Intelligent fault detection and diagnosis method and system

Abstract

The invention relates to the technical field of fault detection, in particular to an intelligent fault detection and diagnosis method and system, and the method comprises the following steps: extracting and classifying fault feature vectors item by item based on the normal state and abnormal state operation data of a vehicle, transmitting the fault feature vectors to a fault coding layer through coding, inputting a potential space, and carrying out the clustering of feature points, and obtaining a preliminary fault identification feature. According to the method, the accuracy and the efficiency of fault diagnosis are remarkably improved by refining fault feature extraction and expanding sample analysis, and the accuracy and the efficiency of fault diagnosis are remarkably improved by analyzing normal and abnormal state data, extracting fine fault feature vectors, performing preliminary identification by utilizing a clustering technology, further expanding a sample set by simulating samples and deepening fault mode learning. And graded fault detection is introduced, response is adjusted according to fault severity, early warning is realized through dynamic monitoring, the equipment operation period is effectively prolonged, the shutdown risk is reduced, the maintenance cost is saved, and the system reliability is improved.

Description

Technical Field

[0001] The present invention relates to the field of technical fault detection, and in particular to an intelligent fault detection and diagnosis method and system. Background Art

[0002] The field of fault detection technology involves monitoring and analyzing the operating status of equipment, systems or vehicles, aiming to identify abnormal phenomena and potential faults. This technology relies on sensors, data acquisition and processing, signal analysis, pattern recognition and other technical means. Through continuous monitoring and feature extraction of machine equipment status data, it can identify signs of faults and judge the health status of equipment. Fault detection can assist in predicting fault time, improve system reliability, extend equipment service life, and reduce maintenance costs, thereby avoiding economic losses and safety risks caused by unexpected equipment downtime.

[0003] Among them, the intelligent fault detection and diagnosis method is a technology that applies advanced artificial intelligence and machine learning algorithms in the field of fault detection. It aims to automatically identify and accurately diagnose the fault status of complex systems such as vehicles. By analyzing massive data during the operation of the equipment, the algorithm automatically learns the fault mode, locates the source of potential faults and estimates the damage. Its main uses include improving the safety, stability and reliability of vehicle operation, reducing the occurrence of sudden failures, optimizing maintenance processes, and realizing predictive maintenance, thereby providing more valuable technical support for vehicle management and maintenance.

[0004] Existing technologies rely on traditional sensor data and relatively static analysis models when processing fault detection, which limits its efficiency and agility in processing massive dynamic data. The lag and roughness of fault detection make it impossible to accurately predict the specific time and type of faults, increasing the uncertainty and potential risks of the methods. Traditional methods are difficult to adapt to rapidly changing equipment states without the comprehensive use of machine learning and pattern recognition technology, resulting in inaccurate fault diagnosis and long fault response time in actual operations. This method that relies on preset fault models is not effective in dealing with unknown or atypical faults, and is difficult to adapt to current fault modes, thereby ignoring initial faults or abnormal phenomena, increasing equipment operation and maintenance challenges and safety risks. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent fault detection and diagnosis method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an intelligent fault detection and diagnosis method, comprising the following steps:

[0007] S1: Based on the normal and abnormal operation data of the vehicle, the fault feature vectors are extracted and classified item by item, transmitted to the fault encoding layer through encoding, input into the latent space, and the feature points are clustered to obtain preliminary fault identification features;

[0008] S2: Based on the preliminary fault identification features, extract the potential fault features and extended samples through the fault simulation samples, analyze the spatial distribution differences between the extended samples and the fault features, screen the extended samples associated with the potential faults, generate a simulated fault sample set, verify the simulated fault sample set from multiple angles, and generate a sample extension structure for fault simulation;

[0009] S3: Based on the sample extension structure of the fault simulation, the samples and abnormal density points are analyzed, and the fault features of the differentiated detection levels are screened by setting the multi-level detection threshold of the fault detection threshold layer, and a multi-level fault detection standard set is established. According to the multi-level fault detection standard set, the abnormal feature distribution of the detection level is matched to obtain the graded fault detection result;

[0010] S4: Based on the hierarchical fault detection results, call the real-time fault monitoring flow, dynamically monitor the signal characteristics during vehicle operation, match the signal characteristics with the detection standards, judge the vehicle status, embed the fault identification into the monitoring configuration, gradually update the identification status, and obtain real-time fault detection and diagnosis solutions.

[0011] As a further solution of the present invention, the step of obtaining the preliminary fault identification feature is specifically as follows:

[0012] S111: extracting the characteristic distribution of the vehicle in the normal state and the abnormal state based on the vehicle operation data in the normal state and the abnormal state, calling the operation parameters of speed, temperature, and vibration frequency and performing state encoding, performing hierarchical classification processing on the state characteristics, and obtaining a feature vector set;

[0013] S112: Perform information entropy analysis on the feature vector set, by calculating the information entropy value of each feature vector and removing feature points whose information entropy value is lower than a set threshold, using the formula:

[0014]

[0015] Calculate the similarity value between feature vectors and establish a feature similarity matrix, where S represents the similarity value between feature vectors, a i is the feature weight coefficient, v i represents the value of the eigenvector, g i is the feature difference parameter, b i For additional adjustment parameters;

[0016] S113: According to the feature homogeneity matrix, apply clustering logic to group the feature points, calculate the cluster center feature of each group, and generate preliminary fault identification features.

[0017] As a further solution of the present invention, the step of acquiring the simulated fault sample set is specifically as follows:

[0018] S211: Based on the preliminary fault identification features, feature parameters in the fault simulation samples are called for matching, potential feature values ​​of features in the fault simulation samples are extracted respectively, and the correlation between the potential features and the fault identification features is determined by calculating the differences between the features, so as to generate a potential fault feature set;

[0019] S212: Perform position distribution analysis on the feature space of the extended sample according to the feature parameters in the potential fault feature set, identify the difference distribution between the extended sample and the fault feature space, and select the extended sample set associated with the potential fault by setting the associated threshold, using the formula:

[0020]

[0021] Calculate the difference distribution value of the feature space and obtain the extended sample associated with the potential fault, where D represents the difference distribution value of the fault feature space, w is the weight coefficient, f and g represent the feature values ​​in the fault potential feature and the extended sample respectively, k is the extended sample offset coefficient, h is the standard deviation of the extended sample feature, and p is the offset adjustment parameter;

[0022] S213: According to the extended samples associated with the potential fault, samples associated with the potential fault are screened by difference distribution values, and parameters of each associated sample are integrated to generate a simulated fault sample set.

[0023] As a further solution of the present invention, the step of obtaining the sample extension structure of the fault simulation is specifically as follows:

[0024] S221: Based on each fault feature of the simulated fault sample set, analyze and extract feature angles, value ranges, and distribution densities, calculate the difference index and distribution weight of each feature, and generate a multi-angle analysis matrix of fault features;

[0025] S222: Based on the feature difference index and weight value of the fault feature multi-angle analysis matrix, calculate the impact score of the feature at the difference angle, using the formula:

[0026]

[0027] Calculate the normalized influence value of the feature and obtain the verification result of the feature, where Y is the normalized influence value of the feature, m and n are the weight coefficients of the angle, x and y are the feature values ​​under the differentiated angle, z is the difference adjustment parameter, and r and q are the additional adjustment parameters of the influence degree;

[0028] S223: Based on the verification result of the feature, the screening result of the multi-angle analysis matrix is ​​called, the feature sets of the correlation are merged, and the sample extension structure of the fault simulation is generated.

[0029] As a further solution of the present invention, the steps of obtaining the multi-level fault detection standard set are specifically as follows:

[0030] S311: extracting fault feature data from sample data based on the sample extension structure of the fault simulation, calculating difference values ​​with abnormal density points, normalizing the calculation results, and generating a fault feature difference vector;

[0031] S312: Based on the fault feature difference vector, a detection threshold layer of a multi-level fault detection standard set is set, and for each feature data, the formula is:

[0032]

[0033] Calculate the fault feature difference value and generate the fault feature difference matrix, where Q represents the fault feature difference value, F i represents the i-th eigenvalue in the fault feature vector, T is the set detection threshold, and W i represents the weight of the feature, A i is the characteristic adjustment coefficient;

[0034] S313: According to the fault feature difference matrix, determine whether each feature difference is greater than a detection threshold, filter fault features that are greater than the threshold, mark features that do not exceed the threshold as low-level features, and establish a multi-level fault detection standard set.

[0035] As a further solution of the present invention, the step of obtaining the hierarchical fault detection result is specifically:

[0036] S321: According to the abnormal feature distribution data of each level in the multi-level fault detection standard set, for each level of abnormal feature distribution, the matching detection level is determined in turn, and the initial hierarchical fault detection feature is obtained by calculating the cumulative deviation of the abnormal feature distribution;

[0037] S322: Based on the initial graded fault detection features, calculate the cumulative deviation of the abnormal feature distribution, set the fault detection level threshold, and judge the cumulative deviation using the formula:

[0038] E=|a·d x-b·d y +c·d z |

[0039] Generate the matching results of hierarchical fault detection features, where E represents the matching degree of abnormal distribution deviation, d x represents the cumulative deviation of the initial classification fault detection features, d y is the set fault detection level threshold, d z is the difference value of the cumulative value of the detection level in the previous step, and a, b, and c are the weight parameters of the cumulative deviation respectively;

[0040] S323: Analyze the matching results of the hierarchical fault detection features, select hierarchical features whose abnormal distribution deviation exceeds the detection level threshold, and generate hierarchical fault detection results by calling the matching conditions and judgments of the fault detection features.

[0041] As a further solution of the present invention, the steps of obtaining the real-time fault detection and diagnosis solution are specifically as follows:

[0042] S411: calling the fault level parameter in the hierarchical fault detection result, extracting the signal characteristic value in the vehicle operation process according to the real-time fault monitoring flow, performing normalization processing on the collected signal data by dynamically collecting the signal characteristics, calculating the deviation between the signal characteristic value and the standard value of the corresponding fault level, and generating a real-time fault signal deviation result;

[0043] S412: According to the real-time fault signal deviation result, the vehicle state judgment condition is called, and the monitoring threshold set by the deviation value is analyzed, and the formula is used:

[0044]

[0045] Calculate the normalized result of the signal deviation and determine whether it exceeds the threshold to generate the vehicle state matching result, where H is the signal deviation state value, δ s is the real-time signal deviation value, δ t is the deviation threshold in the fault level parameter, δ u is the initial deviation value of the hierarchical fault detection result, w a 、w b 、w c is the signal feature weight parameter;

[0046] S413: Dynamically monitor the vehicle status matching result, update the current fault identification according to the vehicle status, gradually embed the fault identification into the monitoring configuration, update and record the real-time status of the vehicle fault, and generate a real-time fault detection and diagnosis solution.

[0047] An intelligent fault detection and diagnosis system, the intelligent fault detection and diagnosis system is used to perform the above intelligent fault detection and diagnosis method, the system comprises:

[0048] The running status analysis module separates various signals in the running process based on the normal and abnormal running data of the vehicle, extracts characteristic signals and fault parameters, performs difference calculation on the correlation between signals, merges the fault characteristic values, and combines the characteristics by category to obtain a multi-dimensional fault characteristic signal set;

[0049] The fault coding analysis module maps the characteristic parameters of the signal set according to the requirements of fault coding based on the multi-dimensional fault characteristic signal set, matches the characteristic signal with the corresponding fault characteristic point in the coding space, calculates the matching relationship between the characteristic value and the spatial distribution, divides the boundary characteristics of the differentiated faults, and obtains the fault aggregation identification characteristics;

[0050] The simulated fault expansion module calls the original data samples based on the fault aggregation identification features, adds samples that meet the fault features, performs multi-dimensional distribution analysis on the samples that meet the features, merges and expands the sample set, and obtains the fault simulation feature structure;

[0051] The hierarchical fault judgment module sets multi-level fault judgment thresholds based on the fault simulation feature structure, compares the differentiated features in the simulation feature structure with the thresholds of each level, selects the feature points of the differentiated levels in turn, merges them one by one, and generates a multi-level fault judgment structure;

[0052] The real-time fault monitoring module matches the real-time monitoring signal characteristics based on the multi-level fault judgment structure, analyzes the signal characteristics with the classification boundaries, determines the matching of the fault levels, embeds the fault status and operating status signals into the configuration in real time, and obtains a real-time fault detection and diagnosis solution.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are:

[0054] In the present invention, by introducing refined fault feature extraction and extended sample analysis methods, the accuracy and efficiency of fault diagnosis are significantly improved. By analyzing normal and abnormal state data, more detailed fault feature vectors are extracted, and on the basis of the vectors, clustering technology is used to preliminarily identify faults. The sample set is further expanded through fault simulation samples, which not only optimizes the fault diagnosis process, but also improves the learning depth of fault modes, making the prediction of potential faults more accurate. The introduction of hierarchical fault detection allows the method to respond according to the severity of the fault, ensuring that monitoring parameters can be dynamically adjusted in real-time monitoring, achieving early fault warning, effectively extending the operation cycle of the equipment, and reducing the risk of unexpected downtime, thereby saving maintenance costs and improving the reliability of the overall method. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0056] Figure 2 A flow chart of the preliminary fault identification features in the present invention;

[0057] Figure 3 It is a flow chart of simulating a fault sample set in the present invention;

[0058] Figure 4 A flow chart of a sample extended structure of fault simulation in the present invention;

[0059] Figure 5 is a flow chart of a multi-level fault detection standard set in the present invention;

[0060] Figure 6 It is a flow chart of the hierarchical fault detection results in the present invention;

[0061] Figure 7 Flow chart of the real-time fault detection and diagnosis scheme in the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution 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.

[0063] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0064] Embodiment 1

[0065] See also Figure 1 The present invention provides a technical solution: an intelligent fault detection and diagnosis method, comprising the following steps:

[0066] S1: Based on the normal and abnormal operation data of the vehicle, the fault feature vectors are extracted and classified item by item, transmitted to the fault encoding layer through encoding, input into the latent space, and the feature points are clustered to obtain preliminary fault identification features;

[0067] S2: Based on the preliminary fault identification features, through the fault simulation samples, extract the potential fault features and extended samples, analyze the spatial distribution differences between the extended samples and the fault features, screen the extended samples associated with the potential faults, generate the simulated fault sample set, verify the simulated fault sample set from multiple angles, and generate the sample extension structure of the fault simulation;

[0068] S3: Based on the sample extension structure of fault simulation, the samples and abnormal density points are analyzed, and the fault features of differentiated detection levels are screened by setting the multi-level detection threshold of the fault detection threshold layer, and a multi-level fault detection standard set is established. According to the multi-level fault detection standard set, the abnormal feature distribution of the detection level is matched to obtain the hierarchical fault detection results;

[0069] S4: Based on the hierarchical fault detection results, call the real-time fault monitoring flow, dynamically monitor the signal characteristics of the vehicle in operation, match the signal characteristics with the detection standards, judge the vehicle status, embed the fault identification into the monitoring configuration, gradually update the identification status, and obtain real-time fault detection and diagnosis solutions.

[0070] Preliminary fault identification features include fault category identification, density distribution characteristics, and cluster feature points. The sample extension structure of fault simulation includes simulated fault samples, potential fault types, and extended sample sets. The multi-level fault detection standard set includes detection threshold levels, fault feature classification, and abnormal distribution standards. The graded fault detection results include fault detection levels, abnormal identification types, and distributed detection results. The real-time fault monitoring stream includes dynamic signal data, detection matching standards, and real-time monitoring identification. The real-time fault detection and diagnosis scheme includes identification status update, monitoring configuration adjustment, and fault identification status.

[0071] See also Figure 2 ,The steps for obtaining the preliminary fault identification features are as follows:

[0072] S111: extracting the characteristic distribution of the vehicle in the normal state and the abnormal state based on the vehicle operation data in the normal state and the abnormal state, calling the operation parameters of speed, temperature, and vibration frequency and performing state encoding, performing hierarchical classification processing on the state characteristics, and obtaining a feature vector set;

[0073] Based on the vehicle operation data, the operation characteristic data of the vehicle in normal and abnormal states are extracted. First, the collected parameters such as speed, temperature, vibration frequency, etc. are preliminarily classified in turn. By comparing the value range and fluctuation frequency of the parameters under differentiated states, the distribution range of each state feature is identified. After obtaining the distribution range, each parameter is marked in state, and a state marking matrix is ​​generated to facilitate the analysis of the boundary relationship between different states. The feature screening rules are further called to compare different parameters one by one, and repeated or highly interfering feature items are eliminated. The feature distribution results under each state are layered through state coding. In the layering process, the features in the state coding set are uniformly managed to ensure that accurate feature distribution data can be called in the subsequent fault feature classification process to obtain a preliminary feature vector set.

[0074] S112: Perform information entropy analysis on the feature vector set by calculating the information entropy value of each feature vector and removing feature points whose information entropy value is lower than a set threshold, using the formula:

[0075]

[0076] Calculate the similarity value between feature vectors and establish a feature similarity matrix, where S represents the similarity value between feature vectors, a i is the feature weight coefficient, v i represents the value of the eigenvector, g i is the feature difference parameter, b i For additional adjustment parameters;

[0077] The benefit of the formula is that through the combination of feature weight coefficient, feature difference parameter and additional adjustment parameter, the calculation of similarity between feature points is more flexible and accurate, so as to more accurately judge the similarity relationship between different features in high-dimensional data;

[0078] Set the weight coefficient a i =0.5, eigenvector value v i =10, characteristic difference parameter g i =2, additional adjustment parameter b i =1.5, substitute the value into the formula to calculate the characteristic homogeneity value;

[0079] First calculate the numerator:

[0080]

[0081] Next, calculate the denominator:

[0082]

[0083] Substituting the numerator and denominator into the formula, we get the homogeneity value:

[0084]

[0085] The result shows that the homogeneity value between feature vectors is 0.97, which is close to 1, indicating that the degree of homogeneity between feature points is high. This value can be used to further analyze the results of feature clustering and provide a reference for the feature center set in subsequent steps.

[0086] S113: according to the feature homogeneity matrix, apply clustering logic to group the feature points, calculate the cluster center feature of each group, and generate preliminary fault identification features;

[0087] On the basis of the feature homogeneity matrix, the homogeneity clustering logic is applied to perform feature classification. By calculating the homogeneity value between each feature point in the feature homogeneity matrix, the feature points with higher homogeneity are classified into the same category according to a certain homogeneity threshold to form a preliminary clustering grouping. Then, a representative central feature point is calculated in each clustering group. The selection of the central feature point is determined by calculating the homogeneity mean of each feature point and feature point in the group. The feature point with the highest mean is selected as the clustering center. After completing the selection of the clustering center of each group, the central feature point of each group is recorded, and each cluster center feature point forms a feature center set, and the features in the set are marked as preliminary fault identification features.

[0088] See also Figure 3 ,The specific steps for obtaining the simulated fault sample set are:

[0089] S211: Based on the preliminary fault identification features, feature parameters in the fault simulation samples are called for matching, potential feature values ​​of the features in the fault simulation samples are extracted respectively, and the correlation between the potential features and the fault identification features is determined by calculating the differences between the features, so as to generate a potential fault feature set;

[0090] Based on the preliminary fault identification features, various parameters in the fault simulation samples are extracted, and the characteristic parameters of the fault simulation samples are compared with the preliminary fault identification features item by item. The correlation between the features is judged by calculating the numerical differences between the characteristic values. First, each feature data contained in the simulation sample is extracted, and the potential feature values ​​of the associated fault features are called. The statistical features such as the value range, standard deviation and variance of each fault feature are compared, and the fault features associated with the preliminary identification features are screened out to establish a preliminary fault feature set. After the potential fault feature set is extracted, the parameters in the fault sample are standardized to ensure that all parameters in the feature set have consistent dimensions and scales. After standardization, the potential fault feature set is generated to ensure the accuracy of feature comparison and calling in the subsequent analysis process.

[0091] S212: Based on the characteristic parameters in the potential fault feature set, position distribution analysis is performed on the feature space of the extended sample to identify the difference distribution between the extended sample and the fault feature space, and the extended sample set associated with the potential fault is screened by setting an associated threshold, using the formula:

[0092]

[0093] Calculate the difference distribution value of the feature space and obtain the extended sample associated with the potential fault, where D represents the difference distribution value of the fault feature space, w is the weight coefficient, f and g represent the feature values ​​in the fault potential feature and the extended sample respectively, k is the extended sample offset coefficient, h is the standard deviation of the extended sample feature, and p is the offset adjustment parameter;

[0094] The benefit of the formula is that by introducing a combination of weight coefficients, offset coefficients and adjustment parameters, the contribution of different features can be more flexibly adjusted in the calculation of the similarity of the feature space, thereby more accurately reflecting the relationship between the potential features of the fault and the extended samples;

[0095] Set the weight coefficient w = 0.8, the potential fault feature value f = 15, the extended sample feature value g = 12, the offset coefficient k = 0.5, the standard deviation of the extended sample feature h = 4, and the adjustment parameter p = 1.0, and substitute them into the formula to calculate the difference distribution value;

[0096] First, the square of the difference of the first term is calculated:

[0097] (w·(fg)) 2 =(0.8·(15-12)) 2 =(0.8·3) 2 =2.4 2 =5.76

[0098] Next we calculate the square of the offset of the second term:

[0099] (k·h) 2 =(0.5·4) 2 =2 2 =4

[0100] Then substitute the values ​​into the formula to calculate the final difference distribution value:

[0101]

[0102] The result shows that the difference distribution value is 3.28, indicating that the difference between the fault potential characteristics and the extended samples is small, which can be used to further screen the extended samples associated with the fault potential characteristics as a component of the simulated fault sample set.

[0103] S213: According to the extended samples associated with the potential fault, the samples associated with the potential fault are screened by the difference distribution value, and each associated sample parameter is integrated to generate a simulated fault sample set;

[0104] On the basis of completing the spatial difference distribution analysis of the fault potential feature set and the extended sample feature set, the threshold of the difference distribution is set to screen the extended samples with higher correlation. First, the various parameters in the obtained extended sample set are analyzed, and the previously calculated difference distribution value is compared with the threshold. Samples with difference distribution values ​​less than or equal to the set threshold are marked as related samples. In this way, various types of related samples are gradually summarized into the final generated simulated fault sample set, so that the set can comprehensively cover the diversity of potential fault characteristics.

[0105] See also Figure 4 ,The steps for obtaining the sample extension structure of the fault simulation are as follows:

[0106] S221: Based on each fault feature of the simulated fault sample set, analyze and extract the feature angle, value range and distribution density, calculate the difference index and distribution weight of each feature, and generate a multi-angle analysis matrix of the fault features;

[0107] Based on the various fault features of the simulated fault sample set, the angle, value range and distribution density of the features are analyzed and extracted item by item. First, the statistical properties of each feature parameter, including the mean, variance and extreme value, are calculated to describe the performance of each fault feature at different angles, and further generate its difference index. Then, adaptation weights are assigned to different angles of each feature. The distribution of weights depends on the variation range of the feature at different angles and the proportion in the overall feature set. Through this process, the feature performance at different angles is compared to determine which features are more representative at specific angles, forming a complete feature multi-angle analysis matrix. The matrix is ​​used to screen and judge the relevance and difference of each feature, which can provide data support for subsequent steps and ensure the diversity and relevance of features at different angles.

[0108] S222: Based on the feature difference index and weight value of the fault feature multi-angle analysis matrix, calculate the impact score of the feature at the differentiated angle using the formula:

[0109]

[0110] Calculate the normalized influence value of the feature and obtain the verification result of the feature, where Y is the normalized influence value of the feature, m and n are the weight coefficients of the angle, x and y are the feature values ​​under the differentiated angle, z is the difference adjustment parameter, and r and q are the additional adjustment parameters of the influence degree;

[0111] The benefit of the formula is that by using different weights and adjustment parameters, the difference calculation between different features can be accurately controlled in multi-angle analysis, and the feature fitness can be flexibly adjusted in features with significant differences;

[0112] Set the weight coefficients m=0.6 and n=0.4, the difference eigenvalue x=18, the other angle eigenvalue y=12, the adjustment parameter z=5, the influence additional parameters r=2 and q=3, and substitute the values ​​into the formula to calculate the multi-angle normalized influence value;

[0113] First calculate the numerator:

[0114] |m·(xy)|=|0.6·(18-12)|=|0.6·6|=3.6

[0115]

[0116] The total numerator is: 3.6 + 0.896 = 4.496;

[0117] Next, calculate the denominator:

[0118] r+q=2+3=5

[0119] Finally, the normalized influence value is obtained:

[0120]

[0121] The result shows that the normalized impact value is 0.899, which means that in the multi-angle verification process, the comprehensive impact value of the feature is high, which meets the requirements of multi-angle analysis and can be used for subsequent comprehensive analysis of multi-angle verification scores to support multi-angle verification and expansion of fault features.

[0122] S223: Based on the verification result of the feature, the screening result of the multi-angle analysis matrix is ​​called, the feature set of the correlation degree is merged, and the sample extension structure of the fault simulation is generated;

[0123] Based on the comprehensive results of the multi-angle verification scores, the feature screening results of each feature in the multi-angle analysis matrix are compared. The correlation between the features is determined through the comparison process, and the features with higher correlation are classified and merged. At the same time, the merged feature set is verified to ensure that there is no loss or deviation of feature information in the multi-angle verification process. The feature set after multi-angle verification is marked as a high-correlation feature, and the feature set with the highest correlation is further integrated to form a sample extension structure for fault simulation. Based on this feature extension structure, multiple fault features can be organically combined to cover more potential fault features, ensuring that the sample extension structure of fault simulation has a more comprehensive fault feature coverage capability.

[0124] See also Figure 5 ,The specific steps for obtaining the multi-level fault detection standard set are:

[0125] S311: based on the sample extension structure of the fault simulation, extracting fault feature data from the sample data, calculating the difference value with the abnormal density point, normalizing the calculation result, and generating a fault feature difference vector;

[0126] Based on the sample extension structure of fault simulation, fault feature data is extracted, and each sample data is compared with the abnormal density points one by one. By calculating the distance difference between each sample point and the abnormal density point, the difference distribution between the sample and the abnormal point is statistically analyzed. Then all the calculation results are normalized to ensure that different feature data are compared under a unified scale. The normalized data is converted into independent values ​​of each fault feature through the feature decomposition method, and a series of standardized feature values ​​are obtained. A fault feature difference vector containing various fault feature components is generated. The difference values ​​of different features can be clearly distinguished for subsequent calculations, providing specific input data support for subsequent multi-level detection.

[0127] S312: Based on the fault feature difference vector, the detection threshold layer of the multi-level fault detection standard set is set, and for each feature data, the formula is:

[0128]

[0129] Calculate the fault feature difference value and generate the fault feature difference matrix, where Q represents the fault feature difference value, F i represents the i-th eigenvalue in the fault feature vector, T is the set detection threshold, and W i represents the weight of the feature, A i is the characteristic adjustment coefficient;

[0130] The formula is useful in that by adding the characteristic adjustment factor A i and feature weight W i , making the calculation results more adaptable, and being able to adaptively adjust the difference calculation under differentiated detection conditions to improve the accuracy of detection;

[0131] Let F i =100 is the characteristic value, T=80 is the detection threshold, and the characteristic adjustment coefficient A i =1.2 Calculated through the previous feature distribution analysis, feature weight W i =0.5 is set based on the feature importance weight distribution;

[0132] Substitute the values ​​into the formula to calculate:

[0133]

[0134] First calculate the numerator:

[0135] |100-80|·1.2=20·1.2=24

[0136] Then calculate the denominator:

[0137]

[0138] Combine the numerator and denominator:

[0139]

[0140] The result shows that the calculated difference degree Q = 19.6 indicates that the difference level of this feature is relatively high under the current threshold value, and can be used as a basis for screening high-level fault features.

[0141] S313: judging whether each feature difference is greater than a detection threshold according to the fault feature difference matrix, screening fault features above the threshold, marking features that do not exceed the threshold as low-level features, and establishing a multi-level fault detection standard set;

[0142] According to each data point in the fault feature difference matrix, each feature difference is compared with the detection threshold. If the difference of a certain feature is greater than the set threshold, the feature is marked as a high-level fault feature and added to the high-level fault feature set. The cycle continues to determine the size of the difference value in each feature data, and all data that meet the high-level features are screened. If the difference does not exceed the threshold, the feature is classified as a low-level feature and its original state is maintained. Through this classification method, the grading and classification of fault features are achieved, so that each feature is arranged in order according to its detection level, forming a multi-level fault detection standard set.

[0143] See also Figure 6 , the specific steps for obtaining the hierarchical fault detection results are:

[0144] S321: according to the abnormal feature distribution data of each level in the multi-level fault detection standard set, for each level of abnormal feature distribution, sequentially determine the matching detection level, and obtain the initial graded fault detection feature by calculating the cumulative deviation of the abnormal feature distribution;

[0145] According to the abnormal feature distribution data of each level in the multi-level fault detection standard set, the abnormal features of each level are retrieved in turn, and the feature quantization processing is performed on the detection content of each level. The abnormal feature data in the classification is converted into numerical data, and compared one by one according to the abnormal data interval of the standard set at each level. The numerical deviation of the distribution feature is calculated through the set detection threshold judgment interval, and the cumulative deviation value of each abnormal feature is recorded, so as to determine the matching of the abnormal features under each level, and the cumulative deviation value is distributed and matched with the set detection standards at each level to obtain the initial hierarchical fault detection features.

[0146] S322: Based on the initial graded fault detection features, calculate the cumulative deviation of the abnormal feature distribution, set the fault detection level threshold, and judge the cumulative deviation using the formula:

[0147] E=|a·d x -b·d y +c·d z |

[0148] Generate the matching results of hierarchical fault detection features, where E represents the matching degree of abnormal distribution deviation, d x represents the cumulative deviation of the initial classification fault detection features, d y is the set fault detection level threshold, d z is the difference value of the cumulative value of the detection level in the previous step, and a, b, and c are the weight parameters of the cumulative deviation respectively;

[0149] The benefit of the formula is that it introduces the weight parameter of the cumulative deviation, so that the detection results can more accurately reflect the actual impact of the abnormal features, and enhance the refinement and accuracy of the hierarchical feature matching results;

[0150] Parameter d x is the cumulative deviation obtained from the initial graded fault detection features, which is specifically calculated by the aforementioned detection threshold deviation cumulative value;

[0151] Parameter d y The detection level threshold is set based on the average of the upper and lower limits of the classification threshold of the multi-level fault detection standard set;

[0152] Parameter d z is the difference in the cumulative values ​​of each detection level, indicating the range of numerical differences of the detection features at each level, which is calculated based on the cumulative deviation values ​​of the abnormal features under the above classification;

[0153] The weight parameters a, b, and c are set based on the weighted influence of each parameter on the degree of influence of the abnormal characteristics, and fluctuate with the upper and lower thresholds of the classification standard;

[0154] Setting x =15, d y =10, d z =5, set a=1.2, b=0.8, c=1.5, then the formula calculation process is as follows:

[0155] Calculate a·d x =1.2×15=18;

[0156] Calculate b·d y =0.8×10=8;

[0157] Calculate c·d z =1.5×5=7.5;

[0158] Formula result E = |18-8+7.5| = |17.5| = 17.5;

[0159] The result shows that the matching degree of the abnormal distribution deviation is 17.5, which has a high degree of consistency with the deviation interval of the fault detection level, and generates a matching result of the graded fault detection feature.

[0160] S323: Analyze the matching results of the hierarchical fault detection features, select hierarchical features whose abnormal distribution deviation exceeds the detection level threshold, and generate hierarchical fault detection results by calling the matching conditions and judgments of the fault detection features;

[0161] According to the abnormal distribution of the aforementioned detection levels, the matching results in the second step are further analyzed. For the fault features in each level whose abnormal distribution deviation exceeds the detection level threshold, the final matching situation of the classification feature is recorded, and the result parameters are compounded with the initial classification fault detection features. According to the increasing trend of the deviation degree of the matching parameters, the fault level threshold is called for further classification processing. By recording the final matching situation of each level, the classified fault detection results are generated.

[0162] See also Figure 7 ,The steps for obtaining real-time fault detection and diagnosis solutions are as follows:

[0163] S411: calling the fault level parameter in the hierarchical fault detection result, extracting the signal characteristic value in the vehicle operation process according to the real-time fault monitoring stream, performing normalization processing on the collected signal data by dynamically collecting the signal characteristics, calculating the deviation between the signal characteristic value and the standard value of the corresponding fault level, and generating a real-time fault signal deviation result;

[0164] The fault level parameters in the hierarchical fault detection results are called, and the signal feature data collected in real time during the operation of the vehicle are sorted out item by item. Each signal feature is normalized in turn, and the signal feature value range is uniformly mapped to the standard scale to make different source signals consistent. Through this normalization process, a standardized representation of each signal feature is generated to ensure that the signal features are within a unified range for subsequent operations. The normalized signal feature data is matched according to the standard value of each level of fault level, and the difference between the actual signal feature value and the standard fault detection value is calculated item by item. By gradually calculating the deviation value of each signal feature, a deviation amount set of the signal feature is formed. Considering the differences in the deviation value set, it is determined whether it exceeds the set range of the fault level parameter, and the signal deviations that meet the conditions are accumulated to form a real-time fault signal deviation result. This deviation result will be used to further determine the matching situation between the vehicle's operating status and the fault level.

[0165] S412: Based on the real-time fault signal deviation result, the vehicle status judgment condition is called, and the monitoring threshold set by the deviation value is analyzed, using the formula:

[0166]

[0167] Calculate the normalized result of the signal deviation and determine whether it exceeds the threshold to generate the vehicle state matching result, where H is the signal deviation state value, δ s is the real-time signal deviation value, δ t is the deviation threshold in the fault level parameter, δ u is the initial deviation value of the hierarchical fault detection result, w a 、w b 、w c is the signal feature weight parameter;

[0168] The benefit of the formula is that it makes the detection results more flexible by adjusting the weights of multiple signals, can adapt to the actual impact of different signals, and help reflect the current state of the vehicle in more detail;

[0169] This formula is used to calculate the signal deviation state value H;

[0170] δ s The real-time signal deviation value is the difference between the vehicle's current operating signal and the fault standard value. It is obtained by collecting real-time signal data on site and comparing it with the preset standard value.

[0171] δ t It indicates the deviation threshold in the set fault level parameters and the tolerance deviation range obtained through data analysis and actual monitoring;

[0172] δ uis the initial deviation value of hierarchical fault detection, obtained through hierarchical detection data processing;

[0173] Weight parameter w a 、w b 、w c Each signal deviation item is weighted separately, and the weight setting is dynamically adjusted according to the actual operating status;

[0174] Substitute numerical calculation example: Let δ s =12, δ t =8, δ u =6, and set w a =1.1, w b =0.9, w c =1.0;

[0175] The calculation process is as follows:

[0176] Calculate w a ·δ s =1.1×12=13.2;

[0177] Calculate w b ·δ t =0.9×8=7.2;

[0178] Calculate w c ·δ u =1.0×6=6;

[0179] Substituting into the formula:

[0180] The result shows that the signal deviation state value is 9.33, which indicates that the current signal deviation is within the detection range and generates a matching vehicle state matching result.

[0181] S413: Dynamically monitor the vehicle status matching result, update the current fault identification according to the vehicle status, gradually embed the fault identification into the monitoring configuration, update and record the real-time status of the vehicle fault, and generate a real-time fault detection and diagnosis plan;

[0182] Based on the vehicle status matching results, the dynamic monitoring mechanism is further used to gradually embed the vehicle fault identification, call the real-time signal deviation status value results generated above, embed the qualified fault identification information into the monitoring configuration system, and update the fault identification in the configuration system for specific signal deviation status values. Each embedding is updated according to the current data of the signal deviation status during the operation of the vehicle, and the matched fault identifications are embedded one by one, gradually increasing the vehicle's fault monitoring status information, embedding the currently monitored fault status results into records, and continuously generating real-time fault detection and diagnosis solutions.

[0183] The intelligent fault detection and diagnosis system is used to execute the above intelligent fault detection and diagnosis method, and the system includes:

[0184] The running status analysis module separates various signals in the running process based on the normal and abnormal running data of the vehicle, extracts characteristic signals and fault parameters, performs difference calculation on the correlation between signals, merges the fault characteristic values, and combines the characteristics by category to obtain a multi-dimensional fault characteristic signal set;

[0185] The fault coding analysis module is based on a multi-dimensional fault feature signal set. It maps the feature parameters of the signal set according to the requirements of fault coding, matches the feature signal with the corresponding fault feature points in the coding space, calculates the matching relationship between the feature value and the spatial distribution, divides the boundary features of differentiated faults, and obtains the fault aggregation recognition features.

[0186] The fault simulation expansion module calls the original data samples based on the fault aggregation identification features, adds the samples that meet the fault characteristics, performs multi-dimensional distribution analysis on the samples that meet the fault characteristics, merges and expands the sample set, and obtains the fault simulation feature structure;

[0187] The hierarchical fault judgment module sets multi-level fault judgment thresholds based on the fault simulation feature structure, compares the differentiated features in the simulation feature structure with the thresholds of each level, selects the feature points of the differentiated levels in turn, merges them one by one, and generates a multi-level fault judgment structure;

[0188] The real-time fault monitoring module is based on a multi-level fault judgment structure, matches the real-time monitoring signal characteristics, analyzes the signal characteristics with the classification boundaries, determines the matching of the fault level, embeds the fault status and operating status signals into the configuration in real time, and obtains a real-time fault detection and diagnosis solution.

[0189] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent fault detection and diagnosis method, characterized in that: The following steps are involved: Based on the vehicle's normal and abnormal operating data, the fault feature vectors are extracted and classified item by item, transmitted to the fault encoding layer through encoding, input into the latent space, and the feature points are clustered to obtain preliminary fault identification features; Based on the preliminary fault identification features, extracting fault potential features and extended samples through fault simulation samples, analyzing the spatial distribution differences between extended samples and fault features, screening extended samples associated with potential faults, generating a simulated fault sample set, verifying the simulated fault sample set from multiple angles, and generating a sample extension structure for fault simulation; Based on the sample extension structure of the fault simulation, the samples and abnormal density points are analyzed, and the fault features of the differentiated detection levels are screened by setting the multi-level detection threshold of the fault detection threshold layer, and a multi-level fault detection standard set is established. According to the multi-level fault detection standard set, the abnormal feature distribution of the detection level is matched to obtain the graded fault detection result; Based on the hierarchical fault detection results, the real-time fault monitoring flow is called to dynamically monitor the signal characteristics of the vehicle in operation, match the signal characteristics with the detection standards, judge the vehicle status, embed the fault identification into the monitoring configuration, gradually update the identification status, and obtain real-time fault detection and diagnosis solutions.

2. The intelligent fault detection and diagnosis method according to claim 1, characterized in that: The steps for obtaining the preliminary fault identification features are specifically as follows: Based on the normal and abnormal state operation data of the vehicle, the characteristic distribution of the vehicle in the normal and abnormal state is extracted, the operation parameters of speed, temperature, and vibration frequency are called and state-encoded, the state characteristics are hierarchically classified and processed, and a feature vector set is obtained; The information entropy analysis is performed on the feature vector set, by calculating the information entropy value of each feature vector and eliminating feature points whose information entropy value is lower than the set threshold, using the formula: Calculate the similarity value between feature vectors and establish a feature similarity matrix, where S represents the similarity value between feature vectors, a i is the feature weight coefficient, v j represents the value of the eigenvector, g i is the feature difference parameter, b i For additional adjustment parameters; According to the feature homogeneity matrix, clustering logic is applied to group the feature points, and the cluster center features of each group are calculated to generate preliminary fault identification features.

3. The intelligent fault detection and diagnosis method according to claim 2, characterized in that: The steps of obtaining the simulated fault sample set are specifically as follows: Based on the preliminary fault identification features, feature parameters in the fault simulation samples are called for matching, potential feature values ​​of features in the fault simulation samples are extracted respectively, and the correlation between the potential features and the fault identification features is determined by calculating the differences between the features, so as to generate a potential fault feature set; According to the characteristic parameters in the potential fault feature set, the position distribution analysis is performed with the feature space of the extended sample to identify the difference distribution between the extended sample and the fault feature space, and the extended sample set associated with the potential fault is screened by setting the associated threshold, using the formula: Calculate the difference distribution value of the feature space to obtain the extended sample associated with the potential fault, where D represents the difference distribution value of the fault feature space, w is the weight coefficient, f and g represent the feature values ​​in the fault potential feature and the extended sample respectively, k is the extended sample offset coefficient, h is the standard deviation of the extended sample feature, and p is the offset adjustment parameter; According to the extended samples associated with the potential fault, samples associated with the potential fault are screened by using difference distribution values, and each associated sample parameter is integrated to generate a simulated fault sample set.

4. The intelligent fault detection and diagnosis method according to claim 3, characterized in that: The steps for obtaining the sample extension structure of the fault simulation are specifically as follows: Based on each fault feature of the simulated fault sample set, the feature angle, value range and distribution density are analyzed and extracted, the difference index and distribution weight of each feature are calculated, and a multi-angle analysis matrix of the fault features is generated; Based on the feature difference index and weight value of the fault feature multi-angle analysis matrix, the impact score of the feature under the differentiation angle is calculated using the formula: Calculate the normalized influence value of the feature and obtain the verification result of the feature, where Y is the normalized influence value of the feature, m and n are the weight coefficients of the angle, x and y are the feature values ​​under the differentiated angle, z is the difference adjustment parameter, and r and q are the additional adjustment parameters of the influence degree; Based on the verification results of the features, the screening results of the multi-angle analysis matrix are called, the feature sets of the correlation are merged, and the sample extension structure of the fault simulation is generated.

5. The intelligent fault detection and diagnosis method according to claim 4, characterized in that: The steps for obtaining the multi-level fault detection standard set are specifically as follows: Based on the sample extension structure of the fault simulation, fault feature data is extracted from the sample data, difference values ​​are calculated with abnormal density points, and the calculation results are normalized to generate a fault feature difference vector; Based on the fault feature difference vector, the detection threshold layer of the multi-level fault detection standard set is set, and for each feature data, the formula is used: Calculate the fault feature difference value and generate the fault feature difference matrix, where Q represents the fault feature difference value, F i represents the i-th eigenvalue in the fault feature vector, T is the set detection threshold, and W i Represents the weight of the feature, Ai is the feature adjustment coefficient; According to the fault feature difference matrix, it is determined whether each feature difference is greater than a detection threshold, fault features above the threshold are screened, features that do not exceed the threshold are marked as low-level features, and a multi-level fault detection standard set is established.

6. The intelligent fault detection and diagnosis method according to claim 5, characterized in that: The steps for obtaining the hierarchical fault detection results are specifically as follows: According to the abnormal feature distribution data of each level in the multi-level fault detection standard set, for each level of abnormal feature distribution, the matching detection level is determined in turn, and the initial graded fault detection feature is obtained by calculating the cumulative deviation of the abnormal feature distribution; Based on the initial hierarchical fault detection features, the cumulative deviation of the abnormal feature distribution is calculated, the fault detection level threshold is set, and the cumulative deviation is judged using the formula: E=|a·d x -b·d y +c·d z | Generate the matching results of hierarchical fault detection features, where E represents the matching degree of abnormal distribution deviation, d x represents the cumulative deviation of the initial classification fault detection features, d y is the set fault detection level threshold, d z is the difference value of the cumulative value of the detection level in the previous step, and a, b, and c are the weight parameters of the cumulative deviation respectively; The matching results of the hierarchical fault detection features are analyzed, hierarchical features whose abnormal distribution deviation exceeds the detection level threshold are selected, and hierarchical fault detection results are generated by calling the matching conditions and judgments of the fault detection features.

7. The intelligent fault detection and diagnosis method according to claim 6, characterized in that: The steps for obtaining the real-time fault detection and diagnosis solution are specifically as follows: Calling the fault level parameter in the hierarchical fault detection result, extracting the signal characteristic value in the vehicle operation process according to the real-time fault monitoring flow, performing normalization processing on the collected signal data by dynamically collecting the signal characteristics, calculating the deviation between the signal characteristic value and the standard value of the corresponding fault level, and generating a real-time fault signal deviation result; According to the real-time fault signal deviation result, the vehicle status judgment condition is called, and the monitoring threshold set by the deviation value is analyzed, and the formula is used: Calculate the normalized result of the signal deviation and determine whether it exceeds the threshold to generate the vehicle state matching result, where H is the signal deviation state value, δ s is the real-time signal deviation value, δ t is the deviation threshold in the fault level parameter, δ u is the initial deviation value of the hierarchical fault detection result, w a 、w b 、w c is the signal feature weight parameter; The vehicle status matching result is dynamically monitored, the current fault identification is updated according to the vehicle status, the fault identification is gradually embedded into the monitoring configuration, the real-time status of the vehicle fault is updated and recorded, and a real-time fault detection and diagnosis solution is generated.

8. An intelligent fault detection and diagnosis system, characterized in that: According to the intelligent fault detection and diagnosis method according to any one of claims 1 to 7, the system comprises: The running status analysis module separates various signals in the running process based on the normal and abnormal running data of the vehicle, extracts characteristic signals and fault parameters, performs difference calculation on the correlation between signals, merges the fault characteristic values, and combines the characteristics by category to obtain a multi-dimensional fault characteristic signal set; The fault coding analysis module maps the characteristic parameters of the signal set according to the requirements of fault coding based on the multi-dimensional fault characteristic signal set, matches the characteristic signal with the corresponding fault characteristic point in the coding space, calculates the matching relationship between the characteristic value and the spatial distribution, divides the boundary characteristics of the differentiated faults, and obtains the fault aggregation identification characteristics; The simulated fault expansion module calls the original data samples based on the fault aggregation identification features, adds samples that meet the fault features, performs multi-dimensional distribution analysis on the samples that meet the features, merges and expands the sample set, and obtains the fault simulation feature structure; The hierarchical fault judgment module sets multi-level fault judgment thresholds based on the fault simulation feature structure, compares the differentiated features in the simulation feature structure with the thresholds of each level, selects the feature points of the differentiated levels in turn, merges them one by one, and generates a multi-level fault judgment structure; The real-time fault monitoring module matches the real-time monitoring signal characteristics based on the multi-level fault judgment structure, analyzes the signal characteristics with the classification boundaries, determines the matching of the fault levels, embeds the fault status and operating status signals into the configuration in real time, and obtains a real-time fault detection and diagnosis solution.

Images