Fault diagnosis method and device

Abstract

The present invention provides a fault diagnosis method and apparatus, relating to the technical field of fault analysis. The method includes: constructing a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network; determining first feature data for characterizing text semantics and second feature data for characterizing text fault topics according to work order sample data, the feature extraction network, and a historical fault database; determining a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network; and determining corresponding work order processing information in the historical fault database according to a similarity calculation between the fault diagnosis model and the historical fault database. The solution of the present invention completes the diagnosis of different cases, improves the accuracy of automated fault diagnosis, enhances the interpretability of model feature learning, and the diagnostic adaptation ability for different fault topics.

Description

Technical Field

[0001] The present invention relates to the technical field of fault analysis, and particularly to a fault diagnosis method and device. Background Art

[0002] In manufacturing industries such as aerospace, automotive, and processing industries, the vast amount of fault records stored has inspired the application of data mining and text mining in history data-driven fault diagnosis technologies. Fault records contain the mechanisms of product failures, the parts involved, and the fault phenomena, which can help carry out product analysis and guide workers to complete the repair of faults. Industrial fault records are divided into structured data (such as the model numbers of parts, operating signals, and common voltage and current observation values, etc., which can be directly used by computers for diagnosis), and unstructured information (usually embedded in text form). Structured numerical data can be directly utilized by computers. However, information retrieval and diagnosis of unstructured fault records usually rely on the experience of professional technicians, which is time-consuming and inefficient. Mining and analyzing such text records can help maintenance workers complete the judgment of fault types, the analysis of fault causes, and retrieve corresponding repair plans. Faults of complex products often involve different components, and each component will also produce different fault phenomena under different fault causes, corresponding to different fault theme categories and solutions. In addition, due to the existence of common text description problems such as professional terms, polysemous words, and modal particles, etc., it may cause deviations in computer translation and feature recognition of text. The differences in the habitual ways of recording texts by different manufacturers and maintenance personnel also affect the extraction effect of fault features. The extraction results of features, as the input of classifiers and similar case retrieval, directly determine the fault diagnosis performance of the model. Therefore, for case records of different products, there is an urgent need for a general method that can accurately extract fault features and carry out intelligent fault diagnosis. Summary of the Invention

[0003] The purpose of the present invention is to provide a fault diagnosis method and device to solve the problems in the prior art solutions, such as insufficient efficiency and inaccurate features in fault text feature extraction technology, resulting in low automation degree and insufficient accuracy of fault diagnosis.

[0004] To achieve the above object, an embodiment of the present invention provides a fault diagnosis method, including:

[0005] Construct a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network;

[0006] Determine first feature data for characterizing text semantics and second feature data for characterizing text fault themes according to work order sample data, the feature extraction network, and a historical fault database;

[0007] Determine a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network;

[0008] Calculate the similarity between the fault diagnosis model and the historical fault database to determine the corresponding work order processing information in the historical fault database.

[0009] Optionally, determining a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network includes:

[0010] Perform vector product interaction on the first feature data and the second feature data to determine the interaction weight coefficient matrix;

[0011] Determine the processed first weight coefficient according to the normalization function of the feature interaction network and the weight coefficient matrix;

[0012] Determine the third feature data by weighted summation of the first weight coefficient and the first feature data; the third feature data is used to represent the feature data that fuses text semantics and themes;

[0013] Determine a trained fault diagnosis model according to the third feature data and the objective loss function of the feature classification network.

[0014] Optionally, the objective loss function of the feature classification network is determined in the following manner:

[0015] Determine the first loss function of the second feature data passing through the feature classification network, the second loss function of the third feature data passing through the feature classification network, and the third loss function of the model parameter regularization loss;

[0016] Determine the objective loss function by weighted summation of the first loss function, the second loss function, and the third loss function.

[0017] Optionally, the first loss function is determined according to a preset first cross-entropy function and the number of training samples;

[0018] The second loss function is determined according to a preset second cross-entropy function and the number of training samples;

[0019] Wherein, both the first cross-entropy function and the second cross-entropy function include: the class label of the sample and the predicted value of the sample classification.

[0020] Optionally, performing vector product interaction on the first feature data and the second feature data to determine the interaction weight coefficient matrix includes:

[0021] Perform a vector product interaction on the first feature data and the second feature data to determine a semantic feature interaction matrix;

[0022] Perform a one-layer convolutional network processing on the semantic feature interaction matrix to determine the weight coefficient matrix.

[0023] Optionally, according to the work order sample data, the feature extraction network, and the historical fault database, determine the first feature data for characterizing the text semantics and the second feature data for characterizing the text fault theme, including:

[0024] Determine the first feature data of the work order sample data through the first preset algorithm of the feature extraction network; the first feature data includes the semantic length and the embedding vector dimension;

[0025] Determine the second feature data of the work order sample data through the second preset algorithm of the feature extraction network according to the work order sample data and the historical fault database; the second feature data includes the number of themes, the fault phenomenon theme, the fault cause theme, and the fault measure theme.

[0026] Optionally, determine the trained fault diagnosis model according to the third feature data and the target loss function of the feature classification network, including:

[0027] Determine the target fault classification loss value according to the third feature data and the target loss function;

[0028] If the target fault classification loss value is lower than the threshold, then optimize the feature extraction network, the feature interaction network, and the feature classification network according to the preset function until the target fault classification loss value is greater than or equal to the threshold, and determine the trained fault diagnosis model.

[0029] Optionally, the above method further includes:

[0030] Obtain N historical fault work orders, and perform preprocessing to determine the fault feature data; N is a positive integer;

[0031] Construct the historical fault database based on the fault feature data; the fault database includes the corresponding relationship between N historical fault work orders and N third feature data; the third feature data is stored in the historical fault database in vector form;

[0032] Wherein, the fault feature data includes one or more of: fault mode, fault cause, fault impact, fault detection method, design improvement measure, and use compensation measure;

[0033] The preprocessing includes one or more of noise information elimination, duplicate data deletion, and sensitive word filtering.

[0034] To achieve the above object, an embodiment of the present invention further provides a fault diagnosis device, including:

[0035] A construction module, configured to construct a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network;

[0036] A first determination module, configured to determine first feature data for characterizing text semantics and second feature data for characterizing text fault topics according to work order sample data, the feature extraction network, and a historical fault database;

[0037] A second determination module, configured to determine a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network;

[0038] A third determination module, configured to calculate a similarity between the fault diagnosis model and the historical fault database to determine corresponding work order processing information in the historical fault database.

[0039] To achieve the above object, an embodiment of the present invention further provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps in the fault diagnosis method described in any one of the above are implemented.

[0040] The beneficial effects of the above technical solution of the present invention are as follows:

[0041] The above technical solution extracts first feature data for characterizing text semantics and second feature data for characterizing text fault topics, combines the first feature data and the second feature data through a feature interaction network, then determines a trained fault diagnosis model through a feature classification network, calculates a similarity between the fault diagnosis model and the historical fault database to determine corresponding work order processing information in the historical fault database, improves the accuracy of automated fault diagnosis, enhances the interpretability of model feature learning, and the diagnostic adaptation ability for different fault topics. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 One of the flowcharts of the fault diagnosis method provided by an embodiment of the present invention;

[0043] Figure 2 The structural diagram of the fault diagnosis network model provided by an embodiment of the present invention;

[0044] Figure 3 The structural diagram of the feature interaction network provided by an embodiment of the present invention;

[0045] Figure 4The second flowchart of the fault diagnosis method provided by the embodiment of the present invention;

[0046] Figure 5 The structural diagram of the fault diagnosis device provided by the embodiment of the present invention. Detailed implementation manners

[0047] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0048] It should be understood that the "one embodiment" or "an embodiment" mentioned throughout the specification means that a specific feature, structure or characteristic related to the embodiment is included in at least one embodiment of the present invention. Therefore, the appearances of "in one embodiment" or "in an embodiment" throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0049] In various embodiments of the present invention, it should be understood that the order numbers of the following processes do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiment of the present invention.

[0050] In addition, the terms "system" and "network" are often used interchangeably in this article.

[0051] In the embodiments provided in the present application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0052] As Figure 1 shown, the embodiment of the present invention provides a fault diagnosis method, including:

[0053] Step 101, construct a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network and a feature classification network.

[0054] The fault diagnosis network model based on text mining feature extraction and interaction constructed in the present invention is called a Topic-context Interaction Model (TCIM). As Figure 2 shown, according to the specific functions of each part of the structure, the TCIM model can be divided into three parts: a feature extraction network (Gg), a feature interaction network (Gf) and a feature classification network (Gd). Here, Figure 2 the network structure of which can be used in any of the following steps, Figure 2The arrow in it represents the flow process of the original fault data during model training.

[0055] Step 102: Determine first feature data for characterizing text semantics and second feature data for characterizing text fault topics according to the work order sample data, the feature extraction network, and the historical fault database. Among them, the work order sample data is fault case text data, such as recorded data in text forms like maintenance logs and manuals.

[0056] In this embodiment, the first feature data can represent the global information of the text mined from the perspective of words or terms, and can also be represented as local semantic features determined by the attention mechanism; the second feature data is the feature data of the text fault topic, representing the key information in the text. If the topic word matches the candidate keyword, it means that the candidate content fully represents the text theme. The Latent Dirichlet Allocation (LDA) topic model is used to construct topic keywords. The basis for the topic feature score is whether the candidate keyword appears in the topic feature words. If it appears, the weight is doubled; otherwise, the weight remains unchanged.

[0057] Step 103: Determine a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network;

[0058] In this embodiment, the first feature data and the second feature data are interacted through the feature interaction network, and then a trained fault diagnosis model is further determined through the feature classification network.

[0059] Step 104: Calculate the similarity with the historical fault database according to the fault diagnosis model, and determine the corresponding work order processing information in the historical fault database.

[0060] In the embodiment of the present invention, after determining the fault diagnosis model, the work order data to be processed is input into the fault diagnosis model. The fault diagnosis model can classify the work order data to be processed, determine the classified data or determine the key information of the work order data to be processed. According to the classified data or the key information of the work order data to be processed, calculate the similarity with the historical fault database, so as to find similar work orders and return the processing opinions of the work orders.

[0061] It should be noted that the process of similarity is as follows: The first feature data and the second feature data are used by the feature interaction network to generate third feature data containing the first two feature information, and calculate the similarity with the third feature data set stored in the historical fault database to find similar fault cases to help solve the current problem. The fault diagnosis model is used to help complete the intelligent diagnosis of fault classification.

[0062] Optionally, the calculation method of similarity calculation adopts the determination method of cosine similarity, and judges the correlation degree of two texts through cosine measurement. If the cosine similarity is larger, it means that the included angle between the two variables is smaller, and the similarity degree of the two texts is higher. If the cosine similarity is smaller, it means that the included angle between the two variables is larger, and the similarity degree of the two texts is lower. Finally, after comparing with the historical fault database, the work order processing information with the highest similarity in the historical fault database is determined.

[0063] The solution of the present invention can complete the diagnosis and retrieval tasks of different cases, improve the accuracy of automatic fault diagnosis, enhance the interpretability of model feature learning and the diagnostic adaptability to different fault themes.

[0064] Optionally, step 102 described above includes:

[0065] Step 104: Determine the first feature data of the work order sample data through the first preset algorithm of the feature extraction network; the first feature data includes semantic length and embedding vector dimension;

[0066] Step 105: Determine the second feature data of the work order sample data according to the work order sample data and the historical fault database through the second preset algorithm of the feature extraction network; the second feature data includes the number of topics, fault phenomenon topics, fault cause topics, and fault measure topics.

[0067] It should be noted that for the number of topics, the prior distribution of the topics is a parameter determined in advance (how many topics are expected to divide the research field), and the prior determination can be determined manually or by machine algorithm. Here, the topics of the second feature data refer to the fault topics included in all historical fault databases. For example, in the automotive field, it includes: engine cylinder intake and exhaust faults, mechanical transmission faults, power battery fault types, etc. After the topic feature module of the feature extraction network learns, all historical data is divided into several categories corresponding to the number of topics, and there will be corresponding words for each topic (i.e., article-topic and topic-word probability distributions). By weighting and averaging the embedding vectors corresponding to the first few high-frequency words under each topic, the final second feature information can be obtained, which represents the main fault phenomena and causes under each topic.

[0068] In the embodiment of the present invention, the feature extraction network (Gg) is used to extract effective fault features from the original fault data text (work order sample data), and is divided into a semantic feature extraction module (such as Figure 2 shown) and a topic feature extraction module (such as Figure 2As shown in the figure). The semantic feature extraction module extracts the semantic information of the text based on the learned word vectors. Through normalization, the rectified linear unit (ReLU function), and the pooling layer, the required text semantic feature vector, that is, the first feature data, can be obtained. The topic feature extraction module performs topic mining on the records in the case database based on the latent Dirichlet allocation (LDA) topic model widely used in text mining, obtains the fault topic distribution that can be observed manually and the high-frequency words under each topic, and performs weighted summation on the high-frequency word vectors under each topic, that is, the topic feature vector under each fault topic is obtained, that is, the first feature data.

[0069] In an optional embodiment, the semantic features and topic category features are extracted through the feature extraction network (Gg), where the semantic features (the first feature data) are the semantic feature embedding vectors D = {d1, d2, …, d e} ∈ R s×e , s is the length of the document, and e is the set embedding vector dimension. The topic category features (the second feature data) are the topic feature vectors T = {t1, t2, …, t k} ∈ R k×e , k is the number of topics, and e is the embedding vector dimension. It can be understood that R k×e represents the vector dimension space, that is, there are k components, and each component is a vector of e dimensions.

[0070] Optionally, step 103 above includes:

[0071] Step 104: Perform vector product interaction on the first feature data and the second feature data to determine the weight coefficient matrix after interaction;

[0072] Step 105: Determine the processed first weight coefficient according to the normalization function of the feature interaction network and the weight coefficient matrix;

[0073] Step 106: Determine the third feature data by weighted summation of the first weight coefficient and the first feature data; the third feature data is used to characterize the feature data that fuses text semantics and topics;

[0074] Step 107: Determine the trained fault diagnosis model according to the third feature data and the target loss function of the feature classification network.

[0075] In this embodiment, in the feature interaction network (Gf), the extracted topic features and semantic features are subjected to vector multiplication to obtain the topic-semantic relationship, that is, the first feature data and the second feature data are subjected to vector multiplication interaction to determine the weight coefficient matrix after interaction; after the weight coefficient matrix passes through the normalization function (softmax activation function), it is the corresponding topic weight value of the text segment, and weighting the semantic features can obtain the third feature data of the text under the guidance of the topic category. The third feature data integrates the topic and semantic information of the case. For the third feature data, it is in the vector form convenient for the computer to store, and the similarity between the current case and the historical fault cases in the case library can be obtained through the calculation of the cosine vector, and the case retrieval is completed to provide the solution of the current case.

[0076] Among them, in step 107, the feature classification network (Gd) is composed of a multi-layer fully connected neural network. Its input is the topic category feature vector extracted by the topic feature extraction module and the third feature data generated by the interaction network. After passing through the target loss function of the feature classification network (Gd), the output is the probability values corresponding to all fault categories, which are used to predict the fault classification according to the case features and complete the intelligent fault diagnosis.

[0077] Optionally, the target loss function of the feature classification network is determined in the following manner:

[0078] Step 108, determine the first loss function of the second feature data passing through the feature classification network, the second loss function of the third feature data passing through the feature classification network, and the third loss function of the model parameter regularization loss;

[0079] Step 109, determine the target loss function according to the weighted sum of the first loss function, the second loss function, and the third loss function.

[0080] In this embodiment, the target loss function is mainly composed of the loss function of the classifier. The total loss function of the fault classifier, that is, the target loss function Lc, is obtained by the weighted sum of the prediction loss of the topic features, the prediction loss of the interaction features, and the model parameter regularization loss, and is defined as:

[0081] L c = L1 + PL2 + λR(w), Formula 1;

[0082] In Formula 1, L1 is the prediction loss obtained by the interaction features passing through the classification module, that is, the second loss function of the third feature data passing through the feature classification network; L2 is the prediction loss obtained by the topic features passing through the classification module, that is, the first loss function of the second feature data passing through the feature classification network; P and λ are the weight values corresponding to the losses of the two, which are between 0 and 1 (including 0 and 1), and R(W)=‖W‖ 2It is the quadratic regularization value of all the parameters to be learned in the model.

[0083] Specifically, the first loss function is determined according to a preset first cross-entropy function and the number of training samples;

[0084] The second loss function is determined according to a preset second cross-entropy function and the number of training samples;

[0085] Among them, both the first cross-entropy function and the second cross-entropy function include: the class label of the sample and the predicted value of the sample classification.

[0086] In this embodiment, the fault prediction loss values of the second loss function L1 and the first loss function L2 are measured by the cross-entropy between the prediction and the true label, and the prediction loss formulas of L1 and L2 are as follows:

[0087]

[0088]

[0089] Among them, n is the number of training samples; is the cross-entropy function, y1 is the fault type of the current fault text, y h = f(H) is the predicted value of the fault type obtained by the interaction feature H passing through the classifier, y t = f(T) represents the predicted value of the fault type obtained by the topic feature vector T passing through the classifier.

[0090] The cross-entropy function is defined as follows:

[0091]

[0092] Among them, y (i) is the class label of the i-th sample, that is, the true value of the sample; is the predicted value of the model for classifying the i-th sample.

[0093] To sum up, the target loss function Lc is equivalently expressed as follows:

[0094]

[0095] Optionally, as Figure 3 shown, step 104 above includes:

[0096] Step 110: Perform vector product interaction on the first feature data and the second feature data to determine a semantic feature interaction matrix.

[0097] In this embodiment, the first feature data and the second feature data after extracting the work order sample data are subjected to vector product interaction, that is, the first feature data and the second feature data are input into the feature interaction network, and the specific network structure of the feature interaction network is as Figure 3 shown. First, vector product is performed to complete the interaction calculation, and the category semantic feature interaction matrix G = DT T ={g1, g2, …, g s} ∈ R s×k is obtained. The semantic feature interaction matrix G enables the model to provide more feature information for the subsequent tasks and retains the interpretability of the model. By introducing prior knowledge of the topic, it helps to re-express the text.

[0098] Step 111: Process the semantic feature interaction matrix through a one-layer convolutional network to determine the weight coefficient matrix.

[0099] To obtain the association of the topic-word pairs, the semantic feature interaction matrix G is further processed through a one-layer convolutional network to capture the non-linear relationship between adjacent words.

[0100] For each G vector centered at d and ranging from d - n to d + n, convolution with a convolution kernel size of n is performed, and the ReLU activation function is used to obtain the convolved vector U = {u1, u2, …, u s} ∈ R s×k . For the d-th convolved vector u d , the formula five for this convolution process is as follows:

[0101] u d = Relu(w d G d-n:d+n + b d ), Formula Five;

[0102] where w d and b d are the parameters to be learned by the convolutional network model, and finally the maximum pooling (maxpooling) of u d is performed to obtain the final weight attention vector V = {v1, v2, …, v S} ∈ R s×1 . V is a vector of length s that stores the attention scores between each word in the current text and the topic. The Softmax function can be used to convert the scores into the topic-word weight coefficient vector β, that is: β = SoftMax(v).

[0103] Among them, the d-th topic-word weight coefficient is calculated as:

[0104]

[0105] After calculating the weight coefficient vector β, weighted corresponding to each word component of the original text semantic features can generate a new interaction feature H = {h1, h2, …, h S} ∈ R S×E , this feature is the newly generated feature with the interaction relationship between the theme-word features, that is, the third feature data, which contains the text information and category information of each case, and is calculated as:

[0106]

[0107] Furthermore, as Figure 2 shown, step 107 of the embodiment of the present invention includes:

[0108] Step 112: Determine the target fault classification loss value according to the third feature data and the target loss function;

[0109] Step 113: If the target fault classification loss value is lower than the threshold, then optimize the feature extraction network, the feature interaction network, and the feature classification network according to a preset function until the target fault classification loss value is greater than or equal to the threshold, and determine the trained fault diagnosis model.

[0110] In the embodiment of the present invention, the semantic feature extraction network, the feature interaction network, and the feature classification network can also be optimized in the determined trained fault diagnosis model. Fix the trained theme feature extraction module, extract semantic features using the original data, complete feature interaction and classification prediction, and obtain the target fault classification loss value after one round of training; if the target fault classification loss value is lower than the threshold, then optimize the feature extraction network, the feature interaction network, and the feature classification network according to a preset function, that is, perform backpropagation of the reverse gradient for the current loss, repeatedly optimize, and repeat step 112 until the target fault classification loss value is greater than or equal to the threshold, and the model loss converges and ends, and determine the trained fault diagnosis model.

[0111] It should be noted that the optimization objective of the model is to minimize the overall classification loss of the fault classifier. On the one hand, the model needs to minimize the L2 loss to constrain the weighted theme features extracted to be as close as possible to the classification category. On the other hand, after adding semantic features, the model needs to minimize the L1 loss to constrain the generated features to be able to represent the complete information of the case, so that the feature classification network can distinguish the types of faults and make the generated fine-grained features have the ability of fault diagnosis. During the training process, the model realizes the above optimization objective by calculating the target fault classification loss value in the feature classification network and backpropagating to iteratively optimize the parameters of the feature extraction model and the feature interaction model in turn.

[0112] It should also be noted that before the above step 103, model initialization is required first. Initialize the weight parameters of the feature extraction network (Gg), the feature interaction network (Gf), and the feature classification network (Gd); secondly, experiment and determine the parameters of the theme feature extraction module in the feature extraction network (Gg); extract theme features using the original data, and use the high-frequency word and theme visualization distribution method to experiment and determine appropriate theme model parameters, making full use of the experience of relevant technical personnel to make the theme feature distribution as close as possible to the real fault feature distribution.

[0113] Optionally, the above method further includes:

[0114] Step 114, obtain N historical fault work orders, and perform preprocessing to determine fault feature data; N is a positive integer;

[0115] Step 115, based on the fault feature data, construct the historical fault database; the fault database includes the correspondence between N historical fault work orders and N third feature data; the third feature data is stored in the historical fault database in vector form;

[0116] Among them, the fault feature data includes one or more of: fault mode, fault cause, fault impact, fault detection method, design improvement measures, and use compensation measures;

[0117] The preprocessing includes one or more of noise information elimination, duplicate data deletion, and sensitive word filtering.

[0118] It should be noted that what is stored in the constructed historical fault database is historical fault cases (historical fault work orders) and their corresponding third features (generated during training, including the first and second feature data), and the third feature data is stored in vector form. Therefore, compared with traditional text storage and similar text retrieval, this method is more convenient for storage and more convenient to complete the similarity calculation between cases.

[0119] In this embodiment, N historical fault work orders are obtained and preprocessed. Among them, the determined fault feature data includes manual diagnosis results and system diagnosis results. The system diagnosis results are obtained based on the pre-stored fault diagnosis system processing the fault feature data. The N historical fault work orders include M manual diagnosis results, where M is a positive integer less than or equal to N, and N is a positive integer. Among them, the fault feature data is obtained by processing the fault detail data of the historical fault work orders, and the fault detail data includes data within a preset time period associated with the occurrence time of the fault in the historical fault work order. The fault feature data includes product information (product name, product model, function, material, environmental load, performance parameters) and fault information (fault mode, fault cause, fault impact, fault detection method, design improvement measures, use compensation measures, etc.)

[0120] In an alternative embodiment: A historical fault database is established, including three database tables: product family / product platform, product detailed information, and function description, and three database tables: fault mode, detailed information, and fault mechanism. Products with the same internal interface and similar functions are organized in the form of product families and product trees, and then different personalized modules are added to the product platform to form product instances. All fault knowledge belongs to a certain product instance or platform. The database includes two database tables: a relational layer table and an application layer table. The relational layer table stores known object relationships, and the application layer table stores the data relationships between product functions and faults, constituting the historical fault database.

[0121] In another specific embodiment, as Figure 4 shown, the present invention also provides an overall flowchart, including:

[0122] Step 1: Extract historical fault case data (historical fault work orders) from the database and perform text preprocessing, including identifying professional words and removing stop words, to obtain a fault case corpus (historical fault database).

[0123] Step 2: Extract the text semantic features of each fault case and the overall fault theme category features of the case library in two stages according to the feature extraction network. Select appropriate network layer structure parameters to generate features through the feature interaction network, and train the model by minimizing the overall loss function of the classifier in the feature classification network.

[0124] Step 3: For a new fault case, directly input the preprocessed case test sample into the trained model to obtain the feature representation of the current case.

[0125] Step 4: The characterized features of this case can be classified for predictive diagnosis, and the similarity with the case features in the historical fault database is calculated to find similar cases, thus completing the diagnosis of the new fault case and the retrieval of the solution.

[0126] Step 5: After analyzing and inspecting this case, add it to the historical fault database to continuously update the database.

[0127] In summary, the solution of the present invention extracts semantic and thematic features in two stages and performs interaction to obtain fine-grained case features considering the fault theme-semantic relationship, completes the diagnosis and retrieval tasks of different cases, improves the accuracy of automatic fault diagnosis, and enhances the interpretability of model feature learning and the diagnostic adaptability to different fault themes.

[0128] As Figure 5 shown, an embodiment of the present invention further provides a fault diagnosis device, including:

[0129] A construction module 501, configured to construct a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network;

[0130] A first determination module 502, configured to determine first feature data for characterizing text semantics and second feature data for characterizing text fault themes according to work order sample data, the feature extraction network, and the historical fault database;

[0131] A second determination module 503, configured to determine a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network;

[0132] A third determination module 504, configured to calculate the similarity with the historical fault database according to the fault diagnosis model to determine the corresponding work order processing information in the historical fault database.

[0133] Optionally, the second determination module 503 includes:

[0134] A first determination sub-module, configured to perform vector product interaction on the first feature data and the second feature data to determine an interaction weight coefficient matrix;

[0135] A second determination sub-module, configured to determine a processed first weight coefficient according to the normalization function of the feature interaction network and the weight coefficient matrix;

[0136] A third determination sub-module, configured to determine third feature data by weighted summation of the first weight coefficient and the first feature data; the third feature data is used to characterize the feature data fusing text semantics and themes;

[0137] A fourth determination sub-module, configured to determine a trained fault diagnosis model according to the third feature data and an objective loss function of the feature classification network.

[0138] Optionally, in the building module 501, the objective loss function of the feature classification network is determined in the following manner:

[0139] A first determination unit, configured to determine a first loss function of the second feature data passing through the feature classification network, a second loss function of the third feature data passing through the feature classification network, and a third loss function of model parameter regularization loss;

[0140] A second determination unit, configured to determine the objective loss function according to a weighted sum of the first loss function, the second loss function, and the third loss function.

[0141] Specifically, the first determination unit is specifically configured to determine the first loss function according to a preset first cross-entropy function and the number of training samples;

[0142] The second determination unit is specifically configured to determine the second loss function according to a preset second cross-entropy function and the number of training samples;

[0143] Wherein, both the first cross-entropy function and the second cross-entropy function include: a class label of a sample and a predicted value of sample classification.

[0144] Optionally, the first determination sub-module includes:

[0145] A third determination unit, configured to perform vector product interaction on the first feature data and the second feature data to determine a semantic feature interaction matrix;

[0146] A fourth determination unit, configured to process the semantic feature interaction matrix through a one-layer convolutional network to determine the weight coefficient matrix.

[0147] Optionally, the first determination module 502 includes:

[0148] A fifth determination unit, configured to determine first feature data of work order sample data through a first preset algorithm of the feature extraction network; the first feature data includes a semantic length and an embedding vector dimension;

[0149] A sixth determination unit, configured to determine second feature data of work order sample data according to the work order sample data and a historical fault database through a second preset algorithm of the feature extraction network; the second feature data includes the number of topics, a fault phenomenon topic, a fault cause topic, and a fault measure topic.

[0150] Optionally, the fourth determination sub-module includes:

[0151] A seventh determination unit, configured to determine a target fault classification loss value according to the third feature data and the target loss function;

[0152] An eighth determination unit, configured to, when the target fault classification loss value is lower than a threshold, optimize the feature extraction network, the feature interaction network, and the feature classification network according to a preset function until the target fault classification loss value is greater than or equal to the threshold, and determine a trained fault diagnosis model.

[0153] In an embodiment of the present invention, the above-mentioned fault diagnosis device further includes:

[0154] An acquisition module, configured to acquire N historical fault work orders, perform preprocessing, and determine fault feature data; N is a positive integer;

[0155] A second construction module, configured to construct the historical fault database based on the fault feature data; the fault database includes the correspondence between N historical fault work orders and N third feature data; the third feature data is stored in the historical fault database in vector form;

[0156] Wherein, the fault feature data includes one or more of a fault mode, a fault cause, a fault impact, a fault detection method, a design improvement measure, and a use compensation measure;

[0157] The preprocessing includes one or more of noise information elimination, duplicate data deletion, and sensitive word filtering.

[0158] Wherein, the implementation embodiments of the above-mentioned fault diagnosis method are all applicable to the embodiments of this fault diagnosis device and can achieve the same technical effects. To avoid repetition, they will not be elaborated here.

[0159] A readable storage medium according to an embodiment of the present invention, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps in the above-mentioned fault diagnosis method are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be elaborated here.

[0160] Wherein, the processor is the processor in the above-mentioned fault diagnosis method. The readable storage medium includes computer-readable storage media, such as a computer read-only memory (Read-Only Memory, abbreviated as ROM), a random access memory (Random Access Memory, abbreviated as RAM), a magnetic disk, or an optical disc, etc.

[0161] In embodiments of the present invention, a module may be implemented in software for execution by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed, for example, as objects, procedures, or functions. Nevertheless, the executable code of the identified module need not be physically located together, but may include different instructions stored in different locations, which, when logically combined, constitute the module and achieve the specified purpose of the module.

[0162] In fact, an executable code module may be a single instruction or many instructions, and may even be distributed over multiple different code segments, distributed among different programs, and spanning multiple memory devices. Similarly, the operating data may be identified within the module and may be implemented in any suitable form and organized within any suitable type of data structure. The operating data may be collected as a single data set, or may be distributed at different locations (including on different storage devices), and may exist at least partially only as electronic signals in a system or network.

[0163] When a module can be implemented in software, considering the level of existing hardware technology, for a module that can be implemented in software, without considering cost, those skilled in the art can build corresponding hardware circuits to implement the corresponding functions. The hardware circuits include conventional very large scale integration (VLSI) circuits or gate arrays, as well as existing semiconductors such as logic chips, transistors, or other discrete components. A module may also be implemented using programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, etc.

[0164] The foregoing exemplary embodiments have been described with reference to the accompanying drawings. Many different forms and embodiments are possible without departing from the spirit and teachings of the invention. Therefore, the invention should not be construed as limited to the exemplary embodiments presented herein. Rather, these exemplary embodiments are provided so that the invention will be complete and will convey the scope of the invention to those skilled in the art. In the drawings, the dimensions and relative dimensions of components may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless the context clearly dictates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms as well. It will be further understood that the terms "comprises" and / or "comprising", when used in this specification, specify the presence of the stated features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. Unless otherwise indicated, when stating a value range, the range includes the upper and lower limits thereof and any sub-ranges therebetween.

[0165] The foregoing is a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A fault diagnosis method, characterized in that, Including: Constructing a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network; Determining first feature data for characterizing text semantics and second feature data for characterizing text fault topics according to work order sample data, the feature extraction network, and a historical fault database, including: determining the first feature data of the work order sample data through a first preset algorithm of the feature extraction network; the first feature data includes semantic length and embedding vector dimension; determining the second feature data of the work order sample data according to the work order sample data and the historical fault database through a second preset algorithm of the feature extraction network; the second feature data includes the number of topics, fault phenomenon topics, fault cause topics, and fault measure topics; Determining a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network, including: performing vector product interaction on the first feature data and the second feature data to determine an interaction weight coefficient matrix; determining a processed first weight coefficient according to the normalization function of the feature interaction network and the weight coefficient matrix; determining third feature data by weighted summing the first weight coefficient and the first feature data; the third feature data is used to characterize feature data integrating text semantics and topics; determining a trained fault diagnosis model according to the third feature data and the target loss function of the feature classification network; Calculating the similarity between the fault diagnosis model and the historical fault database to determine the corresponding work order processing information in the historical fault database.

2. The method according to claim 1, characterized in that The target loss function of the feature classification network is determined in the following manner: Determining a first loss function of the second feature data passing through the feature classification network, a second loss function of the third feature data passing through the feature classification network, and a third loss function of model parameter regularization loss; Determining the target loss function according to the weighted summation of the first loss function, the second loss function, and the third loss function.

3. The method according to claim 2, characterized in that, The first loss function is determined according to a preset first cross-entropy function and the number of training samples; The second loss function is determined according to a preset second cross-entropy function and the number of training samples; Wherein, both the first cross-entropy function and the second cross-entropy function include: the class label of the sample and the predicted value of sample classification.

4. The method according to claim 1, wherein Performing vector product interaction on the first feature data and the second feature data to determine an interaction weight coefficient matrix, including: Performing vector product interaction on the first feature data and the second feature data to determine a semantic feature interaction matrix; Performing one-layer convolutional network processing on the semantic feature interaction matrix to determine the weight coefficient matrix.

5. The method according to claim 1, wherein Determining a trained fault diagnosis model according to the third feature data and the target loss function of the feature classification network, including: Determining a target fault classification loss value according to the third feature data and the target loss function; When the target fault classification loss value is lower than the threshold, optimize the feature extraction network, the feature interaction network, and the feature classification network according to a preset function until the target fault classification loss value is greater than or equal to the threshold, and then determine the trained fault diagnosis model.

6. The method according to claim 1, characterized in that, The method further includes: Obtain N historical fault work orders, perform preprocessing on them, and determine fault feature data; N is a positive integer; Based on the fault feature data, construct the historical fault database; the fault database includes the correspondence between N historical fault work orders and N third feature data; the third feature data is stored in the historical fault database in vector form; Among them, the fault feature data includes one or more of: fault mode, fault cause, fault impact, fault detection method, design improvement measures, and use compensation measures; The preprocessing includes one or more of noise information elimination, duplicate data deletion, and sensitive word filtering.

7. A fault diagnosis device, characterized in that, It includes: A construction module for constructing a fault diagnosis network model; the fault diagnosis network model includes a feature extraction network, a feature interaction network, and a feature classification network; A first determination module for determining first feature data for characterizing text semantics and second feature data for characterizing text fault topics according to work order sample data, the feature extraction network, and the historical fault database; The first determination module includes: a fifth determination unit for determining the first feature data of the work order sample data through a first preset algorithm of the feature extraction network; the first feature data includes semantic length and embedding vector dimension; a sixth determination unit for determining the second feature data of the work order sample data through a second preset algorithm of the feature extraction network according to the work order sample data and the historical fault database; the second feature data includes the number of topics, fault phenomenon topics, fault cause topics, and fault measure topics; A second determination module for determining a trained fault diagnosis model according to the first feature data, the second feature data, the feature interaction network, and the feature classification network; the second determination module includes: a first determination sub-module for performing vector product interaction on the first feature data and the second feature data to determine an interaction weight coefficient matrix; a second determination sub-module for determining a processed first weight coefficient according to the normalization function of the feature interaction network and the weight coefficient matrix; a third determination sub-module for determining third feature data by weighted summing the first weight coefficient and the first feature data; the third feature data is used to characterize the feature data that fuses text semantics and topics; a fourth determination sub-module for determining a trained fault diagnosis model according to the third feature data and the target loss function of the feature classification network; A third determination module for calculating the similarity between the fault diagnosis model and the historical fault database to determine the corresponding work order processing information in the historical fault database.

8. A readable storage medium, on which a program or instructions are stored, characterized in that, When the program or instruction is executed by a processor, it implements the steps in the fault diagnosis method according to any one of claims 1-6.

Images