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

Abstract

Under the transition towards electrification and intelligence in modern automotive industry, the health status of low-voltage wiring harnesses directly affects vehicle performance and safety. To address the challenge of predicting performance degradation caused by multi-physics coupling effects during wiring harness aging, this study proposes a CNN-BiLSTM-Attention hybrid neural network model. By capturing voltage, current, and temperature parameters during low-voltage system operation, the model combines CNN’s local feature extraction, BiLSTM’s temporal sequence analysis, and attention mechanisms to predict aging levels. Accelerated aging experiments were conducted to obtain wiring harnesses with different degradation levels from new to 720 h aged states, and a dedicated experimental platform was built for data collection and verification. The results show the system achieves a mean absolute error (MAE) of 0.02806, with 32.50% and 62.06% error reduction compared to LSTM and Random Forest models, respectively, demonstrating effective prediction performance.