Vehicle-mounted power unit health management method and system

Abstract

The invention discloses a vehicle-mounted power unit health management method and system, and the method comprises the steps: collecting the operation data of a vehicle-mounted power unit in real time, carrying out the feature extraction of the collected data, and carrying out the abnormal value removal, interpolation and noise reduction of the collected data through a Z-Score method and a lightweight convolutional neural network; the method comprises the following steps: constructing a BKA-Pathform fault diagnosis model, optimizing Pathform model hyper-parameters based on a meta learning framework by using a BKA optimization algorithm, calculating a performance degradation index according to voltage polarity change, fusing the performance degradation index with a diagnosis result to form a health index, comparing the health index with a set threshold value, if the performance degradation index exceeds the set threshold value, giving an alarm to a user side, and storing data locally at the same time. And generating a hash value based on a block chain technology and storing the hash value in a cloud platform. According to the invention, real-time health monitoring and early warning of the vehicle-mounted power unit can be realized, and the safety and reliability of vehicle operation are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle power unit health management, and particularly relates to a vehicle power unit health management method and system. Background Art

[0002] A vehicle power unit includes an engine, a transmission and related drive systems, and is widely used in multiple fields such as automobiles and industry. However, with the continuous in-depth development of mechanization, as an important part of the power source, the importance of the vehicle power unit has become increasingly prominent. Especially in high-precision and high-speed power systems, the safety and reliability issues of the vehicle power unit have become the key factors restricting the further improvement of the performance of the entire power system. Vehicles are in a complex and changeable environment for a long time, and the load and pressure borne by the power system also increase accordingly, which further exacerbates the demand for the safety and reliability of the vehicle power unit. In recent years, with the increasing global attention to environmental protection and sustainable development, automobile manufacturers are facing the dual pressures of reducing emissions and improving energy efficiency. New energy vehicles have become the mainstream means of transportation, and the rise of new energy vehicles has put forward new requirements for vehicle power units. Electric vehicles use high-efficiency power battery packs and electric motors to replace traditional internal combustion engines, while hybrid vehicles need to achieve smooth power switching and collaborative work between internal combustion engines and electric motors.

[0003] Traditional vehicle power unit health management methods mainly rely on the experience judgment of operators and instrument detection. There are differences in the structure, materials and performance of vehicle power units of different models. Therefore, manual diagnosis can only be carried out one by one according to experience and specific models. This method has hysteresis, low efficiency, and is easily affected by subjective factors, resulting in inaccurate diagnosis results. In addition, due to the technical barriers and data non-sharing between different brand automobile manufacturers, many same faults need to be comprehensively and repeatedly checked, wasting a lot of manpower and material resources. Secondly, the vehicle power unit consists of many components. Most of the existing health management methods only use a single digital model for a single component and cannot manage the complete vehicle power unit, and it is difficult to give accurate diagnosis results for complex and changeable fault phenomena.

[0004] Therefore, a health management method that can cope with complex working conditions and is aimed at the complete vehicle power unit is needed. At the same time, a health management method that can utilize cloud platforms such as vehicle networking and blockchain to improve the health management efficiency and save resources is needed. This method should be able to comprehensively consider various fault factors, utilize advanced sensor technologies and data analysis methods to achieve rapid and accurate diagnosis of vehicle power unit faults, and provide a strong guarantee for the safe and stable operation of the power system. Summary of the Invention

[0005] Objective of the Invention: The present invention provides a method and system for health management of an in-vehicle power unit, realizing intelligent processing of in-vehicle power unit data and accurate identification of fault types.

[0006] Technical Solution: A method for health management of an in-vehicle power unit according to the present invention includes the following steps:

[0007] (1) Real-time collect the operation data of the in-vehicle power unit;

[0008] (2) Extract features from the collected data, and remove outliers, interpolate, and denoise the collected data through the Z-Score method and a lightweight convolutional neural network;

[0009] (3) Construct a BKA-Pathformer fault diagnosis model; the model is based on a meta-learning framework, uses the Black Kite Optimization Algorithm BKA to optimize the hyperparameters of the Pathformer model for fault diagnosis; for voltage polarization changes, generate a performance degradation index, fuse the fault diagnosis result with the performance degradation index to obtain a health index, and output the diagnosis result;

[0010] (4) Store the diagnosis result locally, and at the same time, based on blockchain technology, generate a hash value, store the hash value in the cloud to verify data integrity and security; and send the result to the user.

[0011] Further, the operation data of the in-vehicle power unit in step (1) includes a temperature sensor, a piezoelectric pressure sensor, a vehicle speed sensor, and a magnetic current sensor.

[0012] Further, the implementation process of step (2) is as follows:

[0013] (21) Extract features from the collected data, perform anomaly marking in the frequency domain, time domain, and Z-Score using sliding window analysis, and normalize and splice the features to use as the input of the lightweight convolutional neural network;

[0014] (22) Perform global average pooling through the lightweight convolutional neural network to reduce the feature dimension;

[0015] (23) Based on the meta-learning framework, combine online gradient descent and backpropagation to update the parameter weights of the lightweight convolutional neural network;

[0016] (24) Remove data noise caused by communication through the optimized lightweight convolutional neural network.

[0017] Further, the hyperparameters of the Pathformer model in step (3) include batch size, learning rate, and regularization coefficient.

[0018] Furthermore, the process of using BKA to optimize the hyperparameters of the Pathformer model in step (3) is as follows:

[0019] S1: Random initialization:

[0020] X i = BK lb + rand(BK ub - BK lb )

[0021] In the formula, i is an integer between 1 and pop, where pop is the total number of the potential parameter set, BK ub and BK lb are the lower and upper bounds of the i-th parameter group in the j-th dimension, and rand is a random number between [0, 1];

[0022] During the initialization process, BKA selects the parameter group with the best composite fitness value F(X i ) and records it as X L . This group of parameters is considered the optimal group. The false alarm rate FAR and the miss rate LR are weighted to form a composite fitness function, and the formula is:

[0023] F best = min(F(X i ))

[0024] X L = X(find(F best == F(X i )))

[0025] F(X i ) = ω1FAR + ω2LR

[0026] ω1 + ω2 = 1

[0027]

[0028] In the formula, ω1 and ω2 are the weights of the false alarm rate and the miss rate, NF is the number of normal samples misreported as faults, NS is the total number of normal samples, NFR is the number of normal samples missed as faults, and NFT is the total number of faults;

[0029] S2: Global search and exploration of the parameter group, and the formula is as follows:

[0030]

[0031] In the formula, and is the position of the i-th group of parameters in the j-th dimension at the (t + 1)-th and t-th iteration steps, r is a random number between 0 and 1, p is a constant, T is the total number of iterations, and t is the number of iterations completed so far;

[0032] S3: When the parameter group that appears is better than the current optimal parameter group, replace the optimal parameter group and perform a specified search according to the search direction between the two parameter groups. The formula is as follows:

[0033]

[0034]

[0035] In the formula, is the parameter group with the smallest composite fitness value in the j-th dimension of the current t-th iteration, F i is the composite fitness value obtained by any parameter group in the t-th iteration, C(0, 1) is the Cauchy mutation;

[0036] S4: Use multiple samples for training and evaluation, use multiple different data sets, and repeat steps S1 to S3;

[0037] S5: Select the optimal parameter group through the results of multiple data sets, output the results, and perform fault diagnosis.

[0038] Furthermore, the implementation process of step (3) is as follows:

[0039] (31) Perform statistical analysis on the analyzed and processed data to obtain the fault diagnosis result. Obtain the performance degradation index of the on-vehicle power unit through the change of voltage polarization and calculate it. Integrate the two to obtain the final health index;

[0040] (32) Set different thresholds according to different conditions such as the model and service life of the vehicle; at the same time, adjust the threshold through the AHP (Analytic Hierarchy Process) according to the usage mode of the vehicle and the driving habits of the owner;

[0041] (33) Compare the fusion result with the set threshold to make the final health diagnosis result of the on-vehicle power unit.

[0042] An on-vehicle power unit health management system according to the present invention includes:

[0043] Data acquisition module: Through various sensors, collect the operation data of the on-vehicle power unit in real time in all directions;

[0044] Data preprocessing module: Remove outliers, noise, and perform interpolation on the collected data;

[0045] Health Diagnosis Module: It conducts fault diagnosis on the preprocessed data. Based on the meta-learning framework, it uses the Black-winged Kite Optimization Algorithm BKA to optimize the hyperparameters of the Pathformer model, trains and evaluates the model through multiple different datasets, and selects the optimal hyperparameter combination; for voltage polarization changes, it generates a performance degradation index, fuses the fault diagnosis result with the performance degradation index to obtain a health index, and outputs the diagnosis result;

[0046] Intelligent Chain Cloud-Local Hybrid Module: Based on the blockchain-local hybrid storage architecture, it combines blockchain technology and the local storage module and operates in coordination with the communication module; it stores the original data, the final diagnosis result, the health index, and the alarm status locally, and at the same time generates a hash value, which is stored in the cloud, i.e., the blockchain platform, for verifying whether the data has been modified;

[0047] Communication Module: It is used to connect local data acquisition and cloud information storage.

[0048] Furthermore, the data preprocessing module consists of the Z-Score method and a lightweight convolutional neural network. A lightweight convolutional neural network is deployed in the module for filtering, and the filtering parameters are adjusted in real time according to the noise characteristics. When the Z-Score method detects outliers, it increases the sampling rate of the sensor and triggers local data encryption storage, storing the data in the blockchain technology node to ensure the integrity of the abnormal segment.

[0049] Advantageous Effects: Compared with the prior art, the advantageous effects of the present invention are as follows: The present invention preprocesses the collected data through a lightweight convolutional neural network and the Z-Score method, filters the data through the lightweight convolutional neural network, and dynamically adjusts the filtering parameters according to the changes in the noise characteristics collected in real time, which can improve the accuracy and working efficiency of the health management system; the present invention constructs an intelligent connection cloud-local hybrid module, which can effectively save data transmission time and storage costs, improve the working efficiency of the in-vehicle power unit health management system, and combine blockchain technology to achieve data traceability and anti-tampering through smart contracts; the present invention is based on the meta-learning framework, uses the Black-winged Kite Optimization Algorithm to optimize the parameters of the Pathformer model, trains the model using multiple datasets, and selects the optimal parameter group, which can ensure that the system has good generalization performance, can handle all health problems in the in-vehicle power unit, and comprehensively realizes health management of the in-vehicle power unit. Description of the Drawings

[0050] Figure 1 It is a flowchart of the health management method for the in-vehicle power unit. Detailed Embodiment

[0051] The present invention will be further described in detail below with reference to the drawings.

[0052] As Figure 1As shown in the figure, the present invention provides a method for health management of vehicle power units, which realizes intelligent processing of vehicle power unit data and accurate identification of fault types, and specifically includes the following steps:

[0053] Step 1: Collect data through a data acquisition module, and collect the running data of the vehicle power unit in real time, including but not limited to: temperature sensors, piezoelectric pressure sensors, vehicle speed sensors, and magnetic current sensors.

[0054] Step 2: Transmit the collected data to the data preprocessing module for feature extraction, and remove outliers, interpolate, and denoise the collected data through the Z-Score method and a lightweight convolutional neural network.

[0055] Extract features from the collected data, perform anomaly marking in the frequency domain, time domain, and Z-Score through sliding window analysis, and normalize and splice the features to use them as the input of the lightweight convolutional neural network; through the lightweight convolutional neural network, perform global average pooling to reduce the feature dimension; based on the meta-learning framework, combined with online gradient descent, backpropagate to update the parameter weights of the lightweight convolutional neural network; through the optimized lightweight convolutional neural network, eliminate the data noise caused by communication.

[0056] Step 3: Transmit the preprocessed data to the vehicle power unit intelligent diagnosis system for fault diagnosis. With the meta-learning ML-Eagle Owl optimization algorithm BKA-Pathformer model as the core, establish a framework based on meta-learning, use BKA to optimize the hyperparameters of the Pathformer model for fault diagnosis. For voltage polarization changes, generate a performance degradation index, fuse the fault diagnosis result with the performance degradation index to obtain a health index, and output the final diagnosis result.

[0057] Perform statistical analysis on the analyzed and processed data to obtain the fault diagnosis result. The performance degradation index of the vehicle power unit can be obtained from the change of voltage polarization through calculation. Fuse the two according to a certain ratio to obtain the final health index. Set different thresholds according to different conditions such as the vehicle model and service life. At the same time, adjust the threshold through the AHP (Analytic Hierarchy Process) according to the vehicle usage pattern and the owner's driving habits. Compare the fusion result with the set threshold to make the final health diagnosis result of the vehicle power unit.

[0058] Perform fault diagnosis based on the meta-learning ML-Eagle Owl optimization algorithm BKA-Pathformer model. Use multiple samples to train and evaluate the optimized model. By learning the model parameters on multiple data sets, select the optimal parameter combination. The parameters of the Pathformer model include the batch size, learning rate, and regularization coefficient of the model. Use the Eagle Owl optimization algorithm to simulate predation behavior, information exchange, and iterative optimization to optimize the parameter group. The specific steps are as follows:

[0059] S1: Random initialization:

[0060] X i = BK lb + rand(BK ub - BK lb )

[0061] Where i is an integer between 1 and pop, pop is the total number of the set of potential parameters, BK ub and BK lb are the lower and upper bounds of the i-th parameter group in the j-th dimension, and rand is a random number between [0, 1].

[0062] During the initialization process, BKA selects the parameter group with the best composite fitness value F(X i ) and records it as X L . This group of parameters is considered the optimal group. The false alarm rate FAR and the miss rate LR are weighted to form a composite fitness function, and the formula is:

[0063] F best = min(F(X i ))

[0064] X L = X(find(F best == F(X i )))

[0065] F(X i ) = ω1FAR + ω2LR

[0066] ω1 + ω2 = 1

[0067]

[0068] Where ω1 and ω2 are the weights of the false alarm rate and the miss rate, NF is the number of normal samples misreported as faults, NS is the total number of normal samples, NFR is the number of normal samples misreported as faults, and NFT is the total number of faults.

[0069] S2: Global search and exploration of parameter groups, and the formula is as follows:

[0070]

[0071] Where and are the positions of the i-th group of parameters in the j-th dimension at the (t + 1)-th and t-th iteration steps, r is a random number between 0 and 1, p is a constant of 0.9, T is the total number of iterations, and t is the number of iterations completed so far.

[0072] S3: When the parameter group that appears is better than the current optimal parameter group, replace the optimal parameter group and perform a specified search according to the search direction between the two parameter groups. The formula is as follows:

[0073]

[0074] In the formula, is the parameter group with the minimum composite fitness value in the j-th dimension of the current t-th iteration, F i is the composite fitness value obtained by any parameter group in the t-th iteration, and C(0, 1) is the Cauchy mutation.

[0075] S4: Use multiple samples for training and evaluation. Use multiple different data sets and repeat steps S1 to S3.

[0076] S5: Select the optimal parameter group through the results of multiple data sets, output the results, and perform fault diagnosis.

[0077] Step 4: Store the final diagnosis result locally. At the same time, based on blockchain technology, generate a hash value and store the hash value in the cloud to verify the integrity and security of the data. According to the diagnosis result, send the alarm and the corresponding result to the user terminal through the communication system.

[0078] The intelligent chain-cloud hybrid technology consists of blockchain-local hybrid storage. According to the diagnosis result, health fusion result, and warning signal, it performs block storage in the cloud and locally. The original data, final diagnosis result, health index, and alarm status are stored locally to reduce the pressure on the cloud to store a large amount of data. At the same time, a hash value is generated based on blockchain technology and stored in the cloud to improve the data exchange efficiency. The hash value is used to verify whether the data has been modified when retrieving the data later.

[0079] The present invention also provides a vehicle power unit health management system, including:

[0080] Data acquisition module: Omnidirectionally and real-time collect the operation data of the vehicle power unit through various sensors, including: temperature measurement sensors, piezoelectric pressure sensors, vehicle speed sensors, and magnetic current sensors.

[0081] Data preprocessing module: Remove outliers, noise, and perform interpolation on the collected data; It consists of the Z-Score method and a lightweight convolutional neural network. Deploy the lightweight convolutional neural network in the module for filtering, and the filtering parameters are adjusted in real time according to the noise characteristics. When the Z-Score method detects outliers, increase the sampling rate of the relevant sensors and trigger local data encryption storage, and store the data in the blockchain technology node to ensure the integrity of the abnormal segment; Thus, realize edge-side adaptive filtering adjustment and complete the removal of outliers, interpolation, and noise reduction of the collected data.

[0082] Intelligent Chain-Cloud Hybrid Module: Based on the blockchain-local hybrid storage architecture, it combines blockchain technology and local storage modules and operates in coordination with the communication module. It stores the original data, final diagnosis results, health index, and alarm status locally, and at the same time generates hash values, which are stored in the cloud, i.e., the blockchain platform, for verifying whether the data has been modified.

[0083] Health Diagnosis Module: Processes the preprocessed data. Based on the meta-learning framework, it uses the Black Kite Optimization Algorithm BKA to optimize the hyperparameters of the Pathformer model, trains and evaluates the model through multiple different datasets, and selects the optimal hyperparameter combination.

[0084] Communication Module: Used to connect local data collection and cloud information storage.

[0085] The above embodiments are only for illustrating the technical concept and features of the present invention, and the purpose is to enable those who are familiar with this technology to understand the content of the present invention and implement it accordingly. It should not be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for health management of a vehicle power unit, characterized in that, It includes the following steps: (1) Collect the running data of the vehicle power unit in real time; (2) Extract the features of the collected data, and remove outliers, interpolate, and denoise the collected data through the Z-Score method and the lightweight convolutional neural network; (3) Construct a BKA-Pathformer fault diagnosis model; the model builds a framework based on meta-learning, uses the Black Kite Optimization Algorithm BKA to optimize the hyperparameters of the Pathformer model for fault diagnosis; for voltage polarization changes, generate a performance degradation index, fuse the fault diagnosis result with the performance degradation index to obtain a health index, and output the diagnosis result; (4) Store the diagnosis result locally, and at the same time, based on blockchain technology, generate a hash value, store the hash value in the cloud to verify the data integrity and security; and send the result to the user.

2. The on-vehicle power unit health management method according to claim 1, wherein The running data of the vehicle power unit in step (1) includes a temperature sensor, a piezoelectric pressure sensor, a vehicle speed sensor, and a magnetic current sensor.

3. The vehicle-mounted power unit health management method according to claim 1, characterized in that, The implementation process of step (2) is as follows: (21) Extract the features of the collected data, perform anomaly marking in the frequency domain, time domain, and Z-Score with sliding window analysis, and normalize and splice the features as the input of the lightweight convolutional neural network; (22) Through the lightweight convolutional neural network, perform global average pooling to reduce the feature dimension; (23) Based on the meta-learning framework, combined with online gradient descent, backpropagate to update the parameter weights of the lightweight convolutional neural network; (24) Eliminate the data noise caused by communication through the optimized lightweight convolutional neural network.

4. A vehicle power unit health management method according to claim 1, characterized in that, The hyperparameters of the Pathformer model in step (3) include batch size, learning rate, and regularization coefficient.

5. The vehicle-mounted power unit health management method according to claim 1, characterized in that The implementation process of using BKA to optimize the hyperparameters of the Pathformer model in step (3) is as follows: S1: Random initialization: X i = BK lb + rand(BK ub - BK lb ) where i is an integer between 1 and pop, pop is the total number of the set of potential parameters, BK ub and BK lb are the lower and upper bounds of the i-th parameter group in the j-th dimension, and rand is a random number between [0, 1]; During the initialization process, BKA selects a parameter group with the best composite fitness value F(X i ) and records it as X L . This group of parameters is considered the optimal group. The false alarm rate FAR and the miss rate LR are weighted to form a composite fitness function, and the formula is: F best = min(F(X i )) X L = X(find(F best = F(X i ))) F(X i ) = ω1FAR + ω2LR ω1 + ω2 = 1 where ω1 and ω2 are the weights of the false alarm rate and the missed alarm rate, NF is the number of normal samples misreported as faults, NS is the total number of normal samples, NFR is the number of normal samples missed as faults, and NFT is the total number of faults; S2: Perform global search and exploration on the parameter group, and the formula is as follows: In the formula, and are the positions of the i-th group of parameters in the j-th dimension at the (t + 1)-th and t-th iteration steps, r is a random number between 0 and 1, p is a constant, T is the total number of iterations, and t is the number of iterations completed so far; S3: When the parameter group that appears is better than the current optimal parameter group, replace the optimal parameter group, and perform specified search according to the search direction between the two parameter groups, and the formula is as follows: In the formula, is the parameter group with the smallest composite fitness value in the j-th dimension of the current t-th iteration, F i is the composite fitness value obtained by any parameter group in the t-th iteration, and C(0, 1) is the Cauchy mutation; S4: Use multiple samples for training evaluation, use multiple different data sets, and repeat steps S1 to S3; S5: Select the optimal parameter group through the results of multiple data sets, output the results, and perform fault diagnosis.

6. The on-vehicle power unit health management method according to claim 1, characterized in that The implementation process of step (3) is as follows: (31) Perform statistical analysis on the analyzed and processed data to obtain a fault diagnosis result, calculate the vehicle power unit performance degradation index from the voltage polarization change, and fuse the two to obtain the final health index; (32) Set different thresholds according to different conditions such as the vehicle model and service life; at the same time, adjust the threshold through the AHP (Analytic Hierarchy Process) according to the vehicle usage mode and the owner's driving habits; (33) Compare the fusion result with the set threshold to obtain the final health diagnosis result of the vehicle power unit.

7. The vehicle-mounted power unit health management system according to claim 1, characterized in that, It includes the following modules: Data acquisition module: Omnidirectionally and real-time collect the operation data of the vehicle power unit through various sensors; Data preprocessing module: Remove outliers, noise and perform interpolation on the collected data; Health diagnosis module: Perform fault diagnosis on the preprocessed data. Based on the meta-learning framework, use the Black-winged Kite Optimization Algorithm BKA to optimize the hyperparameters of the Pathformer model. Train and evaluate the model through multiple different datasets, and select the optimal hyperparameter combination; for voltage polarization changes, generate a performance degradation index, fuse the fault diagnosis result with the performance degradation index to obtain a health index, and output the diagnosis result; Intelligent chain-cloud hybrid module: Based on the blockchain-local hybrid storage architecture, combine blockchain technology and the local storage module, and operate in coordination with the communication module; Store the original data, the final diagnosis result, the health index and the alarm status locally, and at the same time generate a hash value, which is stored in the cloud, that is, the blockchain platform, for verifying whether the data has been modified; Communication module: Used to connect local data acquisition and cloud information storage.

8. A vehicle power unit health management system according to claim 7, wherein, The data preprocessing module consists of the Z-Score method and a lightweight convolutional neural network. Deploy a lightweight convolutional neural network in the module for filtering, and the filtering parameters are adjusted in real time according to the noise characteristics. When the Z-Score method detects an outlier, increase the sampling rate of the sensor and trigger local data encryption storage, and store the data in the blockchain technology node to ensure the integrity of the abnormal segment.

Images