A Fresh Spin on Rolling Bearing Fault Diagnosis

Rolling bearings play a crucial role in the safety and performance of rotating machines. While recent years have seen advances in intelligent fault diagnosis, noise in vibration signals remains a challenge. This is where GMSCNN comes in—a method that combines Gram Matrix (GM) and Multi-Scale Convolutional Neural Network (MSCNN) to diagnose bearing faults effectively. First off, GM takes care of the noise in vibration signals, making the data cleaner for analysis. Next, MSCNN kicks in, using different-sized convolutional kernels to spot patterns in the signals at various frequencies and time scales. This is like having multiple pairs of eyes looking at the data from different angles.

To make the model even more powerful, two feature enhancement branches are added. These branches use the original, noisy signals as input to uncover more nuanced features. This diversity boosts the model's ability to understand and respond to different situations. Experiments on two bearing datasets showed that GMSCNN is pretty robust against noise. It can handle signals that aren't perfect, making it a reliable tool for real-world conditions.

Actions