Automobile Fault Diagnosis Using Lightweight Cetacean Optimization and Multi-Scale Residual Neural Networks

Abstract

With the advancement of automotive technologies, accurate and real-time fault diagnosis is essential for ensuring vehicle safety and reducing maintenance costs. This paper proposes an automobile fault diagnosis system based on the Lightweight Whale Optimization Algorithm (LWA) integrated with a Multi-Scale Residual Unit deep neural network. The proposed model is trained on a real-world automobile sensor dataset containing 2,000 labeled samples spanning four fault types: engine, brake system, battery, and transmission failures. The model demonstrates a diagnostic accuracy of 95.4%, outperforming baseline methods such as Support Vector Machine (90.1%), Random Forest (92.3%), and CNN-based models (94.2%). Additionally, the LWA achieves faster convergence and a 25% reduction in inference time compared to traditional MSRU models, with a response time under 2.5 seconds. The lightweight design also reduces model parameters by 64%, enabling real-time deployment in embedded vehicle systems. Experimental results show that the proposed method not only enhances diagnostic accuracy but also improves computational efficiency, offering a practical solution for intelligent automotive maintenance.