A motorcycle engine fault detection method based on vibration effect monitoring

Abstract

The present invention relates to the field of fault detection technology, and specifically relates to a motorcycle engine fault detection method based on vibration effect monitoring. First, first vibration characteristic data, state characteristic data, and first position data of the engine are obtained from a sensor group; the first position data and the first vibration characteristic data are input into a position adjustment module, and the first position data is updated according to the output position adjustment coefficient to obtain second position data; then, the state characteristic data and second vibration characteristic data corresponding to the second position data are input into a fault detection influence weight prediction module to obtain a fault detection influence weight; then, the second vibration characteristic data is classified using a multi-task fault diagnosis module to obtain fault characteristics and non-fault characteristics, and these characteristics and the fault detection influence weight are input into the module for analysis to obtain a fault detection value for fault judgment; this method can improve the accuracy and effectiveness of motorcycle engine fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, in particular to a motorcycle engine fault detection method based on vibration effect monitoring. Background Art

[0002] With the development of artificial intelligence, a machine learning algorithm is used to establish the correspondence between vibration characteristics and engine status characteristics, and then the RVM classifier is used to diagnose the faults of the test samples. To address the interference problem during driving, the fault analysis strategy to be executed is selected based on the fuzzy category of the time domain waveform image segment.

[0003] At present, the existing technology for motorcycle engine fault detection methods still has shortcomings; on the one hand, the existing technology does not take into account that the sensor position will affect the complexity of the interference characteristics in the data, resulting in a large deviation between the actual value obtained and the reference value, thereby affecting fault analysis and evaluation; on the other hand, the existing technology does not consider monitoring parameters such as ambient temperature, speed, and load as interference factors in the fault analysis process, thereby reducing the accuracy and effectiveness of motorcycle engine fault detection.

[0004] Therefore, a motorcycle engine fault detection method based on vibration effect monitoring is proposed. Summary of the Invention

[0005] The present invention aims to provide a motorcycle engine fault detection method based on vibration effect monitoring. First, the engine's first vibration characteristic data, state characteristic data, and first position data are obtained from a sensor group; the first position data and the first vibration characteristic data are input into a position adjustment module, and the first position data is updated according to the output position adjustment coefficient to obtain second position data; then, the state characteristic data and the second vibration characteristic data corresponding to the second position data are input into a fault detection influence weight prediction module to obtain a fault detection influence weight; then, the second vibration characteristic data is classified using a multi-task fault diagnosis module to obtain fault characteristics and non-fault characteristics, and these characteristics and the fault detection influence weight are input into the module for analysis to obtain a fault detection value for fault judgment; this method can improve the accuracy and effectiveness of motorcycle engine fault detection.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A motorcycle engine fault detection method based on vibration effect monitoring, comprising:

[0008] Acquire first vibration characteristic data, state characteristic data and first position data of the engine from the sensor group;

[0009] Inputting the first position data and the corresponding first vibration characteristic data into a position adjustment module to obtain a position adjustment coefficient; updating the first position data using the position adjustment coefficient to obtain second position data;

[0010] Constructing a fault detection influence weight prediction module, inputting the state feature data and the second vibration feature data corresponding to the second position data into the fault detection influence weight prediction module to obtain a fault detection influence weight;

[0011] Using a multi-task fault diagnosis module to classify the second vibration feature data to obtain fault features and non-fault features; then, inputting the fault features, the non-fault features, and the fault detection influence weights into the multi-task fault diagnosis module for analysis to obtain a fault detection value; the fault detection value is used for fault determination;

[0012] The fault detection value is compared with a fault detection threshold to obtain a fault detection result.

[0013] Furthermore, the first vibration characteristic data and the second vibration characteristic data include fault vibration characteristic data and non-fault vibration characteristic data; and the state characteristic data includes: speed data, load data and temperature data.

[0014] Furthermore, the first position data and the corresponding first vibration characteristic data are input into a position adjustment module to obtain a position adjustment coefficient; and the first position data is updated using the position adjustment coefficient to obtain the second position data. The process includes:

[0015] Training the position adjustment module using historical position data and historical vibration feature data to obtain a pre-trained position adjustment module;

[0016] Wherein, a position penalty term is introduced into the loss function of the position adjustment module to penalize position data that exceeds a limit range;

[0017] Inputting the first position data and the corresponding first vibration characteristic data into the pre-trained position adjustment module for prediction to obtain the position adjustment coefficient;

[0018] The first position data is adjusted using the position adjustment coefficient to obtain the second position data.

[0019] Furthermore, the state feature data and the second vibration feature data corresponding to the second position data are input into the fault detection influence weight prediction module, and the process of obtaining the fault detection influence weight is as follows:

[0020] Constructing the fault detection impact weight prediction module, including: an input processing layer, a feature extraction layer, an impact weight prediction layer and a fusion output layer;

[0021] Utilizing the input processing layer to receive the state feature data and the second vibration feature data, and performing frequency domain transformation and wavelet transformation on the second vibration feature data to obtain time domain feature data, frequency domain feature data, and wavelet feature data;

[0022] Inputting the state feature data, the time domain feature data, the frequency domain feature data and the wavelet feature data into the feature extraction layer for feature extraction to obtain state features, time domain features, frequency domain features and wavelet features;

[0023] The time domain feature, the frequency domain feature and the wavelet feature are respectively combined with the state feature, and are respectively input into the corresponding sub-network in the influence weight prediction layer for weight prediction to obtain the time domain prediction weight, the frequency domain prediction weight and the wavelet prediction weight;

[0024] The time domain prediction weight, the frequency domain prediction weight, and the wavelet prediction weight are processed using the fusion output layer to obtain the fault detection influence weight.

[0025] Furthermore, the multi-task fault diagnosis module includes: an input layer, a feature extraction and classification layer, a fault analysis layer, a weight correction layer and an output layer;

[0026] The input layer is used to transform input data into data features;

[0027] The feature extraction and classification layer is used to extract and classify the data features;

[0028] The fault analysis layer is used to output fault analysis results in combination with classification features;

[0029] The weight correction layer is used to correct the fault analysis result using the fault detection impact weight;

[0030] The output layer is used to map the corrected fault analysis result into the fault detection value.

[0031] Furthermore, the second vibration characteristic data is classified using a multi-task fault diagnosis module to obtain fault characteristics and non-fault characteristics; then, the fault characteristics, the non-fault characteristics, and the fault detection influence weight are input into the multi-task fault diagnosis module for analysis to obtain a fault detection value. The process includes:

[0032] Processing the second vibration feature data using the input layer and feature extraction and classification layer of the multi-task fault diagnosis module to obtain the fault feature and the non-fault feature;

[0033] Inputting the fault characteristics and the non-fault characteristics into the fault analysis layer of the multi-task fault diagnosis module for processing to obtain fault analysis characteristics;

[0034] Inputting the fault analysis feature and the fault detection impact weight into the weight correction layer of the multi-task fault diagnosis module for processing to obtain a fault analysis correction feature;

[0035] The fault analysis and correction features are transformed using the output layer of the multi-task fault diagnosis module to obtain the fault detection value; wherein the fault detection value is used for fault judgment.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention proposes a position adjustment method for adjusting the position of a sensor; this method utilizes a position adjustment module to learn the relationship between the sensor position and the complexity of vibration characteristic data in historical data, and introduces a position penalty term in the loss function to limit the actual range of the output position data; the position adjustment coefficient obtained by the position adjustment module is used to adjust the real-time position, which can increase the proportion of fault vibration characteristic data in complex environments, reduce interference, and facilitate subsequent data classification, thereby improving the accuracy and effectiveness of motorcycle engine fault detection.

[0038] 2. The present invention proposes a fault detection influence prediction method for obtaining the fault detection influence weight; the method utilizes the fault detection influence weight prediction module to divide the input data into time domain, frequency domain and time-frequency domain features, and then uses the influence weight prediction layer to predict the degree of influence of the real-time state on the vibration characteristics in different domains and fuse them to obtain the fault detection influence weight; the fault detection influence weight can effectively reflect the influence of the operating state on the detection during the fault detection process, and thus the combination of this weight can improve the accuracy and effectiveness of motorcycle engine fault detection.

[0039] 3. The present invention proposes a fault detection method for fault judgment; the method first classifies vibration feature data into fault features and non-fault features, and then uses the fault analysis layer of the multi-task fault diagnosis module to process it to obtain fault analysis features; the fault analysis features are corrected in combination with the fault analysis features and the fault detection influence weight to obtain a fault detection value; the fault detection value can reflect the degree of fault under the actual operating state, and using this value for fault judgment is conducive to improving the accuracy and effectiveness of motorcycle engine fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of a flow chart of a motorcycle engine fault detection method based on vibration effect monitoring according to the present invention;

[0041] Figure 2 This is a schematic diagram of the structure of the fault detection impact weight prediction module of the present invention;

[0042] Figure 3 This is a flow chart of the multi-task fault diagnosis module of the present invention obtaining fault detection values. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] See also Figures 1 to 3 The present invention provides a motorcycle engine fault detection method based on vibration effect monitoring, and the technical solution is as follows:

[0045] Example 1: In order to reduce the operation and maintenance costs of motorcycle engines, a company uses a motorcycle engine fault detection method based on vibration effect monitoring proposed in this invention. The process of this method is shown in the following figure: Figure 1 As shown, specifically including:

[0046] Acquire first vibration characteristic data, state characteristic data and first position data of the engine from the sensor group;

[0047] Furthermore, the first vibration characteristic data and the second vibration characteristic data include fault vibration characteristic data and non-fault vibration characteristic data; the state characteristic data includes: speed data, load data and temperature data;

[0048] Furthermore, fault vibration characteristic data is generated by faults in components such as pistons, crankshafts, and valves; non-fault vibration characteristic data includes data caused by periodic motion (such as piston reciprocation, crankshaft rotation, etc.) and combustion impact under normal conditions, as well as interference data generated by the external environment (such as road conditions, aerodynamic noise, external machinery, etc.);

[0049] Furthermore, the position data is three-dimensional coordinate data of a sensor installed on the device.

[0050] By using the multi-dimensional data of the sensor, including position data, state characteristic data and vibration characteristic data, a data basis is provided for subsequent sensor position adjustment and fault detection, thereby ensuring the accuracy and effectiveness of motorcycle engine fault detection.

[0051] Inputting the first position data and the corresponding first vibration characteristic data into a position adjustment module to obtain a position adjustment coefficient; updating the first position data using the position adjustment coefficient to obtain second position data;

[0052] Furthermore, the first position data and the corresponding first vibration characteristic data are input into a position adjustment module to obtain a position adjustment coefficient; and the first position data is updated using the position adjustment coefficient to obtain the second position data. The process includes:

[0053] The position adjustment module is trained using the historical position data and the historical vibration characteristic data to obtain a pre-trained position adjustment module;

[0054] Among them, a position penalty term is introduced into the loss function of the position adjustment module to penalize position data that exceeds the limit range;

[0055] Inputting the first position data and the corresponding first vibration characteristic data into a pre-trained position adjustment module for prediction to obtain a position adjustment coefficient;

[0056] Adjusting the first position data using the position adjustment coefficient to obtain second position data;

[0057] Furthermore, the position adjustment module uses a GNN model. The implementation process of this module is as follows: each sensor position and each vibration feature is regarded as a node; connections are established between sensor position nodes and vibration feature nodes, connections between vibration feature nodes, and connections between sensor position nodes; the constructed graph structure is input into the GNN model to obtain the position adjustment coefficient;

[0058] Furthermore, the sensor position is a three-dimensional coordinate (x, y, z), so each position node has three coordinate features;

[0059] Furthermore, GNN models include: graph convolutional networks, graph attention networks, gated graph neural networks, etc.

[0060] Furthermore, the loss function of the position adjustment module includes: maximizing the fault feature energy ratio, minimizing information entropy, and the position penalty term; the position penalty term can be expressed as:

[0061] ;

[0062] in, is the position penalty term; Represents the number of training samples for location data; represents the maximization operation; Represents the coordinate value of the i-th sample on the j-axis; Indicates the maximum coordinate value of the i-th sample on the j-axis; Indicates the minimum coordinate value of the i-th sample on the j-axis.

[0063] By using the position adjustment coefficient obtained by the position adjustment module to adjust the real-time position, the proportion of fault vibration characteristic data can be increased in complex environments, interference can be reduced, and it is beneficial to subsequent data classification, thereby improving the accuracy and effectiveness of motorcycle engine fault detection.

[0064] Constructing a fault detection influence weight prediction module, inputting the state feature data and the second vibration feature data corresponding to the second position data into the fault detection influence weight prediction module to obtain the fault detection influence weight;

[0065] Furthermore, the state feature data and the second vibration feature data corresponding to the second position data are input into the fault detection influence weight prediction module, and the process of obtaining the fault detection influence weight is as follows:

[0066] Construct a fault detection impact weight prediction module, the structure of which is as follows Figure 2 As shown, it includes: input processing layer, feature extraction layer, influence weight prediction layer and fusion output layer;

[0067] Using the input processing layer to receive the state feature data and the second vibration feature data, and performing frequency domain transformation and wavelet transformation on the second vibration feature data to obtain time domain feature data, frequency domain feature data and wavelet feature data;

[0068] Inputting state feature data, time domain feature data, frequency domain feature data and wavelet feature data into the feature extraction layer for feature extraction to obtain state features, time domain features, frequency domain features and wavelet features;

[0069] Combine the time domain features, frequency domain features and wavelet features with the state features respectively, and input them into the corresponding sub-networks in the influence weight prediction layer for weight prediction, and obtain the time domain prediction weight, frequency domain prediction weight and wavelet prediction weight;

[0070] The fusion output layer is used to process the time domain prediction weight, frequency domain prediction weight and wavelet prediction weight to obtain the fault detection influence weight;

[0071] Furthermore, the input processing layer obtains time domain feature data, frequency domain feature data and wavelet feature data by directly extracting from the original data, performing frequency domain transformation using fast Fourier transform and decomposing using continuous wavelet transform.

[0072] Furthermore, the feature extraction layer uses MLP to extract features from the state feature data, and uses autoencoders to extract features from the time domain feature data, frequency domain feature data, and wavelet feature data respectively;

[0073] Furthermore, the influence weight prediction layer contains three sub-networks, which fuse the state features with the time domain features, frequency domain features, and wavelet features respectively and perform weight prediction on each fused feature; among them, the Transformer layer is used for feature fusion, and the MLP network is used for weight prediction;

[0074] Furthermore, the fusion output layer uses the gating mechanism and normalization function to output the expression of the fault detection influence weight:

[0075] ;

[0076] in, is the fault detection impact weight; represents the softmax function; 、 and Represent the gating coefficients of time domain, frequency domain and wavelet respectively, and the value range is between (0,1); 、 and Represent the prediction weights of time domain, frequency domain and wavelet respectively;

[0077] Furthermore, the gating coefficient is obtained by sequentially processing the prediction weights using MLP and Sigmoid functions.

[0078] By utilizing the fault detection influence weight prediction module to divide the input data into time domain, frequency domain and time-frequency domain features, the influence weight prediction layer is used to predict the influence of the real-time state on the vibration characteristics in different domains and fuse them to obtain the fault detection influence weight; this can effectively reflect the influence of the operating state on the detection during the fault detection process, and thus, combined with this weight, the accuracy and effectiveness of motorcycle engine fault detection can be improved.

[0079] The second vibration characteristic data is classified using a multi-task fault diagnosis module to obtain fault characteristics and non-fault characteristics. The fault characteristics, non-fault characteristics, and fault detection influence weights are then input into the multi-task fault diagnosis module for analysis to obtain a fault detection value. The fault detection value is used for fault determination.

[0080] Furthermore, the multi-task fault diagnosis module includes: an input layer, a feature extraction and classification layer, a fault analysis layer, a weight correction layer, and an output layer;

[0081] The input layer is used to transform the input data into data features using convolution kernels;

[0082] The feature extraction and classification layer is used to extract and classify data features;

[0083] Furthermore, the feature extraction and classification layer uses two parallel CNN classifiers, namely the fault classifier and the non-fault classifier, to identify fault features and non-fault features;

[0084] The fault analysis layer is used to output fault analysis results by combining classification features;

[0085] Furthermore, the fault analysis layer uses a ConvLSTM network to analyze fault features, while using non-fault features as guiding features to assist the network in fault analysis;

[0086] The weight correction layer is used to correct the fault analysis results using the fault detection impact weight;

[0087] Furthermore, the weight correction layer uses the fault detection impact weight to weight the output features of the fault analysis layer, and performs a residual connection between the output features of the fault analysis layer and the weighted features;

[0088] The output layer is used to map the corrected fault analysis results into fault detection values.

[0089] By utilizing the multi-task fault diagnosis module, the tasks of feature classification, feature analysis and feature analysis can be realized. The fault analysis features are corrected by combining the fault analysis features and the fault detection influence weights to obtain the fault detection value. The module can obtain the fault conditions that conform to the actual operating state, thereby improving the accuracy and effectiveness of motorcycle engine fault detection.

[0090] Furthermore, the process of obtaining fault detection values ​​by the multi-task fault diagnosis module is as follows: Figure 3 As shown, the details are as follows:

[0091] The second vibration feature data is processed using the input layer and feature extraction and classification layer of the multi-task fault diagnosis module to obtain fault features and non-fault features;

[0092] Input the fault features and non-fault features into the fault analysis layer of the multi-task fault diagnosis module for processing to obtain the fault analysis features;

[0093] The fault analysis features and fault detection impact weights are input into the weight correction layer of the multi-task fault diagnosis module for processing to obtain the fault analysis correction features;

[0094] The output layer of the multi-task fault diagnosis module is used to transform the fault analysis correction features to obtain a fault detection value; wherein the fault detection value is used for fault judgment.

[0095] The fault detection value is compared with the fault detection threshold to obtain a fault detection result.

[0096] To illustrate the fault detection value proposed in this invention, three data sets from different equipment models were randomly selected for fault detection testing, designated as Test 1, Test 2, and Test 3. The data included vibration characteristic data and state characteristic data collected by the repositioned sensor. Each set of data was fed into a pre-trained fault detection influence weight prediction module to obtain its own fault detection influence weight. The specific process of obtaining fault detection values ​​in the multi-task fault diagnosis module was combined to obtain the fault detection value for each group. The fault detection threshold was set to 0.75, and the fault detection value of each group was compared with the fault detection threshold to obtain the fault detection test results, as shown in Table 1.

[0097] Table 1 Fault detection test results

[0098] test Fault detection value Is there a fault? Test 1 0.59 no Test 2 0.78 yes Test Three 0.61 no

[0099] By using the fault detection value obtained by the multi-task fault diagnosis module and combining it with the fault detection influence weight, it can reflect the fault degree under the actual operating state. Using this value for fault judgment is conducive to improving the accuracy and effectiveness of motorcycle engine fault detection.

[0100] This embodiment proposes a motorcycle engine fault detection method based on vibration effect monitoring. The method first obtains first vibration characteristic data, state characteristic data, and first position data of the engine from a sensor group; inputs the first position data and the first vibration characteristic data into a position adjustment module, and updates the first position data based on the output position adjustment coefficient to obtain second position data; then, inputs the state characteristic data and second vibration characteristic data corresponding to the second position data into a fault detection influence weight prediction module to obtain a fault detection influence weight; then, uses a multi-task fault diagnosis module to classify the second vibration characteristic data to obtain fault characteristics and non-fault characteristics, and inputs these characteristics and the fault detection influence weight into the module for analysis to obtain a fault detection value for fault judgment. This method can improve the accuracy and effectiveness of motorcycle engine fault detection.

[0101] Example 2: This invention proposes a motorcycle engine fault detection method based on vibration effect monitoring. To further verify the effectiveness of the proposed fault detection impact weight prediction module and fault detection value acquisition process, this invention conducts module ablation tests and process comparison tests for different module selections and acquisition processes. This invention selects two companies, A and B, for each of these two sets of tests.

[0102] The present invention selects the historical data of Company A over the past three years as the data set of the module, wherein the data from the first two years is used as the training set of the model, and the data from the third year is used as the validation set; wherein the historical data includes historical state data and historical vibration characteristic data; the sampling of the model data set refers to the following rules: taking the week as the unit, extracting data for 7 days of each week; wherein, two groups of data are extracted each day in the morning, noon and evening; and all data come from the same type of motorcycle engine.

[0103] The present invention inputs the training sets collected from Company A into different fault detection influence weight prediction modules for training to obtain respective pre-training modules; then inputs the verification sets into each pre-training module to obtain the fault detection influence weight of each module; then, the fault detection influence weight is compared with the actual situation through manual verification to obtain the proportion of the fault detection influence weight of each module within a reasonable range.

[0104] Each test module is: the fault detection influence weight prediction module proposed in the present invention, recorded as module one; remove the gating mechanism in the fusion output layer, and directly merge and normalize the prediction weights of the time domain, frequency domain and wavelet, recorded as module two; only retain the time domain data for processing, that is, the feature extraction layer and the influence weight prediction layer have only one network and the output layer has no fusion part, recorded as module three.

[0105] The module effectiveness test results are shown in Table 2.

[0106] Table 2 Module effectiveness test results

[0107] Test Module The proportion within a reasonable range Model 1 91.08% Model 2 89.37% Model 3 85.91%

[0108] From the results in Table 2, it can be seen that the fault detection influence weight prediction module proposed in the present invention has better effectiveness test results than the test results of other modules; thus, it can be shown that the module proposed in the present invention can output accurate fault detection influence weights by combining multi-domain feature prediction and gated fusion, further improving the accuracy and effectiveness of motorcycle engine fault detection.

[0109] To further test the effectiveness of the fault detection value acquisition process, this example collected nearly one year of historical data from Company B. Following the same data sampling rules, the sampled data set was then divided into three groups for testing, designated as test sample one, test sample two, and test sample three. Each group of test data was processed according to process one and process two to obtain its own fault detection value. Process one is the process proposed in this invention for obtaining fault detection values ​​using a multi-task fault diagnosis module. Process two removes the feature classification operation and weight correction layer and directly inputs the vibration feature data into the fault analysis layer for output. Finally, manual verification was performed to verify the rationality of the fault detection values ​​obtained by each process. The process effectiveness test results are shown in Table 3.

[0110] Table 3 Process effectiveness test results

[0111] Test data Proportion within a reasonable range (Process 1) Proportion within a reasonable range (Process 2) Test sample 1 90.71% 88.64% Test sample 2 89.88% 87.43% Test sample three 90.29% 87.98%

[0112] From the results in Table 3, it can be seen that the effectiveness test results obtained by using Process 1, that is, the fault detection value acquisition scheme proposed in the present invention, are better than those obtained by using Process 2. This illustrates the necessity of correcting the fault analysis in combination with the fault detection influence weight, which can improve the accuracy and effectiveness of motorcycle engine fault detection.

[0113] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A motorcycle engine fault detection method based on vibration effect monitoring, characterized in that: include: Acquire first vibration characteristic data and state characteristic data and first position data of the engine from the sensor group; wherein the position data is three-dimensional coordinate data of the sensor installed on the equipment; Inputting the first position data and the corresponding first vibration characteristic data into a position adjustment module to obtain a position adjustment coefficient; updating the first position data using the position adjustment coefficient to obtain second position data; Constructing a fault detection influence weight prediction module, inputting the state feature data and the second vibration feature data corresponding to the second position data into the fault detection influence weight prediction module to obtain a fault detection influence weight; The first vibration characteristic data and the second vibration characteristic data include fault vibration characteristic data and non-fault vibration characteristic data; the state characteristic data includes: speed data, load data and temperature data; The process of obtaining the fault detection impact weight is as follows: Constructing the fault detection impact weight prediction module, including: an input processing layer, a feature extraction layer, an impact weight prediction layer and a fusion output layer; Utilizing the input processing layer to receive the state feature data and the second vibration feature data, and performing frequency domain transformation and wavelet transformation on the second vibration feature data to obtain time domain feature data, frequency domain feature data, and wavelet feature data; Inputting the state feature data, the time domain feature data, the frequency domain feature data and the wavelet feature data into the feature extraction layer for feature extraction to obtain state features, time domain features, frequency domain features and wavelet features; The time domain feature, the frequency domain feature and the wavelet feature are respectively combined with the state feature, and are respectively input into the corresponding sub-network in the influence weight prediction layer for weight prediction to obtain the time domain prediction weight, the frequency domain prediction weight and the wavelet prediction weight; Using the fusion output layer to process the time domain prediction weight, the frequency domain prediction weight, and the wavelet prediction weight to obtain the fault detection influence weight; Using a multi-task fault diagnosis module to classify the second vibration feature data to obtain fault features and non-fault features; then, inputting the fault features, the non-fault features, and the fault detection influence weights into the multi-task fault diagnosis module for analysis to obtain a fault detection value; the fault detection value is used for fault determination; The fault detection value is compared with a fault detection threshold to obtain a fault detection result.

2. A motorcycle engine fault detection method based on vibration effect monitoring according to claim 1, characterized in that: Inputting the first position data and the corresponding first vibration characteristic data into a position adjustment module to obtain a position adjustment coefficient; and updating the first position data using the position adjustment coefficient to obtain second position data includes: Training the position adjustment module using historical position data and historical vibration feature data to obtain a pre-trained position adjustment module; Wherein, a position penalty term is introduced into the loss function of the position adjustment module to penalize position data that exceeds a limit range; Inputting the first position data and the corresponding first vibration characteristic data into the pre-trained position adjustment module for prediction to obtain the position adjustment coefficient; The first position data is adjusted using the position adjustment coefficient to obtain the second position data.

3. The motorcycle engine fault detection method based on vibration effect monitoring according to claim 1 is characterized in that: The multi-task fault diagnosis module includes: an input layer, a feature extraction and classification layer, a fault analysis layer, a weight correction layer and an output layer; The input layer is used to transform input data into data features; The feature extraction and classification layer is used to extract and classify the data features; The fault analysis layer is used to output fault analysis results in combination with classification features; The weight correction layer is used to correct the fault analysis result using the fault detection impact weight; The output layer is used to map the corrected fault analysis result into the fault detection value.

4. The motorcycle engine fault detection method based on vibration effect monitoring according to claim 1, characterized in that: The process of classifying the second vibration feature data using a multi-task fault diagnosis module to obtain fault features and non-fault features; then, inputting the fault features, the non-fault features, and the fault detection influence weights into the multi-task fault diagnosis module for analysis to obtain a fault detection value includes: Processing the second vibration feature data using the input layer and feature extraction and classification layer of the multi-task fault diagnosis module to obtain the fault feature and the non-fault feature; Inputting the fault characteristics and the non-fault characteristics into the fault analysis layer of the multi-task fault diagnosis module for processing to obtain fault analysis characteristics; Inputting the fault analysis feature and the fault detection impact weight into the weight correction layer of the multi-task fault diagnosis module for processing to obtain a fault analysis correction feature; The fault analysis and correction features are transformed using the output layer of the multi-task fault diagnosis module to obtain the fault detection value; wherein the fault detection value is used for fault judgment.

Images