CN114137915A - A kind of fault diagnosis method of industrial equipment - Google Patents

Abstract

The invention discloses a fault diagnosis method of industrial equipment, which comprises the following steps: collecting samples, wherein the samples comprise normal samples and fault samples; constructing a parameter evaluation model, and training the parameter evaluation model by adopting a normal sample; evaluating the monitoring parameters of the equipment to be diagnosed by adopting the trained parameter evaluation model to obtain a monitoring parameter evaluation value; updating the fault sample according to the monitoring parameter evaluation value and the corresponding monitoring parameter; constructing a fault recognition model, and training the fault recognition model by adopting a fault sample; and acquiring real-time monitoring parameters of the equipment to be diagnosed, and identifying the real-time monitoring parameters through the trained fault identification model to obtain a fault diagnosis result. The invention improves the fault diagnosis efficiency and the fault diagnosis accuracy of the industrial equipment.

Description

A fault diagnosis method for industrial equipment

technical field

The invention belongs to the field of fault diagnosis, in particular to a fault diagnosis method for industrial equipment.

Background technique

Fault diagnosis mainly refers to the detection, separation and identification of faults that occur during the operation of the system or equipment, that is, to determine whether the fault occurs, and to locate the location and type of the fault. At present, the main fault diagnosis methods include the following:

1. Fault diagnosis method based on expert system

The expert system-based fault diagnosis method is to use the experience accumulated by domain experts in long-term practice to establish a knowledge base, and design a set of computer programs to simulate the reasoning and decision-making process of human experts for fault diagnosis. Expert system is mainly composed of knowledge base, reasoning engine, comprehensive database, human-machine interface and interpretation module.

The fault diagnosis method based on expert system can make use of the rich experience and knowledge of experts, without the need for mathematical modeling of the system and the diagnosis results are easy to understand, so it has been widely used. However, it is difficult to obtain the expert knowledge required by such methods, which becomes the main bottleneck in the development of expert systems; secondly, the accuracy of diagnosis depends on the richness of expert experience and the level of knowledge in the knowledge base; finally, when the rules When there are many cases, there are problems such as matching conflict and combination explosion in the reasoning process, which makes the reasoning speed slow and inefficient.

2. Fault diagnosis method based on graph theory

The fault diagnosis methods based on graph theory mainly include the Signed directed graph (SDG) method and the fault tree (Fault tree) method. SDG is a widely used graphical model for describing system causality. A fault tree is a special kind of logic diagram. The diagnosis method based on fault tree is an analysis process from effect to cause. It starts from the fault state of the system, conducts reasoning analysis step by step, and finally determines the basic cause, influence degree and occurrence probability of the fault.

The fault diagnosis method based on graph theory has the characteristics of simple modeling, easy to understand results and wide application range. However, when the system or equipment is more complex, the search process of such methods will become very complicated, and the diagnosis accuracy rate is not high, which may give invalid fault diagnosis results. Therefore, in practical applications, most of them are used in combination with other methods. .

3. Fault diagnosis method based on analytical model

The analytical model-based fault diagnosis method uses the accurate mathematical model of the system and the observable input and output to construct residual signals to reflect the inconsistency between the expected behavior of the system and the actual operating mode, and then performs fault diagnosis based on the analysis of the residual signals.

The fault diagnosis based on the analytical model utilizes the deep understanding of the inside of the system and has a good diagnosis effect. However, this kind of method relies on the accurate mathematical model of the object to be diagnosed. In practice, it is often difficult to establish an accurate mathematical model of the object. At this time, the fault diagnosis method based on the analytical model is no longer applicable.

4. Fault diagnosis method based on multivariate statistical analysis

The fault diagnosis method based on multivariate statistical analysis is to use the correlation between multiple variables of the process to diagnose the fault of the process. According to the historical data of the process variables, this method uses the multivariate projection method to decompose the multivariate sample space into the main component variables. A lower-dimensional projection subspace and a corresponding residual subspace are formed, and statistics that can reflect spatial changes are constructed in these two spaces respectively, and then the observation vector is projected to the two subspaces respectively, and the corresponding Statistical indicators are used for process monitoring.

The fault diagnosis method based on multivariate statistical analysis does not require in-depth understanding of the structure and principle of the system. It is completely based on the measurement data of sensors during the operation of the system, and the algorithm is simple and easy to implement. However, the physical meaning of the faults diagnosed by this kind of method is unclear and difficult to explain, and due to the complexity of the actual system, there are still many problems in this kind of method to be further studied, such as the nonlinearity between the process variables, and the dynamics of the process. Sex and time-varying, etc.

5. Fault diagnosis method based on signal processing

The fault diagnosis method based on signal processing uses various signal processing methods to analyze and process the measurement signal, and extract the time domain or frequency domain features of the fault-related signal for fault diagnosis, mainly including spectrum analysis method and wavelet transform method.

The fault diagnosis method based on signal processing needs to manually design the signal processing process and details for fault diagnosis. However, for different problems, the signal selection and processing methods are different, which makes the application scope of such methods very small; Meanwhile, the manually designed signal processing methods are highly dependent on domain knowledge, which makes it difficult for the extracted signal features to accurately diagnose faults.

6. Fault diagnosis method based on machine learning

The basic idea of fault diagnosis methods based on machine learning is to use the historical data of the system under normal and various fault conditions to train machine learning algorithms such as neural networks or support vector machines for fault diagnosis. In fault diagnosis, neural network is mainly used to classify the extracted fault features.

The fault diagnosis method based on machine learning takes the correct rate of fault diagnosis as the learning target, and has a wide range of applications. However, machine learning algorithms require sample data in the case of process failures, and the accuracy is closely related to the completeness and representativeness of the samples, so it is difficult to apply to those industrial processes where a large amount of failure data cannot be obtained.

SUMMARY OF THE INVENTION

Aiming at the above deficiencies in the prior art, the present invention provides a fault diagnosis method for industrial equipment, which solves the problems of inaccurate fault diagnosis and low fault diagnosis efficiency in the prior art.

In order to achieve the above purpose of the invention, the technical solution adopted in the present invention is: a fault diagnosis method for industrial equipment, comprising:

collecting samples, the samples include normal samples and fault samples;

Build a parameter evaluation model, and use normal samples to train the parameter evaluation model;

Use the trained parameter evaluation model to evaluate the monitoring parameters of the equipment to be diagnosed, and obtain the evaluation value of the monitoring parameters;

Update the fault samples according to the evaluation value of the monitoring parameters and the corresponding monitoring parameters;

Build a fault identification model, and use fault samples to train the fault identification model;

Obtain the real-time monitoring parameters of the equipment to be diagnosed, and identify the real-time monitoring parameters through the trained fault identification model to obtain the fault diagnosis result.

Further, the normal sample is:

D _P ={(X _tk ,Y _t )|t∈{k,k+1,...,m},k>0}

Among them, D _P represents the normal sample set, the input X _tk of the normal sample represents the monitoring parameter value of the industrial equipment from the monitoring start time to the time tk, k represents the prediction time interval, m is a positive integer, m>k, the output of the normal sample Y _t represents the monitoring parameter value of the industrial equipment at time t; when k=1, the input X _tk and output Y _t of the normal sample respectively represent the monitoring parameter value matrix of the industrial equipment from the monitoring start time to the previous time and the monitoring at the current time. parameter value.

3. The fault diagnosis method of industrial equipment according to claim 2, wherein the fault sample is:

D _E ={(X _t ,Y _t ),t∈{0,1,...,m}}

Among them, D _E represents the set of fault samples, X _t represents the monitoring parameter value of the industrial equipment at time t, and Y _t represents the one-hot vector of the fault label of the industrial equipment at time t.

Further, the parameter identification model includes an input layer, several convolutional neural network layers, a flat layer, an LSTM layer, a fully connected layer and an output layer connected in sequence; the convolutional neural network layer includes a first convolutional layer connected in sequence. layer, the first pooling layer, the second convolution layer and the second pooling layer; the fully connected layer uses the Relu function as the activation function.

Further, according to the monitoring parameter evaluation value and the corresponding monitoring parameter, the fault sample is updated, including:

According to the monitoring parameter evaluation value, the difference σ between the monitoring parameter value of the equipment to be diagnosed and the monitoring parameter evaluation value is obtained as:

σ=f(Y,Y')

Wherein, f(Y, Y') represents the difference value calculation function, Y represents the monitoring parameter value of the device to be diagnosed, and Y' represents the monitoring parameter evaluation value, and the monitoring parameter value and the monitoring parameter evaluation value are both at continuous time points. sequence, |*| means to take the absolute value;

According to the difference value σ, the stability H of the equipment to be diagnosed is obtained as:

Among them, e represents a natural constant, W represents the weight coefficient of the monitoring parameter, and the monitoring parameter includes n sub-monitoring parameters, W=[w ₁ w ₂ ... _wn ], w ₁ w ₂ ... _wn respectively represents the weight of the n sub-monitoring parameters coefficient;

According to the stability H, the monitoring parameters corresponding to the failure of the industrial equipment are obtained, and the corresponding monitoring parameters are added to the fault sample set to complete the update of the fault samples.

Further, according to the degree of stability H, obtaining the monitoring parameters corresponding to the failure of the equipment to be diagnosed, including:

Obtain the sample mean and standard deviation of the normal sample;

Sum the sample mean and three standard deviations to get the sum, and use the sum as the upper control limit;

Calculate the difference between the sample mean and three times the standard deviation to obtain the difference, and use the difference as the lower control limit;

Obtain the average of the upper and lower control limits, and use the average as the center line;

Construct fault judgment conditions according to the upper control limit, lower control limit and center line;

The stability of the equipment to be diagnosed at successive time points is obtained, and the monitoring parameters corresponding to the failure of the equipment to be diagnosed are obtained according to the stability at the successive time points and fault judgment conditions.

Further, according to the upper control limit, lower control limit and center line, construct fault judgment conditions, including:

Divide the area between the upper control limit and the center line into three equal parts, and divide the area between the lower control limit and the center line into three equal parts;

Take the area close to the upper control limit and the area close to the lower control limit as area A, the area close to the center line as area C, and the area between area A and area C as area B;

According to the A, B and C areas, construct the fault judgment condition.

Further, the fault judgment conditions include:

Condition 1, if the stability is lower than the lower control limit or higher than the upper control limit, the equipment to be diagnosed is faulty;

Condition 2, if the stability at nine consecutive time points is located in zone C and on one side of the center line, the equipment to be diagnosed is faulty;

Condition 3, if the stability increases or decreases at six consecutive time points, the equipment to be diagnosed is faulty;

Condition 4: If there are 14 consecutive time points where the stability is alternately up and down, the equipment to be diagnosed is faulty;

Condition 5: If two of the stability at three consecutive time points are located in the A area, the equipment to be diagnosed is faulty;

Condition 6, if four of the stability at five consecutive time points are located outside the C area and on one side of the center line, the equipment to be diagnosed is faulty;

Condition 7, if the stability at fifteen consecutive time points is located in zone C and distributed on both sides of the center line, the equipment to be diagnosed is faulty;

Condition 8: If the stability at eight consecutive time points is located in zone A or zone B and distributed on both sides of the center line, the device to be diagnosed is faulty.

Further, building a fault identification model and using fault samples to train the fault identification model, including:

The XGBoost model is used as the fault identification model;

The fault identification model is trained with fault samples.

The beneficial effects of the present invention are:

(1) The present invention provides a fault diagnosis method for industrial equipment, which improves the fault diagnosis efficiency and fault diagnosis accuracy of the industrial equipment.

(2) The present invention builds a parameter evaluation model, and selects the normal operation data of industrial equipment as training sample data, so as to realize the estimation of the current detection parameters of industrial equipment under the premise of not needing fault sample data, combined with the current actual situation of industrial equipment Monitor the parameters, obtain their current stability, and then obtain the stability sequence at continuous time points, then construct the fault judgment condition, and monitor the obtained stability sequence in real time through the fault judgment condition to judge whether the operation of the industrial equipment is abnormal.

(3) The parameter evaluation model in the present invention only uses normal data as a training sample, and the fault identification model only uses fault data as a training sample, so that the normal data and the fault data do not interfere with each other. In addition, only by judging the abnormal operation of industrial equipment through the parameter evaluation model, the fault identification model is used to identify the fault data, so as to avoid the fault identification operation of the normal operation data and improve the efficiency of fault diagnosis.

(4) After the invention adopts the fault identification model to obtain the corresponding fault type, and combines the fault type and the fault detailed information association table, the detailed information such as the specific fault location and cause can be obtained, thereby supporting the operation and maintenance personnel to solve the fault quickly and accurately. question.

Description of drawings

FIG. 1 is a flowchart of a fault diagnosis method for an industrial equipment provided by an embodiment of the present application.

FIG. 2 is a structural diagram of a parameter evaluation model provided by an embodiment of the present application.

FIG. 3 is a schematic diagram of condition 1 provided by an embodiment of the present application.

FIG. 4 is a schematic diagram of condition 2 provided by the embodiment of the present application.

FIG. 5 is a schematic diagram of condition 3 provided by the embodiment of the present application.

FIG. 6 is a schematic diagram of condition 4 provided by the embodiment of the present application.

FIG. 7 is a schematic diagram of condition 5 provided by an embodiment of the present application.

FIG. 8 is a schematic diagram of the sixth condition provided by the embodiment of the present application.

FIG. 9 is a schematic diagram of the seventh condition provided by the embodiment of the present application.

FIG. 10 is a schematic diagram of condition eight provided by this embodiment of the present application.

Detailed ways

The specific embodiments of the present invention are described below to facilitate those skilled in the art to understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Such changes are obvious within the spirit and scope of the present invention as defined and determined by the appended claims, and all inventions and creations utilizing the inventive concept are within the scope of protection.

The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

Example 1

As shown in Figure 1, a fault diagnosis method for industrial equipment includes:

collecting samples, the samples include normal samples and fault samples;

Build a parameter evaluation model, and use normal samples to train the parameter evaluation model;

Use the trained parameter evaluation model to evaluate the monitoring parameters of the equipment to be diagnosed, and obtain the evaluation value of the monitoring parameters;

Update the fault samples according to the evaluation value of the monitoring parameters and the corresponding monitoring parameters;

Build a fault identification model, and use fault samples to train the fault identification model;

Obtain the real-time monitoring parameters of the equipment to be diagnosed, and identify the real-time monitoring parameters through the trained fault identification model to obtain the fault diagnosis result.

Optionally, the industrial equipment may include equipment such as machine tools and stamping equipment, but the technical solutions provided by the present invention are not limited thereto, and the technical solutions provided by the present invention may also be applied to industrial systems.

Optionally, the monitoring parameter values may include temperature values, rotational speed values, current values, and voltage values, etc. After data cleaning and parameter standardization operations are performed on the monitoring parameter values, they may be divided into normal samples and fault samples.

Data cleaning can include data consistency checking, null value processing, missing data processing, invalid value processing, and duplicate value processing. Normalization is to eliminate the differences between features and facilitate weight learning during model training. Normalization can include min-max normalization, Z-score normalization (zero-mean normalization), or decimal scaling normalization.

In a possible implementation, the normal sample is:

D _P ={(X _tk ,Y _t ),t∈{k,k+1,...,m},k>0}

Among them, D _P represents the normal sample set, the input X _tk of the normal sample represents the monitoring parameter value of the industrial equipment from the monitoring start time to the time tk, k represents the prediction time interval, m is a positive integer, m>k, the output of the normal sample Y _t represents the monitoring parameter value of the industrial equipment at time t; when k=1, the input X _tk and output Y _t of the normal sample respectively represent the monitoring parameter value matrix of the industrial equipment from the monitoring start time to the previous time and the monitoring at the current time. parameter value.

Optionally, if the monitoring parameter values are multiple parameter values, the input X _tk of the normal sample represents the monitoring parameter value matrix of the industrial equipment from the monitoring start time to the time tk, and the output Y _t of the normal sample represents the monitoring of the industrial equipment at the time t. A vector of parameter values.

In this embodiment, the monitoring parameter value in the normal sample is the time series multidimensional data of the industrial equipment during the normal operation period.

In a possible implementation, the fault sample is:

D _E ={(X _t ,Y _t ),t∈{0,1,...,m}}

Among them, D _E represents the fault sample set, X _t represents the monitoring parameter value of the industrial equipment at time t, and Y _t represents the one-hot (one-hot) vector of the fault label of the industrial equipment at time t.

The monitoring parameter value and the one-hot vector of the fault label when the industrial equipment fails are taken as the fault sample.

Optionally, an association table between fault labels and fault details can be constructed, and after identifying the fault type (fault label) through the fault identification model, the fault details can be located through the association table.

As shown in Figure 2, the parameter identification model includes an input layer, several convolutional neural network layers, a flat layer, an LSTM (Long Short-Term Memory, long short-term memory network) layer, a fully connected layer, and an output layer that are connected in sequence; The convolutional neural network layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; the fully connected layer uses a Relu function (linear rectification function) as activation function.

Optionally, the training of the parameter evaluation model is supervised training.

Among them, the convolution layer is a two-dimensional convolution layer, which is used to perform convolution operation on the input data to realize feature extraction. R ^h×m×1 , let the number of convolution kernels K, the specification F of each convolution kernel, the stride S of convolution, and the number of zero-value padding P. For each convolution operation, first perform zero-value padding on the data X to convert the data volume into X'∈R ^{(h+p)×(m+p)×1} , and then perform the convolution operation. The convolution operation is as follows shown.

Kernel ∈ R ^F×F×1

T=Kernel×X'

Repeat the above calculation for the input X in the order from left to right and from top to bottom, according to the convolution step size S, and re-integrate the y values obtained each time into a matrix, that is, the output Y∈R ^{h'×m' ×d} , where h'=(h-F+2*P)/S+1, m'=(m-F+2*P)/S+1, d=1.

The pooling layer can choose two methods: Max polling and Average pooling. The calculation formula is as follows:

Maxpolling:y=max(X _c )

Aceragepooling: y=mean(X _c )

The input X _c is a matrix whose width and height are F, and the pooling operation is performed on the output of the convolution layer in the order from left to right and top to bottom, that is, the output Y'∈R ^h”×m” is obtained, where , h"=(hF)/S+1, m"=(mF)/S+1.

At the same time, considering that the monitoring parameters of the system or equipment are usually positive values, Relu is selected as the activation function in the fully connected layer, and the image of the Relu function definition domain function is shown below.

f(x)=max(0,x).

In a possible implementation manner, the updating of the fault sample according to the monitoring parameter evaluation value and the corresponding monitoring parameter includes:

According to the monitoring parameter evaluation value, the difference σ between the monitoring parameter value of the equipment to be diagnosed and the monitoring parameter evaluation value is obtained as:

σ=f(Y,Y')

Wherein, f(x,y) represents the difference value calculation function, Y represents the monitoring parameter value of the device to be diagnosed, Y' represents the monitoring parameter evaluation value, and the monitoring parameter value and the monitoring parameter evaluation value are both sequences on consecutive time points .

Optionally, if the monitoring parameter value is multiple parameter values, the parameter evaluation value is a vector, and the parameter evaluation value includes the evaluation values corresponding to the multiple parameter values.

The monitoring parameter evaluation value corresponds to the monitoring parameter value, the monitoring parameter evaluation value represents the monitoring parameter predicted value of the device to be diagnosed at successive time points, and the monitoring parameter value represents the real value of the monitoring parameter of the device to be diagnosed at continuous time points. According to the monitoring parameters and the evaluation values of the monitoring parameters, the sequence of difference values of the equipment to be diagnosed at successive time points can be obtained.

According to the difference value σ, the stability H of the equipment to be diagnosed is obtained as:

Among them, e represents a natural constant, W represents the weight coefficient of the monitoring parameter, and the monitoring parameter includes n sub-monitoring parameters, W=[w ₁ w ₂ ... _wn ], w ₁ w ₂ ... _wn respectively represents the weight of the n sub-monitoring parameters coefficient;

According to the stability H, the monitoring parameters corresponding to the failure of the industrial equipment are obtained, and the corresponding monitoring parameters are added to the fault sample set to complete the update of the fault samples.

In a possible implementation manner, according to the degree of stability H, the acquisition of monitoring parameters corresponding to the failure of the equipment to be diagnosed includes:

Obtain the sample mean and standard deviation of the normal sample;

Sum the sample mean and three standard deviations to get the sum, and use the sum as the upper control limit;

Calculate the difference between the sample mean and three times the standard deviation to obtain the difference, and use the difference as the lower control limit;

Obtain the average of the upper and lower control limits, and use the average as the center line;

Construct fault judgment conditions according to the upper control limit, lower control limit and center line;

The stability of the equipment to be diagnosed at successive time points is obtained, and the monitoring parameters corresponding to the failure of the equipment to be diagnosed are obtained according to the stability at the successive time points and fault judgment conditions.

In this embodiment, the monitoring parameter value is the time series multidimensional data of the industrial equipment during the normal operation period. By identifying the time series multidimensional data, the stability of the equipment to be diagnosed at consecutive time points can be obtained, that is, a stability sequence can be obtained. .

In a possible implementation, according to the upper control limit, the lower control limit and the center line, the fault judgment condition is constructed, including:

Divide the area between the upper control limit and the center line into three equal parts, and divide the distance between the lower control limit and the center line into three equal parts;

Take the area close to the upper control limit and the area close to the lower control limit as area A, the area close to the center line as area C, and the area between area A and area C as area B;

According to the A, B and C areas, construct the fault judgment condition.

In a possible implementation manner, the fault judgment conditions include condition one to condition eight.

As shown in Figure 3, the first condition is: if the stability is lower than the lower control limit or higher than the upper control limit, the equipment to be diagnosed is faulty;

As shown in Figure 4, the second condition is: if the stability at nine consecutive time points are all located in zone C and on one side of the center line, the equipment to be diagnosed is faulty;

As shown in Figure 5, the third condition is: if the stability increases or decreases at six consecutive time points, the equipment to be diagnosed is faulty;

As shown in Figure 6, the fourth condition is: if there are fourteen consecutive time points where the stability is alternately up and down, the device to be diagnosed is faulty;

As shown in Figure 7, the fifth condition is: if two of the stability degrees at three consecutive time points are located in the A area, the equipment to be diagnosed is faulty;

As shown in Figure 8, the sixth condition is: if four of the stability degrees at five consecutive time points are located outside the C area and are located on one side of the center line, the device to be diagnosed is faulty;

As shown in Figure 9, the seventh condition is: if the stability at fifteen consecutive time points is located in the C zone and distributed on both sides of the center line, the equipment to be diagnosed is faulty;

As shown in FIG. 10 , the eighth condition is: if the stability at eight consecutive time points is located in the A area or the B area, and is distributed on both sides of the center line, the equipment to be diagnosed is faulty.

In this embodiment, when the equipment to be diagnosed fails, the stability at the last time point in the consecutive time points is used as the stability violation point (X in Fig. 3-10), and the corresponding stability violation point is used as the stability violation point. The monitoring parameters and the one-hot vector of the fault labels corresponding to the monitoring parameters are used as fault samples.

In a possible implementation manner, the building a fault identification model, and using fault samples to train the fault identification model, includes:

The XGBoost model is used as the fault identification model;

The fault identification model is trained with fault samples.

When an industrial equipment fault is detected, it is necessary to identify the specific fault type, so as to locate the fault and facilitate the maintenance and troubleshooting of the operation and maintenance personnel. When the fault sample data accumulates to a certain extent, XGBoost (Extreme Gradient Boosting, gradient boosting model) is used to build a fault identification model, select fault sample data, and divide the sample data accordingly (part of it as training data and part of it as test data). data), perform supervised training, and save the training result model. XGBoost is the product of the combination of the decision tree model and AdaBoost (Adaptive Boost), and is an improvement of the GDBT (Gradient Boosting Decision Tree, gradient boosting tree) model. Its specific implementation is as follows.

If the decision tree of the t-1th round is known, and the decision tree of the t-th round is constructed based on this, the objective function Obj ^(t) of the t-th round is:

Among them, y _i is the actual value, f _t-1 ( _xi ) is the output result of the first t-1 decision trees, f( _xi ) is the output result of constructing the decision tree in the t round, L(y _i ,f _t-1 (x _i )+f(x _i )) is the loss function, that is, Ω(f(x)) is regular.

Perform a second-order Taylor expansion of the loss function at the t-1th tree:

L(y _i ,f _t-1 (x _i )+f _t (x _i ))=L(y _i ,f _t-1 (x _i ))+g _i f(x _i )+h _i f(x _i ) ² /2

Among them, L(y _i , f _t-1 (x _i )) is the loss function value corresponding to the first t-1 trees, that is, the prediction error of the learning model composed of the first t-1 trees, which is a constant, g _i , h _i are the first-order and second-order derivatives of the prediction error to the current model, respectively.

For decision trees, there are the following formulas:

f _k (x)=w _q(x)

为叶子节点向量的模。Among them, q(x) represents the output leaf node serial number, which is an index mapping, which is used to map the input to the leaf index number, w _q(x) represents the value of the corresponding leaf node serial number, and r represents the node segmentation. Difficulty, T is the number of leaf nodes, λ is the L2 regularization coefficient, w _j is the predicted value of the jth leaf node,

is the modulus of the leaf node vector.

Therefore, the objective function can be written in the following form:

Substitute into the decision tree formula to get:

Convert the first part of the above formula from accumulating all pairs of training sample sets to accumulating all leaf nodes, you can get:

Among them, I _t represents the sample set, and G _j and H _j represent the sum of the first-order derivatives and the second-order derivatives of all input samples mapped to the leaf node j, respectively.

Taking the derivative of W _j and bringing it into the above formula, we get:

Based on the above formula, XGBoost can construct the predicted value of each leaf node of the decision tree, and use the greedy algorithm to calculate the gain corresponding to the objective function of the decision tree after splitting, until the maximum decision tree depth is reached, or the Gain value is less than 0. The calculation formula of Gain is shown in the following formula.

G_L、G_R分别为左右节点对应的G值，H_L、H_R分别为左右节点对应的H值，I指当前节点实例集合。in,

GL and _GR are the G values corresponding to the left and right nodes, respectively, _HL and _HR are the _H values corresponding to the left and right nodes, respectively, and I refers to the current node instance set.

Obtain the fault parameters of the equipment to be diagnosed, and standardize the fault parameters. According to the fault parameters, the fault type is identified through the fault identification model, and the fault details are obtained through the fault type and fault details association table, so that the operation and maintenance personnel can troubleshoot and repair the equipment to be diagnosed based on the obtained fault details. The fault details can include the fault location and the fault cause.

Example 2

In this embodiment, a fault diagnosis apparatus for industrial equipment is provided, which includes an acquisition module, a first training module, an evaluation module, an update module, a second training module, and an identification module.

The collection module is used to collect samples, and the samples include normal samples and fault samples;

The first training module is used to construct a parameter evaluation model, and use normal samples to train the parameter evaluation model;

The evaluation module is used to evaluate the monitoring parameters of the to-be-diagnosed equipment by using the trained parameter evaluation model to obtain the evaluation value of the monitoring parameters;

The update module is used to update the fault sample according to the evaluation value of the monitoring parameter and the corresponding monitoring parameter;

The second training module is used to construct a fault identification model, and use fault samples to train the fault identification model;

The identification module is used to obtain the real-time monitoring parameters of the equipment to be diagnosed, and identify the real-time monitoring parameters through the trained fault identification model to obtain the fault diagnosis result.

Example 3

In this embodiment, a fault diagnosis device for industrial equipment is provided, including a processor and a memory;

the memory stores computer-executable instructions;

The processor executes the computer-executed instructions stored in the memory, so that the processor executes the method for diagnosing a fault of an industrial device as described in Embodiment 1.

Example 4

In this embodiment, a computer-readable storage medium is provided, where computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, are used to implement the industrial process described in Embodiment 1 Troubleshooting methods for equipment.

Example 5

In this embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the method for diagnosing the fault of the industrial equipment described in Embodiment 1.

The invention provides a fault diagnosis method for industrial equipment, which improves the fault diagnosis efficiency and fault diagnosis accuracy of the industrial equipment.

The present invention constructs a parameter evaluation model, selects the normal operation data of the industrial equipment as the training sample data, so as to realize the estimation of the current detection parameters of the industrial equipment under the premise of not needing the fault sample data, and combined with the current actual monitoring parameters of the industrial equipment, Obtain its current stability, and then obtain the stability sequence at continuous time points, then construct the fault judgment condition, and monitor the obtained stability sequence in real time through the fault judgment condition to judge whether the operation of the industrial equipment is abnormal.

The parameter evaluation model in the present invention only uses normal data as a training sample, and the fault identification model only uses fault data as a training sample, so that the normal data and the fault data do not interfere with each other. In addition, only the abnormal operation of industrial equipment is judged by the parameter evaluation model, and the fault identification model is used to identify the fault data, so as to avoid the fault identification operation of the normal operation data and improve the efficiency of fault diagnosis.

The present invention uses the fault identification model to obtain the corresponding fault type, and combines the fault type and the fault detailed information association table to obtain the specific fault location, cause and other detailed information, so as to support the operation and maintenance personnel to solve the fault problem quickly and accurately.

Claims (9)

1. A method of diagnosing a fault in an industrial device, comprising:

collecting samples, wherein the samples comprise normal samples and fault samples;

constructing a parameter evaluation model, and training the parameter evaluation model by adopting a normal sample;

evaluating the monitoring parameters of the equipment to be diagnosed by adopting the trained parameter evaluation model to obtain a monitoring parameter evaluation value;

updating the fault sample according to the monitoring parameter evaluation value and the corresponding monitoring parameter;

constructing a fault recognition model, and training the fault recognition model by adopting a fault sample;

and acquiring real-time monitoring parameters of the equipment to be diagnosed, and identifying the real-time monitoring parameters through the trained fault identification model to obtain a fault diagnosis result.

2. The method for diagnosing a malfunction of an industrial device according to claim 1, wherein the normal sample is:

D_P＝{(X_t-k,Y_t)|t∈{k,k+1,...,m},k>0}

wherein D is_PRepresenting a set of normal samples, input X of normal samples_t-kRepresenting the value of a monitoring parameter of the industrial equipment from the monitoring starting moment to the t-k moment, k representing the prediction time interval, m being a positive integer, m>k, output of Normal sample Y_tMonitoring parameter for indicating industrial equipment at time tA numerical value; when k is 1, input X of normal sample_t-kAnd output Y_tRespectively representing a monitoring parameter value matrix of the industrial equipment from the monitoring starting moment to the previous moment and a monitoring parameter value of the industrial equipment at the current moment.

3. The method for fault diagnosis of an industrial device according to claim 2, wherein the fault samples are:

D_E＝{(X_t,Y_t),t∈{0,1,...,m}}

wherein D is_ERepresenting a set of fault samples, X_tValue of a monitoring parameter representing the industrial plant at time t, Y_tAnd a fault label one-hot vector of the industrial equipment at the moment t.

4. The fault diagnosis method of industrial equipment according to claim 1, wherein the parameter identification model comprises an input layer, a plurality of convolutional neural network layers, a flat layer, an LSTM layer, a full connection layer and an output layer which are connected in sequence; the convolutional neural network layer comprises a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer which are sequentially connected; the full connection layer adopts Relu function as an activation function.

5. The method for diagnosing the fault of the industrial equipment according to claim 1, wherein the updating the fault sample according to the evaluated value of the monitoring parameter and the corresponding monitoring parameter comprises:

acquiring a difference value sigma between a monitoring parameter value and a monitoring parameter evaluation value of the equipment to be diagnosed according to the monitoring parameter evaluation value, wherein the difference value sigma is as follows:

σ＝f(Y,Y')

wherein f (Y, Y ') represents a difference value calculation function, Y represents a monitoring parameter value of the equipment to be diagnosed, Y' represents a monitoring parameter evaluation value, the monitoring parameter value and the monitoring parameter evaluation value are both sequences on continuous time points, and | represents an absolute value;

according to the difference value sigma, obtaining the stability H of the equipment to be diagnosed as follows:

wherein e represents a natural constant, W represents a weight coefficient of the monitoring parameter, the monitoring parameter includes n sub-monitoring parameters, and W ═ W₁ w₂…w_n]，w₁ w₂…w_nRespectively representing the weight coefficients of the n sub-monitoring parameters;

and acquiring corresponding monitoring parameters when the industrial equipment fails according to the stability H, and adding the corresponding monitoring parameters to the failure sample set to complete the updating of the failure sample.

6. The method for diagnosing the fault of the industrial equipment according to claim 5, wherein the step of obtaining the corresponding monitoring parameter when the equipment to be diagnosed has the fault according to the stability H comprises the steps of:

obtaining a sample mean value and a standard deviation of a normal sample;

summing the sample mean value and triple standard deviation to obtain a sum value, and taking the sum value as a control upper limit;

calculating the difference between the sample mean value and the triple standard deviation to obtain a difference value, and taking the difference value as a lower control limit;

acquiring an average value of the upper control limit and the lower control limit, and taking the average value as a central line;

constructing a fault judgment condition according to the upper control limit, the lower control limit and the center line;

and acquiring the stability of the equipment to be diagnosed at the continuous time point, and acquiring corresponding monitoring parameters when the equipment to be diagnosed fails according to the stability and the fault judgment condition at the continuous time point.

7. The method of diagnosing a fault in an industrial apparatus according to claim 6, wherein constructing the fault determination condition based on the upper control limit, the lower control limit, and the center line includes:

equally dividing the area between the upper control limit and the central line into three parts, and equally dividing the area between the lower control limit and the central line into three parts;

taking a region close to the upper control limit and a region close to the lower control limit as a region A, taking a region close to the central line as a region C, and taking a region between the region A and the region C as a region B;

and constructing a fault judgment condition according to the area A, the area B and the area C.

8. The fault diagnosis method of an industrial apparatus according to claim 7, wherein the fault determination condition includes:

if the stability is lower than the lower control limit or higher than the upper control limit, the equipment to be diagnosed fails;

if the stability of the nine continuous time points is located in the area C and on one side of the central line, the equipment to be diagnosed fails;

if the stability at six continuous time points is increased or decreased progressively, the equipment to be diagnosed fails;

if the stability of fourteen continuous time points is alternately up and down, the equipment to be diagnosed fails;

if two of the three stability degrees at the continuous time points are located in the area A, the equipment to be diagnosed fails;

if four of the stability degrees on the five continuous time points are located outside the C area and on one side of the center line, the equipment to be diagnosed breaks down;

if fifteen stabilities at continuous time points are located in the C area and distributed on two sides of the center line, the equipment to be diagnosed fails;

and if the stabilities of eight continuous time points are positioned in the area A or the area B and distributed on two sides of the central line, the equipment to be diagnosed fails.

9. The method for fault diagnosis of industrial equipment according to claim 1, wherein the constructing a fault recognition model and training the fault recognition model with fault samples comprises:

an XGboost model is adopted as a fault identification model;

and training the fault recognition model by adopting the fault sample.

Images