Fault diagnosis method based on neural network analysis
Through the fault diagnosis method based on neural network analysis, convolutional neural network is used to extract the characteristics of complex equipment systems and build a decision tree model, which solves the shortcomings of traditional fault diagnosis methods in complex equipment systems and achieves more efficient and accurate faults diagnosis.
Patent Information
- Application Number
- CN202411567141.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-11-01
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-25
AI Technical Summary
Traditional fault diagnosis methods cannot fully adapt to the fault diagnosis tasks of complex equipment systems, and have problems such as low diagnostic accuracy, long diagnosis time, and high requirements for system data integrity.
Using a fault diagnosis method based on neural network analysis, feature extraction is performed through convolutional neural networks, information gain is calculated as the splitting standard, and a decision tree model is built to be used for fault diagnosis of complex equipment systems.
It improves the accuracy and efficiency of fault diagnosis, shortens diagnosis time, reduces the requirements for data integrity, and improves the reliability, repairability and safety of complex equipment systems.
Smart Images
Figure CN119513740A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method based on neural network analysis. Background Art
[0002] With the development of modern industry, the safety, reliability and maintainability of complex equipment systems such as aviation, aerospace, and ships have gradually become the focus of attention. In addition, these systems have complex structures, and once a failure occurs, it will cause huge economic losses and even casualties.
[0003] However, traditional fault diagnosis methods, such as experience-based methods and model-based methods, cannot fully meet the requirements of fault diagnosis tasks for complex equipment systems. These methods often have problems such as low diagnostic accuracy, long diagnostic time, and high requirements for system data integrity.
[0004] Convolutional neural network (CNN) is a type of feedforward neural network with deep structure and convolutional calculation. Although convolutional neural network has achieved great success in the field of image processing, its application in fault diagnosis is still in its infancy. Existing fault diagnosis methods based on convolutional neural network usually have problems such as long network training time, high requirements on system data integrity, and insufficient diagnostic accuracy.
[0005] Therefore, how to improve the accuracy and efficiency of fault diagnosis and improve the reliability, repairability and safety of complex equipment systems has become a problem that needs to be solved urgently. Summary of the invention
[0006] In order to solve the above problems, the present invention proposes a fault diagnosis method based on neural network analysis, which can perform feature extraction through convolutional neural network, improve the accuracy of fault diagnosis, shorten the diagnosis time, reduce the requirements for data integrity, and thus improve the reliability, repairability and safety of complex equipment systems.
[0007] In order to achieve the above object, the present invention is implemented by the following technical solutions:
[0008] The present invention is a fault diagnosis method based on neural network analysis, comprising the following operations:
[0009] Collect historical system data in complex equipment systems;
[0010] Performing preprocessing operations on the historical system data to generate corresponding target system data;
[0011] Extracting target system features corresponding to the target system data based on a convolutional neural network;
[0012] Calculating the information gain corresponding to the target system feature as a splitting criterion, and constructing a decision tree model for fault diagnosis according to the splitting criterion;
[0013] The current system data in the complex equipment system is obtained and input into a decision tree model for fault diagnosis, and the corresponding fault diagnosis results are output and displayed to the user end.
[0014] A further improvement of the present invention is that the historical system data includes historical sensor data, operation logs and maintenance records in the complex equipment system.
[0015] A further improvement of the present invention is that the preprocessing operation on the historical system data to generate corresponding target system data specifically includes:
[0016] Removing abnormal values and duplicate values from the historical system data to generate corresponding first system data;
[0017] Using interpolation to fill missing values in the first system data to generate corresponding second system data;
[0018] The second system data is standardized by using discrete standardization to obtain the target system data.
[0019] A further improvement of the present invention is that: the extraction of target system features corresponding to the target system data based on a convolutional neural network specifically includes:
[0020] Determine the size, number, and stride of the convolution kernels corresponding to the convolutional layers in the convolutional neural network;
[0021] A convolution operation is performed on the target system data based on the convolution kernel to extract a corresponding first output feature, where the calculation formula of the first output feature is:
[0022]
[0023] In the formula, S1(i,j) represents the value of the first output feature at position (i,j); I(i+m,j+n) represents the value of the target system data at position (i+m,j+n); K(m,n) represents the weight of the convolution kernel at position (m,n); i and j represent the coordinates on the first output feature; m and n represent the coordinates on the convolution kernel; Σ represents the summation symbol;
[0024] Perform a maximum pooling operation on the first output feature, and select the maximum value in the corresponding pooling window as the second output feature. The calculation formula of the second output feature is:
[0025]
[0026] In the formula, S2(i,j) represents the value of the second output feature at position (i,j); max represents the maximum value operation; G represents the pooling window; S1(i,j) represents the value of the first output feature at position (i,j); i and j represent the coordinates on the first output feature; m and n represent the coordinates on the convolution kernel.
[0027] A further improvement of the present invention is that: the first output feature and the second output feature are used as fusion input features, the weights corresponding to the fusion input features are determined and weighted summed to generate corresponding fusion output features, and the calculation formula of the fusion output features is:
[0028]
[0029] In the formula, F d represents the fusion output feature; w k represents the weight corresponding to the kth fusion input feature; F k represents the kth fusion input feature; Σ represents the summation symbol;
[0030] Based on the convolutional neural network, the fused output feature is the target system feature corresponding to the target system data.
[0031] A further improvement of the present invention is that the information gain corresponding to the target system feature is calculated as a splitting criterion, and a decision tree model for fault diagnosis is constructed according to the splitting criterion, specifically including:
[0032] The calculation formula of the information gain corresponding to the target system feature is:
[0033]
[0034] Where D represents the data set consisting of all fault diagnosis results; A represents the target system feature; IG(D,A) represents the information gain of the target system feature A to the data set D consisting of all fault diagnosis results; E(D) represents the entropy of the data set D consisting of all fault diagnosis results; V(A) represents all possible values of the target system feature A; D v' represents the subset of the target system feature A whose value is v'; N(D v' ) represents the subset D of the target system feature A whose value is v' v' N(D) represents the number of samples in the dataset D consisting of all fault diagnosis results; E(D v' ) represents the subset D of the target system feature A whose value is v' v' The entropy of ; Σ represents the summation symbol;
[0035] The calculation formula of the entropy E(D) of the data set D composed of all fault diagnosis results is:
[0036]
[0037] Where E(D) represents the entropy of the data set D composed of all fault diagnosis results; Q represents the number of fault diagnosis categories in the data set D composed of all fault diagnosis results; p q represents the probability of the qth fault diagnosis category in the data set D composed of all fault diagnosis results; log2() represents the logarithmic function with base 2; Σ represents the summation symbol;
[0038] The target system feature A takes the value of v' as a subset D v' The entropy E(D v' ) is calculated as:
[0039]
[0040] In the formula, E(D v' ) represents the subset D of the target system feature A whose value is v' v' Entropy; Q' represents the subset D of the target system feature A whose value is v' v' The number of fault diagnosis categories in p q' The subset D that represents the target system feature A with the value v' v' The probability of the q'th fault diagnosis category in ; log2() represents the logarithmic function with base 2; Σ represents the summation symbol;
[0041] The information gain corresponding to the target system feature is used as a splitting criterion, and a decision tree model for fault diagnosis is constructed according to the splitting criterion.
[0042] A further improvement of the present invention is that the current system data in the complex equipment system is obtained and input into a decision tree model for fault diagnosis, and the corresponding fault diagnosis result is output and displayed to the user end, specifically including:
[0043] Acquire current system data in the complex equipment system, repeat the above preprocessing and feature extraction steps, generate corresponding current system features and input them into a decision tree model for fault diagnosis, and output the corresponding fault diagnosis result, wherein the fault diagnosis result includes the fault type and fault degree;
[0044] The output fault diagnosis result is displayed to the user terminal in a visual manner, wherein the visual manner includes a chart, a curve and an image;
[0045] Collect feedback results of the user terminal on the fault diagnosis result, and optimize the decision tree model used for fault diagnosis according to the feedback results.
[0046] The beneficial effects of the present invention are as follows: the present invention proposes a fault diagnosis method based on neural network analysis, by collecting historical system data in a complex equipment system; preprocessing the historical system data to generate corresponding target system data; extracting target system features corresponding to the target system data based on a convolutional neural network; calculating the information gain corresponding to the target system features as a splitting criterion, and constructing a decision tree model for fault diagnosis according to the splitting criterion; obtaining the current system data in the complex equipment system and inputting it into the decision tree model for fault diagnosis, outputting the corresponding fault diagnosis results and displaying them to the user end. The present invention uses a convolutional neural network to perform feature extraction, which can extract more complex and accurate feature information, thereby improving the accuracy of fault diagnosis; and can quickly perform fault diagnosis on complex equipment systems, shortening the diagnosis time; at the same time, the method of the present invention can process incomplete data or data with special values, reducing the requirements for data integrity; and by timely discovering and diagnosing faults, corresponding maintenance measures can be taken, thereby improving the reliability, repairability and safety of complex equipment systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 A schematic diagram of a flow chart of a fault diagnosis method based on neural network analysis provided by an embodiment of the present invention;
[0048] Figure 2 A schematic diagram of target system feature extraction of a fault diagnosis method based on neural network analysis provided by an embodiment of the present invention;
[0049] Figure 3 A schematic diagram showing the fault diagnosis results of a fault diagnosis method based on neural network analysis provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0051] See also Figure 1 , a fault diagnosis method based on neural network analysis in this embodiment includes the following steps:
[0052] S1, collects historical system data in complex equipment systems;
[0053] S2, performing a preprocessing operation on the historical system data to generate corresponding target system data;
[0054] S3, extracting target system features corresponding to the target system data based on a convolutional neural network;
[0055] S4, calculating the information gain corresponding to the target system feature as a splitting criterion, and constructing a decision tree model for fault diagnosis according to the splitting criterion;
[0056] S5, obtaining the current system data in the complex equipment system and inputting it into a decision tree model for fault diagnosis, outputting the corresponding fault diagnosis result and displaying it to the user end.
[0057] In order to ensure the comprehensiveness and accuracy of the data, sensor technology and data recording systems can be used to capture the historical operation data of complex equipment systems from multiple dimensions and levels, including vibration signals, temperature distribution, pressure changes, current and voltage fluctuations, and real-time readings of various sensors.
[0058] After obtaining the historical system data, data preprocessing operations can be performed on these historical system data based on filtering technology, interpolation algorithms, normalization methods, feature engineering and other means, so as to eliminate noise in the data, fill missing values and standardize the data range, and generate cleaner, more standardized target system data suitable for subsequent analysis, ensuring the quality and consistency of the data.
[0059] Furthermore, convolutional neural networks (CNNs) can be used to extract features from the preprocessed target system data. Through its powerful local feature perception ability and hierarchical structure, convolutional neural networks can automatically learn and extract the characteristic information implicit in the target system data that is critical to fault diagnosis, namely, target system features. These features not only contain the operating status information of complex equipment systems, but also reflect their potential failure modes and trends.
[0060] Information gain is an important indicator to measure the contribution of features to classification tasks. By comparing the amount of information reduction brought by different features when dividing the data set, the optimal splitting nodes and order can be determined. Therefore, by further calculating the information gain of the target system features and using it as the splitting criterion of the decision tree model, an efficient and accurate decision tree model for fault diagnosis can be constructed.
[0061] Finally, the current system data in the complex equipment system can be collected in real time and input into the trained fault diagnosis decision tree model. The decision tree model can make step-by-step judgments along the branches of the decision tree according to the characteristics of the input data, and output the corresponding fault diagnosis results, including the fault type, possible causes, and recommended maintenance measures. In addition, these fault diagnosis results can be displayed to the user end in a visual way, which is conducive to the rapid response and efficient maintenance of complex equipment systems.
[0062] The historical system data includes historical sensor data, operation logs and maintenance records in complex equipment systems.
[0063] It is understandable that historical system data specifically includes historical sensor data, equipment operation logs, and maintenance records. Specifically, historical sensor data usually contains real-time parameters collected by various sensors during equipment operation, such as temperature, pressure, and vibration, which are essential for analyzing equipment status and predicting potential failures; equipment operation logs record in detail the operation information of each stage such as equipment startup, operation, and shutdown, including operation time, operator, and operation results, etc., which provides an important basis for tracing the equipment operation history; and maintenance records record in detail the time, content, replaced parts, and test results of previous equipment maintenance, which plays a key role in evaluating equipment maintenance effects and formulating subsequent maintenance plans.
[0064] The preprocessing operation on the historical system data to generate corresponding target system data specifically includes:
[0065] Removing abnormal values and duplicate values from the historical system data to generate corresponding first system data;
[0066] Using interpolation to fill missing values in the first system data to generate corresponding second system data;
[0067] The second system data is standardized by using discrete standardization to obtain the target system data.
[0068] It should be noted that the historical system data can first be cleaned, including removing outliers and duplicate values in the historical system data. Outliers usually include data points that deviate significantly from the normal range, values that may be generated by measurement errors or special events, and duplicate values are records that appear repeatedly in the data set. Through data cleaning techniques, such as setting thresholds to filter outliers and using hash algorithms to identify and remove duplicates, the corresponding first system data can be generated, laying the foundation for subsequent processing.
[0069] Interpolation can be used to fill in the missing values that may exist in the first system data. Interpolation is a method of estimating unknown data points based on known data points, including linear interpolation, polynomial interpolation, and nearest neighbor interpolation. According to the characteristics of the data and the distribution of missing values, a suitable interpolation strategy can be selected to generate the corresponding second system data, thereby maintaining the integrity and continuity of the data.
[0070] In order to make the second system data conform to the target system data specifications, discrete normalization can be used for normalization. Discrete normalization is a technique that scales data to a specific range, which helps eliminate the impact between data of different dimensions and improves the generalization ability of the model.
[0071] Through the above steps, standardized target system data can be obtained, ensuring that data can be seamlessly connected and compatible between different systems, facilitating subsequent data analysis and machine learning model training.
[0072] See also Figure 2 , the extracting target system features corresponding to the target system data based on a convolutional neural network specifically includes:
[0073] Determine the size, number, and stride of the convolution kernels corresponding to the convolutional layers in the convolutional neural network;
[0074] A convolution operation is performed on the target system data based on the convolution kernel to extract a corresponding first output feature, where the calculation formula of the first output feature is:
[0075]
[0076] In the formula, S1(i,j) represents the value of the first output feature at position (i,j); I(i+m,j+n) represents the value of the target system data at position (i+m,j+n); K(m,n) represents the weight of the convolution kernel at position (m,n); i and j represent the coordinates on the first output feature; m and n represent the coordinates on the convolution kernel; Σ represents the summation symbol;
[0077] Perform a maximum pooling operation on the first output feature, and select the maximum value in the corresponding pooling window as the second output feature. The calculation formula of the second output feature is:
[0078]
[0079] In the formula, S2(i,j) represents the value of the second output feature at position (i,j); max represents the maximum value operation; G represents the pooling window; S1(i,j) represents the value of the first output feature at position (i,j); i and j represent the coordinates on the first output feature; m and n represent the coordinates on the convolution kernel.
[0080] The first output feature and the second output feature are used as fusion input features, the weights corresponding to the fusion input features are determined and weighted summed to generate corresponding fusion output features. The calculation formula of the fusion output features is:
[0081]
[0082] In the formula, Fd represents the fusion output feature; w k represents the weight corresponding to the kth fusion input feature; F k represents the kth fusion input feature; Σ represents the summation symbol;
[0083] Based on the convolutional neural network, the fused output feature is the target system feature corresponding to the target system data.
[0084] It is understandable that the core parameters of the convolution layer, namely the size, number and step size of the convolution kernel, can be determined first. These parameters directly determine the effect of the convolution operation and the accuracy and efficiency of feature extraction. Specifically, the size of the convolution kernel determines the range of the receptive field, which can affect the size of the captured local features; the number determines the number of different features extracted, which increases the expressive power of the model; the step size controls the speed at which the convolution operation slides on the input data, which can affect the size of the output feature map.
[0085] Based on the configured convolution kernel, the target system data can be convolved to extract the corresponding first output feature. Specifically, the value of the first output feature S1(i,j) at position (i,j) is: S1(i,j) = ∑ m ∑ n I(i+m,j+n)K(m,n), where I(i+m,j+n) represents the value of the target system data at position (i+m,j+n); K(m,n) represents the weight of the convolution kernel at position (m,n); i, j and m, n represent the coordinates of the first output feature and the convolution kernel respectively. This process can effectively extract local features from the input target system data.
[0086] Subsequently, the generated first output feature can be subjected to a maximum pooling operation to further reduce the size of the feature map while retaining the most important information. Maximum pooling selects the maximum value within the pooling window as the second output feature. The specific calculation formula is as follows: Among them, S2(i,j) represents the value of the second output feature at position (i,j); max represents the maximum value operation; G represents the pooling window; S1(i,j) represents the value of the first output feature at position (i,j); i, j and m, n represent the coordinates of the first output feature and the convolution kernel respectively.
[0087] Finally, the first output feature and the second output feature can be used as fusion input features. By determining the weights corresponding to the fusion input features and performing weighted summation, the corresponding fusion output features can be generated, thereby integrating feature information at different levels and improving the comprehensiveness and effectiveness of feature representation. The specific calculation formula for the fusion output feature is: d =∑ k wk F k , where F d represents the fusion output feature; w k represents the weight corresponding to the kth fusion input feature; F k represents the kth fused input feature.
[0088] Through convolution layer configuration, convolution operation, pooling operation and feature fusion strategy, the corresponding target system features can be effectively extracted from the target system data. These features can reflect the inherent structure of the data and provide a solid foundation for subsequent model training, classification or regression tasks.
[0089] The calculating the information gain corresponding to the target system feature as a splitting criterion and constructing a decision tree model for fault diagnosis according to the splitting criterion specifically includes:
[0090] The calculation formula of the information gain corresponding to the target system feature is:
[0091]
[0092] Where D represents the data set consisting of all fault diagnosis results; A represents the target system feature; IG(D,A) represents the information gain of the target system feature A to the data set D consisting of all fault diagnosis results; E(D) represents the entropy of the data set D consisting of all fault diagnosis results; V(A) represents all possible values of the target system feature A; D v' represents the subset of the target system feature A whose value is v'; N(D v' ) represents the subset D of the target system feature A whose value is v' v' N(D) represents the number of samples in the dataset D consisting of all fault diagnosis results; E(D v' ) represents the subset D of the target system feature A whose value is v' v' The entropy of ; Σ represents the summation symbol;
[0093] The calculation formula of the entropy E(D) of the data set D composed of all fault diagnosis results is:
[0094]
[0095] Where E(D) represents the entropy of the data set D composed of all fault diagnosis results; Q represents the number of fault diagnosis categories in the data set D composed of all fault diagnosis results; p q represents the probability of the qth fault diagnosis category in the data set D composed of all fault diagnosis results; log2() represents the logarithmic function with base 2; Σ represents the summation symbol;
[0096] The target system feature A takes the value of v' as a subset Dv' The entropy E(D v' ) is calculated as:
[0097]
[0098] In the formula, E(D v' ) represents the subset D of the target system feature A whose value is v' v' Entropy; Q' represents the subset D of the target system feature A whose value is v' v' The number of fault diagnosis categories in p q' The subset D that represents the target system feature A with the value v' v' The probability of the q'th fault diagnosis category in ; log2() represents the logarithmic function with base 2; Σ represents the summation symbol;
[0099] The information gain corresponding to the target system feature is used as a splitting criterion, and a decision tree model for fault diagnosis is constructed according to the splitting criterion.
[0100] In practical applications, we can first set D to represent the data set consisting of all fault diagnosis results; A to represent the target system characteristics; IG(D,A) represents the information gain of the target system characteristics A to the data set D consisting of all fault diagnosis results. The specific calculation formula is:
[0101] Among them, entropy is a basic concept in information theory, which is used to quantify the uncertainty or amount of information in a data set. The calculation formula of the entropy E(D) of the data set D composed of all fault diagnosis results is: Where Q represents the number of fault diagnosis categories in the data set D consisting of all fault diagnosis results; p q It represents the probability of the qth fault diagnosis category in the data set D composed of all fault diagnosis results. By calculating the probability of each category, the overall entropy value of the data set D composed of all fault diagnosis results can be obtained, and the larger the entropy value, the higher the uncertainty in the data set.
[0102] The target system feature A takes the value of v' as a subset D v' The entropy E(D v' ) is calculated as: Among them, E(D v' ) represents the subset D of the target system feature A whose value is v' v' Entropy; Q' represents the subset D of the target system feature A whose value is v' v' The number of fault diagnosis categories in p q' The subset D that represents the target system feature A with the value v' v' The probability of the q'th fault diagnosis category in .
[0103] It should be noted that when constructing a decision tree model, the feature with the largest information gain can be preferentially selected as the splitting criterion for the current node. This process is repeated recursively until the stopping condition is met, such as the number of samples contained in the node is less than a predetermined threshold, all samples belong to the same category, or there are no more features to split. At this time, a decision tree model for fault diagnosis can be constructed.
[0104] Through the above method, the most critical information can be efficiently screened out from a large number of features, thereby improving the accuracy and efficiency of fault diagnosis.
[0105] The obtaining of current system data in the complex equipment system and inputting it into a decision tree model for fault diagnosis, outputting corresponding fault diagnosis results and displaying them to a user terminal specifically includes:
[0106] Acquire current system data in the complex equipment system, repeat the above preprocessing and feature extraction steps, generate corresponding current system features and input them into a decision tree model for fault diagnosis, and output the corresponding fault diagnosis result, wherein the fault diagnosis result includes the fault type and fault degree;
[0107] The output fault diagnosis result is displayed to the user terminal in a visual manner, wherein the visual manner includes a chart, a curve and an image;
[0108] Collect feedback results of the user terminal on the fault diagnosis result, and optimize the decision tree model used for fault diagnosis according to the feedback results.
[0109] In practical applications, current equipment operating parameters and status information can be collected in real time from complex equipment systems, including sensor readings, equipment temperature, pressure values, and vibration frequencies. Subsequently, these data can be preprocessed, including eliminating noise, filling missing values, and standardizing data formats to ensure data quality. Using convolutional neural networks, feature vectors that have a significant impact on fault diagnosis can be extracted from preprocessed data, and these feature vectors can accurately reflect the current status of complex equipment systems.
[0110] Furthermore, the generated current system features can be input into a pre-built and trained decision tree model. By recursively dividing the current system features, the corresponding fault diagnosis results can be output, including the fault type and fault degree, to provide a scientific basis for subsequent maintenance decisions. Specifically, the fault type can be mechanical wear, electrical fault, etc., and the fault degree can be slight, moderate, severe, etc.
[0111] See also Figure 3In order to improve user experience and diagnostic efficiency, the fault diagnosis results on the terminal device can be displayed to the user in a visual way. This visualization method includes using bar charts, line charts, scatter charts and other charts to intuitively display fault distribution and trends; using time series curves to reflect the changes in system status over time; and using three-dimensional images or thermal maps and other technologies to deeply analyze the location and impact range of the fault.
[0112] In addition, the feedback mechanism can be used to collect user feedback on fault diagnosis results, including accuracy evaluation, false alarm rate and missed alarm rate statistics, etc. These feedback results are used to further optimize the decision tree model used for fault diagnosis, and through strategies such as incremental learning or retraining, the diagnostic accuracy and generalization ability of the model are continuously improved to ensure continuous optimization of the model.
[0113] The present invention will be further described below in conjunction with specific embodiments:
[0114] When the method of the present invention is used to perform fault diagnosis on an aviation equipment system, the collected historical system data can first be preprocessed, including operations such as removing outliers and filling missing values. Then, a convolutional neural network is used to extract features from the preprocessed data, and the extracted features are input into a decision tree model for fault diagnosis to perform fault diagnosis. Finally, the fault diagnosis results can be displayed to the user in the form of a chart, curve or image, so that the user can take corresponding maintenance measures in a timely manner.
[0115] Through the above-mentioned method, the aviation equipment system can be quickly and accurately diagnosed, thereby improving the reliability, repairability and safety of the aviation equipment system.
[0116] Through the introduction of the above embodiments, the present invention can collect historical system data in complex equipment systems through a fault diagnosis method based on neural network analysis; perform preprocessing operations on historical system data to generate corresponding target system data; extract target system features corresponding to target system data based on convolutional neural networks; calculate the information gain corresponding to the target system features as a splitting criterion, and build a decision tree model for fault diagnosis according to the splitting criterion; obtain current system data in a complex equipment system and input it into a decision tree model for fault diagnosis, output the corresponding fault diagnosis results and display them to the user end. The present invention can extract more complex and accurate feature information through feature extraction using a convolutional neural network, thereby improving the accuracy of fault diagnosis; and can quickly perform fault diagnosis on complex equipment systems, shortening the diagnosis time; at the same time, the method of the present invention can process incomplete data or data with special values, reducing the requirements for data integrity; and by timely discovering and diagnosing faults, corresponding maintenance measures can be taken, thereby improving the reliability, repairability and safety of complex equipment systems.
[0117] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have the same meaning as in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined similarly here.
[0118] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A fault diagnosis method based on neural network analysis, characterized in that: The steps include: Collect historical system data in complex equipment systems; Performing preprocessing operations on the historical system data to generate corresponding target system data; Extracting target system features corresponding to the target system data based on a convolutional neural network; Calculating the information gain corresponding to the target system feature as a splitting criterion, and constructing a decision tree model for fault diagnosis according to the splitting criterion; The current system data in the complex equipment system is obtained and input into a decision tree model for fault diagnosis, and the corresponding fault diagnosis results are output and displayed to the user end.
2. A fault diagnosis method based on neural network analysis according to claim 1, characterized in that: The historical system data includes historical sensor data, operation logs and maintenance records in complex equipment systems.
3. A fault diagnosis method based on neural network analysis according to claim 1, characterized in that: The preprocessing operation on the historical system data to generate corresponding target system data specifically includes: removing abnormal values and duplicate values in the historical system data to generate corresponding first system data; Using interpolation to fill missing values in the first system data to generate corresponding second system data; The second system data is standardized by using discrete standardization to obtain the target system data.
4. A fault diagnosis method based on neural network analysis according to claim 1, characterized in that: The extracting the target system features corresponding to the target system data based on the convolutional neural network specifically includes: determining the size, number and step length of the convolution kernel corresponding to the convolution layer in the convolutional neural network; A convolution operation is performed on the target system data based on the convolution kernel to extract a corresponding first output feature, where the calculation formula of the first output feature is: Wherein, S1(i, j) represents the value of the first output feature at position (i, j); I(i+m, j+n) represents the value of the target system data at position (i+m, j+n); K(m, n) represents the weight of the convolution kernel at position (m, n); i, i represent the coordinates on the first output feature; m, n represent the coordinates on the convolution kernel; ∑ represents the summation symbol; Perform a maximum pooling operation on the first output feature, and select the maximum value in the corresponding pooling window as the second output feature. The calculation formula of the second output feature is: In the formula, S2(i, j) represents the value of the second output feature at position (i, j); max represents the maximum value operation; G represents the pooling window; S1(i, j) represents the value of the first output feature at position (i, j); i and j represent the coordinates on the first output feature; m and n represent the coordinates on the convolution kernel.
5. A fault diagnosis method based on neural network analysis according to claim 4, characterized in that: The first output feature and the second output feature are used as fusion input features, the weights corresponding to the fusion input features are determined and weighted summed to generate corresponding fusion output features. The calculation formula of the fusion output features is: In the formula, F d represents the fusion output feature; w k represents the weight corresponding to the kth fusion input feature; F k represents the kth fusion input feature; ∑ represents the summation symbol; Based on the convolutional neural network, the fused output feature is the target system feature corresponding to the target system data.
6. A fault diagnosis method based on neural network analysis according to claim 1, characterized in that: The calculating the information gain corresponding to the target system feature as a splitting criterion and constructing a decision tree model for fault diagnosis according to the splitting criterion specifically includes: The calculation formula of the information gain corresponding to the target system feature is: Where D represents the data set consisting of all fault diagnosis results; A represents the target system feature; IG(D, A) represents the information gain of the target system feature A to the data set D consisting of all fault diagnosis results; E(D) represents the entropy of the data set D consisting of all fault diagnosis results; V(A) represents all possible values of the target system feature A; D v′ represents the subset of the target system feature A whose value is v′; N(D v′ ) represents the subset D of the target system feature A whose value is v′ v′ N(D) represents the number of samples in the dataset D consisting of all fault diagnosis results; E(D v′ ) represents the subset D of the target system feature A whose value is v′ v′ The entropy of ; ∑ represents the summation symbol; The calculation formula of the entropy E(D) of the data set D composed of all fault diagnosis results is: Where E(D) represents the entropy of the data set D composed of all fault diagnosis results; Q represents the number of fault diagnosis categories in the data set D composed of all fault diagnosis results; p q represents the probability of the qth fault diagnosis category in the data set D composed of all fault diagnosis results; log2( ) represents the logarithmic function with base 2; ∑ represents the summation symbol; The target system feature A takes the value of v′ as a subset D v The entropy E(D v The calculation formula of ′) is: In the formula, E(D v ′) represents the subset D of the target system feature A whose value is v′ v ′; Q′ represents the subset D of the target system feature A whose value is v′ v The number of fault diagnosis categories in ′; q′ Denotes the subset D of the target system feature A whose value is v′ v′ The probability of the q′th fault diagnosis category in ; log2( ) represents the logarithmic function with base 2; ∑ represents the summation symbol; The information gain corresponding to the target system feature is used as a splitting criterion, and a decision tree model for fault diagnosis is constructed according to the splitting criterion.
7. A fault diagnosis method based on neural network analysis according to claim 1, characterized in that: The obtaining of current system data in the complex equipment system and inputting it into a decision tree model for fault diagnosis, outputting corresponding fault diagnosis results and displaying them to a user terminal specifically includes: Acquire current system data in the complex equipment system, repeat the above preprocessing and feature extraction steps, generate corresponding current system features and input them into a decision tree model for fault diagnosis, and output the corresponding fault diagnosis result, wherein the fault diagnosis result includes the fault type and fault degree; The output fault diagnosis result is displayed to the user terminal in a visual manner, wherein the visual manner includes a chart, a curve and an image; Collect feedback results of the user terminal on the fault diagnosis result, and optimize the decision tree model used for fault diagnosis according to the feedback results.