Gear Fault Diagnosis Method Based on Transfer Learning

Transfer learning offers a novel approach to address complex feature processing and human factors in traditional gearbox fault diagnosis. First, one-dimensional vibration signals are decomposed into IMF components via Empirical Mode Decomposition and Hilbert transform, with prominent components selected to form a 2D image sample set. The pre-trained model is fine-tuned and combined with a support vector machine optimized using Particle Swarm Optimization for classification. Experimental results show an average classification accuracy of 98.54% with a small sample size, fast convergence, low computational requirements, and strong adaptability to various scenarios with personalized outputs.