New energy vehicle fault identification based on improved activation functions and parameter-free attention mechanisms

The drive system of New Energy Vehicles (NEVs) is a core component ensuring the safe and efficient operation of vehicles. Key components such as batteries, motors, and inverters are prone to faults under complex operating conditions, severely affecting vehicle stability and reliability. Traditional fault diagnosis methods based on rules or statistical analysis struggle to adapt to complex nonlinear fault patterns, while deep learning provides a new pathway for intelligent fault diagnosis. This paper proposes an improved CNN-based fault diagnosis method integrating the ReLUT activation function, SimAM parameter-free attention mechanism, and residual modules: ReLUT combines the advantages of ReLU and tanh to enhance the model's nonlinear expression capability and avoid gradient stagnation; SimAM adaptively focuses on key fault features without additional parameters; and residual modules alleviate the gradient vanishing problem and accelerate model convergence. Extensive comparative experiments were conducted on NEVData, a dedicated fault diagnosis dataset for NEVs acquired from online resources. This dataset includes 8 core fields: voltage (V), current (A), motor speed (RPM), temperature (℃), vibration (g), ambient temperature (℃), humidity (%), and fault label, with a total of 11,000 samples. It covers normal operating conditions (label 0, 5,000 samples) and 3 types of fault conditions (labels 1-3, 2,000 samples each), featuring practical scenario characteristics such as class imbalance and outliers in some variables. Results show that the proposed method achieves a fault diagnosis accuracy of 0.992 and precision of 0.989, representing a 1.5% improvement in both metrics compared to the traditional KNN method. While adapting to the characteristics of real-world data, it realizes higher diagnostic reliability.