Fault Diagnosis of Rotating Equipment Unbalance Problem Based on Denoising Stacked Autoencoders

Rotating equipment is crucial in various industries, from manufacturing to aerospace. A flaw in even one component can cause system-wide vibrational stress, accelerating wear on the machinery. Current research on equipment fault diagnosis typically focuses on fault detection, not on the severity, which can be critical given the commonality of minor faults. Addressing this, our study breaks new ground by categorizing these unbalanced fault types and exploring different data segmentation strategies to tackle the issue. We’ve developed a stacked autoencoder-based signal extractor for its superior feature extraction and interpretability. Testing our model under diverse data splits, we’ve also integrated a novel data fusion technique using the Kalman filter, improving accuracy by 3% in five-category classification.