Fault diagnosis method based on assembly cascade graph attention and neural network

Abstract

The invention provides a fault diagnosis method based on assembly cascade graph attention and a neural network. The method comprises the following steps: step 1, establishing a complex equipment cascade assembly model; 2, establishing a graph attention model based on a cascade assembly model, and preprocessing the feature vectors; and step 3, establishing a fault diagnosis service-oriented neural network model to realize product fault diagnosis service training. According to the method, the cascade assembly relation of the product is creatively fused into the graph attention mechanism, the fault diagnosis method based on the assembly cascade graph attention mechanism and the neural network is constructed, and assistance can be provided for workshop quality management and control personnel to carry out complex equipment assembly quality fault diagnosis.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of product processing, and in particular to a fault diagnosis method based on an assembly cascading graph attention and a neural network. BACKGROUND

[0002] At present, product processing quality fault positioning methods mainly include an artificial experience-based method, a mathematical model-based method and a big data analysis-based method. The artificial experience-based method is that workshop personnel analyzes unqualified product causes by using work experience. The method has defects such as strong subjectivity, low accuracy, time-consuming and laborious, and the like. Especially, current production modes of most enterprises are changed into multi-variety and small-batch production modes. If the traditional artificial experience method is still used to analyze product quality faults, the phenomenon of missing judgment and wrong judgment will occur. The mathematical model-based method is that the correlation between product fault signs and fault sources is described by a certain mathematical means, so as to determine the fault type. The big data analysis-based method is that the correlation information between data is extracted from historical data by using a data analysis method, and then fault analysis of an actual sample is realized. With the advent of the big data era, the method is widely applied in various fields. Compared with the traditional artificial experience-based method and the mathematical model-based method, the method has higher accuracy, stronger feasibility and greatly saves time.

[0003] At present, many in-depth studies have been made on the quality fault positioning method based on big data analysis, mainly including a particle swarm algorithm, an SVM algorithm, a quantum genetic algorithm and a BP neural network algorithm. However, most of the current product processing fault diagnosis researches focus on positioning the fault cause to the final product layer, and lack of positioning the fault cause of parts considering the product assembly relationship. SUMMARY

[0004] The application provides a fault diagnosis method based on an assembly cascading graph attention and a neural network to solve the problems in the background art. The technical scheme provided by the application is as follows.

[0005] A fault diagnosis method based on an assembly cascading graph attention and a neural network comprises the following steps.

[0006] Step 1, establishing a complex equipment cascading assembly model, including a bill of materials model and a product quality performance characteristic model;

[0007] Step 2, establishing a graph attention model based on the cascading assembly model, and pre-processing a characteristic vector;

[0008] Step 3, establishing a neural network model facing a fault diagnosis business, containing an input layer, a hidden layer and an output layer, and performing product fault diagnosis business training.

[0009] Preferably, the bill of material model is: a complex equipment cascade assembly system collectively contains products , wherein is the final assembly, and is represented by the following formula:

[0010]

[0011]

[0012]

[0013] The first formula represents that for any non-final assembly, there is and only one upper product in the assembly relationship; the second formula represents that for the final assembly, there is at least one lower product in the assembly relationship; and the third formula is the definition of the Boolean variable .

[0014] Preferably, the product quality performance characteristic model is:

[0015] The above formula represents that the product collectively contains key quality characteristics.

[0016] Preferably, step 2 specifically comprises:

[0017] Step 21, establishing a product adjacency matrix based on the cascade assembly relationship, and the value of each element in the matrix is .

[0018] Step 22, calculating the correlation between any product and other products .

[0019]

[0020] In the above formula, is a shared linear conversion layer used to convert the dimension of product characteristics to . The function is used to map two one-dimensional vectors into a scalar value;

[0021] Step 23, normalizing the correlation obtained in step 22 by using the softmax function;

[0022] Step 24, obtaining a new product characteristic vector according to the normalized correlation between products:

[0023]

[0024] In the above formula, It is an activation function.

[0025] Preferably, step 3 specifically includes:

[0026] Step 31: Build a neural network model containing one input layer, one hidden layer, and one output layer;

[0027] Step 32: Determine the number of nodes in the input and output layers of the neural network. and :

[0028]

[0029]

[0030] In the above formula, Indicates the type of product fault label;

[0031] Step 33: Determine the number of hidden layer nodes in the neural network model using empirical formulas. ;

[0032] Step 34: Determine the activation function;

[0033] Step 35: Based on the established neural network model, establish a fault diagnosis model:

[0034]

[0035]

[0036] In the above formula, and These represent the weights between the input layer and the hidden layer, and between the hidden layer and the output layer of the neural network, respectively. and Indicates the bias of the corresponding layer;

[0037] Step 36: Update the weights using the backpropagation principle according to the chain rule, and perform final training on the model.

[0038] Preferably, in step 34, the activation function of the hidden layer is set to the ReLU function, and the activation function of the output layer is set to the Softmax function.

[0039] Preferably, in step 35, the bias is... and All are set to 0.

[0040] Compared with the prior art, the beneficial effects achieved by this invention are: it innovatively integrates the cascaded assembly relationship of products into the graph attention mechanism, and constructs a fault diagnosis method based on the assembly cascaded graph attention mechanism and neural network. This invention can provide assistance to workshop quality control personnel in diagnosing faults in the assembly quality of complex equipment. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0042] Figure 1 This is a schematic diagram of the cascaded assembly relationship of the product of this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Example 1: A fault diagnosis method based on assembly cascaded graph attention and neural networks, comprising:

[0046] Step 1: Establish a cascaded assembly model for complex equipment;

[0047] Step 2: Establish a graph attention model based on the cascaded assembly model and preprocess the feature vectors;

[0048] Step 3: Establish a neural network model for fault diagnosis business and implement product fault diagnosis business training.

[0049] Step 1 further includes:

[0050] Step 11: Establish a bill of materials model for the cascaded assembly of complex equipment. The cascaded assembly system of complex equipment includes... Product ,in For final assembly, Figure 1 This is a schematic diagram when n=7. The remaining products are components with a cascaded assembly relationship, and the assembly relationship between the products can be determined by the following formula:

[0051]

[0052]

[0053]

[0054] The first equation states that for any non-final assembly, there must exist one and only one upper-level product in an assembly relationship. The second equation states that for a final assembly, there must exist at least one lower-level product in an assembly relationship. The third equation is a Boolean variable. Definition.

[0055] Step 12: Establish a product quality performance characteristic model:

[0056] The above formula indicates that the product Total of Key quality characteristics.

[0057] Step 2 also includes:

[0058] Step 21: Establish a product adjacency matrix based on cascaded assembly relationships; the product adjacency matrix based on cascaded assembly relationships is a... A matrix, where the value of each element is... .

[0059] Step 22: For any product Calculate its correlation with other products according to the following formula. ;

[0060]

[0061] In the above formula, It is a shared linear transformation layer, that is, a A matrix is ​​used to transform the dimensions of product features to... . The function is used to combine two one-dimensional vectors ( and The size of the vector is The function is mapped to a scalar value, and here it can be set as a fully connected layer of a neural network.

[0062] Step 23: Normalize the correlation obtained in Step 22 using the softmax function:

[0063]

[0064] Step 24: Obtain new product feature vectors based on the normalized inter-product correlations.

[0065]

[0066] In the above formula, The calculation result is still a one-dimensional object with a size of The vectors, respectively with coefficients The result of multiplying and then summing It is also a one-dimensional object with a size of The vector, It is an activation function.

[0067] Step 3 also includes:

[0068] Step 31: Build a neural network model that includes an input layer, a hidden layer, and an output layer;

[0069] Step 32: Determine the number of nodes in the input and output layers of the neural network: and .

[0070]

[0071]

[0072] In the above formula, Indicates the type of product fault label.

[0073] Step 33: Determine the number of hidden layer nodes in the neural network model using empirical formulas. .

[0074]

[0075] Step 34: Determine the activation functions for each layer. For the complex equipment fault diagnosis problem proposed in this invention, the activation function for the hidden layer is set to the ReLU function, and the activation function for the output layer is set to the Softmax function.

[0076] Step 35: Based on the established neural network model, establish a fault diagnosis model:

[0077]

[0078]

[0079] In the above formula, , , and These are the parameters of the neural network, which need to be learned from the training data. and These represent the weights between the input layer and the hidden layer, and between the hidden layer and the output layer of the neural network, respectively. and This indicates the bias of the corresponding layer. In this invention, the bias... and All are set to 0. This is the product feature vector after preprocessing in step 2.

[0080] Step 36: Update the weights using the backpropagation principle according to the chain rule, and perform final training on the model.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fault diagnosis method based on assembly cascaded graph attention and neural networks, characterized in that, include: Step 1: Establish a cascaded assembly model for complex equipment, including a bill of materials model and a product quality and performance characteristic model; Step 2: Establish a graph attention model based on the cascaded assembly model and preprocess the feature vectors; Step 3: Establish a neural network model for fault diagnosis business, including an input layer, hidden layer and output layer, and train it for product fault diagnosis business.

2. The fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 1, characterized in that, The bill of materials model is as follows: a complex equipment cascade assembly system includes... Product ,in For the final assembly, it is represented by the following formula: ； ； ； The first equation indicates that for any non-final assembly, there exists one and only one upper-level product in an assembly relationship; the second equation indicates that for a final assembly, there exists at least one lower-level product in an assembly relationship; the third equation is a Boolean variable. Definition.

3. The fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 2, characterized in that, The product quality performance characteristic model is as follows: ; The above formula indicates that the product Total of Key quality characteristics.

4. The fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 3, characterized in that, Step 2 specifically includes: Step 21: Establish a product adjacency matrix based on cascaded assembly relationships. The value of each element in the matrix is... ; Step 22: Calculate any product Relevance with other products ; ； In the above formula, It is a shared linear transformation layer used to transform the dimensions of product features to... ; The function is used to map two one-dimensional vectors to a scalar value; Step 23: Normalize the correlation obtained in step 22 using the softmax function; Step 24: Obtain new product feature vectors based on the normalized inter-product correlations. ； In the above formula, It is an activation function.

5. The fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 4, characterized in that, Step 3 specifically includes: Step 31: Build a neural network model consisting of an input layer, a hidden layer, and an output layer; Step 32: Determine the number of nodes in the input and output layers of the neural network. and : ； ； In the above formula, Indicates the type of product fault label; Step 33: Determine the number of hidden layer nodes in the neural network model using empirical formulas. ; Step 34: Determine the activation function; Step 35: Based on the established neural network model, establish a fault diagnosis model: ； ； In the above formula, and These represent the weights between the input layer and the hidden layer, and between the hidden layer and the output layer of the neural network, respectively. and Indicates the bias of the corresponding layer; Step 36: According to the chain rule, update the weights using the backpropagation principle and perform final training on the model.

6. The fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 5, characterized in that, Step 34 sets the activation function of the hidden layer to ReLU and the activation function of the output layer to Softmax.

7. In the fault diagnosis method based on assembly cascaded graph attention and neural network according to claim 5, in step 35, the bias... and All are set to 0.

Images