JP2018173948A - Malfunction diagnosis device, malfunction diagnosis method, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a malfunction diagnosis device that is able to accurately diagnose a malfunctioning component, a malfunction diagnosis method, and a computer program.SOLUTION: A malfunction diagnosis device has a model creation server 12 that acquires a sensor value of each sensor, computes, on the basis of a sensor value when a component is normal and a sensor value when the component is malfunctioning from among the acquired sensor values, a degree of contribution of each of the acquired sensor values to a malfunction of each component, and creates a table indicating the relations of the sensor values with degrees of contributions to component malfunctions. In addition, for each component, the device analyzes a main constituent by using a degree of contribution, and creates a malfunction component prediction model. When a sensor value is sent from a transceiver 15, the device refers to the malfunction component prediction model and computes the malfunction probability of each component. When there is a component whose malfunction probability exceeds a threshold probability, the device displays this component on a display terminal 16.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a failure diagnosis apparatus, a failure diagnosis method, and a computer program for diagnosing a failure of each component provided in a device, and more particularly to a technique for accurately diagnosing a failure even when a plurality of components fail.

  2. Description of the Related Art Conventionally, a method of diagnosing failures of a plurality of parts such as a motor and a solenoid valve mounted on a device such as an outdoor unit of an air conditioner by using detection values of various sensors attached to the device has been adopted. For example, in order to measure various physical quantities such as temperature, humidity, motor rotation speed, vibration, etc. of an outdoor unit, various sensors such as a temperature sensor, a humidity sensor, a rotation speed sensor, and a vibration sensor are provided. The detected value is used to diagnose whether or not the component has failed.

  As a conventional example of such a device failure diagnosis apparatus, for example, one disclosed in Patent Document 1 is known. In Patent Document 1, it is disclosed that, when a diagnosis report of a device is obtained, the degree of similarity with past failure data is calculated and used for diagnosis, so that even an engineer with little experience can make an accurate diagnosis. Yes.

  However, the technique disclosed in Patent Document 1 has a problem that it is difficult to accurately diagnose a failure of a component when a failure has occurred in a plurality of components provided in the device. For example, as shown in FIG. 12, a device such as an outdoor unit is provided with three parts P0, P1, and P2, and the physical quantities of the parts P0, P1, and P2 are detected by four sensors A to D. Suppose. When only one component has failed, the failed component can be identified with high accuracy. For example, when the component P0 is out of order, an abnormality is detected by the sensors A and D. Therefore, the engineer can determine that the component P0 is faulty when an abnormal value is detected by both the sensors A and D. Similarly, when an abnormality is detected in both sensors A and B, it can be determined that a failure has occurred in component P1.

JP 2016-110448 A

  However, in the conventional example disclosed in Patent Document 1 described above, when the components P0 and P1 shown in FIG. 12 are simultaneously broken, abnormality is detected simultaneously by the three sensors A, B, and D. There is a problem that it is difficult to diagnose a faulty part with high accuracy.

  The present invention has been made to solve such a conventional problem, and an object of the present invention is to provide a failure diagnosis apparatus, a failure diagnosis method, and a computer program capable of accurately diagnosing a failed component. Is to provide.

  One aspect of the present invention is a failure diagnosis device for diagnosing a failure of a plurality of components mounted on a device, acquires a sensor value of a sensor that detects a physical quantity of the component, and among the acquired sensor values, A failure feature that calculates the contribution of each sensor value to each component failure based on the sensor value when the component is normal and the sensor value at the time of failure, and creates a table showing the relationship between the contribution of the sensor value to the component failure A table creation unit, a failed component prediction model creation unit that creates a failed component prediction model by performing principal component analysis using the degree of contribution for each component, and when the sensor value of the sensor is acquired, the failed component A failure probability calculation unit that calculates a failure probability of each component with reference to the prediction model, and a presentation control unit that presents this component when there is a component whose failure probability exceeds a predetermined threshold probability And wherein the door.

  One aspect of the present invention is the failure diagnosis apparatus according to claim 1, further comprising an operation data acquisition unit that acquires operation data of the device, wherein the failure feature table creation unit includes the normal state sensor. In addition to the value and the sensor value at the time of the failure, the contribution degree is calculated based on the operation data.

  One aspect of the present invention is the failure diagnosis apparatus according to claim 2, wherein the failure feature table creation unit acquires, as the operation data, an elapsed time after the device starts driving, The contribution is calculated for each elapsed time, and the failure feature table is created for each elapsed time.

  One aspect of the present invention is the failure diagnosis apparatus according to claim 2, wherein the failure feature table creation unit calculates, as the operation data, a continuous stop time of the device when the device starts driving. Acquiring, calculating the contribution for each continuous stop time, and creating the failure feature table for each continuous stop time.

  One aspect of the present invention is the failure diagnosis apparatus according to claim 1, further comprising an ambient temperature acquisition unit that acquires ambient temperature data of the device, and the failure feature table creation unit includes the normal time In addition to the sensor value and the sensor value at the time of the failure, the contribution is calculated based on the ambient temperature data, and the failure feature table is created for each ambient temperature.

  One aspect of the present invention is the failure diagnosis apparatus according to any one of claims 1 to 5, wherein the presentation control unit includes a display terminal capable of screen display, and the failure probability is a predetermined threshold value. Parts larger than the probability are displayed on the display terminal in a ranking format.

  One aspect of the present invention is the failure diagnosis apparatus according to any one of claims 1 to 5, wherein among the plurality of parts, a sensor value when a failure occurs in an arbitrary part; A failure case table that stores cases indicating a relationship with a component in which a failure has occurred; and a failure case search unit that searches past failure cases from the failure case table based on the failure probability. It is characterized by that.

One aspect of the present invention, there is provided a fault diagnosis apparatus as claimed in claim 1, wherein the degree of contribution, and characterized in that a contributing factor R ² of the squared correlation coefficient R To do.

  One aspect of the present invention is the failure diagnosis apparatus according to any one of claims 1 to 5, wherein the failure component prediction model creation unit performs principal component analysis and uses MSPC of each sensor value. The failure probability is calculated.

  One aspect of the present invention is a failure diagnosis method for diagnosing a failure of a plurality of components mounted on a device, acquiring a sensor value of a sensor that detects a physical quantity of the component, and among the acquired sensor values, Calculating a contribution of each sensor value to each component failure based on a sensor value when the component is normal and a sensor value at the time of failure, and creating a table indicating a relationship between the contribution of the sensor value to the component failure; and For each component, perform a principal component analysis using the contribution degree to create a failed component prediction model, and when the sensor value of the sensor is acquired, refer to the failed component prediction model And a step of presenting the component when the component has a failure probability exceeding a predetermined threshold probability.

  One embodiment of the present invention acquires a sensor value of a sensor that detects a physical quantity of a component with respect to a computer that functions as a failure diagnosis device that diagnoses a failure of a plurality of components mounted on a device, and acquires the acquired sensor value Among these, based on the sensor value when the part is normal and the sensor value at the time of failure, the degree of contribution of each sensor value to each part failure is calculated, and a table showing the relationship between the contribution of the sensor value to the part failure A step of creating a principal component analysis using the contribution degree for each component, creating a failure component prediction model, and referring to the failure component prediction model when the sensor value of the sensor is acquired. A step for calculating a failure probability of each component and a step for presenting the component when the failure probability exceeds a predetermined threshold probability. Is a computer program.

  With the failure diagnosis apparatus according to the present invention, it is possible to diagnose a failed component with high accuracy.

FIG. 1 is a block diagram illustrating a configuration of a failure diagnosis apparatus according to the first embodiment of the present invention and peripheral devices thereof. FIG. 2A is an explanatory diagram illustrating an example of data stored in a normal sensor value table. FIG. 2B is an explanatory diagram illustrating a data example of a failure part table and an abnormal sensor value table stored in the failure case table. FIG. 2C is an explanatory diagram illustrating an example of data stored in the failure feature table. FIG. 3 is a flowchart showing a processing procedure in the training phase. FIG. 4 is a flowchart showing a processing procedure in the test phase. FIG. 5 is a sequence diagram showing a processing procedure in the training phase. FIG. 6 is a sequence diagram showing a processing procedure in the test phase. FIG. 7 is a characteristic diagram showing a data distribution obtained by principal component analysis of each sensor value, and shows an example of determining a failure using USPC. FIG. 8 is a characteristic diagram showing a data distribution obtained by principal component analysis of each sensor value, and shows an example in which a failure is determined using MSPC. FIG. 9 is a block diagram showing the configurations of the failure diagnosis apparatus according to the second and third embodiments of the present invention and its peripheral devices. FIG. 10 is an explanatory diagram illustrating a failure feature table of the failure diagnosis apparatus according to the second embodiment. FIG. 11 is a block diagram showing a configuration of a failure diagnosis apparatus according to the fourth embodiment of the present invention and its peripheral devices. FIG. 12 is an explanatory diagram illustrating a relationship between a failed part and a sensor value in the related art.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
(Description of the first embodiment)
[Description of configuration]
FIG. 1 is a block diagram illustrating a configuration of a failure diagnosis apparatus according to the first embodiment of the present invention and peripheral devices thereof. The failure diagnosis apparatus 100 according to the present embodiment diagnoses a failure of each component P0, P1, P2, P3 mounted on the device 51. The device 51 is, for example, an outdoor unit of an air conditioner, and the parts P0 to P3 are, for example, a motor and a solenoid valve provided in the outdoor unit. The device 51 includes sensors P0 to P3 and sensors A to D for measuring surrounding physical quantities. Examples of the sensors A to D include a temperature sensor, a humidity sensor, a rotation speed sensor that detects the rotation speed of a motor, and a vibration sensor. In the present embodiment, the four components P0 to P3 and the four sensors A to D will be described as examples, but the present invention is not limited to this.

  As shown in FIG. 1, the failure diagnosis apparatus 100 receives measurement values (referred to as “sensor values”) detected by sensors A to D mounted on the device 51, and uses the received sensor values. It is connected to a transmitter / receiver 15 for transmission. That is, the sensor values of the sensors A to D are transmitted to the transceiver 15, and further transmitted from the transceiver 15 to the DB server 11 and the fault component diagnosis server 13.

  The DB server 11 includes a normal sensor value table 111, a failure feature table 112, and a failure case table 113.

  The normal sensor value table 111 stores sensor values of the sensors A to D when the components P0, P1, P2, and P3 mounted on the device 51 are operating normally. Specifically, as shown in FIG. 2A, sensor values of sensors A to D are stored.

  The failure case table 113 stores sensor values of the sensors A to D in association with failure cases when an abnormality has occurred in at least one of the components P0, P1, P2, and P3. That is, when an abnormality occurs in an arbitrary part, the sensor values of the sensors A to D are stored and stored in association with the failed part.

  As an example, a failure part table and an abnormal sensor value table as shown in FIG. 2B are set. When the part P0 fails as the case ID “0”, the sensor value of the sensor A is “8.5”, the sensor value of the sensor B is “−2”, the sensor value of the sensor C is “0.06”, the sensor It is stored that the sensor value of D is “1”. In this case, the sensor A shows a numerical value “8.5” higher than usual. That is, it is understood that the sensor value of the sensor A shows a high numerical value when a failure has occurred in the component P0.

  Further, when two parts P1 and P3 fail as the case ID “1”, the sensor value of the sensor A is “1.3”, the sensor value of the sensor B is “−48”, and the sensor value of the sensor C is “ 0.3 "and the sensor value of sensor D is stored as" 0 ". In this case, the sensor A shows a numerical value “1.3” lower than usual, and the sensor B shows a value “−48” lower than usual. That is, it is understood that the sensor values of the sensors A and B show a low numerical value when a failure has occurred in the parts P1 and P3. Similarly, sensor values of case IDs “2”, “3”, and “4” are stored.

As will be described later, the failure feature table 112 stores a contribution rate R ² (contribution) of each component P0 to P3 created by the model creation server 12 to each sensor value. “R” is a correlation coefficient, and “R” is squared to obtain a contribution rate. Specifically, as shown in FIG. 2C, for each sensor to D, and stores the failure contribution ^{R 2} in each part P0-P3. As shown in FIG. 2C, for example, in the sensor A, the failure contribution rate of the component P0 is “0.54”, the failure contribution rate of the component P1 is “0.01”, and the failure contribution rate of the component P2 is “0.07”. The failure contribution rate of the component P3 is “0.35”. From this data, it is understood that the failure contribution rate of the parts P0 and P3 is high for the sensor A, for example.

When the model creation server 12 is executing a training phase (a phase in which each model is created from detected data), the sensor values of the sensors A to D stored in the normal sensor value table 111 (FIG. 2A). Reference) and sensor values (see FIG. 2B) of the sensors A to D stored in the failure case table 113, the sensor values at the time of failure of the components P0 to P3 are patterned, and the sensors A to D are patterned. for, it calculates the contribution ratio ^{R 2} of each part P0-P3. Then, the calculated contribution rate R ² is stored in the failure feature table 112. As a result, the failure feature table shown in FIG. 2C described above is obtained. Further, the model creation server 12 creates a failure part prediction model, which will be described later, and stores it in the failure part prediction model storage unit 131 of the failure part diagnosis server 13.

  That is, the model creation server 12 acquires the sensor values of the sensors A to D that detect the physical quantities of the components P0 to P3, and among the acquired sensor values, the sensor value when the components P0 to P3 are normal and the failure value Based on the sensor value, a function as a failure feature table creation unit that calculates the contribution of each sensor value to each component failure and creates a table indicating the relationship between the contribution of the sensor value to the component failure is provided.

  Further, the model creation server 12 performs a principal component analysis, which will be described later, using the contributions stored in the failure feature table 112 for each component P0 to P3, and creates a failure component prediction model creation unit that creates a failure component prediction model. It has the function as.

  The failure component diagnosis server 13 uses a failure component prediction model stored in the failure component prediction model storage unit 131 when a sensor value of each of the sensors A to D is acquired from the transmitter / receiver 15 by a method described later. Perform fault diagnosis. Specifically, the failure probability of each of the components P0 to P3 is calculated, and when there is a component having a failure probability higher than a preset threshold, it is estimated that a failure has occurred in this component. Data indicating the part estimated to be faulty is transmitted to the fault case search server 14 (failure case search unit) and the display terminal 16. That is, the failure component diagnosis server 13 functions as a failure probability calculation unit that calculates the failure probability of each of the components P0 to P3 with reference to the failure component prediction model when the sensor values of the sensors A to D are acquired. It has.

  The failure case search server 14 refers to the failure case table 113 on the basis of the sensor values of the sensors A to D at the time of the failure and the parts estimated to have a failure, and searches for similar failure cases in the past. get. The acquired data is transmitted to the display terminal 16.

  The display terminal 16 can display a screen and displays each transmitted data on the screen. That is, the display terminal 16 has a function as a presentation control unit that presents a part when the part has a failure probability exceeding a predetermined threshold probability.

  The DB server 11, model creation server 12, failure part diagnosis server 13, and failure case search server 14 shown in FIG. 1 are, for example, a general-purpose computer including a CPU, a memory, and an input / output unit, and are installed in advance. Each function is achieved by executing a computer program.

[Procedure for creating a failure feature table]
Next, specific processing by the model creation server 12, that is, failure feature table creation processing and failure component prediction model creation processing will be described in detail.

同様にして、上記（２）式により、標準化した各センサ値Ａｓ、Ｂｓ、Ｃｓ、Ｄｓの、各部品Ｐ０、Ｐ１、Ｐ２、Ｐ３に対する故障寄与率Ｒ^２（α、β）（但し、０≦Ｒ^２≦１）を演算する。但し、αはＡｓ、Ｂｓ、Ｃｓ、Ｄｓ、βはＰ０〜Ｐ３である。 First, the procedure for creating the failure feature table (see FIG. 2C) will be described in detail. The model creation server 12 is stored in the sensor values (see FIG. 2A) of the sensors A to D at the normal time stored in the normal sensor value table 111 and the abnormal value sensor value table of the failure case table 113. The sensor values (see FIG. 2B) of each of the sensors A to D at the time of failure are read and further standardized. When the sensor values of the sensors A to D are indicated by the same symbols A to D as the respective sensors and the standardized sensor values are indicated by As to Ds, the standardized data Asi of the i-th sensor value of the sensor A is (1).
Asi = (Ai−μ) / σ (1)
Where μ is an average value and σ is a standard deviation. Then, the contribution rate R ² (As, P0) (where 0 ≦ R ² ≦ 1) of the standardized sensor value As to the failure of the component P0 is calculated by the following equation (2).

Similarly, the failure contribution rate R ² (α, β) for each component P0, P1, P2, P3 of the standardized sensor values As, Bs, Cs, Ds according to the above equation (2) (where 0 ≦ R ² ≦ 1) is calculated. However, (alpha) is As, Bs, Cs, Ds, and (beta) is P0-P3.

Then, the contribution rate R ² to the sensor value of the component failure obtained by the calculation is stored in the failure feature table 112. As a result, as shown in FIG. 2C described above, a failure feature table indicating the failure contribution rates of the components P0 to P3 for the sensors A to D is created.

[Procedure for creating a failure part prediction model]
Further, the model creation server 12 includes each sensor value stored in the normal sensor value table 111, each sensor value stored in the failure case table 113, and the contribution ratio R stored in the failure feature table 112. Based on ² , a failure part prediction model is created. Details will be described below.

  First, a first threshold Th1 and a second threshold Th2 are set. The first threshold value Th1 is a threshold value for selecting a sensor with a high contribution rate from a plurality of sensors. The second threshold Th2 is a threshold for the cumulative contribution rate when setting the number of principal components used for principal component analysis. The first threshold Th1 is set to 95%, for example. For example, 95% is set as the second threshold Th2.

  Next, based on the failure feature table (see FIG. 2C), the sensors having a high failure contribution rate are sorted for each component Px (x = 0 to 4). For example, taking the part P0 as an example, in the example shown in FIG. 2C, the sensor A is “0.54”, the sensor B is “0.02”, the sensor C is “0.04”, and the sensor D is “0. Since these are “01”, when these are sorted, the order is A → C → B → D. Then, the sensor value of the first threshold Th1 (for example, 95%), which is the upper level, is used as an input. Here, although an example using four sensors A to D is shown, for example, when 20 sensors are used, the top 19 sensors of 95% are selected.

  Further, the plurality of sorted sensor values are sequentially added from the upper side, and the sensor value that becomes the second threshold Th2 is selected. For example, when the sum of the failure contribution rates for the top two sensors is 0.95 (95%), the two sensors are selected. Then, the principal component analysis is performed with the sensor values of the two sensors as the principal components. When the number of principal components is three or more, principal component analysis is performed using the three or more sensor values as principal components.

Further, an average value vector μ, a variance covariance matrix Σ, and a threshold value θth are calculated for the main components of the sensor value at the normal time (see FIG. 2A). The average value vector μ, the variance covariance matrix Σ, and the threshold value θth are calculated by the following equations (3) to (5).

Then, the average value vector μ, the variance covariance matrix Σ, and the threshold value θth calculated by the above formula are stored in the faulty part prediction model storage unit 131 of the faulty part diagnosis server 13 as a faulty part prediction model for the part P0. To do. As a result, for example, as shown in FIG. 8, normal data when the horizontal axis is the first process amount and the vertical axis is the second process amount is obtained as elliptical region data. In addition, the same processing is performed for the parts P1, P2, and P3, and a faulty part prediction model is created and stored in the faulty part prediction model storage unit 131.
In the training phase of the present embodiment, a failure feature table (see FIG. 2C) and a failure component prediction model are created by the above processing.

  Next, the test phase will be described. In the test phase, the failure part prediction model created in the training phase and the failure feature table are used, and further, a process for estimating a part in which a failure has occurred is performed based on the sensor values detected by the sensors A to D. .

但し、ｘｉはセンサ値を示す。 The faulty component diagnosis server 13 illustrated in FIG. 1 acquires sensor values detected by the sensors A to D from the transceiver 15. Then, the Mahalanobis distance θ with the sensor value is calculated using the average value vector μ of the failed part prediction model stored in the failed part prediction model storage unit 131 and the variance-covariance matrix Σ. As is well known, the Mahalanobis distance θ can be calculated by the following equation (6).

However, xi shows a sensor value.

Then, it is determined from the magnitude relationship between the Mahalanobis distance θ and the threshold value θth whether the sensor value is normal or abnormal.
Specifically, when θ ≦ θth, it is determined to be normal, and when θ> θth, it is determined to be abnormal.
Further, when the failure probability p is calculated as a sigmoid function of the difference between θ and θth, it can be expressed by the following equation (7).

  If it is determined that the value is an abnormal value, a failure probability is calculated for each failed part and transmitted to the display terminal 16 as a failure diagnosis result. The failure case search server 14 reads the failure diagnosis result and the failure case table, and outputs similar failure case data.

[Description of processing procedure]
Next, a processing procedure of the failure diagnosis apparatus 100 according to the first embodiment will be described with reference to flowcharts shown in FIGS. 3 and 4 and sequence diagrams shown in FIGS. 5 and 6.

  First, the processing procedure in the training phase will be described with reference to FIGS. In step S11 of FIG. 3, the model creation server 12 shown in FIG. 1 reads the sensor value at the normal time from the normal sensor value table 111 of the DB server 11 (see L11 in FIG. 5). Specifically, as shown in FIG. 2A, sensor values detected by the sensors A to D when the components P0 to P3 are operating normally are read.

  In step S12, the model creation server 12 reads failure case data from the failure case table 113 of the DB server 11 (see L12 in FIG. 5). Specifically, as shown in FIG. 2B, sensor values detected by the sensors A to D when at least one of the components P0 to P3 is abnormal are read.

  In step S13, the model creation server 12 creates a failure part prediction model. In this process, a principal component analysis is performed as described above to create a failure part prediction model. Then, the failure part prediction model is stored in the failure part prediction model storage unit 131 of the failure part diagnosis server 13 (see L13 in FIG. 5).

  In step S <b> 14, the model creation server 12 creates a failure feature table 112. Specifically, as shown in FIG. 2C, a table is created in which the failure contribution rates of the components P0 to P3 are associated with the sensors A to D (see L14 in FIG. 5).

  Thus, in the training phase, a failure feature table and a failure component prediction model can be created by the above processing.

  Next, the processing procedure in the test phase will be described with reference to FIGS. First, in step S31 in FIG. 4, the faulty component diagnosis server 13 receives the sensor values of the sensors A to D from the transceiver 15 (see L21 in FIG. 6).

  In step S32, the fault component diagnosis server 13 executes fault diagnosis based on each sensor value. Specifically, the detected sensor value is input to xi in the above-described equation (6), and the Mahalanobis distance θ is calculated. If it is determined that “θ> threshold θth”, it is determined that the sensor value is abnormal, and the failure probability p for this component is calculated using the above-described equation (7). Then, the principal component analysis is performed to obtain the data distribution of FIG. 8, and further abnormality diagnosis is performed by variable statistical process control (MSPC). In FIG. 8, the area where data is concentrated (area indicated by an ellipse) is an area that is determined to be normal, and is abnormal when the distance of the detected data from the ellipse area is large (a failure has occurred). Judgment) For example, it is determined that the data Q1 is abnormal.

  In step S33, the fault component diagnosis server 13 determines whether or not the fault probability p exceeds a threshold value. If the threshold value is exceeded, the failure diagnosis result is output to the display terminal 16 in step S34 (see L22 in FIG. 6).

  In step S35, the failed component diagnosis server 13 reads various failure cases stored in the failure case table 113 of the DB server 11 (see L23 in FIG. 6).

  In step S36, the fault component diagnosis server 13 refers to the past fault cases stored in the fault case table 113, and searches for past fault cases similar to the fault case currently occurring. In this process, the cosine similarity for each part is calculated based on the past failure cases stored in the failure case table 113 and the failure case that occurred this time. Then, failure cases are searched in descending order of similarity. In step S37, the data is transmitted to the display terminal 16 (see L24 in FIG. 6). The display terminal 16 displays parts having a high failure probability in a ranking format. Therefore, the operator can recognize a component having a high failure probability by looking at the display content of the display terminal 16.

  Thus, it is possible to diagnose a failure of each of the components P0 to P3 mounted on the device 51 based on the sensor values detected by the sensors A to D.

  In this manner, the failure diagnosis apparatus 100 according to the present embodiment creates the failure feature table 112 based on the sensor values detected in the past by the sensors A to D in the training phase, and further predicts the failure part. Create a model. In the test phase, when the sensor values of the sensors A to D when the device 51 is operating are obtained, a component with a high failure probability is detected by the failure component prediction model based on the sensor value. To do. Therefore, when the sensor values of the sensors A to D are abnormal, it is possible to estimate the part where the failure has occurred. Further, even when a plurality of parts are simultaneously failed, it is possible to estimate the failure by specifying the plurality of parts. Therefore, it is possible to diagnose a faulty part with high accuracy regardless of the experience and skill level of the operator.

  In the present embodiment, as shown in FIG. 8, failure diagnosis is performed using MSPC based on the data distribution. For example, when the data indicated by the symbol Q1 in FIG. 8 is obtained, the data Q1 is away from the elliptical area of the data distribution, so it can be determined that it is abnormal.

  On the other hand, for example, when using well-known USPC (Univariate Statistical Process Control), as shown in FIG. 7, the detected data is within the upper limit of the first process amount on the horizontal axis, the range of the lower limit, and the first of the vertical axis. 2 If it is within the upper and lower limits of the process amount, it is not determined to be abnormal, so the data Q1 is determined to be normal. Therefore, it cannot be determined with high accuracy that an abnormality has occurred in the sensor value. In the present embodiment, as shown in FIG. 8, since the MSPC is used, the failure determination can be performed with high accuracy.

  In addition, since a part having a high possibility of failure is predicted from a plurality of sensor values and ranked, it is possible to immediately recognize the failed part. Furthermore, since past similar failure cases are searched from a plurality of sensor values and displayed on the display terminal 16, it is possible to compare with failure cases that have occurred in the past.

  Further, since the principal component analysis is performed, the failed part can be recognized even when the relation between each part and each sensor is not obvious. Furthermore, even when the sensor value is not recognized (anonymity), it is possible to present a component failure.

(Description of Second Embodiment)
Next, a second embodiment will be described. FIG. 9 is a block diagram illustrating a configuration of the failure diagnosis apparatus according to the second embodiment. As shown in FIG. 9, the failure diagnosis apparatus 100 a according to the second embodiment is different from the first embodiment described above in that it includes an operation data acquisition unit 17.

  The operation data acquisition unit 17 acquires operation data from the control device 52 that controls the driving of the device 51 via the transceiver 15 and further stores the operation data. The operation data includes an elapsed time after the device 51 starts driving. For example, when the device 51 (air conditioner or the like) is stopped and the switch is turned on to start driving, the elapsed time data from the start of driving is included. The acquired elapsed time data is output to the model creation server 12 and the fault component diagnosis server 13.

Modeling server 12, based on the elapsed time data, calculates the fault contribution R ² in each elapsed time. For example, the elapsed time is divided into “10 minutes or less”, “30 minutes or less”, “60 minutes or less”, and “60 minutes or more”, and a failure part prediction model is created for each elapsed time in the test phase described above. To do.

  Specifically, the failure contribution rate when 60 minutes or more have elapsed is the failure contribution rate shown in the first embodiment described above. Then, by multiplying the coefficient set for each elapsed time and each component P0 to P3 using this as a reference value, the failure contribution rate of “10 minutes or less”, “30 minutes or less”, and “60 minutes or less” Is calculated. As a result, for example, as shown in FIG. 10, failure feature tables X1 to X4 for each elapsed time are created. That is, the failure feature table shown in FIG. 2C is created for each elapsed time.

Moreover, defective parts diagnostic server 13, based on the elapsed time data output from the operation data acquisition unit 17, refers to the failure characteristics table 112, acquires the failure contribution R ^2. Specifically, a plurality of fault feature table X1~X4 shown in FIG. 10, acquires the failure contribution R ² with reference to the failure characteristics table corresponding to the elapsed time at that time. Then, to create the failed component prediction model based on the acquired fault contribution ratio R ^2. The method for creating the failure part prediction model is the same as that in the first embodiment.

That is, the behavior of components such as a motor mounted on the device 51 may vary greatly depending on the elapsed time since the start of driving, such as fluctuations in rotational speed and generation of vibrations. In the second embodiment, to create a plurality of fault characteristic table for each elapsed time, select the fault characteristic table according to the current elapsed time to set the fault contribution R ^2, predicts a failed component. Therefore, when the driving of the device 51 is started, even when the behavior of the device 51 changes according to the elapsed time from the start of driving, it is possible to predict a failed part with high accuracy.

  Note that the failure feature table creation procedure can be performed and created for each elapsed time in the same manner as in the first embodiment described above. Moreover, although 2nd Embodiment demonstrated the example which divides elapsed time into four, 10 minutes or less, 30 minutes or less, 60 minutes or less, and 60 minutes or more, this invention is not limited to this. It is also possible to classify for a long period of time, such as 1 day or less, 1 week or less, 1 month or less.

(Description of Modified Example of Second Embodiment)
Next, a modification of the second embodiment will be described. In the second embodiment described above, an example in which the elapsed time is divided into four preset times, for example, “10 minutes or less”, “30 minutes or less”, “60 minutes or less”, “60 minutes or more” is described. did. The modification differs from the second embodiment described above in that the elapsed time is automatically set. Details will be described below.

  In the invention according to the modified example of the second embodiment, a pre-phase is provided, and the elapsed time from when the device 51 starts operation to when it stops is acquired from the operation data acquisition unit 17 shown in FIG. Then, the model creation server 12 calculates an average value of elapsed times for a preset number of times of operation N1 (for example, N1 = 10 times). As a result, for example, “6 hours” is calculated as the average elapsed time.

  Further, the average elapsed time is divided by a preset division number M1 (for example, M1 = 4). As a result, there are four categories: “1.5 hours or less”, “3 hours or less”, “4.5 hours or less”, and “4.5 hours or more”. In the test phase described above, a failure part prediction model is created for each elapsed time. Subsequent processing is the same as that of the second embodiment described above, and a description thereof will be omitted.

  And in the modification of 2nd Embodiment, by setting beforehand the frequency | count N1 (for example, 10 times) and the division | segmentation number M1 (for example, 4 division | segmentation) at the time of calculating an average elapsed time, a prior phase Is executed, a failure component prediction model is created for each divided elapsed time. At this time, as shown in the second embodiment, it is not necessary to divide the time, such as “10 minutes or less”, “30 minutes or less”, and the time division is automatically set according to the elapsed time. Therefore, it becomes possible to improve convenience.

  In the above-described modification, the example in which the number of operations N1 used for calculating the average value is “10” and the division number M1 of elapsed time is “4” has been described, but the present invention is not limited to this. .

(Description of the third embodiment)
Next, a third embodiment will be described. The configuration of the failure diagnosis apparatus according to the third embodiment is the same as that shown in FIG.

  The operation data acquisition unit 17 acquires operation data of the device 51 via the transceiver 15 and further stores it. The operation data includes a continuous stop time in which the device 51 (air conditioner or the like) is continuously stopped. The continuous stop time is the time that the device 51 has been continuously stopped when the drive is started, and is the elapsed time from the time when the device 51 is stopped to the time when the drive is started this time. is there. The acquired continuous stop time is output to the model creation server 12 and the fault component diagnosis server 13.

Modeling server 12, based on the continuous stop time data, calculates the fault contribution R ² each continuation downtime. For example, the continuation stop time is classified into “30 minutes or less”, “2 hours or less”, “24 hours or less”, “1 week or less”, and “1 week or more”. A failure part prediction model is created for each time.

  Specifically, the failure contribution rate when the continuous stop time is 30 minutes or less is the failure contribution rate shown in the first embodiment described above. Then, by using this as a reference value and multiplying by a coefficient set for each continuous stop time and for each component P0 to P3, “2 hours or less”, “24 hours or less”, “1 week or less”, “1 The failure contribution rate of “weeks or more” is calculated. As a result, five types of failure feature tables are created for each continuous stop time. That is, the failure feature table shown in FIG. 2C is created for each continuous stop time.

Moreover, defective parts diagnostic server 13, based on the continuous stop time data output from the operation data acquisition unit 17, refers to the failure characteristics table 112, acquires the failure contribution R ^2. Specifically, the five types of fault characteristic table described above, to obtain the failure contribution R ² with reference to the failure characteristics table corresponding to the continuation downtime that time. Then, to create the failed component prediction model based on the acquired fault contribution ratio R ^2. The method for creating the failure part prediction model is the same as that in the first embodiment.

That is, the behavior of components such as a motor mounted on the device 51 may greatly change depending on the continuous stop time before starting driving, such as fluctuations in the number of rotations and occurrence of vibrations. In the third embodiment, a plurality of failure feature tables are created for each continuous stop time, a failure feature table is selected according to the continuous stop time at the start of driving, a failure contribution rate R ² is set, and a failure part is predicted. To do. Therefore, even when the behavior of the device 51 changes according to the continuous stop time of the device 51, it is possible to predict a faulty part with high accuracy. In the third embodiment, the example in which the continuous stop time is divided into five parts of 30 minutes or less, 2 hours or less, 24 hours or less, 1 week or less, 1 week or more has been described, but the present invention is not limited thereto. Not.

(Description of Modified Example of Third Embodiment)
Next, a modification of the third embodiment will be described. In the third embodiment described above, the continuous stop time is set in advance, for example, “30 minutes or less”, “2 hours or less”, “24 hours or less”, “1 week or less”, and “1 week or more”. The example divided into two has been described. The modification is different from the third embodiment described above in that the continuous stop time is automatically set. Details will be described below.

  In the invention according to the modification of the third embodiment, a pre-phase is provided, and the time during which the device 51 is continuously stopped is acquired from the operation data acquisition unit 17 illustrated in FIG. Then, the model creation server 12 calculates an average value of the continuous stop times for a preset number of operations N2 (for example, N2 = 10 times). As a result, for example, “5 hours” is calculated as the average continuous stop time.

  Further, the average continuous stop time is divided by a preset division number M2 (for example, M2 = 5). As a result, there are five categories: “1 hour or less”, “2 hours or less”, “3 hours or less”, “4 hours or less”, and “4 hours or more”. In the test phase described above, a failure component prediction model is created for each continuous stop time. Subsequent processing is the same as that of the third embodiment described above, and a description thereof will be omitted.

  In the modification of the third embodiment, the number of operations N2 (for example, 10 times) and the number of divisions M2 (for example, 5 divisions) for calculating the average continuous stop time are set in advance. When the phase is executed, a failure part prediction model is created for each divided continuous stop time. At this time, as shown in the third embodiment, it is not necessary to divide the time, such as “30 minutes or less”, “2 hours or less”, and the time division is automatically set according to the continuous stop time. Therefore, convenience can be further improved.

  In the above-described modification, the example in which the number of operations N2 used for calculating the average value is “10” and the division number M2 of the continuous stop time is “5” has been described, but the present invention is not limited to this. Absent.

(Explanation of Fourth Embodiment)
Next, a fourth embodiment will be described. FIG. 11 is a block diagram illustrating a configuration of a failure diagnosis apparatus according to the fourth embodiment. As shown in FIG. 11, the failure diagnosis apparatus 100 b according to the fourth embodiment is different from the first embodiment described above in that a temperature data acquisition unit 18 is provided.

  The temperature data acquisition unit 18 acquires the ambient temperature data of the device 51 detected by the temperature sensor 53 (ambient temperature acquisition unit) provided in the device 51 via the transceiver 15 and further stores it. The ambient temperature data is, for example, the outside air temperature around the device 51. The acquired ambient temperature data is output to the model creation server 12 and the fault component diagnosis server 13.

Modeling server 12, based on the elapsed time data, calculates the fault contribution R ² for each ambient temperature. For example, the ambient temperature is classified into “0 ° C. or lower”, “0-10 ° C.”, “10-20 ° C.”, “20-30 ° C.”, and “30 ° C. or higher”. Create a failure part prediction model for each temperature.

  Specifically, the failure contribution rate when the ambient temperature is 20 to 30 ° C. is the failure contribution rate shown in the first embodiment described above. And by multiplying the coefficient set for each ambient temperature and each component P0 to P3 using this as a reference value, “0 ° C. or less”, “0 to 10 ° C.”, “10 to 20 ° C.”, “ The failure contribution rate of “30 ° C. or higher” is calculated. As a result, five types of failure feature tables are created for each ambient temperature. That is, the failure feature table shown in FIG. 2C is created for each ambient temperature.

Moreover, defective parts diagnostic server 13, based on the ambient temperature data outputted from the temperature data acquisition unit 18, refers to the failure characteristics table 112, acquires the failure contribution R ^2. Specifically, the failure contribution rate R ² is acquired from the five types of failure feature tables described above with reference to the failure feature table corresponding to the ambient temperature at that time. Then, to create the failed component prediction model based on the acquired fault contribution ratio R ^2. The method for creating the failure part prediction model is the same as that in the first embodiment.

In other words, the behavior of components such as a motor mounted on the device 51 may vary greatly depending on the ambient temperature, such as fluctuations in rotational speed and occurrence of vibration. In the fourth embodiment, to create a plurality of fault characteristic table for each ambient temperature, by selecting the failed feature table sets the fault contribution R ² according to the ambient temperature at the start of driving to predict the faulty part. Therefore, even when the behavior of the device 51 changes according to the ambient temperature of the device 51, it is possible to predict a faulty part with high accuracy. In the fourth embodiment, the example in which the ambient temperature of the device 51 is divided into five ranges of 0 ° C. or lower, 0 to 10 ° C., 10 to 20 ° C., 20 to 30 ° C., and 30 ° C. or higher is described. The present invention is not limited to this.

  The failure diagnosis apparatus, failure diagnosis method, and computer program of the present invention have been described based on the illustrated embodiment. However, the present invention is not limited to this, and the configuration of each part has the same function. It can be replaced with any configuration. For example, instead of the display terminal 16, it is also possible to present information on the failed part by voice or the like.

DESCRIPTION OF SYMBOLS 11 DB server 12 Model creation server 13 Fault component diagnostic server 14 Fault case search server 15 Transmitter / receiver 16 Display terminal 17 Operation data acquisition part 18 Temperature data acquisition part 51 Apparatus 52 Control apparatus 53 Temperature sensor 100 Fault diagnosis apparatus 111 Normal sensor Value table 112 Failure feature table 113 Failure case table 131 Failure component prediction model storage unit P0 to P3

Claims (7)

A failure diagnosis device for diagnosing a failure of a plurality of parts mounted on a device,
The sensor value of the sensor that detects the physical quantity of the component is acquired, and the contribution degree of each sensor value to each component failure based on the sensor value when the component is normal and the sensor value when the component is out of the acquired sensor values A failure feature table creation unit that creates a failure feature table that indicates the relationship between the sensor value and the contribution to the component failure,
For each part, a principal component analysis is performed using the contribution, and a faulty part prediction model creating unit that creates a faulty part prediction model,
When the sensor value of the sensor is acquired, a failure probability calculation unit that calculates a failure probability of each component with reference to the failure component prediction model;
When there is a component with a failure probability exceeding a predetermined threshold probability, a presentation control unit that presents the component;
A failure diagnosis apparatus comprising: An operation data acquisition unit for acquiring operation data of the device,
The failure diagnosis apparatus according to claim 1, wherein the failure feature table creation unit calculates the contribution based on the operation data in addition to the normal sensor value and the failure sensor value. . The failure feature table creation unit acquires, as the operation data, an elapsed time after the device starts driving, calculates the contribution for each elapsed time, and the failure feature table for each elapsed time The failure diagnosis apparatus according to claim 2, wherein: The failure feature table creation unit acquires, as the operation data, a continuous stop time of the device when the device starts driving, calculates the contribution for each continuous stop time, and the continuous stop time The failure diagnosis apparatus according to claim 2, wherein the failure feature table is created every time. An ambient temperature acquisition unit that acquires ambient temperature data of the device;
The failure feature table creation unit calculates the contribution based on the ambient temperature data in addition to the normal sensor value and the failure sensor value, and creates the failure feature table for each ambient temperature. The failure diagnosis apparatus according to claim 1, wherein: A failure diagnosis method for diagnosing a failure of a plurality of parts mounted on a device,
The sensor value of the sensor that detects the physical quantity of the component is acquired, and the contribution degree of each sensor value to each component failure based on the sensor value when the component is normal and the sensor value when the component is out of the acquired sensor values A step of creating a failure feature table indicating the relationship between the sensor value and the contribution to the component failure,
For each part, perform a principal component analysis using the contribution, and create a failure part prediction model,
When the sensor value of the sensor is acquired, referring to the failure component prediction model, calculating the failure probability of each component;
Presenting a component when there is a component with a failure probability exceeding a predetermined threshold probability; and
A failure diagnosis method comprising:   The computer program for functioning the computer as a failure diagnosis apparatus of any one of Claims 1-5.

Images