科研之友
微信
新浪微博
Facebook
分享链接
摘要
Remaining useful life (RUL) prediction is the core link with prognostic and health management (PHM). Compared with the stable service environment, the degradation trend of key parts of electromechanical equipment under dynamic service environment has greater uncertainty. Dynamic changes of the service environment in the subsequent work tasks can result in inaccurate or even invalid current prediction results. To this end, considering that tasks can be known in advance and affect the service environment, a novel RUL prediction method is proposed which updates the predicting results as the tasks change based on dynamic correlation of planned tasks, service environment, degradation characteristics, and RUL. Firstly, the influence of the changes in service environment on the RUL of key parts is analyzed, and a task-driven classification and matching method of dynamic service environment is presented. Secondly, the key service environment variables affecting RUL of parts are screened employing grey relation analysis (GRA) and Pearson correlation coefficient. Thirdly, a task-driven deep long-short time memory (DLSTM) neural network prediction model under dynamic service environment is established, and a regularization method for optimizing the DLSTM model is given to improve the prediction accuracy. Finally, the RUL prediction of a machine tool spindle is taken as an example to verify the effectiveness of the proposed method.
