Research On Remaining Useful Life Prediction Of Finishing Slot Cutter For Turbine Based On Deep Learning

Turbine rotor is the core component of the steam turbine.The profile of rotor slot is complex,which is the most critical and difficult processing part in the rotor machining process.Its machining quality affects the overall performance of the steam turbine directly.In finish machining,tool wear directly affects the surface quality of the workpiece.In the actual machining of workshop,tool life is dynamically and variably affected by the process conditions,while the tool replacement largely depends on judgement based on manual experience,which lacks theoretical guidance and data support.It is inevitable that the tool replacement is too early or too late,which leads to the increased production costs,poor quality of parts or even scrap.Therefore,accurate prediction of remaining useful life of finish cutter is of practical significance for the guarantee of machining quality,improvement of tool utilization and reduction of production cost.For the above problem,this paper conducts a research on the prediction of the remaining useful life(RUL)of the finishing tools in turbine rotor processing.The main research contents are as follows:First,in view of the lack of date on the machining process of tool and the lack of association between production factors on the manufacturing site,including machine,tool,cutting parameters,etc.,the IDEF0 method was utilized to analyze the machining process of rotor slot.An intelligent monitoring system for NC machining is designed thereafter,to acquire the data of the machining process of tool.The UML approach was utilized to model the NC machining system and the full life cycle of cutting tool,to realize the association between production factors on the manufacturing site,which provides a data basis for subsequent tool RUL prediction.Secondly,in view of the difference in the life decay modes of different tool individuals,a hybrid information model was proposed.The time series features was extracted from the acoustic emission signals of tool condition monitoring,based on long short-term memory network,which were combined with tool attributes and process information to predict tool RUL.Experiment has shown the accurate prediction of tool RUL.In addition,tool condition monitoring and RUL prediction have provided the data support for tool replacement in the machining site.Compared with the advanced tool replacement based on manual experience,the tool utilization has been improved.Finally,aiming at the problem of the failure of the original prediction model under new process conditions,a dynamic adversarial domain adaptation method was proposed.Based on the data distribution discrepancy between the process monitoring signals under historical process conditions and the new process conditions,the RUL prediction model was adjusted to improve the accuracy of RUL prediction under new process condition.Experiment has shown the significant improvement of prediction accuracy,where the accuracy of the tool RUL prediction under the new process conditions was improved from 56.66% to 91.26%.