Abstract: Aiming at the problems of lagged fault response, long shutdown time, difficulty in accumulating maintenance knowledge and others caused by the reliance on manual inspection and empirical diagnosis of roadheaders in Pingshu Coal Industry, a cloud edge collaborative intelligent fault diagnosis and maintenance system is constructed and implemented. The system adopts a dual-layer architecture of "edge computing+cloud deep analysis", and integrates CNN-Transformer deep learning model and intelligent maintenance knowledge closed-loop mechanism, effectively solving the pain points of data delay, sensor stability and others under complex working conditions in downholes. The on-site application shows that the comprehensive diagnostic accuracy rate of the system reaches 92.5%, the average maintenance time is shortened by 40.8%, the unplanned shutdown of the roadheader is decreased from 30 h per month to 18 h per month, and the annual direct economic benefit is about 2.16 million yuan, significantly improving the intelligent operation and maintenance level and operational safety of the excavation system. This system provides a technical demonstration for fault precise management and control and maintenance knowledge accumulation of excavation equipment in coal mines.