CN-122017548-A - Early fault diagnosis method for high-voltage relay and computer program product

Abstract

The application discloses a high-voltage relay early fault diagnosis method and a computer program product, which comprise the steps of obtaining high-voltage relay historical fault data and vehicle bus historical data, carrying out rule characterization processing on the historical fault data to obtain rule characteristic data, training the rule characteristic data and the vehicle bus historical data to obtain a hybrid neural network model, obtaining high-voltage relay current data and vehicle bus real-time data, inputting the hybrid neural network model to obtain a fault early warning signal, realizing dynamic modeling of the high-voltage relay running state by fusing the rule characteristic and the multi-source bus data, capturing early degradation characteristics such as contact resistance micro-variation, action delay and the like, completing model training and prediction without labeling a large number of early fault samples, overcoming the defects of traditional threshold early warning hysteresis and machine learning dependence labeling data, and improving early fault recognition precision and system reliability.

Inventors

• WEI BANGHONG

Assignees

• 东风汽车有限公司东风日产乘用车公司

Dates

Publication Date

20260512

Application Date

20260213

Claims (10)

1. 1. A method for diagnosing an early failure of a high voltage relay, comprising: acquiring historical fault data of a high-voltage relay and historical data of a vehicle bus related to the historical operating state of the high-voltage relay; Performing rule characterization processing on the high-voltage relay historical fault data to obtain rule characteristic data; Training the rule characteristic data and the vehicle bus historical data to obtain a hybrid neural network model; and acquiring current data of the high-voltage relay and real-time data of the vehicle bus, and inputting the current data of the high-voltage relay and the real-time data of the vehicle bus into the hybrid neural network model to obtain a fault early warning signal.
2. 2. The method for diagnosing early failure of a high voltage relay according to claim 1, wherein said training with said rule feature data and said vehicle bus history data to obtain a hybrid neural network model comprises: Constructing a rule characteristic neural network and an original signal neural network; determining original bus data according to the vehicle bus history data; Inputting the rule characteristic data into a rule characteristic neural network for training to obtain a trained rule characteristic neural network and a soft target label; Inputting the original bus data into an original signal neural network for training, and carrying out knowledge distillation based on the soft target label to obtain a trained original signal neural network; and determining a hybrid neural network model according to the trained rule characteristic neural network and the trained original signal neural network.
3. 3. The method for diagnosing early failure of a high voltage relay according to claim 2, wherein said determining raw bus data from said vehicle bus history data comprises: screening a signal for judging faults of the high-voltage relay in the vehicle bus historical data as a fault list trigger signal; determining fault list supplementary signals related to the attribute, the category, the naming and the feedback relation according to the fault list trigger signals; And acquiring a real vehicle environment signal, and taking the fault list trigger signal, the fault list supplement signal and the real vehicle environment signal as original bus data.
4. 4. The method for diagnosing the early failure of the high-voltage relay according to claim 2, wherein the construction of the rule characteristic neural network comprises the steps of initializing a gating cycle unit network and constructing the rule characteristic neural network; the step of inputting the rule characteristic data into a rule characteristic neural network for training specifically comprises the following steps: performing data preprocessing on the rule characteristic data; training the rule characteristic neural network by utilizing the rule characteristic data after data preprocessing; predicting the rule feature data through the trained rule feature neural network to obtain rule prediction probability; determining rule characteristic neural network loss according to the rule prediction probability and the rule real label; if the rule characteristic neural network loss is smaller than the preset rule loss, the rule characteristic neural network training is completed, the rule prediction probability is used as a soft target label, and otherwise, parameters of the rule characteristic neural network are adjusted based on the rule characteristic neural network loss.
5. 5. The method for diagnosing early failure of a high voltage relay according to claim 2, wherein said constructing an original signal neural network includes initializing a convolutional neural network to construct an original signal neural network; the step of inputting the original bus data into an original signal neural network for training and carrying out knowledge distillation based on the soft target tag comprises the following steps: Performing data preprocessing on the original bus data; Training the original signal neural network by utilizing the original bus data after data preprocessing; predicting the original bus data through the trained original signal neural network to obtain an original prediction probability; determining distillation loss according to the soft target label and the original prediction probability; Determining a predicted original loss according to the original prediction probability and the original real label; Multiplying the distillation loss by the sum of the first preset weight and the predicted original loss multiplied by the second preset weight to obtain the comprehensive loss of the original signal neural network; If the comprehensive loss of the original signal neural network is smaller than a preset original loss threshold, training the original signal neural network is completed, otherwise, parameters of the original signal neural network are adjusted based on the comprehensive loss of the original signal neural network.
6. 6. The method for diagnosing an early failure of a high voltage relay according to claim 2, wherein said determining a hybrid neural network model based on the trained rule characteristic neural network and the trained raw signal neural network further comprises: inputting rule characteristic data with a rule true label into the trained rule characteristic neural network to obtain the confidence coefficient of the rule characteristic neural network; inputting the original bus data with the original real tag into the trained original signal neural network to obtain the confidence coefficient of the original signal neural network; Determining the dynamic fusion weight of the hybrid neural network model according to the rule characteristic neural network confidence and the original signal neural network confidence; and carrying out weighted fusion on the rule characteristic neural network and the original signal neural network according to the dynamic fusion weight to obtain a hybrid neural network model.
7. 7. The method for diagnosing an early failure of a high voltage relay according to claim 6, further comprising, before said obtaining a hybrid neural network model: obtaining the model precision of the hybrid neural network model; If the model precision reaches a preset precision threshold, training is finished, and the hybrid neural network model is obtained; and if the model precision does not reach the preset precision threshold, training the rule characteristic neural network and the original signal neural network again.
8. 8. The method for diagnosing early failure of a high-voltage relay according to claim 1, wherein the rule characterization processing is performed on the historical failure data of the high-voltage relay, and specifically comprises: Acquiring expected threshold data corresponding to the historical fault data of the high-voltage relay one by one; And generating rule characteristic data according to the one-to-one corresponding deviation between the historical fault data of the high-voltage relay and the expected threshold data.
9. 9. The method for diagnosing an early failure of a high voltage relay according to claim 1, further comprising, after the obtaining of the failure early warning signal: acquiring the contribution degree of each input data in the rule characteristic data and the vehicle bus historical data according to the hybrid neural network model; if the contribution degree of the input data is larger than a preset contribution threshold, marking the input data as key features; and generating maintenance display suggestions according to the key characteristics.
10. 10. A computer program product comprising computer program/instructions which, when executed by a processor, implements a method for diagnosing early faults of a high voltage relay according to any of claims 1-9.

Description

Early fault diagnosis method for high-voltage relay and computer program product Technical Field The application relates to the technical field of high-voltage relays, in particular to a method for diagnosing early faults of a high-voltage relay and a computer program product. Background Along with the development of new energy automobiles, the high-voltage relay is used as a key execution and protection element, and the reliability of the high-voltage relay directly determines the safety and stability of the whole high-voltage architecture. At present, a diagnosis method based on a rule threshold or a traditional machine learning model is generally adopted, namely, the detection and alarm of obvious faults are realized by monitoring the switching time, the voltage and current parameters of a high-voltage relay and comparing the switching time, the voltage and current parameters with a preset threshold or training a classification model by utilizing historical fault data. However, the rule threshold method in the prior art depends on a fixed threshold, and cannot capture early potential fault characteristics such as slow degradation of switching characteristics, slight change of contact resistance and the like, so that early warning lag is caused. And models based on historical faults rely on a large amount of manually noted data for training, while early fault data are scarce and difficult to obtain in practice. The prior art also lacks self-adaptive adjustment capability under complex operation conditions, and the problem of false alarm and missing report is easy to occur. Disclosure of Invention The object of the present application is to overcome the above problems and to provide a method and a computer program product for diagnosing an early failure of a high voltage relay. The technical scheme of the application provides a method for diagnosing early faults of a high-voltage relay, which comprises the following steps: acquiring historical fault data of a high-voltage relay and historical data of a vehicle bus related to the historical operating state of the high-voltage relay; Performing rule characterization processing on the high-voltage relay historical fault data to obtain rule characteristic data; Training the rule characteristic data and the vehicle bus historical data to obtain a hybrid neural network model; and acquiring current data of the high-voltage relay and real-time data of the vehicle bus, and inputting the current data of the high-voltage relay and the real-time data of the vehicle bus into the hybrid neural network model to obtain a fault early warning signal. Further, the training by using the rule feature data and the vehicle bus history data to obtain a hybrid neural network model specifically includes: constructing a rule characteristic neural network and an original signal neural network; determining original bus data according to the vehicle bus history data; Inputting the rule characteristic data into a rule characteristic neural network for training to obtain a trained rule characteristic neural network and a soft target label; Inputting the original bus data into an original signal neural network for training, and carrying out knowledge distillation based on the soft target label to obtain a trained original signal neural network; and determining a hybrid neural network model according to the trained rule characteristic neural network and the trained original signal neural network. Further, the determining the original bus data according to the vehicle bus history data specifically includes: screening a signal for judging faults of the high-voltage relay in the vehicle bus historical data as a fault list trigger signal; determining fault list supplementary signals related to the attribute, the category, the naming and the feedback relation according to the fault list trigger signals; And acquiring a real vehicle environment signal, and taking the fault list trigger signal, the fault list supplement signal and the real vehicle environment signal as original bus data. Further, the construction of the rule characteristic neural network comprises initializing a gating circulation unit network and establishing the rule characteristic neural network; the step of inputting the rule characteristic data into a rule characteristic neural network for training specifically comprises the following steps: performing data preprocessing on the rule characteristic data; training the rule characteristic neural network by utilizing the rule characteristic data after data preprocessing; predicting the rule feature data through the trained rule feature neural network to obtain rule prediction probability; determining rule characteristic neural network loss according to the rule prediction probability and the rule real label; if the rule characteristic neural network loss is smaller than the preset rule loss, the rule characteristic neural network training is completed, and the rule prediction pr