Intelligent fault diagnosis of bearings using multi-sensor spectrogram fusion and machine learning models

Bearing components are essential to the operation of machinery, and their failure can lead to significant operational disruptions. Predictive maintenance, powered by artificial intelligence, holds promise for early fault diagnosis, reducing both maintenance costs and risks. However, existing approaches often rely on single-sensor data, which may miss critical fault indicators. To address this challenge, this study proposed an intelligent fault diagnosis model that compares single-sensor and multi-sensor approaches, using various machine learning models, including convolutional neural networks (CNN), random forest (RF), and support vector machine (SVM). The study examined 450 samples across five fault types: healthy, cage fault, ball fault, inner-ring fault, and outer-ring fault. A multi-sensor approach combining vibration and acoustic signals through feature-level fusion based on spectrograms derived from Short-Time Fourier Transform was utilized. The results indicated that the multi-sensor approach significantly outperformed the single-sensor method, with accuracy improvements of 5.68% for CNN, 10.11% for RF, and 3.33% for SVM. These findings highlighted the effectiveness of integrating multiple signal modalities and underscored the potential of multi-sensor approaches to enhance diagnostic accuracy in predictive maintenance systems.