The invention discloses a ''prediction-verification-feedback-optimization'' closed-loop system of online early warning and offline diagnosis. The ''prediction-verification-feedback-optimization'' closed-loop system comprises online early warning and offline diagnosis closed-loop systems. The online early warning closed-loop system comprises online prediction, verification, feedback and optimization, fault type prediction is achieved through online prediction, whether online prediction is correct or not is judged through online verification, fault information is fed back to a fault early-warning system through online feedback, an online error self-checking algorithm is optimized through online optimization, and a fault case library is perfected. The offline diagnosis closed-loop system comprises off-line prediction, verification, feedback and optimization. A precision degradation model is established through offline prediction, offline prediction of a health condition of a robot is achieved, offline prediction accuracy is verified offline, offline prediction parameters are fed back to the precision degradation model through offline feedback, parameter influences and an evolution lawof the precision degradation model are optimized and analyzed offline, and improvement and parameter optimization are performed on a digital-analog linkage algorithm. In the invention, accurate faultearly warning under complex working conditions is realized, and more accurate predictive maintenance is realized too.
