Attention-embedded multi-scale quadratic convolutional neural network for early bearing fault diagnosis

Bearings are vital components in various industries such as chemical, textile, food, and tobacco. Timely fault diagnosis is crucial for maintenance, performance, safety, and cost efficiency. Traditional methods rely on manual feature extraction and are time-consuming. This study proposes an attention-embedded multi-scale quadratic convolutional neural network (MQCNN) based on the attention-embedded quadratic network (QCNN). It utilizes an attention mechanism from quadratic neurons (qttention) to enhance noise resistance and automatically extracts frequency-specific signal features through varied convolution kernels and depths. This method improves diagnostic accuracy and offers significant economic advantages in engineering applications.