Primary equipment defect diagnosis method, system, electronic equipment and storage medium

By building a device defect database and using a convolutional neural network for diagnosis, the problem of incomplete equipment fault diagnosis information in the existing technology is solved, the accuracy and detail of fault diagnosis are improved, and the safety of equipment is enhanced.

CN113590396BActive Publication Date: 2024-11-01SHENZHEN COMTOP INFORMATION TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110835929.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-07-23
Publication Date
2024-11-01
Estimated Expiration
2041-07-23

AI Technical Summary

Technical Problem

Existing equipment fault diagnosis methods rely on on-site patrol to extract fault information, resulting in incomplete input information and low accuracy of fault diagnosis.

Method used

By obtaining device attribute data, historical defect data and historical operation data for preprocessing, a defect database is built, and a diagnostic model is trained using convolutional neural network algorithm to diagnose equipment defects.

Benefits of technology

It improves the accuracy of equipment fault diagnosis, provides detailed defect severity, diagnosis causes and management measures, and enhances the safe operation of the equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113590396B_ABST
    Figure CN113590396B_ABST
Patent Text Reader

Abstract

The present application discloses a method, system, electronic device and storage medium for diagnosing defects of a primary device. The method for diagnosing defects of a primary device of the present application comprises: obtaining device attribute data, historical defect data and historical operation data of the primary device to be diagnosed, preprocessing the acquired device attribute data, historical defect data and historical operation data to obtain multiple device defect data, constructing a defect database according to the multiple device defect data, constructing a defect diagnosis model based on the defect database, training the defect diagnosis model to obtain a trained defect diagnosis model, inputting multiple device defect data into the trained defect diagnosis model, and obtaining a defect diagnosis result of the primary device.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of equipment diagnosis, and in particular to a defect diagnosis method, system, electronic device and storage medium for a primary equipment. Background Art

[0002] During the operation of the equipment, primary equipment fault diagnosis is of great significance to ensure the safe operation of the equipment, because once an accident occurs in the primary equipment, it may cause economic losses and casualties. In recent years, due to the rapid development of sensor technology and computers, fault diagnosis has gradually received attention in industry and academia. At present, traditional feature extraction models and learning classification models are usually used to study primary equipment fault diagnosis. However, the above methods all use the extracted equipment fault information as the input of the machine learning model. Since the equipment fault information is usually extracted by the staff through on-site patrols, the probability of the staff discovering the primary equipment on site is low, resulting in the input equipment fault information is not comprehensive enough, and the result of the primary equipment fault diagnosis is not comprehensive enough, resulting in a low accuracy rate of primary equipment fault diagnosis. Summary of the invention

[0003] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a primary equipment defect diagnosis method, system, electronic device and storage medium, which can improve the accuracy of primary equipment fault diagnosis.

[0004] A defect diagnosis method based on a primary device according to an embodiment of the first aspect of the present application includes:

[0005] Obtain equipment defect data of the primary equipment to be diagnosed;

[0006] Inputting the equipment defect data into a trained defect diagnosis model to obtain a defect diagnosis result of the primary equipment;

[0007] The training of the defect diagnosis model includes:

[0008] Acquire multiple training attribute data of multiple training devices to be trained;

[0009] Preprocessing the plurality of training attribute data to obtain a plurality of training defect data;

[0010] The defect diagnosis model is obtained by training the preset diagnosis model according to the plurality of training defect data. The defect diagnosis method of a primary device according to the embodiment of the present application has at least the following beneficial effects:

[0011] Acquire device attribute data, historical defect data and historical operation data of a primary device to be diagnosed, pre-process the acquired device attribute data, historical defect data and historical operation data to obtain multiple device defect data, build a defect database based on the multiple device defect data, build a defect diagnosis model based on the defect database, train the defect diagnosis model to obtain a trained defect diagnosis model, input the multiple device defect data into the trained defect diagnosis model, and obtain a defect diagnosis result of the primary device.

[0012] According to some embodiments of the present application, the training attribute data includes equipment attribute data, historical defect data, and historical operation data;

[0013] Correspondingly, the preprocessing of the plurality of training attribute data to obtain a plurality of training defect data includes:

[0014] Processing the equipment attribute data, the historical defect data and the historical operation data to obtain a plurality of equipment defect data;

[0015] The plurality of equipment defect data are cleaned to remove abnormal data, thereby obtaining a plurality of cleaned equipment defect data.

[0016] According to some embodiments of the present application, the method further includes: constructing a defect database according to the plurality of device defect data;

[0017] The step of constructing a defect database according to the plurality of device defect data comprises:

[0018] Calculating text similarity of the plurality of equipment defect data using text distribution;

[0019] extracting a plurality of key defect data from the plurality of device defect data according to the text similarity;

[0020] Obtain multiple historical defect data;

[0021] Performing structured processing on the plurality of historical defect data according to a preset standard to obtain a plurality of processed historical defect data;

[0022] According to the plurality of historical defect data and the plurality of preset defect features, text annotation is performed on the plurality of key defect data to obtain a plurality of annotated key defect data;

[0023] A defect database is constructed according to the marked key defect data.

[0024] According to some embodiments of the present application, the method further includes: constructing a preset diagnosis model based on the defect database;

[0025] The step of constructing a preset diagnosis model based on the defect database includes:

[0026] Obtain a preset business diagnosis model;

[0027] Extracting preset diagnostic indicators from the business diagnostic model;

[0028] Get the preset device word segmentation rules;

[0029] Perform word segmentation processing on the defect indicator description based on the equipment word segmentation rule to obtain a standard diagnostic indicator;

[0030] Get the preset defect corpus model;

[0031] Inputting the standard diagnostic indicators into the defect corpus model to obtain trained standard diagnostic indicators;

[0032] Using the convolutional neural network algorithm, a convolutional neural network is constructed according to the trained standard diagnostic indicators to obtain a preset diagnostic model.

[0033] According to some embodiments of the present application, the step of training a preset diagnosis model according to the plurality of training defect data to obtain the defect diagnosis model includes:

[0034] Extracting historical domain samples and target domain samples from the preset diagnostic model;

[0035] Inputting the historical domain samples and the target domain samples into the preset diagnostic model respectively through forward propagation to obtain target features;

[0036] Optimizing the target feature by using a cross entropy loss function to obtain an optimized target feature;

[0037] The preset diagnosis model is trained according to the optimized target features to obtain a defect diagnosis model.

[0038] According to some embodiments of the present application, inputting the device defect data into a trained defect diagnosis model to obtain a defect diagnosis result of the target device includes:

[0039] Inputting the plurality of equipment defect data into the trained defect diagnosis model;

[0040] The equipment defect input is classified by using the classifier in the defect diagnosis model to obtain a defect diagnosis result of the primary equipment, wherein the defect diagnosis result includes defect severity, defect diagnosis cause and defect management measures.

[0041] According to some embodiments of the present application, the method further includes:

[0042] Classifying the equipment defect degree of the primary equipment according to the defect severity, the defect diagnosis cause and the defect management measure to obtain a classification result;

[0043] Obtaining an equipment risk level of the primary equipment according to the classification result;

[0044] According to the equipment risk level, the development trend of the primary equipment within a preset time range is predicted.

[0045] A defect diagnosis system based on a primary device according to a second aspect of the present application includes:

[0046] Acquisition module: the acquisition module is used to acquire equipment defect data of the primary equipment to be diagnosed;

[0047] Generation module: The generation module is used to input the equipment defect data into the trained defect diagnosis model to obtain the defect diagnosis result of the primary equipment.

[0048] An electronic device according to an embodiment of a third aspect of the present application includes:

[0049] at least one processor, and

[0050] a memory communicatively connected to the at least one processor; wherein,

[0051] The memory stores instructions, and the instructions are executed by the at least one processor so that when the at least one processor executes the instructions, a defect diagnosis method for a primary equipment as described in any one of the embodiments of the first aspect of the present application is implemented.

[0052] A computer-readable storage medium according to a fourth aspect of the present application includes:

[0053] The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the defect diagnosis method for a primary equipment as described in the embodiment of the first aspect of the present application.

[0054] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present application is further described below with reference to the accompanying drawings and embodiments, wherein:

[0056] Figure 1 A first flow chart of a method for diagnosing a defect of a primary device provided in some embodiments of the present application;

[0057] Figure 2A specific flow chart of step S200 in the method for diagnosing defects of a primary device provided in some embodiments of the present application;

[0058] Figure 3 A specific flow chart of step S300 in the defect diagnosis method for a primary device provided in some embodiments of the present application;

[0059] Figure 4 A specific flow chart of step S400 in the defect diagnosis method for a primary device provided in some embodiments of the present application;

[0060] Figure 5 A specific flow chart of step S500 in the defect diagnosis method for a primary device provided in some embodiments of the present application;

[0061] Figure 6 A specific flow chart of step S600 in the defect diagnosis method for a primary device provided in some embodiments of the present application;

[0062] Figure 7 A second flow chart of a method for diagnosing defects in a primary device provided in some embodiments of the present application. DETAILED DESCRIPTION

[0063] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0064] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0065] During the operation of the equipment, primary equipment fault diagnosis is of great significance to ensure the safe operation of the equipment, because once an accident occurs in the primary equipment, it may cause economic losses and casualties. In recent years, due to the rapid development of sensor technology and computers, fault diagnosis has gradually received attention in industry and academia. At present, traditional feature extraction models and learning classification models are usually used to study primary equipment fault diagnosis. However, the above methods all use the extracted equipment fault information as the input of the machine learning model. Since the equipment fault information is usually extracted by the staff through on-site patrols, the probability of the staff discovering the primary equipment on site is low, resulting in the input equipment fault information is not comprehensive enough, and the result of the primary equipment fault diagnosis is not comprehensive enough, resulting in a low accuracy rate of primary equipment fault diagnosis.

[0066] Based on this, the present application proposes a defect diagnosis method, system, electronic device and storage medium for a primary device, which can obtain the device attribute data, historical defect data and historical operation data of the primary device to be diagnosed, pre-process the acquired device attribute data, historical defect data and historical operation data to obtain multiple device defect data, build a defect database based on the multiple device defect data, build a defect diagnosis model based on the defect database, train the defect diagnosis model to obtain a trained defect diagnosis model, input multiple device defect data into the trained defect diagnosis model, and obtain the defect diagnosis result of the primary device.

[0067] In a first aspect, an embodiment of the present application provides a defect diagnosis method based on a primary device.

[0068] In some embodiments, the defect diagnosis method of the embodiment of the present application includes: obtaining equipment defect data of a primary device to be diagnosed, inputting the equipment defect data into a trained defect diagnosis model, and obtaining a defect diagnosis result of the primary device; wherein, the training of the defect diagnosis model includes: obtaining multiple training attribute data of multiple training devices to be trained; preprocessing the multiple training attribute data to obtain multiple training defect data; training a preset diagnosis model according to the multiple training defect data to obtain a defect diagnosis model.

[0069] The training attribute data mentioned in the embodiments of the present application include equipment attribute data, historical defect data and historical operation data.

[0070] Reference Figure 1 , Figure 1 The first flow chart of the defect diagnosis method of a primary device provided in some embodiments of the present application specifically includes the steps of:

[0071] S100, obtaining device attribute data, historical defect data and historical operation data of a primary device to be diagnosed;

[0072] S200, preprocessing the equipment attribute data, historical defect data and historical operation data to obtain a plurality of equipment defect data;

[0073] S300, constructing a defect database according to defect data of multiple devices;

[0074] S400, builds a defect diagnosis model based on the defect database;

[0075] S500, training the defect diagnosis model to obtain a trained defect diagnosis model;

[0076] S600, multiple equipment defect data are input into the trained defect diagnosis model to obtain a defect diagnosis result of the equipment.

[0077] In step S100, the device attribute data, historical defect data and historical operation data of the primary device to be diagnosed are obtained. Among them, electrical equipment that is directly used to produce and use electric energy and has a higher voltage than the control circuit, that is, the secondary device, is called a primary device. It should be noted that "primary" mainly refers to the "main circuit" in power production, and "secondary" is mainly used to control the "primary". The use of "primary" to modify the device is mainly to reflect that the device belongs to the main circuit or the voltage level of the device is higher than the voltage level of the secondary device. The device attribute data of the embodiment of the present application mainly includes the device type, device name, device code, device manufacturer and device model, etc. By obtaining different device attribute data, the present application can classify and manage different types of equipment according to different device attribute data, and can perform targeted defect diagnosis for different devices, thereby improving the accuracy of defect diagnosis. The historical defect data of the embodiment of the present application refers to the historical monitoring records of the primary equipment to be diagnosed, mainly including the equipment name, defect type, defect description, major category, manufacturer, year and month of manufacture, equipment model, commissioning date, defect cause category, defect cause, defect appearance, discovery time, defect location and treatment measures of the primary equipment. By obtaining the historical monitoring records of the primary equipment, the present application can refer to and correct the equipment defect diagnosis of the primary equipment, and focus on marking the defect locations and defect causes that have appeared before, which can improve the efficiency of some equipment defect detection. The historical operation data of the embodiment of the present application mainly includes the voltage, three-phase unbalanced current, voltage level, dielectric loss, equivalent capacitance, reference voltage alarm, three-phase unbalanced current alarm, dielectric loss alarm, full current alarm, equivalent capacitance alarm, communication status, operation status, equipment self-test abnormality, partial discharge and core current of the primary equipment, etc., which can reflect the operation status of the primary equipment, among which the equipment attribute data, historical defect data and historical operation data of the primary equipment are used as the data source of the defect database.

[0078] In step S200, the equipment attribute data, historical defect data and historical operation data are preprocessed to obtain multiple equipment defect data. The preprocessing mainly includes data fusion of the equipment attribute data, historical defect data and historical operation data, and processing the fused data to obtain standardized equipment defect data required by the user. The present application fuses the data to enhance the dimension of the indicator system of the defect status of a primary equipment, and can dynamically and real-timely perform online detection of a primary equipment.

[0079] In some embodiments, Figure 2 As shown, step S200 specifically includes the following steps:

[0080] S210, processing the equipment attribute data, the historical defect data and the historical operation data to obtain a plurality of equipment defect data;

[0081] S220, cleaning the plurality of device defect data, removing abnormal data, and obtaining the cleaned plurality of device defect data.

[0082] In step S210, the equipment attribute data, historical defect data and historical operation data are processed to obtain multiple equipment defect data. The processing here includes data fusion of the equipment attribute data, historical defect data and historical operation data. In the embodiment of the present application, three fusion methods can be adopted: front-end fusion (early-fusion) or data level fusion (data-level fusion), back-end fusion (late-fusion) or decision level fusion (decision-level fusion) and intermediate fusion (intermediate-fusion). Front-end fusion is to fuse multiple independent data sets into a single feature vector, which is then input into the machine learning classifier. Since the front-end fusion of multimodal data often cannot fully utilize the complementarity between multiple modal data, and the original data of front-end fusion usually contains a lot of redundant information, the multimodal front-end fusion method is often combined with feature extraction methods to eliminate redundant information, such as principal component analysis (PCA), maximum relevance minimum redundancy algorithm (mRMR), automatic decoders (Autoencoders), etc. Taking PCA as an example, n-dimensional features are mapped to k dimensions. This k dimension is a new orthogonal feature also called principal component. It is a k-dimensional feature reconstructed on the basis of the original n-dimensional feature. PCA A is to sequentially find a set of mutually orthogonal coordinate axes from the original space. The selection of new coordinate axes is closely related to the data itself. The first new coordinate axis is selected in the direction with the largest variance in the original data. The second new coordinate axis is selected in the plane orthogonal to the first coordinate axis with the largest variance. The third axis is the plane orthogonal to the first and second axes with the largest variance. By analogy, n coordinate axes can be obtained. Since most of the variance is contained in the first k coordinate axes, only the first k coordinate axes containing most of the variance are retained, thereby achieving dimensionality reduction of data features. PCA can be used to reduce the dimensionality of the original data fused at the front end and remove a large amount of redundant information contained in the original data.Back-end fusion is to fuse the output scores of classifiers trained separately for different modal data. Since the errors of the fusion model come from different classifiers, and the errors from different classifiers are often unrelated and do not affect each other, they will not cause further accumulation of errors. Back-end fusion methods include max-fusion, averaged-fusion, Bayes' rule based fusion, and ensemble learning. Among them, ensemble learning, as a typical representative of back-end fusion methods, is widely used in research fields such as communication, computer recognition, and speech recognition. Ensemble learning is a model that obtains the final result based on the results of multiple trainers. For example, the random forest model is a typical ensemble learning method. In the random forest, n trees are constructed and the final result is obtained based on the results of all trees. Intermediate fusion refers to converting different modal data into high-dimensional feature expressions first, and then fusing them in the intermediate layer of the model. Taking neural networks as an example, intermediate fusion first uses neural networks to convert original data into high-dimensional feature expressions, and then obtains the commonalities of different modal data in high-dimensional space. The intermediate fusion method flexibly selects the fusion location. This application can choose the appropriate fusion method according to actual needs, such as front-end fusion, back-end fusion and intermediate fusion, etc., to fuse equipment attribute data, historical defect data and historical operation data, and clean and deduplicate the fused data to obtain the standardized equipment defect data required by users.

[0083] In step S220, multiple device defect data are cleaned, abnormal data are removed, and multiple cleaned device defect data are obtained, wherein the device defect data are mainly cleaned and deduplicated for the duplication, missing and garbled of the device defect data. The duplication of the device defect data mainly refers to the same content of multiple device defect data, and the repeated device defect data needs to be removed. The missing of the device defect data mainly refers to the situation that there are missing fields in the device defect data, and the device defect data with missing fields needs to be located and the missing fields need to be repaired. The garbled of the device defect data mainly refers to the use of an inappropriate character set during the processing of the device defect data, which causes some or all characters to be unreadable. It is necessary to locate the device defect data with garbled fields and repair the garbled fields. In addition, there may be noise data in the device defect data, and the device defect data may also have problems such as full-width to half-width conversion and English uppercase and lowercase errors. It should be noted that full-width refers to a character occupying two standard character positions, and half-width refers to a character occupying one standard character position. In response to the above problems, this application needs to clean the data of multiple device defect data, and the cleaning process is as follows:

[0084] 1. Ensure the data integrity of the equipment defect data. Clean the equipment defect data according to the preset data cleaning rules. On the data obtained by the cleaning rules, perform an improved cleaning algorithm. For example, on the basis of the basic nearest neighbor sorting algorithm and the multi-pass nearest neighbor sorting algorithm, assign weights to each field and calculate the weighted similarity of similar duplicate records. It should be noted that the fields with higher weights will also have high similarity. The data integrity mentioned in this application mainly refers to the lack of integrity of data records and information. Due to the lack of data information and data records, both will lead to inaccurate statistical results in the absence of master records. Therefore, integrity is the basis for ensuring the quality of equipment defect data. Due to the repeated information contained in a large number of databases, a lot of time and crawling costs are consumed. Therefore, in the process of data screening, duplicate records should be removed first, and only one valid data should be retained. In the actual process of network data crawling, the data stored in the background can be sorted in advance, and then the newly generated equipment defect data can be deduplicated to ensure the data integrity of the equipment defect data. In actual applications, multiple deduplication methods can be considered, such as deduplication by primary key or deduplication by rules. Deduplication by primary key means using SQL or Excel to remove duplicate records of equipment defect data. Deduplication by rules means writing a series of rules to deduplicate data with complex duplications. If it is necessary to deduplicate equipment defect data from different channels, the equipment defect data can be matched by the same key information and then merged for deduplication.

[0085] 2. Solve the data consistency problem of equipment defect data, where data consistency mainly includes the standardization of data records and the consistency of data logic. The main criteria for identifying data record consistency are data encoding and formatting issues and the consistency of data constraints. To solve the data consistency problem, it is necessary to first establish a data system and focus on the measurement of the indicator system. The indicator system refers to an organic whole composed of several relatively independent and interrelated statistical indicators that reflect the overall quantitative characteristics of the data. In statistical research, if you want to explain the overall picture of the data, it is often not enough to use only one indicator, because it can only reflect the quantitative characteristics of a certain aspect of the whole. At this time, it is necessary to use multiple related indicators at the same time, and the unified whole composed of these multiple related and independent indicators is the indicator system. In the actual operation process, the database often has inconsistent data record content. Some of the equipment defect data can use their association with the external to manually solve the consistency problem. For example, data entry errors can generally be corrected by comparing with historically stored data records. Since the same attribute is not named in a standardized manner in different databases, inconsistencies often occur during data integration. Therefore, in the actual operation process, it is necessary to standardize the naming of each attribute value to reduce the problem of inconsistent equipment defect data.

[0086] 3. Solve the problem of noise data in equipment defect data. Noise refers to a random error and change in the measured variable. The equipment defect data can be smoothed and denoised through Bin method, cluster analysis method, human-machine combined inspection method and regression method. The Bin method is to smooth a set of sorted data by using the surrounding points of the data points to be smoothed. The sorted data is assigned to several buckets, that is, Bins. There are generally two methods for dividing Bins. One is the equal height method, that is, the number of elements in each Bin is equal, and the other is the equal width method, that is, the value interval of each Bin is the same. First, sort the price data, and then divide it into several Bins of equal height, that is, each Bin contains 3 values. Finally, you can use the mean of each Bin or the boundary of each Bin for smoothing. When smoothing with the mean, the equipment defect data in the first Bin is replaced by the mean of the Bin. When smoothing with the boundary, for a given Bin, its maximum and minimum values ​​constitute the boundary of the Bin. The boundary value of each Bin can replace all values ​​in the Bin, thereby obtaining the denoised equipment defect data. The cluster analysis method is to aggregate similar or adjacent equipment defect data together to form various cluster sets, mark the equipment defect data outside these cluster sets as abnormal data, and remove the marked abnormal data to obtain the denoised equipment defect data. The human-machine combined inspection method is to use information theory-based methods to help identify abnormal patterns in the handwritten symbol library. The identified abnormal patterns can be output to a list, and then the user checks each abnormal pattern in this list and finally confirms the useless pattern. According to the confirmed useless pattern, the abnormal equipment defect data is removed to obtain the denoised equipment defect data. The regression method is to use the linear regression method, including the multivariate regression method, to obtain the fitting relationship between multiple variables, so as to achieve the purpose of using multiple variable values ​​to predict the value of another variable. The fitting function obtained by the regression analysis method can help smooth the equipment defect data and remove the noise of the equipment defect data.

[0087] In step S300, a defect database is constructed based on a plurality of equipment defect data, and the defect database lays a foundation for a subsequent equipment defect diagnosis model.

[0088] In some embodiments, Figure 3 As shown, step S300 specifically includes the following steps:

[0089] S310, using text distribution to calculate text similarity of multiple equipment defect data;

[0090] S320, extracting a plurality of key defect data from the plurality of device defect data according to text similarity;

[0091] S330, obtaining a plurality of historical defect data;

[0092] S340, performing structured processing on the plurality of historical defect data according to a preset standard to obtain the processed plurality of historical defect data;

[0093] S350, performing text annotation on multiple key defect data according to multiple historical defect data and multiple preset defect features to obtain multiple annotated key defect data;

[0094] S360, constructing a defect database based on the annotated multiple key defect data.

[0095] In step S310, the text similarity of multiple equipment defect data is calculated using text distribution. Since the text is composed of words and punctuation, computers cannot efficiently process real text. In order to solve this problem, a formal method is needed to represent the real text. Usually, the text is converted into a vector for representation. This application uses hierarchical softmax and negative sampling to make the neural network language model with many parameters and huge calculation amount easy to calculate, including two models and two methods. It should be noted that these two models are applied to the neural network language model. The two models refer to the CBOW (continuous bag-of-words) model and the Skip-Gram model, where the CBOW model predicts the middle word through the context words, and the skip-gram model predicts the words that may appear before and after a specific word. The two methods refer to the hierarchical method and the negative sampling method. The hierarchical method is a method of accelerating the calculation of the probability distribution of words by constructing an effective tree structure, such as the Huffman tree, while the negative sampling method is a method of reducing the amount of calculation by randomly extracting negative samples and participating in each iteration together with the positive samples, turning it into a binary classification problem. The above two models and two methods can calculate the text similarity of multiple equipment defects, and train a language model represented by the word vector of each word in the equipment defect data. Each dimension of the word vector represents the semantic features of the word learned by the model, which is conducive to the establishment of the subsequent defect diagnosis model.

[0096] In step S320, multiple key defect data are extracted from multiple equipment defect data based on text similarity. Since text similarity reflects key information about equipment defects of a device to a certain extent, multiple key defect data are extracted from multiple equipment defect data based on text similarities, that is, key defect data with higher text similarity are selected, which can improve the efficiency of equipment defect detection.

[0097] In step S330, multiple historical defect data are obtained, that is, data of multiple primary devices detected in previous equipment inspection records are obtained, mainly including the equipment name, defect type, defect description, major professional category, manufacturer, year and month of manufacture, equipment model, commissioning date, defect cause category, defect cause, defect symptom, discovery time, defect location and treatment measures of the primary equipment.

[0098] In step S340, multiple historical defect data are structured according to preset standards to obtain multiple processed historical defect data. The structured processing includes text preprocessing of the multiple historical defect data, such as word segmentation processing, etc. The present application adopts NLP (Natural Language Processing) word segmentation algorithm to segment multiple historical defect data. The NLP word segmentation algorithm is mainly divided into two types according to its core idea. The first is dictionary-based word segmentation, which first divides the sentence into words according to the dictionary, and then finds the best combination of words; the second is character-based word segmentation, that is, forming words by characters, first dividing the sentence into individual characters, and then combining the characters into words to find the optimal segmentation strategy, which can also be converted into a sequence labeling problem. The shortest path word segmentation algorithm can be used for dictionary-based word segmentation. The shortest path word segmentation algorithm first matches all the words in a sentence to form a word graph, and then finds the shortest path from the starting point to the end point as the best combination. Assume that the weight of each word in the word graph is equal. When solving the shortest path problem of the DAG graph, it is necessary to use the property that "the shortest path between two points also includes the shortest paths between other vertices on the path". For example, S->A->B->E is the shortest path from S to E, and S->A->B is the shortest path from S to B. Otherwise, there will be C such that d(S->C->B)<d(S-> A->B), then the shortest path from S to E will also become S->C->B->E, which contradicts the assumption. Therefore, the above-mentioned optimal substructure property can be used to further optimize the word segmentation algorithm based on the greedy algorithm or dynamic programming two solving algorithms. The word segmentation based on characters can use the HMM hidden Markov model. The HMM model believes that there are two sequences when solving the sequence labeling problem. One is the observation sequence, that is, the sentence that people observe explicitly, and the sequence label is the hidden state sequence, that is, the observation sequence is X, the hidden state sequence is Y, and the causal relationship is Y->X. Therefore, to obtain the labeling result Y, the probability of X, the probability of Y, and P(X|Y) must be calculated, that is, the probability distribution model of P(X,Y) is established, and the word segmentation of multiple historical defect data is performed according to the probability distribution model.

[0099] In step S350, multiple key defect data are text-annotated based on multiple historical defect data and multiple preset defect features to obtain multiple annotated key defect data. The preset multiple defect features refer to judging which parts of a primary device may have defects based on the properties of the primary device, or judging possible defects of the primary device based on the operating status of the primary device. These defect problems are deeply analyzed, and multiple key defect data are text-annotated using multiple defect features, so as to improve the accuracy of primary equipment defect locating.

[0100] In step S360, a defect database is constructed based on the labeled multiple key defect data. Since the labeled multiple key defect data have been standardized, they can generate data formats that conform to database storage, for example, the labeled multiple key defect data are constructed into a defect database in a format where the variable name corresponds to the data type. In addition, before constructing the defect database, manual annotation can be performed, such as manually annotating the defect appearance, defect location, defect cause and treatment measures of the primary equipment based on the historical defect report of the primary equipment. Manual annotation is mainly based on the defect description, defect cause, treatment description and other text content in the defect record, combined with the experience of business experts to make judgments and screen out multiple key defect data that meet business needs.

[0101] In step S400, a defect diagnosis model is constructed based on the defect database, and the defect diagnosis model is used to perform intelligent diagnosis on defects of primary equipment.

[0102] In some embodiments, Figure 4 As shown, step S400 specifically includes the following steps:

[0103] S410, obtaining a preset business diagnosis model;

[0104] S420, extracting preset diagnostic indicators from the business diagnosis model;

[0105] S430, obtaining a preset device word segmentation rule;

[0106] S440, performing word segmentation processing on the defect indicator description based on the equipment word segmentation rule to obtain a standard diagnostic indicator;

[0107] S450, obtaining a preset defect corpus model;

[0108] S460, inputting the standard diagnostic indicator into the defect corpus model to obtain the trained standard diagnostic indicator;

[0109] S470, using a convolutional neural network algorithm, constructing a convolutional neural network according to the trained standard diagnostic indicators to obtain a defect diagnosis model.

[0110] In step S410, a preset business diagnosis model is obtained. The preset business diagnosis model is a defect diagnosis model pre-constructed based on possible equipment defects of a preset primary device. By combining the specific operating conditions of the primary device and the business diagnosis model, a defect diagnosis model that meets actual needs can be trained.

[0111] In step S420, preset diagnostic indicators are extracted from the business diagnosis model. The preset diagnostic indicators mainly include the equipment type, defect type, defect location, defect component, etc. of the primary equipment. Since the obtained preset diagnostic indicators are not necessarily standardized, the preset diagnostic indicators need to be further processed.

[0112] In step S430, the preset equipment word segmentation rules are obtained. Since Chinese text is different from English text and there is no natural space boundary between words, the Chinese text needs to be segmented before text representation. The user can set appropriate word segmentation based on actual defect detection experience, and collect multiple word segmentations to build a word segmentation library, set appropriate equipment word segmentation rules, and provide a word segmentation basis for subsequent word segmentation. Since there are differences in word segmentation in different fields, pre-building a word segmentation library can avoid the problem of deviation in word segmentation results due to different fields.

[0113] In step S440, the defect indicator description is segmented based on the equipment segmentation rule to obtain standard diagnostic indicators. After the segmentation process, keywords of multiple defect indicators can be obtained, which can be used as input of the defect corpus model.

[0114] In step S450, a preset defect corpus model is obtained. The defect corpus model is constructed based on a large number of pre-processed primary equipment defect records, laying a foundation for subsequent training of standard diagnostic indicators.

[0115] In step S460, the standard diagnostic indicators are input into the defect corpus model to obtain trained standard diagnostic indicators. Since the defect corpus model already contains multiple primary equipment defect records, the standard diagnostic indicators are input into the defect corpus, and then the word vectors and dimensions of the words of the standard diagnostic indicators are trained to obtain trained standard diagnostic indicators.

[0116] In step S470, a convolutional neural network algorithm is used to construct a convolutional neural network according to the trained standard diagnostic indicators to obtain a defect diagnosis model. In the embodiment of the present application, a four-layer convolutional neural network is constructed, including an input layer, a convolutional layer, a pooling layer and an output layer. It should be noted that in the embodiment of the present application, the input layer, the convolutional layer, the pooling layer and the fully connected layer are all one-dimensional CNNs (Convolutional Neural Networks). The defect diagnosis model is obtained according to the trained standard diagnostic indicators and the constructed convolutional neural network. In addition, the convolution layer of the present application includes a set of trainable filters, which are characterized by weight sharing (Weig hts sharing), that is, the same convolution kernel will traverse the input once with a fixed step size, and weight sharing reduces the network parameters of the convolution layer, avoiding overfitting due to too many parameters, and reducing the memory required by the system, reducing the load on the computer; the pooling layer performs a downsampling operation, the main purpose of which is to reduce the parameters of the neural network while retaining the more important features, preventing overfitting, and improving the generalization ability of the model; the fully connected layer classifies the features extracted previously, and plays the role of a "classifier" in the entire neural network. The specific approach is to first spread the output of the last pooling layer into a one-dimensional feature vector as the input of the fully connected layer.

[0117] In step S500, the defect diagnosis model is trained to obtain a trained defect diagnosis model.

[0118] In some embodiments, Figure 5 As shown, step S500 specifically includes the following steps:

[0119] S510, extracting historical domain samples and target domain samples from the defect diagnosis model;

[0120] S520, inputting the historical domain samples and the target domain samples into the defect diagnosis model through forward propagation to obtain target features;

[0121] S530, optimizing the target feature by using a cross entropy loss function to obtain an optimized target feature;

[0122] S540, training the defect diagnosis model according to the optimized target features to obtain a trained defect diagnosis model.

[0123] In step S510, historical domain samples and target domain samples are extracted from the defect diagnosis model, the historical domain samples include multiple health states of a device detected once in history, and the target domain samples include multiple health states of a device detected once in real time.

[0124] In step S520 to step S540, the historical domain samples and the target domain samples are respectively input into the defect diagnosis model through forward propagation to obtain the target feature, and the target feature is optimized by the cross entropy loss function to obtain the optimized target feature. The defect diagnosis model is trained according to the optimized target feature to obtain a trained defect diagnosis model. The target feature is obtained by the classifier pre-set in the defect diagnosis model, and the classification error of the target feature can be optimized by the cross entropy loss function. The specific cross entropy loss function formula is as follows:

[0125]

[0126] Where m is the batch size of historical domain samples; j is the fault category; I[·] is the indicator function, and its value rule is: I[value is true] = 1, I[value is false] = 0. The optimized target features are then input into the defect model for training to obtain a trained defect diagnosis model.

[0127] In step S600, multiple device defect data are input into the trained defect diagnosis model to obtain a defect diagnosis result of the device. That is, the processed defect indicator data is used as the input layer of the convolutional neural network, and the vectorized defect text is classified through the classifier of the convolutional neural network. The corresponding classification results are output to form a final defect diagnosis model. The defect diagnosis model is then trained so that the loss function of the training set shows a downward trend and does not overfit.

[0128] In some embodiments, Figure 6 As shown, step S600 specifically includes the following steps:

[0129] S610, inputting a plurality of equipment defect data into a trained defect diagnosis model;

[0130] S620, using the classifier in the defect diagnosis model to classify the equipment defect input, and obtain a defect diagnosis result of the equipment.

[0131] In step S610, a plurality of equipment defect data are input into the trained defect diagnosis model.

[0132] In step S620, the classifier in the defect diagnosis model is used to classify the equipment defect input to obtain a defect diagnosis result of the equipment, which includes the defect severity, defect diagnosis cause and defect management measures.

[0133] In some embodiments, Figure 7 As shown, the defect diagnosis method of a primary device mentioned in the present application specifically includes the steps of:

[0134] S700, classify the equipment defect degree of the primary equipment according to the defect severity, defect diagnosis cause and defect management measures, and obtain the classification result;

[0135] S800, obtain the equipment risk level of the primary equipment according to the classification result;

[0136] S900, predicts the development trend of a device within a preset time range based on the device risk level.

[0137] In step S700, the degree of equipment defects of the primary equipment is classified according to the severity of the defects, the cause of the defect diagnosis and the defect management measures to obtain a classification result. The classification can adopt the method of evaluating the weight of the index. For example, there are multiple weight indexes corresponding to different parameters. The user can perform weighted scoring based on the specific defect situation of the primary equipment and multiple weight indexes. The degree of equipment defects of the primary equipment is classified according to the weighted scoring to obtain a classification result. In actual applications, the equipment can be divided according to the degree of loss of the primary equipment, etc. to obtain a classification result, which will not be repeated here.

[0138] In step S800, the equipment risk level of the primary equipment is obtained according to the classification result. In practical applications, the entropy method can be combined to evaluate the risk of the primary equipment, and the risk of the primary equipment can be divided into high, medium and low, providing reference value for equipment maintenance.

[0139] In step S900, the development trend of the primary equipment within a preset time range is predicted according to the equipment risk level, so that defects of the primary equipment can be predicted in advance and processed in time to increase the service life of the primary equipment.

[0140] In an embodiment of the present application, device attribute data, historical defect data, and historical operation data of a primary device to be diagnosed are obtained, the acquired device attribute data, historical defect data, and historical operation data are preprocessed to obtain a plurality of device defect data, a defect database is constructed based on the plurality of device defect data, a defect diagnosis model is constructed based on the defect database, the defect diagnosis model is trained to obtain a trained defect diagnosis model, the plurality of device defect data are input into the trained defect diagnosis model, and a defect diagnosis result of the primary device is obtained.

[0141] On the second aspect, an embodiment of the present application also provides a defect diagnosis system based on a primary device, including an acquisition module and a generation module, wherein the acquisition module is used to acquire the device defect data of the primary device to be diagnosed; the generation module is used to input the device defect data into a trained defect diagnosis model to obtain the defect diagnosis result of the primary device.

[0142] In a third aspect, an embodiment of the present application also provides an electronic device.

[0143] In some embodiments, the electronic device includes: at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions, and the instructions are executed by the at least one processor so that when the at least one processor executes the instructions, any one of the defect diagnosis methods for a primary device in the embodiments of the present application is implemented.

[0144] The processor and the memory may be connected via a bus or other means.

[0145] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs, such as the primary device defect diagnosis method described in the embodiment of the present application. The processor implements the primary device defect diagnosis method described above by running the non-transitory software program and instructions stored in the memory.

[0146] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store a defect diagnosis method for executing the above-mentioned primary device. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0147] The non-transient software programs and instructions required to implement the above-mentioned primary equipment defect diagnosis method are stored in the memory, and when executed by one or more processors, the primary equipment defect diagnosis method mentioned in the above-mentioned first aspect embodiment is executed.

[0148] In a fourth aspect, embodiments of the present application also provide a computer-readable storage medium.

[0149] In some embodiments, a computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the defect diagnosis method for a primary equipment mentioned in the embodiment of the first aspect.

[0150] In some embodiments, the storage medium stores computer executable instructions, which are executed by one or more control processors, for example, by a processor in the above-mentioned electronic device, so that the above-mentioned one or more processors can execute the defect diagnosis method of the above-mentioned primary device.

[0151] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0152] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0153] The embodiments of the present application are described in detail above in conjunction with the accompanying drawings, but the present application is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in the relevant technical field without departing from the purpose of the present application. In addition, the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

Claims

1. A defect diagnosis method based on primary equipment, characterized in that: include: Obtain equipment defect data of the primary equipment to be diagnosed; Inputting the equipment defect data into a trained defect diagnosis model; classifying the equipment defect data using a classifier in the defect diagnosis model to obtain a defect diagnosis result of the primary equipment; The defect diagnosis results include the severity of the defect, the cause of the defect diagnosis, and the defect management measures; Performing weighted scoring based on the severity of the defect, the cause of the defect diagnosis, the defect management measures, and a plurality of preset weight indicators, classifying the degree of the equipment defect of the primary equipment to obtain a classification result; Obtaining an equipment risk level of the primary equipment according to the classification result; Predicting the development trend of the primary equipment within a preset time range based on the equipment risk level; The training of the defect diagnosis model includes: Acquire a plurality of training attribute data of a plurality of training devices to be trained; the training attribute data includes device attribute data, historical defect data, and historical operation data; Processing the device attribute data, the historical defect data, and the historical operation data to obtain a plurality of training defect data; Cleaning the plurality of training defect data to remove abnormal data, thereby obtaining a plurality of cleaned training defect data; Building a defect database based on the plurality of training defect data; Obtain the preset business diagnosis model; Extracting preset diagnostic indicators from the business diagnostic model; Get the preset device word segmentation rules; Perform word segmentation processing on the defect indicator description based on the equipment word segmentation rules to obtain standard diagnostic indicators; Get the preset defect corpus model; Inputting the standard diagnostic indicator into the defect corpus model, training the word vectors and dimensions of the words of the standard diagnostic indicator to obtain a trained standard diagnostic indicator; Using a convolutional neural network algorithm, a convolutional neural network is constructed based on the trained standard diagnostic indicators to obtain a preset diagnostic model; the convolutional neural network includes an input layer, a convolution layer, a pooling layer, and an output layer, the convolution layer includes a set of trainable filters, and the pooling layer is used to perform a downsampling operation; The preset diagnosis model is trained according to the plurality of training defect data to obtain the defect diagnosis model.

2. The defect diagnosis method based on primary equipment according to claim 1, characterized in that: The step of constructing a defect database according to the plurality of training defect data comprises: Calculating text similarity of the plurality of training defect data using text distributed computing; extracting a plurality of key defect data from the plurality of training defect data according to the text similarity; Obtain multiple historical defect data; Performing structured processing on the plurality of historical defect data according to a preset standard to obtain a plurality of processed historical defect data; Performing text annotation on the multiple key defect data according to the multiple historical defect data and the multiple preset defect features to obtain multiple annotated key defect data; A defect database is constructed based on the marked key defect data.

3. The defect diagnosis method based on primary equipment according to claim 2, characterized in that: The step of training the preset diagnosis model according to the plurality of training defect data to obtain the defect diagnosis model includes: extracting historical domain samples and target domain samples from the preset diagnostic model; Inputting the historical domain samples and the target domain samples into the preset diagnostic model through forward propagation respectively to obtain target features; Optimizing the target feature by using a cross entropy loss function to obtain an optimized target feature; The preset diagnosis model is trained according to the optimized target feature to obtain a defect diagnosis model; the target feature is obtained by a preset classifier in the defect diagnosis model, and the classification error of the target feature is optimized by a cross entropy loss function.

4. A defect diagnosis system based on primary equipment, characterized by: include: Acquisition module: The acquisition module is used to acquire equipment defect data of the primary equipment to be diagnosed; Generation module: The generation module is used to input the equipment defect data into a trained defect diagnosis model; use the classifier in the defect diagnosis model to classify the equipment defect data to obtain the defect diagnosis result of the primary equipment; the defect diagnosis result includes the defect severity, defect diagnosis cause, and defect management measures; After obtaining the defect diagnosis result of the primary equipment, the defect diagnosis system is further configured to: perform weighted scoring based on the defect severity, the defect diagnosis cause, the defect management measures, and a plurality of preset weight indicators, classify the equipment defect degree of the primary equipment, and obtain a classification result; obtain an equipment risk level of the primary equipment based on the classification result; and predict a development trend of the primary equipment within a preset time range based on the equipment risk level; The training of the defect diagnosis model includes: Acquire a plurality of training attribute data of a plurality of training devices to be trained; the training attribute data includes device attribute data, historical defect data, and historical operation data; Processing the device attribute data, the historical defect data, and the historical operation data to obtain a plurality of training defect data; Cleaning the plurality of training defect data to remove abnormal data, thereby obtaining a plurality of cleaned training defect data; Building a defect database based on the plurality of training defect data; Obtain a preset business diagnosis model; extract preset diagnostic indicators from the business diagnosis model; obtain preset equipment word segmentation rules; perform word segmentation processing on the defect indicator description based on the equipment word segmentation rules to obtain standard diagnostic indicators; obtain a preset defect corpus model; input the standard diagnostic indicators into the defect corpus model, train the word vectors and dimensions of the words of the standard diagnostic indicators to obtain trained standard diagnostic indicators; use a convolutional neural network algorithm to construct a convolutional neural network based on the trained standard diagnostic indicators to obtain a preset diagnostic model; the convolutional neural network includes an input layer, a convolution layer, a pooling layer and an output layer, the convolution layer includes a set of trainable filters, and the pooling layer is used to perform a downsampling operation; The preset diagnosis model is trained according to the plurality of training defect data to obtain the defect diagnosis model.

5. An electronic device, characterized in that include: at least one processor, and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions, and the instructions are executed by the at least one processor, so that the at least one processor implements the primary equipment-based defect diagnosis method according to any one of claims 1 to 3 when executing the instructions.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the primary equipment-based defect diagnosis method according to any one of claims 1 to 3.