Tool Condition Monitoring Based On Cutting Temperatures

Machining processes like heavy duty cutting and hard cutting have been widely applied,which increase the processing efficiency a lot.However,the consequent complex working environments bring cutting tools more risks of failure,making tool condition monitoring an important issue.Cutting temperatures are closely tied to deformations,extrusions,frictions in machining processes,directly reflecting the loads exerted on the tool and also providing abundant information about cutting status.In this study,we take advantages of a novel fiberoptic temperature measuring apparatus as well as its' nice responsiveness and transmission stability,proposing a tool condition monitoring method.There are two problems remained unsolved for the target.Firstly,to provide operators with intuitive cognition of the thermal loads,the measuring is required to be as accurate as possible.Considering that the measuring probe is hard to reach the heat source,several finite element simulations are conducted to investigate the spatial distribution of the temperatures on a tool and evaluate the differences between the measuring values and the maximum temperatures.To ensure the reliability of the FEA model,some temperature experiments using thermo-couple arrays are carried out as a comparison of the simulation and as a reference for parameter calibration.Based on the mappings between measuring values and the maximum temperatures,measuring compensations during cutting processes are discussed.Secondly,to warn in time when instabilities happen to the cutting processes,the correlation between the signals and the tool wears is required to be constructed.A transient thermal simulation and an intermittent cutting experiment are carried out,verifying that the measuring apparatus possesses good responsiveness facing fast undulating temperatures.Then the effects on the frequency domain temperatures brought by tool wears are examined,based on which feature extractions are implemented and a four-layer feedforward neural network is designed for wear distinguishing.A learning rate adaption algorithm is used for optimizing the training processes,greatly decreasing the dependence on hyper parameters and initial conditions,which allows operators to train customized networks conveniently.The accuracy of classification stays 90% and higher in variable arguments experiments,and the output of the neural network also stays in step with the vibration contrasts during the whole life examination,which indicates the practicability of the method.Finally,a cutting process visualization software is introduced,which contains modules like multichannel measurement,wave form display and tool wear caution.The software also supports network training,which is successfully applied in wheel set lathing,affording the industry a new idea about tool condition monitoring.