Unified Diagnostic and Matching Framework of Fault and Quality for Robotic Grinding System

This article explores the corresponding relationship between the equipment fault and grinding quality in a robotic grinding system (RGS), and establishes a unified and lightweight monitoring and matching framework, providing a perceptual basis for accurate tracking and effective control of grinding quality. First, a multichannel vibration imaging method named Wavetrizorn is developed based on vibration signals, and the images generated were used to train a fault diagnosis model for the equipment. Particularly, a lossy reconstruction algorithm based on wavelet packet and convolutional autoencoder (WPCAE) is proposed for vibration signals with strong noise, which can help networks to extract the fault information. Then, a regression model mapping from vibration signal to force signal is established based on the reconstructed signal graphs to monitor the grinding quality. Finally, to match the fault type and the grinding quality, a unique canonical correlation feature (CCF) is proposed and calculated, which can achieve precise quality traceability. Consequently, during the online monitoring, it is only necessary to use vibration signals to regress the CCF to accurately match the fault type and grinding quality with significant efficiency. The effectiveness of the framework is verified on an RGS in the laboratory.