Condition Monitoring Of Cutting Tools With Multiple Failure Forms During NC Machining

Tool deterioration is a common fault in NC machining,which directly affects part quality,production efficiency and manufacturing cost.As the process of tool deterioration during machining process is complicated and changeable,and a variety of tool failure forms coexist,leading to the difficulty in accurate prediction of tool condition.To solve the problem mentioned above,the monitoring of multifailure forms of NC cutting tools is studied in this thesis.The research achievements are as follows:(1)Aiming at the problem on the complex and changeable process of tool degradation during machining,the impact of tool failure forms and cutting conditions on monitoring signals is studied.The significance test method based on analysis of variance is used to analyze the influence of different tool failure forms and cutting conditions on the monitoring signals.It laid the foundation for accurate monitoring of multiple tool failure forms.(2)Aiming at the problem of extracting the monitoring signal features of multiple tool failure forms,a method of processing the monitoring signals for multiple failure forms is studied.The feature selection method based on gray correlation analysis and information gain is studied for different failure forms.The correlation between signal features and influencing factors and the relationship between the information entropy of the feature set and the tool condition are used to reduce the influence of cutting conditions.The key information of the signal is extracted under different cutting tool failure forms.(3)Aiming at the problem of tool condition monitoring caused by the mutual coupling of the monitoring signal features and the mutual influence of condition monitoring in the multiple failure forms,a multitask learning-based accurate prediction method for tool condition is proposed.A prediction model for each failure form is constructed by using the back-propagation network as base model.A method for establishing a low-rank constraint on a shared information tensor is proposed,and then a multitask learning prediction model is established,Compared with the traditional method,the prediction errors of each of the three tasks is reduced by more than 23%.(4)A tool condition monitoring system based on multitask learning is developed on Lab VIEW based on the above research,and it is verified by experiments.