidates for replacement parts by registering information in advance that corresponds to the malfunction content, repair method, and replacement parts that may be used for repair.

JP 2007-304935 A

In the repair and replacement part instruction system described in Patent Document 1, if the device has a simple structure and a limited number of parts, the device's status and the faulty part are uniquely linked, making it possible to accurately estimate the faulty part and replacement part. However, in the case of large devices with many parts, or devices with complex configurations that combine multiple different functions, there are often multiple fault causes and faulty parts even if the device's status is the same. For this reason, multiple fault causes and faulty parts are linked to one fault phenomenon, making it difficult to estimate the faulty part. In addition, because there are many parts linked to multiple fault causes and faulty parts, many candidates for replacement parts appear, making it difficult to estimate the replacement part required for repair.

The present disclosure has been made to solve the above-mentioned problems, and aims to provide a faulty part/replacement part estimation system, method, and program that can accurately estimate the faulty part and replacement part.

In order to achieve the above-mentioned object, the faulty part/replacement part estimation system according to the present disclosure includes a faulty part estimation unit that estimates the faulty part of the equipment based on equipment status data including a fault status of the equipment and equipment information data including the model of the equipment, a part category search unit that searches for a part category of a replacement part according to the faulty part estimated by the faulty part estimation unit, and a replacement part search unit that searches for data of candidate replacement parts in accordance with the faulty part estimated by the faulty part estimation unit and the part category searched by the part category search unit.

According to the present disclosure, it is possible to provide a faulty part/replacement part estimation system that can accurately estimate the faulty part and replacement part based on data such as equipment information and equipment status.

FIG. 1 is a diagram showing a configuration of a faulty part/replacement part estimation system according to a first embodiment of the present disclosure; FIG. 1 is a diagram showing an example of a display screen of a terminal device according to the first embodiment; FIG. 1 is a diagram showing an example of a past case display according to the first embodiment; FIG. 1 is a diagram showing an overview of a learning unit of a server according to a first embodiment; FIG. 1 is a diagram showing an overview of a failure part estimation unit of a server according to a first embodiment; FIG. 1 is a diagram showing an overview of a part category search unit of a server according to a first embodiment; FIG. 1 is a diagram showing an overview of a replacement part search unit of a server according to a first embodiment; FIG. 1 is a diagram showing an overview of a past case search unit of a server according to a first embodiment; FIG. 1 is a diagram showing a table of customer master data according to the first embodiment; FIG. 1 is a diagram showing a table of device master data according to the first embodiment; FIG. 1 is a diagram showing a table of construction report master data according to the first embodiment. FIG. 11B is a continuation of the table of the construction report master data shown in FIG. 11A. FIG. 1 is a diagram showing a table of document data according to the first embodiment; FIG. 1 is a diagram showing an example of a word conversion table according to the first embodiment; FIG. 1 is a diagram showing an example of part master data according to the first embodiment; FIG. 1 is a diagram showing a procedure for generating a word conversion table according to the first embodiment; FIG. 1 is a diagram showing a procedure for generating a word conversion table according to the first embodiment; FIG. 1 is a diagram showing a procedure for generating a word conversion table according to the first embodiment; FIG. 1 is a diagram showing an overview of a neural network according to a first embodiment; FIG. 1 is a diagram showing an example of a hardware configuration of a server according to a first embodiment; FIG. 1 is a diagram showing an example of a hardware configuration of a terminal device according to a first embodiment; Flowchart of learning process according to the first embodiment A flowchart of a process for acquiring basic learning data in the learning process shown in FIG. A flowchart of the word conversion process in the learning process shown in FIG. A flowchart of a process for generating learning data in the learning process shown in FIG. 21. Flowchart of a trained model generation process in the learning process shown in FIG. Flowchart of part estimation process according to the first embodiment A flowchart of a process for acquiring basic data for estimation in the part estimation process shown in FIG. A flowchart of a process for generating estimation data in the part estimation process shown in FIG. A flowchart of a failure part estimation process in the part estimation process shown in FIG. Flowchart of part category search processing in the part estimation processing shown in FIG. 26 A flowchart of a replacement part search process in the part estimation process shown in FIG. Flowchart of past case search processing in the part estimation processing shown in FIG. 26 FIG. 1 is a diagram showing a configuration of a faulty part/replacement part estimation system according to a second embodiment of the present disclosure; Flowchart of part recommendation process according to the second embodiment Flowchart of recommended part display process according to the second embodiment FIG. 13 is a diagram showing the configuration of a faulty part/replacement part estimation system according to a third embodiment of the present disclosure. FIG. 13 is a diagram showing an overview of a learning unit of a server according to a third embodiment; FIG. 13 is a diagram showing an overview of a failure part estimation unit of a server according to a third embodiment. FIG. 13 is a diagram showing a table of device operation status data according to the third embodiment. FIG. 13 is a diagram showing a table of abnormality determination data according to the third embodiment. Flowchart of device operation status data generation process according to the third embodiment Flowchart of abnormality determination data generation process according to the third embodiment

(Embodiment 1)
Hereinafter, a faulty part and replacement part estimation system 100 according to a first embodiment of the present disclosure will be described with reference to the drawings. Note that the same reference numerals are used to denote the same or equivalent parts.

The faulty part/replacement part estimation system 100 is a device that can learn data such as customer information for equipment suppliers, equipment information, and equipment status to generate a learning model for estimating the faulty part, estimate the faulty part using the generated learning model, and estimate replacement parts according to the estimated faulty part.

Figure 1 shows an overview of a faulty part/replacement part estimation system 100. The faulty part/replacement part estimation system 100 comprises a server 1 with the function of learning and estimating the faulty part, and a terminal device 2 for presenting replacement parts estimated based on customer information, the state of the equipment, etc. The server 1 and the terminal device 2 are connected to each other via a network 3. The terminal device 2 is installed in a call center, and a receptionist can input customer information, the fault status of the equipment contacted by the customer, etc., to present the estimated faulty part and replacement parts.

The server 1 includes a learning unit 11 that generates a learning model for estimating the faulty part, a faulty part estimation unit 12 that estimates the faulty part, a part category search unit 13 that roughly identifies a replacement part according to the faulty part, a replacement part search unit 14 that searches for replacement part candidates, a past case search unit 15 that searches for past fault cases, and a memory unit 16 that stores various data.

The learning unit 11 learns data such as customer information of the customer who delivered the equipment, equipment information, and equipment status, and generates a learning model for estimating the faulty part. The faulty part estimation unit 12 estimates the faulty part using the learning model generated by the learning unit 11. The part category search unit 13 searches for a part category for roughly identifying a replacement part according to the faulty part selected by the call center receptionist on the screen of the terminal device 2. The part category refers to the classification name of a part group that includes old parts that are no longer available and parts that are currently available. By checking the part category, the call center receptionist can roughly identify a replacement part based on the classification name of the part group.

The replacement part search unit 14 searches for replacement part candidates according to the faulty part and part category selected by the call center receptionist on the screen of the terminal device 2. The past case search unit 15 searches for past failure cases according to the faulty part, part category, and replacement part selected by the call center receptionist on the screen of the terminal device 2.

The terminal device 2 is equipped with a failure status input unit 21 for inputting the customer information of the customer to whom the equipment was delivered, the status of the equipment, etc., a failure part display unit 22 for displaying the failure part of the equipment, a part category display unit 23 for displaying the part category of replacement parts, a replacement part display unit 24 for displaying candidate replacement parts, and a past case display unit 25 for displaying past failure cases.

Figure 2 shows an example of a display screen of the terminal device 2. The display screen includes a failure status input section 21 for inputting various information, a failure part display section 22 that displays the failure part of the equipment, a part category display section 23 that displays the part category of replacement parts, and a replacement part display section 24 that displays candidate replacement parts.

The failure status input section 21 includes a customer information input section 211 for inputting the customer information of the customer to whom the equipment was delivered, a reception number section 212 for inputting the reception number of the call center, a failure status input section 213 for inputting the failure status of the equipment, and a failure part estimation button 214 for estimating the failure part. The customer information input section 211 includes various input sections for inputting a customer ID (Identification) for uniquely specifying the equipment owned by the customer, the installation date of the equipment, the model name, the model classification, the control device for inputting the type of the control device of the equipment, the type of power supply, and the software version. Note that the various information such as the installation date, model, and control device of the equipment may be directly input by the receptionist of the call center, or may be automatically input using the data contained in the customer master 161 and the equipment master 162 stored in the memory section 16 described later.

The reception number field 212 is an input field into which the call center receptionist enters the reception number when an inquiry is received from a customer. The reception number is a number that is determined in advance by the manufacturer and is assigned based on naming rules. The malfunction status input field 213 is an input field into which the call center receptionist enters the malfunction status of the equipment that they have heard from the customer.

The fault part estimation button 214 is a button for executing the processing of the fault part estimation unit 12 shown in Figure 1. Specifically, after all the items in the fault status input unit 21 are filled in, the call center receptionist presses the fault part estimation button 214. This causes the processing of the fault part estimation unit 12 of the server 1 shown in Figure 1 to be executed.

The faulty part display unit 22 includes an estimation result display unit 221 that displays the faulty part estimated by the estimation processing unit 123 of the server 1 described below, and a part category display button 222 for displaying part categories. The estimation result display unit 221 displays the estimation results estimated by the estimation processing unit 123 of the server 1 shown in FIG. 1. The estimation results display each estimated faulty part in order of probability. A check box is provided for each faulty part. The call center receptionist checks the check box for the estimated faulty part based on the equipment status heard from the customer and the estimation result of the faulty part.

The parts category display button 222 is a button for causing the parts category search unit 13 of the server 1 shown in Figure 1 to search for the parts category of the replacement parts and display the search results. Specifically, the call center receptionist checks the check box of the faulty part displayed in the estimation result display unit 221 and presses the parts category display button 222. This causes the processing of the parts category search unit 13 of the server 1 shown in Figure 1 to be executed.

The part category display unit 23 includes a part category search result display unit 231 that displays the part category search results searched by the part category search unit 13 of the server 1 shown in FIG. 1, and a replacement part display button 232 for displaying replacement parts.

The part category search result display unit 231 displays the part category search results found by the part category search unit 13 of the server 1 shown in FIG. 1 in the order of most frequently searched part categories for each fault part checked in the estimation result display unit 221. A check box is provided for each part category. The call center receptionist checks the check box for the estimated part category based on the equipment status received from the customer, the fault part shown in the estimation result display unit 221, and the part category search results.

The replacement parts display button 232 is a button for causing the replacement parts search unit 14 of the server 1 shown in Figure 1 to search for replacement parts and display the search results. Specifically, a call center receptionist checks the check box of the parts category displayed in the parts category search result display unit 231 and presses the replacement parts display button 232. This causes the processing of the parts category search unit 13 of the server 1 shown in Figure 1 to be executed.

The replacement parts display unit 24 includes a replacement parts search result display unit 241 that displays replacement parts searched for by the replacement parts search unit 14 of the server 1 shown in FIG. 1, and a past case display button 242 for displaying past failure cases.

The replacement parts search result display unit 241 displays the search results of replacement parts searched for by the replacement parts search unit 14 in order of frequency of searches for each faulty part checked in the estimation result display unit 221 and each part category checked in the part category search result display unit 231. A check box is provided for each replacement part. The call center receptionist checks the check box for the estimated replacement part based on the equipment status received from the customer, the faulty part shown in the estimation result display unit 221, the part category search results, and the replacement parts search results.

The past case display button 242 is a button for causing the past case search unit 15 of the server 1 shown in Figure 1 to search for past failure cases and display the search results. Specifically, a call center receptionist checks the check box of the replacement part displayed in the replacement part search result display unit 241 and presses the past case display button 242. This causes the processing of the past case search unit 15 of the server 1 shown in Figure 1 to be executed.

When the past case search unit 15 of the server 1 searches for past failure cases, it causes the search results to be displayed on the display screen of the terminal device 2 in the past case display unit 25 of the terminal device 2 shown in Figure 1. Figure 3 shows a past case search result display 251 in which the search results for past failure cases are displayed on the display screen of the terminal device 2. The past case search result display 251 includes, as display items, the report number, the phenomenon that occurred due to the failure, the cause of the failure, the measures to be taken to resolve the failure, and replacement parts.

For example, as shown in Figure 3, in the case of report number "XXXXX," the symptom of the malfunction was "the operation screen does not light up," and the cause of the malfunction was "a broken wire for the display." In this case, the measure to resolve the malfunction was "replacement of the wiring," and "AB1234" and "AB1256" were used as replacement parts.

Next, the functions of the server 1 shown in FIG. 1 will be described below with reference to FIG. 4 to FIG. 8. First, FIG. 4 is a diagram showing the configuration of the learning unit 11. The learning unit 11 includes a learning basic data acquisition unit 111 that acquires various data that form the basis of learning, a word conversion unit 112 that converts words included in documents into numerical values, a learning data generation unit 113 that generates learning data, and a trained model generation unit 114 that creates a learning model.

As shown in Figure 4, the learning basic data acquisition unit 111 acquires various data that form the basis of learning from the customer master 161, equipment master 162, and construction report master 163 stored in the storage unit 16 described below, and generates learning basic data. The word conversion unit 112 extracts documents from document data 165 stored in the storage unit 16 described below, and divides sentences contained in the documents into words. The word conversion unit 112 then generates a word conversion table 164 for converting each word into a numerical value. The word conversion unit 112 stores the generated word conversion table 164 in the storage unit 16.

The learning data generation unit 113 acquires the learning basic data generated by the learning basic data acquisition unit 111 and the word conversion table 164 generated by the word conversion unit 112, and generates learning data. The trained model generation unit 114 trains a neural network on the learning data, and generates a trained model 166. The trained model generation unit 114 stores the generated trained model 166 in the memory unit 16. The learning in the trained model generation unit 114 will be described in detail later.

Next, Fig. 5 shows an overview of the failure part estimation unit 12 of the server 1 shown in Fig. 1. The failure part estimation unit 12 includes an estimation basic data acquisition unit 121 that acquires various data input to the failure situation input unit 21 of the terminal device 2, an estimation data generation unit 122 that generates estimation data, and an estimation processing unit 123 that estimates the failure part. The estimation basic data acquisition unit 121 acquires various data such as device information and device status input by the call center receptionist to the failure situation input unit 21 shown in Fig. 2 as estimation basic data.

The estimation data generation unit 122 generates estimation data using the basic data for estimation acquired by the basic data for estimation acquisition unit 121 and the word conversion table 164 stored in the memory unit 16. The estimation processing unit 123 acquires the learned model 166 stored in the memory unit 16, and sets the learned model 166 in the constructed neural network. The estimation processing unit 123 inputs the estimation data generated by the estimation data generation unit 122 to the neural network, and estimates the faulty part of the equipment. The estimation processing unit 123 displays the estimated faulty part of the equipment on the faulty part display unit 22 of the terminal device 2.

Figure 6 is a diagram showing an overview of the part category search unit 13 of the server 1 shown in Figure 1. The part category search unit 13 acquires from the terminal device 2 various data such as equipment information and equipment status input by the call center receptionist to the failure status input unit 21 shown in Figure 2, and data on the failure parts with a check mark in the check box among the failure parts displayed in the estimation result display unit 221 of the failure part display unit 22 shown in Figure 2. Next, the part category search unit 13 acquires the data of the part master 167 and the construction report master 163 stored in the memory unit 16. The part category search unit 13 searches for a part category using the acquired various data, and displays the search result on the part category display unit 23 of the terminal device 2.

Figure 7 shows an overview of the replacement part search unit 14. The replacement part search unit 14 acquires from the terminal device 2 various data such as equipment information and equipment status that the call center receptionist has input to the failure status input unit 21 shown in Figure 2. Furthermore, the replacement part search unit 14 acquires from the terminal device 2 data on the failure parts with the checkboxes checked among the failure parts displayed in the estimation result display unit 221 of the failure part display unit 22 shown in Figure 2 by the call center receptionist, and data on the product categories with the checkboxes checked among the part categories displayed in the part category search result display unit 231 of the part category display unit 23.

Next, the replacement part search unit 14 acquires the data of the parts master 167 and the construction report master 163 stored in the memory unit 16. The replacement part search unit 14 searches for replacement parts using the acquired data, and displays the search results on the replacement part display unit 24 of the terminal device 2.

8 is a diagram showing an overview of the past case search unit 15. The past case search unit 15 acquires various data such as device information and device status input by the call center receptionist to the failure status input unit 21 shown in FIG. 2 from the terminal device 2. Furthermore, the past case search unit 15 acquires data of the failure parts with the checkboxes checked among the failure parts displayed in the estimation result display unit 221 of the failure part display unit 22 shown in FIG. 2 by the call center receptionist, data of the product categories with the checkboxes checked among the part categories displayed in the part category search result display unit 231 of the part category display unit 23, and data of replacement parts with the checkboxes checked among the replacement parts displayed in the replacement part search result display unit 241 of the replacement part display unit 24.

Next, the past case search unit 15 acquires the data of the construction report master 163 stored in the memory unit 16. The past case search unit 15 searches for past failure cases using the acquired data, and displays the search results on the past case display unit 25 of the terminal device 2.

Next, the configuration of the various data stored in the memory unit 16 will be described below with reference to Figs. 9 to 14. First, Fig. 9 shows a table of the customer master 161. The customer master 161 is data that uniquely links a combination of the customer to whom the equipment was delivered and data related to the purchased equipment. The customer master 161 includes items such as a customer ID indicating the customer to whom the equipment was delivered, a model indicating the model name of the equipment, an installation date indicating the date the equipment was installed at the customer's premises, a control device built into the equipment, a power supply indicating the model name of the power supply unit built into the equipment, and software built into the equipment.

For example, as shown in Figure 9, customer ID "A0001" is linked to the model "AB", the installation date "2017/12/16", the control device "CT789", the power supply "VA3456", and the software "SA689".

Figure 10 shows a table of equipment master 162. Equipment master 162 includes a model indicating the model name of the equipment, and an equipment model classification. The equipment model classification is an item for inputting the implementation method for realizing the functions of the equipment. For example, the equipment model classification may be motor type, electric type, etc. For example, as shown in Figure 10, in the case of model "AB", "motor type" is input as the equipment model classification.

11A and 11B show the table of the construction report master 163. The construction report master 163 is a table for recording the status of equipment, the inspection and repair methods, etc., when an engineer inspects and repairs equipment malfunctions, abnormalities, failures, etc. that occur in the equipment.

The construction report master 163 contains multiple items, including the report number, the customer ID of the customer who delivered the equipment, the model indicating the model name of the equipment, the construction date when the inspection or repair was performed, the date and time when the equipment failure occurred, the phenomenon which is the condition of the equipment such as malfunction or abnormality, the cause of the phenomenon, the action taken by the engineer, the major and minor failure categories which are codes indicating the parts of the failure, and the part code which indicates the part code of the replacement part. The contents of the items of the construction report master 163, namely, the phenomenon, cause, and action, are written in natural language.

Among the items in the construction report master 163, the major failure part category and minor failure part category represent the failure part by combining the codes entered in both categories. The major failure part category is the system into which the equipment parts are classified, for example, the mechanical control system, the electrical control system, etc. The minor failure part category is the specific name of the equipment part. For example, system items such as "mechanical control system" and "power supply system" are set as the major failure part category, and part items such as "servo amplifier" and "power supply wiring cable" are set as the minor failure part category.

Additionally, a numerical value corresponding to each item is set for the system item in the major failure part category and the part item in the minor failure part category. For example, if the system item in the major failure part category is "machine control system", the code is set to "1", and if the system item is "power supply system", the code is set to "2". For example, if the part item in the minor failure part category is "servo amplifier", the code is set to "5", and if the part item is "power supply cable", the code is set to "4". The numerical code values corresponding to each item in the major failure part category and minor failure part category are arbitrarily set values.

Figure 12 shows a table of document data 165. Document data 165 includes a document number indicating the document number, and a document content item which is text related to the equipment. The text written in the document content item is written in natural language. The document content item contains various text data indicating how a given word is used in equipment maintenance and inspection work. The text data is composed of documents such as equipment installation history, call center reception response history, and equipment instruction manuals.

Figure 13 is a table of word conversion table 164. Word conversion table 164 includes items of words and vectors indicated by the words. The vectors indicated by the words include data for the number of dimensions that the vector has. The generation of word conversion table 164 will be described in detail later. Figure 14 is a table of part master 167. Part master 167 includes items of part code, which is a code set for a part, part name, and part category, which categorizes the parts.

Here, the data of the sentences listed in the document content section included in the table of document data 165 shown in FIG. 12 is used in the word conversion section 112 of the learning section 11 shown in FIG. 4 to generate the word conversion table 164 shown in FIG. 13. The word conversion table 164 is generated using word vectors calculated from the weights of the hidden layer of the neural network that performs text processing. The processing of the word conversion section 112 will be described below with reference to FIGS. 15 and 16. The word conversion section 112 obtains the data of the sentences listed in the document content section from the table of document data 165 shown in FIG. 12, and performs morphological analysis processing to divide it into morphemes.

For example, as shown in FIG. 15, the word conversion unit 112 acquires the document content data of document number "1" from the document data 165 shown in FIG. 12, "Device A has a lever in front of the operation screen, so...", and divides it into morphemes. Specifically, the word conversion unit 112 divides the sentence "Device A has a lever in front of the operation screen, so..." into words, such as "Device", "A", "is", "operation screen", "of", "in front", "to", "lever", "is", "there", and "because". If the sentence contains punctuation marks, the punctuation marks are divided into one word. Furthermore, if there are duplicate words, they are combined into one.

However, neural networks that process text cannot directly process the text itself, i.e., the words themselves. In order to process with a neural network, the words to be processed must be converted into fixed-length vectors. One method for converting into fixed-length vectors is to convert a string of characters into a one-hot vector. A one-hot vector is a vector in which only one of the vector's elements is "1" and the rest are all "0". To generate a one-hot vector, a vector with elements equal to the number of morphemes into which a sentence is divided is prepared, and a "1" is assigned to a different vector element for each morpheme.

For example, as shown in FIG. 15, the words "equipment," "A," "is," "operation screen," "of," "in front," "to," "lever," "ga," "aru," and "no de" divided by the word conversion unit 112 are 11 in number. Therefore, each word is converted into a one-hot vector having 11 elements. For example, the first element of "operation screen" is set to "1," and the rest are set to "0." Similarly, the second element of "ga" is set to "1," and the rest are set to "0." Below, all morphemes are converted into one-hot vectors in order.

Next, the generated one-hot vector is input to the input layer of the neural network. An overview of the neural network is shown in FIG. 16. When an arbitrary word is input to the input layer of the neural network, the value is multiplied by weight W1 and input to the hidden layer, and the result is further multiplied by weight W2 and output from the output layer. The output result of the output layer changes depending on the values of weights W1 and W2. In this embodiment 1, the neural network learns the probability of surrounding words for an arbitrary word by so-called supervised learning. The surrounding words are words placed immediately before and after the arbitrary word. The neural network learns by inputting an arbitrary word to the input layer and adjusting weights W1 and W2 so that the result output from the output layer approaches the surrounding words of the word.

Words input to the input layer of a neural network are input in the form of a one-hot vector. Specifically, each element of the one-hot vector is input to each neuron in the input layer. For example, as shown in Figure 16, when "equipment" is input to the input layer, the one-hot vector "10000000000" shown in Figure 15, whose first element is "1" and the rest are "0", is input to each neuron in the input layer.

Here, the data of the document content of document number "1" shown in Figure 12 is "Device A has a lever in front of the operation screen, so...". In this case, the word immediately before and after "device", i.e., the surrounding word, is "A". For this reason, when "device" is input to the input layer, the neural network adjusts the weights W1 and W2 so that the probability of "A" appearing in the output layer is maximized.

The neural network is trained to learn the probability of surrounding words for every word. The weights of the neural network's hidden layer are matrices, and can be used as word vectors to quantify words. The word vectors corresponding to each word can be extracted by multiplying the weights of the neural network's hidden layer by the one-hot vector of each word.

For example, as shown in Figure 17, by multiplying the one-hot vector "10000000000" of the word "equipment" by the weight matrix of the hidden layer of the neural network, it is possible to extract the row of the weight matrix corresponding to 1 in the one-hot vector. Here, the row of the weight matrix corresponding to 1 in the one-hot vector of the word "equipment" is "12 28 ... 34". Therefore, "12 28 ... 34" becomes the word vector of the word "equipment".

Furthermore, if the morpheme "equipment" appears in a subsequent sentence included in the document content data of document number "1" shown in Figure 12, the surrounding words in that case are also learned in the same way. After this, all words that appear in the document content data are learned, and the weight W1 of the intermediate layer is calculated. Each row of the weight W1 of the intermediate layer becomes the word vector of each morpheme. The word vectors are associated with words and stored in the memory unit 16 of the server 1 shown in Figure 4 as the word conversion table 164 shown in Figure 13.

4, the learning unit 11 of the server 1 quantifies various data acquired from the memory unit 16 by the learning basic data acquisition unit 111 using the word conversion table 164 stored in the memory unit 16 by the learning data generation unit 113 to generate learning data. Specifically, the learning data generation unit 113 divides the learning basic data acquired by the learning basic data acquisition unit 111 into sentences related to the state of the device into morphemes and converts each morpheme into a word vector using the word conversion table 164.

Next, the learning data generation unit 113 digitizes the model data and installation date data contained in the customer information and equipment information contained in the basic learning data. The model data is converted into a one-hot vector with an element of the model type. For example, model A can be digitized as (1, 0, 0), model B as (0, 1, 0), and model C as (0, 0, 1).

The installation date data is calculated by using the construction date, which is also included in the learning basic data, to calculate the number of days that have passed since the installation date. Specifically, the learning data generation unit 113 acquires the construction date data entered in the table of the construction report master 163 shown in FIG. 11A. The learning data generation unit 113 subtracts the date of the equipment installation date data entered in the table of the customer master 161 shown in FIG. 9 from the date of the acquired construction date data. This subtracted value is the number of days that have passed since the equipment was installed. Note that the number of days that have passed is expressed as a decimal. The learning data generation unit 113 links the quantified customer information, equipment information, and equipment status data with the data of the first failure part classification and the second failure part classification to generate learning data.

Next, the learning data generation unit 113 inputs the generated learning data to the learned model generation unit 114 of the learning unit 11 shown in FIG. 4. The learned model generation unit 114 trains the learning data in a neural network constructed therein, and generates weights for the intermediate layer and the output layer as a learned model 166. Then, the learned model generation unit 114 stores the learned model 166 in the memory unit 16.

Specifically, the trained model generation unit 114 inputs data from the training data that is a concatenation of a one-hot vector having an element representing the model type, which is equipment information, the number of days since the installation date, and a sentence vector representing the equipment status, into the input layer of the neural network constructed by the trained model generation unit 114. Note that hereinafter, this concatenated data is referred to as training input data. In addition, the trained model generation unit 114 sets the fault part classification from the training data as a one-hot vector having elements equal to the total number of fault part classifications, into the output layer of the neural network.

FIG. 18 shows the neural network constructed in the trained model generating unit 114. The neural network is composed of an input layer, an intermediate layer, and an output layer, each including a plurality of neurons. In the first embodiment, the intermediate layer is assumed to be one layer. For example, in the case of a three-layer neural network shown in FIG. 18, when a plurality of inputs are input to the input layers X ₁ to X _n , the values are multiplied by weights W ₁₁ to W _nm and input to the intermediate layers Y ₁ to Y _m . The results input to the intermediate layers Y ₁ to Y _m are further multiplied by weights V ₁₁ to V _km and output from the output layers Z ₁ to Z _k . The output results from the output layers Z ₁ to Z _k vary depending on the values of the weights W ₁₁ to W _nm and the weights V ₁₁ to V _km . Here, n and k are integers of 4 or more, and m is an integer of 3 or more.

In the first embodiment, when learning input data is input to each neuron of the input layer X ₁ to X _n , the weights W ₁₁ to W _nm and the weights V ₁₁ to V _km are adjusted so that the results output from the output layers Z ₁ to Z _k approach the data of the fault part classification of the learning data. Adjusting the weights W ₁₁ to W _nm and the weights V ₁₁ to V _km is learning in the neural network. The weights W ₁₁ to W _nm and the weights V ₁₁ to V _km are updated using, for example, a back propagation method. In the following, adjusting the weights W ₁₁ to W _nm and the weights V ₁₁ to V _km is referred to as learning in the learned model generation unit 114. The learned model generation unit 114 stores the adjusted weights W ₁₁ to W _nm and weights V ₁₁ to V _km as a learned model 166 in the storage unit 16 shown in FIG.

5 reads out the learned model 166 stored in the storage unit 16 and sets it in the neural network constructed in the estimation processing unit 123. Specifically, the weights W ₁₁ to W _nm and weights V ₁₁ to V _km , which are the learned model 166, are set in the neural network. As a result, when a one-hot vector having an element of the model type, which is equipment information, and data linking the number of days since the installation date, etc. are input to each neuron in the input layer X ₁ to X _n , the neural network constructed in the estimation processing unit 123 can output data of the fault part classification most related to the input content from the output layer Z ₁ to Z _k .

In this embodiment, if the structure of the device is complex, the number of fault part classifications may become enormous, and the amount of data to be trained may be insufficient for the number of fault part classifications. In this case, the fault part classifications are classified into a first fault part classification, a second fault part classification, etc. based on the device structure, and the data is trained separately for each classified fault part classification.

For example, the input layer of the neural network is input with learning data that concatenates a one-hot vector having an element representing the model type, which is equipment information, the number of days since the installation date, and a sentence vector representing the equipment condition. The output layer of the neural network is set with a one-hot vector having elements for the total number of types included in the first fault part classification. This creates a trained model that estimates the first fault part classification.

After that, a trained model that estimates the first fault part classification is generated for all training data. Then, the total number of types of data included in the first fault part classification is divided into data by type, and a trained model that estimates the second fault part classification for each type of data is created.

Specifically, for example, assume that there are nine types of first fault part classifications, A1, A2, ..., A9. In this case, all nine types of first fault part classifications are first set in the output layer, and a trained model that estimates the first fault part classification is created. In this case, data that is a concatenation of a one-hot vector having an element of the model type, which is equipment information, the number of days since the installation date, and a sentence vector representing the equipment status, from the training data, is input to the input layer of the neural network.

Next, the data included in A1 of the first fault part classification is collected, and a one-hot vector of the second fault part classification is set in the output layer. For example, assume that there are a total of 1000 pieces of data entered into the table of the construction report master 163 shown in Figures 11A and 11B. Of these, assume that there are 300 pieces of data with "1" set in the "major fault part classification" entered into the table of the construction report master 163 shown in Figure 11B. And assume that "1" in the "major fault part classification" includes nine types of classifications, A1, A2, ..., A9. In this case, if there are 50 pieces of "A1" data, collect the 50 pieces of "A1" data. Then, generate a one-hot vector with elements of the collected "A1" data, and set it in the output layer of the neural network.

This generates a "trained model that estimates the second fault part classification in the case of A1 of the first fault part classification". The above procedure is performed for all first fault part classifications up to A9, and a total of 10 trained models are created. At the time of estimation, the trained model for estimating the second fault part classification corresponding to the fault part classification with the highest probability among those estimated in the first fault part classification is operated. For example, at the time of estimation, the first fault part classification is first estimated as the fault part major classification in a state in which the correct fault part major classification and fault part minor classification are not known. The estimation result is that the classification A1 included in the first fault part classification is 50%, the classification A2 is 20%, the classification A3 is 10%, etc. In this case, a trained model that estimates the second fault part classification included in the classification A1 with the highest probability is selected and used. As a result, it is possible to estimate the second fault part classification included in the fault part major classification with the highest probability, i.e., the fault part minor classification, among the estimated first fault part classifications.

Each function executed by the server 1 and terminal device 2 of the faulty part/replacement part estimation system 100 shown in Figure 1 is a function realized by software. In this embodiment 1, software that executes processing to realize each function of the learning unit 11 of the server 1 is referred to as a learning processing program. Also, software that executes processing to realize each function of the faulty part estimation unit 12, part category search unit 13, replacement part search unit 14, and past case search unit 15 of the server 1 is referred to as a part estimation processing program. An example of the hardware configuration of the server 1 for executing the learning processing program and the part estimation processing program is shown in Figure 19.

The server 1 includes a storage device 101 that stores various programs and various data, a connection unit 102 for connecting to the terminal device 2, a memory 103 for expanding the various programs, and a processor 104 for executing the various programs. The storage device 101, the connection unit 102, the memory 103, and the processor 104 are interconnected via a data bus 105.

The storage device 101 is a device that functions as the storage unit 16 shown in Figure 1. The storage device 101 stores various programs executed by the processor 104, and various data such as the customer master 161, the equipment master 162, and the construction report master 163 shown in Figures 4 to 8. The storage device 101 can be configured using a storage device such as a hard disk drive (HDD) or a solid state drive (SSD).

The connection unit 102 is a connection port that can be connected to the terminal device 2. The connection unit 102 functions as the network 3 shown in FIG. 1. The connection unit 102 can be configured using various ports to which devices can be connected, such as a USB (Universal Serial Bus) port, an IEEE 1394 port, etc.

The memory 103 is a device for expanding the various programs stored in the storage device 101. The memory 103 can be configured using a storage element and a storage medium, for example, a volatile or non-volatile semiconductor memory such as a RAM (Random Access Memory) or a flash memory.

The processor 104 reads various programs stored in the storage device 101, expands them in the memory 103, and executes them. The processor 104 can be configured using a processing device such as a CPU (Central Processing Unit) or an MPU (Micro-processing Unit).

Furthermore, the terminal device 2 displays on the display screen the failure status input section 21 shown in Fig. 2, the failure part display section 22 that displays the failure part of the equipment, etc. In this embodiment 1, the display control program is software that executes processes to realize each function, such as the function of displaying various information on the display screen of the terminal device 2 and the function of accepting data input by a call center receptionist into the failure status input section 21. An example of the hardware configuration of the terminal device 2 for executing the display control program is shown in Fig. 20.

The terminal device 2 includes a storage device 201 that stores various programs and various data, a connection unit 202 for connecting to the server 1, an input device 203 that accepts input of various data, a display device 204 that displays various data, a display controller 205 that generates display data to be displayed on the display device 204, a memory 206 for expanding the various programs, and a processor 207 that executes the various programs. The storage device 201, the connection unit 202, the input device 203, the display controller 205, the memory 206, and the processor 207 are connected to each other via a data bus 208.

The storage device 201 stores various programs executed by the processor 207, and display data such as images and characters to be displayed on the display device 204. The storage device 201 can be configured using a storage device such as a hard disk drive (HDD) or a solid state drive (SSD).

The connection unit 202 is a connection port that can be connected to the server 1. The connection unit 202 functions as the network 3 shown in FIG. 1. The connection unit 202 can be configured using various ports to which devices can be connected, such as a USB (Universal Serial Bus) port, an IEEE 1394 port, etc.

The input device 203 is an input unit through which the call center receptionist inputs various data. The input device 203 can be configured using, for example, a keyboard, a mouse, a touch panel, etc. The display device 204 displays a display screen including the failure status input unit 21 and the failure part display unit 22 shown in FIG. 2. The display device 204 also displays various data input by the call center receptionist via the input device 203. The display device 204 can be configured using, for example, an LCD (Liquid Crystal Display), an organic EL (Electroluminescence) monitor, etc.

The display controller 205 is a controller that outputs a video signal to the display device 204 to display display data including characters and images. The display controller 205 can be configured using a video signal output device such as a video card, a GPU (Graphics Processing Unit), a graphics board, etc.

The memory 206 is a device for expanding the various programs stored in the storage device 201. The memory 206 can be configured using storage elements and storage media, such as volatile or non-volatile semiconductor memory such as RAM (Random Access Memory) and flash memory.

The processor 207 reads various programs stored in the storage device 201, expands them in the memory 206, and executes them. The processor 207 can be configured using a processing device such as a CPU (Central Processing Unit) or an MPU (Micro-processing Unit).

Next, the flow of operation of the faulty part/replacement part estimation system 100 according to the first embodiment will be described below with reference to the flowcharts shown in FIGS. 21 to 32. The faulty part/replacement part estimation system 100 first generates a trained model 166 in the learning unit 11 of the server 1 shown in FIG. 1. The processing performed by the learning unit 11 is stored in the memory unit 16 of the server 1 as a learning processing program. The faulty part/replacement part estimation system 100 causes the processor 104 of the server 1 shown in FIG. 19 to read the learning processing program stored in the storage device 101 into the memory 103 and execute it at the timing of generating the trained model 166. The processing of the learning processing program will be described below with reference to the flowcharts shown in FIGS. 21 to 25.

First, in FIG. 21, the learning basic data acquisition unit 111 shown in FIG. 4 executes a learning basic data acquisition process (step S10). The learning basic data acquisition process will be described with reference to the flowchart shown in FIG. 22. The learning basic data acquisition unit 111 acquires customer information data from the customer master 161 shown in FIG. 4 (step S101). The customer information data acquired includes the customer ID, model, installation date, control device, power supply, and software data contained in the table of the customer master 161 shown in FIG. 9. In the following, the data acquired here will be collectively referred to as customer information data.

Next, the learning basic data acquisition unit 111 acquires from the equipment master 162 model category data corresponding to the "model" acquired from the customer master 161 (step S102). The model category data is acquired from the table of the equipment master 162 shown in Figure 10, which corresponds to the model data acquired from the customer master 161. Note that, hereinafter, the model data and the acquired model category data are collectively referred to as equipment information data.

The learning basic data acquisition unit 111 acquires report number data, equipment status data, and faulty part data corresponding to the customer ID and model information acquired from the customer master 161 from the table of the construction report master 163 shown in Figures 11A and 11B (step S103). Here, the equipment status data is the data entered in the phenomenon item of the construction report master 163. Moreover, the faulty part data is the data in the faulty part major category item and the faulty part minor category item of the construction report master 163.

The learning basic data acquisition unit 111 generates learning basic data by linking the acquired customer information data, equipment information data, equipment status data, and faulty part data using the report number data acquired from the construction report master 163 as a key (step S104). The learning basic data acquisition unit 111 stores the generated learning basic data in the memory unit 16 (step S105).

Now, let us return to FIG. 21. The word conversion unit 112 of the learning unit 11 shown in FIG. 4 executes a word conversion process (step S11). The word conversion process will be described with reference to the flowchart shown in FIG. 23. The word conversion unit 112 acquires text data entered in the "Document Content" field of the table of the document data 165 shown in FIG. 12 from the document data 165 stored in the storage unit 16 shown in FIG. 4 (step S111). The word conversion unit 112 executes a morphological analysis process on the acquired text data and divides it into morphemes (step S112). Specifically, the word conversion unit 112 divides the text data into words, including punctuation marks.

The word conversion unit 112 generates a one-hot vector for each word divided into morphemes. The word conversion unit 112 inputs the generated one-hot vector to the input layer of the neural network, and causes it to learn the probability of surrounding words for any word. After learning, the word conversion unit 112 extracts word vectors by multiplying the weights of the hidden layer of the neural network by the one-hot vector of each word (step S113).

The word conversion unit 112 associates the words with the extracted word vectors and generates a word conversion table 164 (step S114). The word conversion unit 112 stores the generated word conversion table 164 in the memory unit 16 (step S115).

Now, let us return to FIG. 21. The learning data generation unit 113 shown in FIG. 4 executes a process for generating learning data (step S12). The learning data generation process will be described below with reference to the flowchart shown in FIG. 24. The learning data generation unit 113 acquires basic learning data from the memory unit 16 (step S121). The learning data generation unit 113 executes a morphological analysis process on sentences included in the device status data contained in the received basic learning data, and divides them into morphemes (step S122).

The learning data generation unit 113 obtains the word conversion table 164 from the storage unit 16. The learning data generation unit 113 replaces each morpheme, i.e., each word constituting the sentence of the device status data, with a word vector included in the word conversion table 164, and combines the replaced word vectors. Hereinafter, the combined word vector is referred to as the sentence vector of the sentence of the device status.

Next, the learning data generation unit 113 generates learning data (step S123). Specifically, first, the learning data generation unit 113 digitizes the model data and installation date data contained in the customer information data and equipment information data contained in the learning basic data. The model data is converted into a one-hot vector having an element of the model type. The installation date data is calculated by subtracting the installation date data from the construction date data contained in the learning basic data to find the number of days that have passed since the installation date.

The learning data generation unit 113 generates learning data by linking the quantified customer information data, equipment information data, and equipment status data with the data of the first failure part classification and the data of the second failure part classification. The learning data generation unit 113 stores the generated learning data in the memory unit 16 (step S124). Now, return to FIG. 21. The trained model generation unit 114 shown in FIG. 4 executes a trained model generation process (step S13). The trained model generation process will be described below with reference to the flowchart shown in FIG. 25.

The trained model generation unit 114 acquires the training data from the storage unit 16 (step S131). The trained model generation unit 114 trains the training data in a neural network constructed in itself, and generates the weighting of the intermediate layer and the output layer as the trained model 166 (step S132). Specifically, the trained model generation unit 114 sets the data of the first fault part classification and the data of the second fault part classification out of the training data in the output layer of the neural network. Next, the trained model generation unit 114 inputs the device information and the device state out of the training data into the input layer of the neural network, and trains the neural network. The trained model generation unit 114 saves the generated trained model 166 in the storage unit 16 (step S133). Returning to FIG. 21, the trained model generation unit 114 ends the training process.

When a call center receptionist receives a call from a customer about a device failure, the call center receptionist uses the faulty part/replacement part estimation system 100 to estimate replacement parts. The faulty part/replacement part estimation system 100 used by the call center receptionist executes a part estimation processing program for estimating parts on the server 1 side, and executes a display control program for displaying the fault status input unit 21, faulty part display unit 22, etc. shown in FIG. 2 on the display screen on the terminal device 2 side. When the server 1 is started, the processor 104 of the server 1 shown in FIG. 19 reads the part estimation processing program stored in the storage device 101 into the memory 103 and executes it. When the terminal device 2 is started, the processor 207 of the terminal device 2 shown in FIG. 20 reads the display control program stored in the storage device 201 into the memory 206 and executes it.

When a call center receptionist receives a call from a customer about a device failure, the call center receptionist inputs the customer ID, reception number, and the failure situation heard from the customer into the failure situation input section 21 displayed on the display screen of the terminal device 2 shown in FIG. 2. When the customer ID is input into the failure situation input section 21, the terminal device 2 may search the customer master 161 stored in the storage unit 16 shown in FIG. 5 by the customer ID, and automatically input the installation date data, model data, control device data, power supply data, and software data extracted as the search results into each item of the failure situation input section 21. The terminal device 2 may also search the device master 162 in the storage unit 16 using the "model data" acquired from the customer master 161, and input the model classification extracted as the search result into the corresponding item of the failure situation input section 21. After completing the input of the necessary items, the call center receptionist presses the failure part estimation button 214. As a result, the terminal device 2 causes the failure part estimation unit 12 of the server 1 shown in FIG. 5 to execute the part estimation process.

The part estimation process will be described below with reference to the flowcharts of Figures 26 to 32. First, in Figure 26, the estimation basic data acquisition unit 121 included in the failure part estimation unit 12 of the server 1 shown in Figure 5 executes the estimation basic data acquisition process (step S20). The estimation basic data acquisition process will be described with reference to the flowchart shown in Figure 27.

The estimation basic data acquisition unit 121 acquires device information data and device status data from the failure status input unit 21 shown in FIG. 5 (step S201). Next, the estimation basic data acquisition unit 121 links the acquired device information data and device status data to generate estimation basic data (step S202). The estimation basic data acquisition unit 121 outputs the generated estimation basic data to the estimation data generation unit 122 shown in FIG. 5.

Returning now to Fig. 26, the estimation data generation unit 122 executes a process for generating estimation data (step S21). The estimation data generation process will be described with reference to the flowchart shown in Fig. 28.

The estimation data generation unit 122 acquires estimation basic data from the estimation basic data acquisition unit 121 (step S211). The estimation data generation unit 122 divides the sentence data of the device status data included in the estimation basic data into morphemes (step S212). The estimation data generation unit 122 acquires the word conversion table 164 from the storage unit 16 (step S213).

The estimation data generation unit 122 converts each morpheme obtained by dividing the sentence data of the device status data in step S212 into a word vector using the word conversion table 164. The estimation data generation unit 122 adds all the word vectors to create a device status vector. The estimation data generation unit 122 concatenates the device information data and the device status vector to generate estimation data (step S214). The estimation data generation unit 122 stores the estimation data in the storage unit 16 (step S215).

Returning now to FIG. 26, the estimation processing unit 123 executes the fault location estimation process (step S22). The fault location estimation process will be described with reference to the flowchart shown in FIG. 29.

The estimation processing unit 123 acquires estimation data from the memory unit 16 (step S221). Next, the estimation processing unit 123 acquires the trained model 166 from the memory unit 16 (step S222). The estimation processing unit 123 constructs a neural network within itself (step S223).

The estimation processing unit 123 inputs the estimation data to the input layer of the neural network (step S224). This multiplies the estimation data by the weight, which is the learned model 166, and the probability of the fault part classification can be output to the output layer. If the fault part classification is one layer, one output may be sufficient. However, if the fault part classification is two layers, the learned model of the second fault part classification corresponding to the highest first fault part classification in the first output is used to construct the neural network again. Then, the probability of the second fault part class is output to the output layer.

The estimation processing unit 123 acquires the faulty part based on the output result of the neural network (step S225). Specifically, the estimation processing unit 123 acquires the faulty part classification with the highest probability as the faulty part among the probabilities of the faulty part classifications output to the output layer of the neural network. The estimation processing unit 123 displays the acquired faulty part on the faulty part display unit 22 of the terminal device 2 shown in FIG. 2 (step S226). Note that the first faulty part classification and the second faulty part classification output by the faulty part estimation unit 12 are displayed on the faulty part display unit 22. These may also be combined into one and displayed as one word. For example, the first faulty part classification is part A and the second faulty part classification is part B. In this case, part A and part B may be listed together, or if part B alone is understandable, only part B may be displayed.

Now, let us return to FIG. 26. The part category search unit 13 shown in FIG. 6 determines whether or not the part category display button 222 shown in FIG. 2 has been pressed (step S23). If the part category display button 222 has not been pressed (step S23; NO), the part category search unit 13 waits until it is pressed. If the part category display button 222 has been pressed (step S23; YES), the part category search unit 13 executes a part category search process (step S24). The part category search process will be described with reference to the flowchart shown in FIG. 30.

The part category search unit 13 acquires device information data from the failure status input unit 21 of the terminal device 2 (step S231). The part category search unit 13 acquires faulty part data from the faulty part display unit 22 of the terminal device 2 (step S232). The part category search unit 13 acquires the part master 167 and the construction report master 163 from the storage unit 16 (step S233).

Next, the part category search unit 13 searches the construction report master 163 using the first fault part classification, the second fault part classification, and the model data included in the data of the fault part acquired from the fault part display unit 22 as a key. The part category search unit 13 converts the part codes included in the multiple records that are the search results into part categories using the part master 167 and tabulates them (step S234). This generates a ranking of past cases of the part category. The part category search unit 13 displays the tabulated part categories on the part category display unit 23 of the terminal device 2 shown in FIG. 2 (step S235). Note that there may be cases where the check boxes of multiple fault part classifications are checked on the fault part display unit 22. In this case, the part category search unit 13 performs tabulation processing for each checked fault part classification and displays each result on the part category display unit 23.

Now, let us return to FIG. 26. The call center receptionist selects a part category that is likely to be useful for repair from among the estimated part categories. The replacement part search unit 14 shown in FIG. 7 then determines whether the replacement part display button 232 shown in FIG. 2 has been pressed (step S25). If the replacement part display button 232 has not been pressed (step S25; NO), the unit waits until it is pressed. If the replacement part display button 232 has been pressed (step S25; YES), the replacement part search unit 14 executes a replacement part search process (step S26). The replacement part search process will be described with reference to the flowchart shown in FIG. 31.

The replacement part search unit 14 acquires device information data from the failure status input unit 21 of the terminal device 2 (step S241). Next, the replacement part search unit 14 acquires data on the failure part from the failure part display unit 22 of the terminal device 2 (step S242). The replacement part search unit 14 acquires a part category from the part category display unit 23 of the terminal device 2 (step S243). The replacement part search unit 14 acquires the part master 167 and the construction report master 163 from the memory unit 16 (step S244).

The replacement part search unit 14 searches the construction report master 163 using the first failure part classification data, the second failure part classification data, and the model data contained in the acquired failure part data as keys. The replacement part search unit 14 tallies the part codes contained in the multiple records that are the search results (step S245). This makes it possible to determine the proportion of replacement parts in past cases and determine rankings.

The replacement part search unit 14 displays the results of the part code aggregation on the replacement part display unit 24 of the terminal device 2 (step S246). Specifically, the replacement part search unit 14 obtains the part names and part categories corresponding to the aggregated part codes from the part master 167, and displays them in order of ranking on the replacement part display unit 24 of the terminal device 2. Note that a check box for a part category may be checked on the part category display unit 23. In this case, the replacement part search unit 14 performs an aggregation process for each checked part category, and displays each result on the part category display unit 23.

Returning to FIG. 26, the call center receptionist selects from the estimated replacement parts those that are likely to be usable for the repair. Thereafter, the past case search unit 15 shown in FIG. 8 determines whether the past case display button 242 shown in FIG. 2 has been pressed (step S27). If the past case display button 242 has not been pressed (step S27; NO), the past case search unit 15 waits until it is pressed. If the past case display button 242 has been pressed (step S27; YES), the past case search unit 15 executes a past case search process (step S28). The past case search process will be explained with reference to the flowchart shown in FIG. 32.

The past case search unit 15 acquires equipment information data from the failure situation input unit 21 shown in FIG. 8 (step S251). The past case search unit 15 acquires failure part data from the failure part display unit 22 shown in FIG. 8 (step S252). The past case search unit 15 acquires replacement part data from the replacement part display unit 24 shown in FIG. 8 (step S253). The past case search unit 15 acquires the construction report master 163 from the memory unit 16 (step S254).

Next, the past case search unit 15 searches for past cases (step S255). Specifically, the past case search unit 15 searches the construction report master 163 using the first and second failure part classifications, the model data, and the replacement part data contained in the acquired failure part data as keys. The past case search unit 15 displays the data entered in each of the fields of report number, phenomenon, cause, treatment, and part code from among the multiple records contained in the search results on the past case display unit 25 of the terminal device 2 (step S256). Here, the past case search unit 15 returns to FIG. 26 and ends the part estimation process.

As described above, according to embodiment 1, a faulty part/replacement part estimation system can be provided that can accurately estimate the faulty part and replacement part based on data such as customer information, equipment information, and equipment status, even for equipment with a complex structure and many parts.

In addition, the device status data written in natural language can be quantified using word vectors. This allows users to input device status data in the language they normally use, rather than using predefined phrases, codes, etc. This reduces the burden on users.

(Embodiment 2)
When a certain part is used to repair a fault in an equipment, it often happens that other parts are replaced as a set at the same time. For example, there are cases where a part that pairs with the part to be replaced exists, or where repair of the fault induces repair of another fault and it is necessary to repair multiple parts at once. Therefore, the faulty part/replacement part estimation system 100A according to the second embodiment has a function of recommending a part to be replaced as a set with the part used for repair, in addition to the functions of the faulty part/replacement part estimation system 100 according to the first embodiment.

The faulty part/replacement part estimation system 100A shown in FIG. 33 includes a server 1A having a function for learning and estimating the faulty part, and a terminal device 2A for presenting replacement parts estimated based on customer information, the state of the equipment, and the like. The server 1A includes a learning unit 11 for generating a learning model for estimating the faulty part, a faulty part estimation unit 12 for estimating the faulty part, a part category search unit 13 for roughly identifying replacement parts according to the faulty part, a replacement part search unit 14 for searching for replacement part candidates, a past case search unit 15 for searching past fault cases, a memory unit 16 for storing various data, and a part recommendation unit 17 for recommending parts to be replaced. The part recommendation unit 17 recommends other parts to be replaced as a set with the part used for repair.

The terminal device 2A also includes a failure status input unit 21 for inputting the customer information of the customer who delivered the equipment, the status of the equipment, etc., a failure part display unit 22 for displaying the failure part of the equipment, a part category display unit 23 for displaying the part category of the replacement part, a replacement part display unit 24 for displaying candidate replacement parts, a past case display unit 25 for displaying past failure cases, and a part display unit 26 for displaying the part names of recommended parts. The part display unit 26 displays the part names of the parts recommended by the part recommendation unit 17. The display format of the part display unit 26 is, for example, a format in which a character string of the part name, a pop-up, etc. is displayed when the mouse pointer is placed over the part name displayed in the replacement part display unit 24 shown in FIG. 2.

The parts recommendation unit 17 of server 1A searches the construction report master 163 shown in Figures 11A and 11B using the part code of each part to be replaced, the data of the first failure part classification, the data of the second failure part classification, and the model data as keys to find the parts to recommend. Here, the data of the first failure part classification is data found from the "major failure part classification" shown in the table of the construction report master 163 shown in Figure 11B. Moreover, the data of the second failure part classification is the "minor failure part classification" shown in the table of the construction report master 163 that is included in the major failure part classification with the highest probability among the data of the first failure part classification.

Specifically, the part recommendation unit 17 first searches the part master 167 shown in Fig. 14 using the part category and part name of the replacement part as keys, and obtains the part code of that part. The part recommendation unit 17 searches the construction report master 163 shown in Figs. 11A and 11B using the first failure part classification data, the second failure part classification data, and the model data as keys. The part recommendation unit 17 narrows down the search results to records that contain the part code obtained from the part master 167.

The part recommendation unit 17 counts the part codes contained in the narrowed-down records. This makes it possible to find the proportion of each part code and determine a ranking. The part recommendation unit 17 finds the part names corresponding to the counted part codes from the part master 167 and generates display data to display in order of ranking. When a call center receptionist places the mouse pointer over a part name displayed in the replacement part display unit 24 shown in Figure 2, the terminal device 2A displays the part names recommended in order of ranking in a display format such as a character string or a pop-up.

The part recommendation by the part recommendation unit 17 is executed after the replacement part search process of step S26 in the flowchart of the part estimation process shown in FIG. 26. The process of the part recommendation unit 17 is stored as a part recommendation process program in the memory unit 16 of the server 1A shown in FIG. 33. The process executed by the part recommendation process program will be described below with reference to the flowchart of the part recommendation process shown in FIG. 34.

The part recommendation unit 17 acquires the part master 167 shown in FIG. 14 and the construction report master 163 shown in FIG. 11A and FIG. 11B from the storage unit 16 of the server 1A (step S301). The part recommendation unit 17 searches the part master 167 using the part category and part name of the replacement part as keys, and acquires the part code of the part (step S302). The part recommendation unit 17 searches the construction report master 163 using the data of the first failure part classification, the data of the second failure part classification, and the model data as keys (step S303).

The part recommendation unit 17 narrows down the search results of step S303 to records that contain the part code found from the part master 167 (step S304). The part recommendation unit 17 tallies up the part codes contained in the narrowed down records and finds the proportion of each part code. The part recommendation unit 17 determines a ranking of the part codes according to the proportion of each part code found (step S305). The part recommendation unit 17 finds part names corresponding to the tallied part codes from the part master 167 and generates display data that displays them in order of ranking (step S306). The part recommendation unit 17 then ends the part recommendation process.

Furthermore, when a call center receptionist places the mouse pointer over a part name displayed in the replacement part display section 24 shown in Fig. 2, the terminal device 2A displays the names of recommended parts in order of ranking in a display format such as a character string or a pop-up. The display process for the names of recommended parts is stored in the memory section 16 of the server 1A shown in Fig. 33 as a recommended parts display processing program. The process executed by the recommended parts display processing program will be described below with reference to the flowchart of the recommended parts display processing shown in Fig. 35.

The terminal device 2A determines whether the mouse pointer is over the part name displayed in the replacement part display section 24 shown in FIG. 2 (step S311). If the mouse pointer is not over the part (step S311; NO), the terminal device 2A repeats step S311. If the mouse pointer is over the part (step S311; YES), the terminal device 2A displays the display data, which displays the recommended part names in order of ranking, generated in step S306 of the flowchart of the part recommendation process shown in FIG. 34, in the part display section 26 shown in FIG. 33 (step S312). The part display section 26 displays the part name in a display format such as a character string or a pop-up that is overlaid on the part name displayed in the replacement part display section 24 shown in FIG. 2.

The terminal device 2A determines whether the mouse pointer has been moved away from the part name displayed in the replacement part display section 24 (step S313). If the mouse pointer has not been moved away (step S313; NO), the terminal device 2A repeats step S313. If the mouse pointer has been moved away (step S313; YES), the terminal device 2A makes the display data displayed in the part display section 26 invisible (step S314). The terminal device 2A returns to step S311 and repeats the processes from step S311 to step S314.

As described above, according to embodiment 2, in addition to the effect of embodiment 1, it is possible to check the parts to be replaced as a set along with the parts to be used for repair.

(Embodiment 3)
In the above-mentioned first and second embodiments, when estimating the faulty part and the replacement part, the receptionist of the call center inputs the fault condition of the customer's device, customer information, etc., into the terminal device 2, 2A. However, if the receptionist of the call center inputs the fault condition of the device, customer information, etc., into the terminal device 2, 2A each time, the estimation process takes time and the burden on the receptionist is large. Therefore, the faulty part/replacement part estimation system 100B according to the third embodiment has, in addition to the function of the faulty part/replacement part estimation system 100 according to the first embodiment, a function of acquiring data on the operating status of the customer's device 4 delivered to the customer in real time from the customer's device 4 and using the data to estimate the faulty part, thereby improving the accuracy of estimating the faulty part.

The faulty part/replacement part estimation system 100B shown in FIG. 36 includes a server 1B having a function for learning and estimating the faulty part, and a terminal device 2A for presenting replacement parts estimated based on customer information, the state of the equipment, and the like. The server 1B is connected to a first device 4A, a second device 4B, ..., an n-th device 4n placed at the customer via a network. Here, the first device 4A is assumed to be a model "AB" delivered to a customer with a customer ID "A0001". The second device 4B is assumed to be a model "CD" delivered to a customer with a customer ID "B0123". The n-th device 4n is assumed to be a model "NM" delivered to a customer with a customer ID "N0001". In the following, the first device 4A, the second device 4B, ..., the n-th device 4n are collectively referred to as customer devices 4.

The customer device 4 is equipped with multiple sensors. An abnormality in the customer device 4 can be determined by detecting changes in the values detected by the sensors, for example, by using a change point detection algorithm when the value changes significantly, or by using a waveform prediction algorithm when there is a change in periodicity. However, if the behavior of the customer device 4 during an abnormality is very variable, it is possible to identify the cause of the failure with higher accuracy by creating a trained model that detects abnormalities in advance and determining whether or not there is an abnormality. Therefore, in this embodiment 3, the learning unit 11B and the failure part estimation unit 12B of the server 1B are provided with a function that can determine whether or not there is an abnormality.

The server 1B includes a learning unit 11B that generates a learning model for estimating the faulty part, a faulty part estimation unit 12B that estimates the faulty part, a part category search unit 13 that roughly identifies a replacement part according to the faulty part, a replacement part search unit 14 that searches for replacement part candidates, a past case search unit 15 that searches for past fault cases, a memory unit 16B that stores various data, and an equipment data acquisition unit 18 that acquires data on the operating status of the customer equipment 4.

The equipment data acquisition unit 18 acquires various measurement, detection, and other data in real time from various sensors attached to the customer equipment 4, such as sensors for measuring voltage, water volume, and the like, and sensors for detecting abnormalities. The equipment data acquisition unit 18 accumulates the acquired measurement, detection, and other data in the memory unit 16B as equipment operation status data 168. Details of the equipment operation status data 168 will be described later.

37, the learning unit 11B of the server 1B includes a learning basic data acquisition unit 111 that acquires various data that are the basis for learning, a word conversion unit 112 that converts words included in a document into numerical values, a learning data generation unit 113B that generates learning data, a trained model generation unit 114 that creates a learning model, and a learning abnormality determination unit 115 that determines abnormal parts of the customer equipment 4. The learning abnormality determination unit 115 determines abnormal parts of the customer equipment 4 based on the equipment operation status data 168 stored in the memory unit 16B for all failure occurrence dates and times of the construction report master 163 shown in FIGS. 11A and 11B.

The learning abnormality determination unit 115 accumulates data on the abnormal parts of the customer device 4 that have been determined as abnormality determination data 169 in the memory unit 16B. The learning abnormality determination unit 115 also inputs the data on the abnormal parts of the customer device 4 that have been determined to be abnormal to the learning data generation unit 113B of the learning unit 11B. The learning data generation unit 113B generates learning data that includes the data on the abnormal parts of the customer device 4 acquired from the learning abnormality determination unit 115.

As shown in FIG. 38, the failure part estimation unit 12B includes an estimation basic data acquisition unit 121 that acquires various data input to the failure status input unit 21 of the terminal device 2, an estimation data generation unit 122B that generates estimation data, an estimation processing unit 123 that estimates the failure part, and an estimation abnormality determination unit 124. The estimation abnormality determination unit 124 determines the abnormal part of the customer device 4 based on the equipment operation status data 168 stored in the memory unit 16B for all failure occurrence dates and times of the construction report master 163 shown in FIG. 11A and FIG. 11B. The estimation abnormality determination unit 124 inputs the data of the abnormal part of the determined customer device 4 to the estimation data generation unit 122B. The estimation data generation unit 122B generates estimation data including the data of the abnormal part of the customer device 4 acquired from the estimation abnormality determination unit 124.

Next, the configuration of the equipment operation status data 168 and the abnormality determination data 169 shown in Fig. 37 will be described with reference to Figs. 39 and 40. First, Fig. 39 shows a table of the equipment operation status data 168. The equipment operation status data 168 includes items such as a customer ID indicating the customer to whom the equipment was delivered, a model indicating the model name of the customer equipment 4, an acquisition date and time indicating the date and time when the data on the operation status of the customer equipment 4 was acquired, and a first sensor, a second sensor, a third sensor, and the like indicating data on various sensors attached to the customer equipment 4.

The equipment operation status data 168 is data in which data on the operation status of the customer equipment 4 acquired at regular intervals by the equipment data acquisition unit 18 of the server 1B shown in FIG. 36 is compiled in a chronological order by customer ID. The equipment operation status data 168 shown in FIG. 39 shows, for example, that for the model "AB" of customer ID "A0001", the first sensor detected "30V", the second sensor detected "20mA", and the third sensor detected "10Ω" at the acquisition date and time "2020/12/4 12:00". Also, for the model "CD" of customer ID "B0123", the first sensor detected "20V", the second sensor detected "15mA", and the third sensor detected "15Ω" at the acquisition date and time "2020/12/14 12:00".

The learning anomaly determination unit 115 searches the equipment operation status data 168 for records having the same customer ID and model as the customer ID and model of the construction report master 163 shown in Figures 11A and 11B that are to be used as the learning target.The learning anomaly determination unit 115 then obtains from the equipment operation status data 168 records for a certain period going back from the date and time of the failure that is linked to the customer ID and model of the construction report master 163.

The learning abnormality determination unit 115 shown in FIG. 37 estimates the part in which an abnormality has occurred for the record acquired from the equipment operation status data 168, for example, according to the change in the values detected from the first to third sensors, and determines the abnormality of the customer equipment 4. Specifically, first, in the construction report master 163 shown in FIG. 11A and FIG. 11B, for customer ID "A0001" and model "AB", the failure occurrence date and time is "2020/12/10 9:23". The learning abnormality determination unit 115 searches for records of customer ID "A0001" and model "AB" from the equipment operation status data 168 shown in FIG. 39. The learning abnormality determination unit 115 acquires records for a certain period of time, for example, a period going back 7 days from the failure occurrence date "2020/12/10" included in the construction report master 163 failure occurrence date and time "2020/12/10 9:23". For example, the learning abnormality judgment unit 115 acquires records for the acquisition date and time period "2020/12/4 to 2020/12/10" for "A0001" and model "AB" of the equipment operational status data 168 shown in FIG.

Here, for example, the first sensor shown in the equipment operation status data 168 is a voltage measurement sensor for detecting voltage changes in the component, the second sensor is a current measurement sensor for detecting current changes in the wiring, and the third sensor is a resistance measurement sensor for detecting wear of the component. The equipment operation status data 168 shown in FIG. 39 shows that for the model "AB" of customer ID "A0001", the first sensor detected "30V", the second sensor detected "20mA", and the third sensor detected "10Ω" at the acquisition date and time "2020/12/4 12:00". Next, it is shown that the values detected by the first to third sensors were the same as those detected previously at the acquisition date and time "2020/12/4 13:00". It is also shown that the values detected by the first to third sensors were the same as those detected previously at the acquisition date and time "2020/12/8 9:00".

At the acquisition date and time of "2020/12/10 9:00", the first sensor detected "25 V" and the second sensor detected "15 mA". The values detected by the third sensor are the same. Furthermore, at the acquisition date and time of "2020/12/10 10:00", the first sensor detected "0 V" and the second sensor detected "5 mA". The values detected by the third sensor are the same.

From the above, it can be seen that the values detected by the first and second sensors changed after the acquisition date and time "2020/12/10 9:00". The fact that a change occurred in the values of the first and second sensors indicates that a malfunction occurred in the part or area to which the first and second sensors were attached when the change occurred. For this reason, the learning abnormality judgment unit 115 shown in FIG. 37 detects a change in the values detected by the first to third sensors, and judges that an abnormality has occurred in the part or area to which the sensors were attached when the change occurred. If the learning abnormality judgment unit 115 judges that there is an abnormality, it outputs "1" as the judgment result, and if it judges that there is no abnormality, it outputs "0" as the judgment result.

The learning abnormality judgment unit 115 stores the judgment result in the abnormality judgment data 169 for each customer device 4. Figure 40 shows a table of the abnormality judgment data 169. The abnormality judgment data 169 includes items such as a report number and the first device 4A, the second device 4B, ..., the nth device 4n located at the customer's site. The report number is the same number as the report number listed in the construction report master 163 shown in Figures 11A and 11B.

For example, the report number "XXXXX" of the abnormality judgment data 169 shown in Figure 40 is a report on customer ID "A0001" and model "AB" as shown in the construction report master 163 shown in Figures 11A and 11B. Therefore, the result of the learning abnormality judgment unit 115's judgment as to whether or not there was an abnormality in the first device 4A, which is model "AB" delivered to the customer with customer ID "A0001", is recorded.

For example, the construction report master 163 shown in Figures 11A and 11B contains the failure history for customer ID "A0001" and model "AB", and the learning abnormality judgment unit 115 judges that there was an abnormality in the first device 4A from the equipment operation status data 168 shown in Figure 39. In this case, "1" is entered in the column for the first device 4A linked to the report number "XXXXX" in the abnormality judgment data 169, and "0" is entered in the columns for the second device 4B, ..., nth device 4n.

The learning abnormality judgment unit 115 also inputs the judgment result to the learning data generation unit 113B of the learning unit 11B shown in Fig. 37. The learning abnormality judgment unit 115 outputs the judgment result to the learning data generation unit 113B in the format of {construction report number, judgment result of the first device 4A, judgment result of the second device 4B, ..., judgment result of the nth device 4n} = {XXXXX, 1, 0, ..., 0}, for example. The learning data generation unit 113B generates learning data including the judgment result acquired from the learning abnormality judgment unit 115.

Similarly to the learning abnormality determination unit 115, the estimation abnormality determination unit 124 of the failure part estimation unit 12B shown in FIG. 38 determines the abnormal part of the customer device 4 based on the device operation status data 168 stored in the memory unit 16B for all the failure occurrence dates and times of the construction report master 163 shown in FIG. 11A and FIG. 11B. The estimation abnormality determination unit 124 outputs the estimation result to the estimation data generation unit 122B in the format of {construction report number, judgment result of the first device 4A, judgment result of the second device 4B, ..., judgment result of the nth device 4n} = {XXXXX, 1, 0, ..., 0}. The estimation data generation unit 122B generates estimation data including the judgment result acquired from the estimation abnormality determination unit 124. The estimation processing unit 123 executes the estimation process using the estimation data generated by the estimation data generation unit 122B.

The equipment operation status data 168 shown in FIG. 39 is generated by the equipment data acquisition unit 18 acquiring various data such as measurements and detections in real time from various sensors of the customer equipment 4 while the server 1B in FIG. 36 is operating. The generation process of the equipment operation status data 168 executed by the equipment data acquisition unit 18 is stored in the memory unit 16B of the server 1B shown in FIG. 36 as an equipment operation status data generation process program. The equipment operation status data generation process program is read from the memory unit 16B to the memory 103 shown in FIG. 19 when the server 1B in FIG. 36 starts operating, and is executed by the processor 104. The process executed by the equipment operation status data generation process program will be described below with reference to the flowchart of the equipment operation status data generation process shown in FIG. 41.

The equipment data acquisition unit 18 acquires various data such as measurement and detection data from various sensors attached to the customer equipment 4, such as sensors for measuring voltage, water volume, etc., and sensors for detecting abnormalities (step S401). The equipment data acquisition unit 18 acquires equipment operation status data 168 from the memory unit 16B of the server 1B. The equipment data acquisition unit 18 adds the data of the various sensors acquired in step S401 to the end of the equipment operation status data 168 (step S402). Specifically, the equipment data acquisition unit 18 associates the acquired data of the various sensors with the customer ID and model from which the data was acquired, and the acquisition date and time of the data, and adds it to the end of the table of equipment operation status data 168 shown in FIG. 39.

The device data acquisition unit 18 stores the device operation status data 168 in the memory unit 16B of the server 1B (step S403). The device data acquisition unit 18 determines whether or not there has been an instruction from the server 1B to end the acquisition of data from various sensors (step S404). For example, this may be the case when the power to the server 1B is turned off or when there has been an instruction from the user to end the acquisition of data. When there has been an instruction to end the acquisition of data (step S404; YES), the device data acquisition unit 18 ends the device operation status data generation processing program. When there has been no instruction to end the acquisition of data (step S404; NO), the device data acquisition unit 18 returns to step S401 and repeats the processes from step S401 to step S404.

In addition, in this embodiment 3, the learning abnormality judgment unit 115 included in the learning unit 11B of the server 1B shown in Fig. 37 judges an abnormal part of the customer device 4 based on the device operation status data 168 stored in the memory unit 16B, and generates abnormality judgment data 169. Furthermore, the learning abnormality judgment unit 115 outputs the judgment result to the learning data generation unit 113B, and as a result, the learning data generation unit 113B generates learning data including the judgment result.

Similarly, the estimated abnormality determination unit 124 included in the fault part estimation unit 12B of the server 1B shown in Fig. 38 determines the abnormal part of the customer device 4 based on the device operation status data 168 stored in the memory unit 16B, and generates abnormality determination data 169. In addition, the estimated abnormality determination unit 124 outputs the determination result to the estimation data generation unit 122B, and thus the estimation data generation unit 122B generates estimation data including the determination result.

The generation and output of abnormality determination data executed by the learning abnormality determination unit 115 of the learning unit 11B and the estimation abnormality determination unit 124 of the fault location estimation unit 12B is stored in the memory unit 16B of the server 1B shown in FIG. 36 as an abnormality determination data generation processing program. The abnormality determination data generation processing program is executed before step S123 in the flowchart of the generation processing of learning data shown in FIG. 24, and the learning data generation unit 113B generates learning data including the judgment result output from the learning abnormality determination unit 115. The abnormality determination data generation processing program is executed before step S213 in the flowchart of the generation processing of estimation data shown in FIG. 28, and the estimation data generation unit 122B generates estimation data including the judgment result output from the estimation abnormality determination unit 124.

The process executed by the abnormality determination data generation process program will be described below with reference to the flowchart of the abnormality determination data generation process shown in FIG. 42. Note that, here, the process will be described as being executed by the learning abnormality determination unit 115 of the learning unit 11B. The learning abnormality determination unit 115 acquires the equipment operation status data 168 from the memory unit 16B shown in FIG. 37 (step S411). The learning abnormality determination unit 115 searches the equipment operation status data 168 for records having the same customer ID and model as the customer ID and model of the construction report master 163 shown in FIG. 11A and FIG. 11B that are to be learned (step S412).

The learning abnormality determination unit 115 acquires records for a certain period going back from the date and time of the failure linked to the customer ID and model in the construction report master 163 from the equipment operation status data 168 (step S413). For example, the learning abnormality determination unit 115 acquires records for a period going back one week from the date and time of the failure from the equipment operation status data 168. The learning abnormality determination unit 115 searches for records in which the value detected by the sensor has changed among the records acquired from the equipment operation status data 168 (step S414).

The learning abnormality judgment unit 115 judges an abnormality in a part or part according to a change in the value detected by the sensor (step S415). For example, the learning abnormality judgment unit 115 judges that an abnormality has occurred in the part or part to which the sensor where the change occurred is attached. The learning abnormality judgment unit 115 sets the judgment result (step S416). For example, if the learning abnormality judgment unit 115 judges that there is an abnormality, it sets the judgment result to "1", and if it judges that there is no abnormality, it sets the judgment result to "0".

The learning abnormality determination unit 115 outputs the determination result to the learning data generation unit 113B of the learning unit 11B shown in Figure 37 (step S417). The learning abnormality determination unit 115 outputs the determination result to the learning data generation unit 113B in the format of {construction report number, determination result of the first device 4A, determination result of the second device 4B, ..., determination result of the nth device 4n} = {XXXXX, 1, 0, ..., 0}, for example.

The learning abnormality judgment unit 115 adds the judgment result to the end of the abnormality judgment data 169 shown in FIG. 40 (step S418). Specifically, the learning abnormality judgment unit 115 associates the report number, which is the same number as the report number described in the construction report master 163 shown in FIG. 11A and FIG. 11B, with the judgment result for each customer device 4, and adds it to the end of the abnormality judgment data 169. The learning abnormality judgment unit 115 saves the abnormality judgment data 169 with the judgment result added in the memory unit 16B shown in FIG. 37 (step S419). The learning abnormality judgment unit 115 ends the abnormality judgment data generation process.

As described above, according to the third embodiment, in addition to the effects of the first and second embodiments, the operating status data of the customer device 4 delivered to the customer is acquired in real time from the customer device 4 and used to estimate the faulty part, thereby improving the accuracy of estimating the faulty part. Furthermore, the learning unit 11B of the server 1B learns the operating status data of the customer device 4 to create a trained model that detects abnormalities, and the faulty part estimation unit 12B uses the trained model to determine the presence or absence of an abnormality, thereby enabling more accurate identification of the cause of the failure. As a result, even if the behavior of the customer device 4 during an abnormality is very variable, the cause of the failure can be identified with high accuracy.

(Variation 1)
In the above first to third embodiments, the neural network constructed in the trained model generation unit 114 shown in Fig. 4 has one intermediate layer. However, the number of intermediate layers is not limited to this, and may be two or more.

(Variation 2)
In the above-mentioned first to third embodiments, the faulty part/replacement part estimation system 100 is a system including the server 1 and the terminal device 2. However, the present invention is not limited to this, and each function operated by the server 1 and the terminal device 2 may be built into one device to form the faulty part/replacement part estimation systems 100, 100A, and 100B.

(Variation 3)
In the above first to third embodiments, the trained model generation unit 114 shown in Fig. 4 is configured to train the faulty part of the device using a neural network. However, instead of the neural network, other learning algorithms such as deep learning, random forest, and support vector machine may be used.

(Variation 4)
In the above-mentioned second embodiment, the part recommendation unit 17 of the server 1A shown in FIG. 33 narrows down the search results obtained by searching the construction report master 163 shown in FIG. 11A and FIG. 11B using the data of the first failure part classification, the data of the second failure part classification, etc., to records including the part code obtained from the part master 167. However, for products with a long period of use, replacement parts are often replaced. For this reason, the part recommendation unit 17 obtains the part category corresponding to all part codes included in the search results obtained by searching the construction report master 163 from the part master 167 shown in FIG. 14. The part recommendation unit 17 counts the obtained part categories, obtains the ratio of each part category, and determines the ranking. The part recommendation unit 17 displays the counted part categories on the part display unit 26 of the terminal device 2A in the order of ranking. As a result, even if a replacement part is replaced, a replacement part can be found from the part category.

In addition, in the first to third embodiments of the present disclosure, the faulty part/replacement part estimation system 100, 100A, and 100B can be realized as a dedicated system. However, they can be realized using a normal computer system without using a dedicated system. For example, a program for realizing each function in the faulty part/replacement part estimation system 100 described above may be stored in a computer-readable recording medium such as a CD-ROM (Compact Disc Read Only Memory) or a DVD-ROM (Digital Versatile Disc Read Only Memory) and distributed, and a computer that can realize each of the above-mentioned functions may be configured by installing this program in a computer. Then, when each function is realized by sharing between an OS (Operating System) and an application, or by cooperation between an OS and an application, only the application may be stored in the recording medium.

Various embodiments and modifications of the present disclosure are possible without departing from the broad spirit and scope of the present disclosure. Furthermore, the above-described embodiments are intended to explain the present disclosure and do not limit the scope of the present disclosure. In other words, the scope of the present disclosure is indicated by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of the disclosure equivalent thereto are deemed to be within the scope of the present disclosure.

This application is based on Japanese Patent Application No. 2021-118914 filed on July 19, 2021. The entire specification, claims, and drawings of Japanese Patent Application No. 2021-118914 are incorporated herein by reference.

This disclosure can be effectively used in fault location and replacement part estimation systems.

1, 1A, 1B Server, 2, 2A Terminal device, 3 Network, 4 Customer device, 4A First device, 4B Second device, 4n Nth device, 11, 11B Learning unit, 12, 12B Faulty part estimation unit, 13 Part category search unit, 14 Replacement part search unit, 15 Past case search unit, 16, 16B Memory unit, 17 Part recommendation unit, 18 Device data acquisition unit, 21 Fault situation input unit, 22 Faulty part display unit, 23 Part category display unit, 24 Replacement part display unit, 25 Past case display unit, 26 Part display unit, 100, 100A, 100B Faulty part/replacement part estimation system, 101, 201 Memory device, 102, 202 Connection unit, 103, 206 Memory, 104, 207 Processor, 105, 208 Data bus, 111 Basic data acquisition unit for learning, 112 Word conversion unit, 113, 113B Learning data generation unit, 114 Learned model generation unit, 115 Learning abnormality determination unit, 121 Estimation basic data acquisition unit, 122, 122B Estimation data generation unit, 123 Estimation processing unit, 124 Estimation abnormality determination unit, 161 Customer master, 162 Equipment master, 163 Construction report master, 164 Word conversion table, 165 Document data, 166 Learned model, 167 Part master, 168 Equipment operation status data, 169 Abnormality determination data, 203 Input device, 204 Display device, 205 Display controller, 211 Customer information input field, 212 Reception number field, 213 Failure status input field, 214 Failure part estimation button, 221 Estimation result display unit, 222 Part category display button, 231 Part category search result display section, 232 replacement part display button, 241 replacement part search result display section, 242 past case display button, 251 past case search result display.

Claims (12)

a failure part estimation unit that estimates a failure part of the equipment based on equipment state data including a failure state of the equipment and equipment information data including a model of the equipment;
a part category search unit that searches for a part category of a replacement part according to the fault part estimated by the fault part estimation unit;
a replacement part search unit that searches for data of the replacement part candidates according to the failure part estimated by the failure part estimation unit and the part category searched by the part category search unit;
A system for estimating faulty parts and replacement parts. a failure part display unit for displaying data on the failure part of the device estimated by the failure part estimation unit,
the part category search unit searches for the part category of the replacement part related to the data of the faulty part of the device selected by a user from the data of the faulty part of the device displayed on the faulty part display unit.
The system for estimating a faulty part and a replacement part according to claim 1 . a part category display unit for displaying part category data of the replacement part related to the data of the failure part of the device searched by the part category search unit,
the replacement part search unit searches for replacement part candidates related to the data of the failure part of the device and the data of the part category selected by a user from among the data of the failure part of the device displayed in the failure part display unit and the data of the part category displayed in the part category display unit;
The system for estimating a faulty part and a replacement part according to claim 2. a replacement part display unit for displaying data of replacement part candidates searched for by the replacement part search unit;
a past case search unit that searches for past failure cases according to the data of the failure part of the device, the data of the part category, and the data of the candidate replacement part selected by a user from among the data of the failure part of the device displayed on the failure part display unit, the data of the part category displayed on the part category display unit, and the data of the candidate replacement part displayed on the replacement part display unit,
The system for estimating a faulty part and a replacement part according to claim 3. Further comprising a learning unit that learns the data of the equipment information and the data of the equipment state and generates a trained model capable of estimating a fault part of the equipment.
The system for estimating a faulty part and a replacement part according to any one of claims 1 to 4. The learning unit includes a neural network, and generates the trained model from weights of an intermediate layer and an output layer in the neural network that are adjusted to output a fault location of the equipment corresponding to the equipment information data and the equipment status data.
The system for estimating a faulty part and a replacement part according to claim 5. the failure part estimation unit includes a neural network in which weights of an intermediate layer and an output layer included in the trained model are set;
The neural network estimates a fault location of the equipment based on the equipment information data and the equipment status data.
The system for estimating a faulty part and a replacement part according to claim 6. A part recommendation unit that recommends a part to be replaced in a pair with the part searched for by the replacement part search unit is further provided.
The system for estimating a faulty part and a replacement part according to any one of claims 1 to 4 . The learning unit learns learning data including a determination result of determining whether or not there is an abnormal part of the equipment based on equipment operation status data indicating an operation status of the delivered equipment obtained from a sensor attached to the equipment, and generates the learned model for outputting a fault part of the equipment.
The system for estimating a faulty part and a replacement part according to claim 6. the failure part estimation unit estimates a failure part of the equipment by using estimation data including a determination result of the presence or absence of an abnormal part of the equipment based on equipment operation status data indicating an operation status of the equipment;
The system for estimating a faulty part and a replacement part according to claim 9. A method executed by a failure part/replacement part estimation system for estimating a failure part and a replacement part of a device, comprising:
estimating a fault location of the device based on device status data including a fault state of the device and device information data including a model of the device;
searching for a part category of the replacement part according to the estimated fault location;
searching for data of the replacement part candidates according to the estimated fault location and the searched part category ;
method. On the computer,
A process of estimating a faulty part of the equipment based on equipment status data including a fault state of the equipment and equipment information data including a model of the equipment;
A process of searching for a part category of a replacement part according to the estimated fault location;
A process of searching for data of the replacement part candidates according to the estimated fault location and the searched part category ;
A program for executing.