Prediction of Automotive Wire Harness Aging Based on CNN-biLSTM-Attention

Area of Science:

• Automotive Engineering
• Materials Science
• Artificial Intelligence

Background:

• The electrification and intelligence of modern vehicles necessitate reliable wiring harness performance.
• Wiring harness degradation, influenced by multi-physics coupling, poses risks to vehicle safety and functionality.
• Predicting aging is crucial for proactive maintenance and ensuring operational integrity.

Purpose of the Study:

• To develop an advanced predictive model for automotive low-voltage wiring harness aging.
• To address the challenge of forecasting performance degradation due to complex aging factors.
• To improve the accuracy of aging level prediction in automotive electrical systems.

Main Methods:

• A hybrid Convolutional Neural Network (CNN)-Bidirectional Long Short-Term Memory (BiLSTM)-Attention neural network model was proposed.
• The model utilized voltage, current, and temperature data from low-voltage systems during operation.
• Accelerated aging experiments generated data from wiring harnesses aged up to 720 hours for model training and validation.

Main Results:

• The CNN-BiLSTM-Attention model demonstrated effective prediction performance for wiring harness aging levels.
• The proposed model achieved a Mean Absolute Error (MAE) of 0.02806.
• Significant error reductions of 32.50% and 62.06% were observed compared to LSTM and Random Forest models, respectively.

Conclusions:

• The hybrid CNN-BiLSTM-Attention model accurately predicts wiring harness aging.
• This predictive capability is vital for ensuring automotive safety and performance in evolving vehicle technologies.
• The study validates the model's superiority over traditional methods in forecasting degradation.