Fault early warning method based on vehicle full life cycle operation data

Abstract

The invention discloses a fault early warning method based on vehicle full life cycle operation data, and the method comprises the steps: carrying out the factory detection before equipment leaves a factory, transmitting the data to a data medium station, generating a data set A, collecting the data in the operation process of the equipment in real time after the equipment leaves the factory, transmitting the data to the data medium station, generating a data set B, and building an after-sales service work order system. The method comprises the following steps: recording and transmitting work orders collected on site to a data medium station, generating a data set C, establishing a full-life-cycle data association relationship, forming a full-life-cycle data set D of equipment by using the data set A, the data set B and the data set C, processing data of the full-life-cycle data set D, training by using an improved random forest model, and establishing a full-life-cycle data association relationship. The LSTM model is utilized to carry out model optimization comparison, equipment faults are predicted, and the method has the advantages of being capable of recognizing potential faults in advance, improving the equipment use efficiency and the maintenance management initiative and the like.

Description

Technical Field

[0001] The invention relates to the technical field of airport ground services, and in particular to a fault early warning method based on vehicle full life cycle operation data. Background Art

[0002] Airport ground special vehicles are indispensable and important equipment in airport operations. They are used to assist airlines and airport management departments in flight operations and ground services. In recent years, with the growth of global air travel and the expansion of airport facilities, the demand for airport ground special vehicles has also been increasing.

[0003] In traditional technology, the vehicle will only trigger an alarm after a failure occurs, which is retroactive and delayed. Maintenance and management are delayed, affecting the use of equipment and increasing the risk of airport operations.

[0004] Therefore, realizing early identification and early warning of potential failures of airport special vehicles is a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] The purpose of the present invention is to solve the above-mentioned deficiencies in the prior art and to provide a fault warning method based on vehicle operation data throughout its life cycle, which can identify potential faults in advance, improve equipment utilization efficiency and maintenance management initiative.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A fault warning method based on vehicle full life cycle operation data, characterized by:

[0008] Step S1: The equipment is tested before leaving the factory, and the data is transmitted to the data center to generate data set A;

[0009] Step S2: After leaving the factory, collect data during the operation of the equipment in real time, transmit the data to the data center, and generate data set B; build an after-sales service work order system, transmit the work order records collected on site to the data center, and generate data set C;

[0010] Step S3: Establish a full life cycle data association relationship, and form a device full life cycle data set D by combining data set A, data set B, and data set C;

[0011] Step S4: Processing the data of the full life cycle data set D;

[0012] Step S5: training using the improved random forest model;

[0013] Step S6: Use the LSTM model to perform model optimization comparison and predict equipment failure;

[0014] After the above steps, early warning can be achieved before equipment failure occurs, which can not only effectively avoid uncontrollable shutdowns, but also avoid excessive maintenance, improve equipment utilization, effectively increase equipment service life, and ensure efficient airport ground services.

[0015] After step S6 of the present invention, it is determined whether the model reaches the expected accuracy. If yes, it jumps to step S7; if not, it jumps to step S5;

[0016] Step S7: Send the warning information to the client;

[0017] Step S8: Model monitoring and updating, continuously monitoring and improving the prediction performance of the model.

[0018] In step S1 of the present invention, the CAN bus of the device is connected via a CAN acquisition board to obtain data of the CAN bus, and the detection content includes electrical detection and mechanical detection.

[0019] In step S3 of the present invention, the full life cycle data set D includes factory inspection data, historical operation data before equipment failure, operation data after equipment failure, historical equipment failure records and historical equipment maintenance plans.

[0020] In step S3 of the present invention, a full life cycle data set D is constructed. The full life cycle data set D is a data warehouse, the vehicle unique ID is used as the associated unique ID primary key, and the data set A detected before leaving the factory is used as the correct data benchmark.

[0021] In step S3 of the present invention, the full life cycle data set D is a data aggregation layer built in a wide table mode.

[0022] In step S4 of the present invention, the method for processing the data of the full life cycle data set D is:

[0023] Step S4.1: missing value processing;

[0024] For continuous numerical data, the median interpolation method is used, and for switch data, the mode interpolation method is used;

[0025]

[0026] where x i is the original data, x′ i is the processed data, Media(X) is the median of feature X, and Mode(X) is the mode of feature X;

[0027] The processed data set is D′, which now forms an m*n two-dimensional matrix X=(X1,X2,…,X n );

[0028]

[0029] Step S4.2: data processing;

[0030] Step S4.2.1: Use Z-score to standardize the data. The formula of Z-score matrix Z is:

[0031]

[0032] Where Z is the normalized m*n two-dimensional matrix, X is the two-dimensional matrix processed in step S4.1, μ=(μ1,μ2,…,μ n ) is the mean of each column feature, σ=(σ1,σ2,…,σ n ) is the standard deviation of each column feature;

[0033] Step S4.2.2: Calculate the covariance matrix C;

[0034]

[0035] Where n is the number of samples, Z T is the transpose of the normalized data;

[0036] Step S4.2.3: Calculate the eigenvalue λ and the eigenvector v;

[0037] Cv=λv

[0038] Step S4.2.4: Select the first k eigenvectors according to the size of the eigenvalue λ;

[0039] Step S4.2.5: construct the projection matrix W;

[0040] W=[v1,v2,…,vi]

[0041] Where vi is the i-th selected eigenvector;

[0042] Step S4.2.6: Data dimensionality reduction;

[0043] Project the two-dimensional matrix X processed in step S4.1 into the new feature space to obtain the reduced-dimensional data Y, which is an m*p two-dimensional matrix, Y=(y1,y2,…,y p );

[0044] Y=ZW

[0045] Step S4.3: Calculate the Gini index of each feature;

[0046] Step S4.3.1: Take the fault code as the multi-objective variable G = (g1, g2, ..., g q), calculate the Gini index;

[0047]

[0048] Where Y is the dataset after dimension reduction in step S4.2.6, G is the target fault classification, and p i is the fault y in the data set Y i The sample proportion of each class;

[0049] Step S4.3.2: For each feature X i Divide the dataset, X i Different values ​​of a k As a subset Y j , respectively calculate the Gini impurity Gini(X i =a k ), calculate its Gini index;

[0050]

[0051] Among them, k is the number of feature values, |Y j | is the data subset Y j , |Y| is the number of samples in the dataset after dimensionality reduction in step S4.2.6;

[0052] Step S4.3.3: Calculate the importance of features;

[0053] Importance(X i )=Gini(Y)-Gini split (X i )

[0054] Step S4.3.4: For each feature X i Calculate its Gini impurity and importance, and finally take the features with high Gini importance for modeling;

[0055] Data quality and analysis accuracy are ensured through data preprocessing.

[0056] In step S5 of the present invention,

[0057] Step S5.1: Based on the fault, a decision tree for the multi-classification problem is established, and a decision tree is constructed for each fault type;

[0058] Step S5.2: Perform hyperparameter tuning according to actual conditions, adjust the number of trees, maximum depth, minimum sample split tree, k-fold cross validation, and optimize model performance;

[0059]

[0060] Among them, Θ represents the set of all parameter combinations, and θ is one of the parameter combinations;

[0061] Step S5.3: Save the parameter combination that performs best in the validation set, and the model training is completed;

[0062] Step S5.4: Evaluate the trained model. Common evaluation criteria include accuracy, precision, recall, and F1 value. These evaluation criteria are combined with k-fold cross validation.

[0063] The formula for k-fold cross validation is as follows:

[0064]

[0065] Accuracy i is the accuracy of the i-th fold;

[0066] The calculation formulas for each indicator are as follows:

[0067]

[0068]

[0069]

[0070]

[0071] Among them, TP is a true positive example, TN is a true negative example, FP is a false positive example, and FN is a false negative example;

[0072] Intelligent predictive analysis of the processed data can be realized, which can identify and warn potential failures of special vehicles at airports in advance, thereby greatly improving the efficiency of equipment use and the initiative of maintenance management, and ensuring the safety of airport operations.

[0073] In step S6 of the present invention, the maximum number of iterations is set to M, and batch_size, timesteps, and features are appropriately adjusted to optimize the model.

[0074] In step S8 of the present invention, the prediction performance of the model is continuously monitored, with a focus on the prediction accuracy and error distribution of the model. Once model drift occurs, the process jumps to step S5 to retrain the model.

[0075] The beneficial effects of the present invention are: early warning before equipment failure can be achieved, which can effectively avoid uncontrollable shutdowns and excessive maintenance, improve equipment utilization, effectively increase equipment service life, and ensure efficient airport ground service work. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is an early warning flow chart of the full life cycle operation data of airport special vehicles of the present invention.

[0077] Figure 2 This is a factory inspection parameter table for airport special vehicles of the present invention.

[0078] Figure 3 These are some of the parameters included in the airport special vehicle full life cycle data set D of the present invention. DETAILED DESCRIPTION

[0079] The present invention is described below in conjunction with the accompanying drawings and embodiments.

[0080] As shown in the accompanying drawings, a fault warning method based on vehicle full life cycle operation data includes the following steps:

[0081] Step S1: The device has multiple built-in and external sensors to monitor the device status in real time. The device is connected to an external CAN acquisition board, which is used to connect the CAN bus of the device to obtain the data of the CAN bus. The device is tested before leaving the factory. The test content includes electrical test and mechanical test. The data is transmitted to the data center through the network to generate data set A. The factory test parameters include but are not limited to Figure 2 Including battery voltage, turning radius, tire pressure, hydraulic system pressure, etc.

[0082] Step S2: After leaving the factory, the data on the CAN bus during the operation of the equipment is collected in real time through the collection device installed on the airport special equipment, and the data is transmitted to the data center through the network to generate data set B; an after-sales service work order system is built to transmit the work order records collected on site to the data center through the network to generate data set C;

[0083] Step S3: Establish the data association relationship of the whole life cycle, and form the equipment life cycle data set D by combining data set A, data set B, and data set C. The whole life cycle data set D includes factory inspection data, historical operation data before equipment failure, operation data after equipment failure, historical equipment failure records, and historical equipment maintenance plans. The whole life cycle data set D is constructed as a data warehouse, and the vehicle unique ID is used as the associated unique ID primary key. The data set A detected before leaving the factory is used as the correct data benchmark. The whole life cycle data set D is a data aggregation layer built in a wide table mode, including some parameters and fault representatives such as Figure 3 As shown;

[0084] Step S4: Process the data of the full life cycle data set D, and proceed to step S5 after processing;

[0085] Step S4.1: missing value processing;

[0086] For continuous numerical data (such as temperature, speed, etc.), the median interpolation method is used. When there may be outliers in the data, median interpolation is more robust. For switch data (such as on / off, normal / abnormal, etc.), the mode interpolation method is used to maintain the classification of categories;

[0087]

[0088] where x i is the original data, x′ i is the processed data, Media(X) is the median of feature X, and Mode(X) is the mode of feature X;

[0089] The processed data set is D′, which now forms an m*n two-dimensional matrix X=(X1,X2,…,X n );

[0090]

[0091] Step S4.2: data processing;

[0092] Step S4.2.1: Use Z-score to standardize the data to ensure that each feature has the same scale. The formula of the Z-score matrix Z is:

[0093]

[0094] Where Z is the normalized m*n two-dimensional matrix, X is the two-dimensional matrix processed in step S4.1, μ=(μ1,μ2,…,μ n ) is the mean of each column feature, σ=(σ1,σ2,…,σ n ) is the standard deviation of each column feature;

[0095] Step S4.2.2: Calculate the covariance matrix C;

[0096]

[0097] Where n is the number of samples, Z T is the transpose of the normalized data;

[0098] Step S4.2.3: Calculate the eigenvalue λ and the eigenvector v;

[0099] Cv=λv

[0100] Step S4.2.4: Select the first k eigenvectors according to the size of the eigenvalue λ;

[0101] Step S4.2.5: construct the projection matrix W;

[0102] W=[v1,v2,…,vi]

[0103] Where vi is the i-th selected eigenvector;

[0104] Step S4.2.6: Data dimensionality reduction;

[0105] Project the two-dimensional matrix X processed in step S4.1 into the new feature space to obtain the reduced-dimensional data Y, which is an m*p two-dimensional matrix, Y=(y1,y2,…,y p );

[0106] Y=ZW

[0107] Step S4.3: Calculate the Gini index of each feature;

[0108] Step S4.3.1: Take the fault code as the multi-objective variable G = (g1, g2, ..., g q ), calculate the Gini index;

[0109]

[0110] Where Y is the dataset after dimension reduction in step S4.2.6, G is the target fault classification, and p i is the fault y in the data set Y i The sample proportion of each class;

[0111] Step S4.3.2: For each feature X i Divide the dataset, X i Different values ​​of a k As a subset Y j , respectively calculate the Gini impurity Gini(X i =a k ), calculate its Gini index;

[0112]

[0113] Among them, k is the number of feature values, |Y j | is the data subset Y j , |Y| is the number of samples in the dataset after dimensionality reduction in step S4.2.6;

[0114] Step S4.3.3: Calculate the importance of features;

[0115] Importance(X i )=Gini(Y)-Gini split (X i )

[0116] Step S4.3.4: For each feature X i Calculate its Gini impurity and importance, and finally take the features with high Gini importance for modeling;

[0117] Preprocess the data to ensure data quality and accuracy of analysis;

[0118] Step S5: Perform training using the improved random forest model, and then proceed to step S6 after training;

[0119] Step S5.1: Based on the fault, a decision tree for the multi-classification problem is established, and a decision tree is constructed for each fault type;

[0120] Step S5.2: Perform hyperparameter tuning according to actual conditions, adjust the number of trees, maximum depth, minimum sample split tree, k-fold cross validation, and optimize model performance;

[0121]

[0122] Among them, Θ represents the set of all parameter combinations, and θ is one of the parameter combinations;

[0123] Step S5.3: Save the parameter combination that performs best in the validation set, and the model training is completed;

[0124] Step S5.4: Evaluate the trained model. Common evaluation criteria include accuracy, precision, recall, and F1 value. These evaluation criteria are combined with k-fold cross validation.

[0125] The formula for k-fold cross validation is as follows:

[0126]

[0127] Accuracy i is the accuracy of the i-th fold;

[0128] The calculation formulas for each indicator are as follows:

[0129]

[0130]

[0131]

[0132]

[0133] Among them, TP is a true positive example, TN is a true negative example, FP is a false positive example, and FN is a false negative example;

[0134] The intelligent prediction and analysis of the processed data can be realized, which can identify and warn the potential failures of special vehicles at the airport in advance, thus greatly improving the efficiency of equipment use and the initiative of maintenance management, and ensuring the safety of airport operations;

[0135] Step S6: Use the LSTM model to perform model optimization comparison, set the maximum number of iterations to M, and appropriately adjust batch_size, timesteps, and features to optimize the model. In this embodiment, the maximum number of iterations is 1000. Predict equipment failures and determine whether the model reaches the expected accuracy. If yes, jump to step S7; if not, jump to step S5.

[0136] Step S7: Send the warning information to the client, and proceed to step S8;

[0137] Step S8: Model monitoring and updating, continuously monitoring and improving the prediction performance of the model, focusing on the prediction accuracy and error distribution of the model. Once model drift occurs, jump to step S5 and retrain the model;

[0138] After model training in this embodiment, the accuracy can reach 83%;

[0139] After the above steps, accurate early warning can be achieved before equipment failure occurs, which can not only effectively avoid uncontrollable shutdowns, but also avoid excessive maintenance, improve equipment utilization, effectively increase equipment service life, and ensure efficient airport ground services.

[0140] In step S5, k-fold cross-validation means that the device has N samples, the N samples are averaged and randomly divided into k equal parts, the first sample is used as a validation set, and the remaining k-1 samples are used as training sets to calculate the accuracy once, then the second sample is used as a validation set, and the remaining k-1 samples are used as training sets to calculate the accuracy again, until each sample has participated in the calculation as a validation set, and the k accuracy rates obtained are averaged as the cross-validation accuracy.

[0141] This embodiment takes 5-fold cross-validation as an example. The number of samples is 100. The 100 samples are averaged and randomly divided into five equal parts. The first sample is used as a validation set, and the remaining four samples are used as training sets to calculate the accuracy once. Next, the second sample is used as a validation set, and the remaining four samples are used as training sets to calculate the accuracy again, until each sample has participated in the calculation as a validation set, and the average of the five accuracy rates is taken as the cross-validation accuracy.

[0142] In this embodiment, the data center adopts a microservice architecture and uses message middleware to ensure high-speed data flow and accurate analysis.

[0143] This application is not limited to early warning of special vehicles at airports before they fail, but is also applicable to vehicles in other scenarios.

Claims

1. A fault warning method based on vehicle full life cycle operation data, characterized in that: Step S1: The equipment is tested before leaving the factory, and the data is transmitted to the data center to generate data set A; Step S2: After leaving the factory, collect data during the operation of the equipment in real time, transmit the data to the data center, and generate data set B; build an after-sales service work order system, transmit the work order records collected on site to the data center, and generate data set C; Step S3: Establish a full life cycle data association relationship, and form a device full life cycle data set D by combining data set A, data set B, and data set C; Step S4: Processing the data of the full life cycle data set D; Step S5: training using the improved random forest model; Step S6: Use the LSTM model to perform model optimization and comparison to predict equipment failures.

2. A fault warning method based on vehicle full life cycle operation data according to claim 1, characterized in that: After step S6, determine whether the model reaches the expected accuracy, if yes, jump to step S7, if no, jump to step S5; Step S7: Send the warning information to the client; Step S8: Model monitoring and updating, continuously monitoring and improving the prediction performance of the model.

3. A fault warning method based on vehicle full life cycle operation data according to claim 1 or 2, characterized in that: In step S1, the CAN bus of the device is connected via a CAN acquisition board to obtain data of the CAN bus, and the detection content includes electrical detection and mechanical detection.

4. A fault warning method based on vehicle full life cycle operation data according to claim 3, characterized in that: In step S3, the full life cycle data set D includes factory inspection data, historical operation data before equipment failure, operation data after equipment failure, historical equipment failure records and historical equipment maintenance plans.

5. A fault warning method based on vehicle full life cycle operation data according to claim 1, 2 or 4, characterized in that: In step S3, a full life cycle data set D is constructed. The full life cycle data set D is a data warehouse. The vehicle unique ID is used as the associated unique ID primary key, and the data set A detected before leaving the factory is used as the correct data benchmark.

6. A fault warning method based on vehicle full life cycle operation data according to claim 5, characterized in that: In step S3, the full life cycle dataset D is a data aggregation layer built in a wide table mode.

7. The fault warning method based on vehicle full life cycle operation data according to claim 5 is characterized by: In step S4, the method for processing the data of the full life cycle data set D is as follows: step S4.1: missing value processing; For continuous numerical data, the median interpolation method is used, and for switch data, the mode interpolation method is used; where x i is the original data, x i ′ is the processed data, Median(X) is the median of feature X, and Mode(X) is the mode of feature X; The processed data set is D′, which now forms an m*n two-dimensional matrix X=(X1,X2,…,X n ); Step S4.2: data processing; Step S4.2.1: Use Z-score to standardize the data. The formula of Z-score matrix Z is: Where Z is the normalized m*n two-dimensional matrix, X is the two-dimensional matrix processed in step S4.1, μ=(μ1,μ2,…,μ n ) is the mean of each column feature, σ=(σ1,σ2,…,σ n ) is the standard deviation of each column feature; Step S4.2.2: Calculate the covariance matrix C; Where n is the number of samples, Z T is the transpose of the normalized data; Step S4.2.3: Calculate the eigenvalue λ and the eigenvector v; Cv=λv Step S4.2.4: Select the first k eigenvectors according to the size of the eigenvalue λ; Step S4.2.5: construct the projection matrix W; W=[v1,v2,…,vi] Where vi is the i-th selected eigenvector; Step S4.2.6: Data dimensionality reduction; Project the two-dimensional matrix X processed in step S4.1 into the new feature space to obtain the reduced-dimensional data Y, which is an m*p two-dimensional matrix, Y=(y1,y2,…,y p ); Y=ZW Step S4.3: Calculate the Gini index of each feature; Step S4.3.1: Take the fault code as the multi-objective variable G = (g1, g2, ..., g q ), calculate the Gini index; Where Y is the dataset after dimension reduction in step S4.2.6, G is the target fault classification, and p i is the fault y in the data set Y i The sample proportion of each class; Step S4.3.2: For each feature X i Divide the dataset, X i Different values ​​of a k As a subset Y j , respectively calculate the Gini impurity Gini(X i =a k ), calculate its Gini index; Among them, k is the number of feature values, |Y j | is the data subset Y j , |Y| is the number of samples in the dataset after dimensionality reduction in step S4.2.6; Step S4.3.3: Calculate the importance of features; Importance(X i )=Gini(Y)-Gini split (X i ) Step S4.3.4: For each feature X i Calculate its Gini impurity and importance, and finally, take the features with high Gini importance for modeling.

8. The fault warning method based on vehicle full life cycle operation data according to claim 5 is characterized by: In step S5, Step S5.1: Based on the fault, a decision tree for the multi-classification problem is established, and a decision tree is constructed for each fault type; Step S5.2: Perform hyperparameter tuning according to actual conditions, adjust the number of trees, maximum depth, minimum sample split tree, k-fold cross validation, and optimize model performance; Among them, Θ represents the set of all parameter combinations, and θ is one of the parameter combinations; Step S5.3: Save the parameter combination that performs best in the validation set, and the model training is completed; Step S5.4: Evaluate the trained model. Common evaluation criteria include accuracy, precision, recall, and F1 value. These evaluation criteria are combined with k-fold cross validation. The formula for k-fold cross validation is as follows: Accuracy i is the accuracy of the i-th fold; The calculation formulas for each indicator are as follows: Among them, TP is a true positive example, TN is a true negative example, FP is a false positive example, and FN is a false negative example.

9. A fault warning method based on vehicle full life cycle operation data according to claim 1 or 2 or 4 or 6 or 7 or 8, characterized in that: In step S6, the maximum number of iterations is set to M, and batch_size, timesteps, and features are appropriately adjusted to optimize the model.

10. The fault warning method based on vehicle full life cycle operation data according to claim 2 is characterized in that: In step S8, the prediction performance of the model is continuously monitored, focusing on the prediction accuracy and error distribution of the model. Once model drift occurs, jump to step S5 and retrain the model.

Images