Research On Intelligent Prediction Of Tool Remaining Useful Life Under Big Data Environment

In the new round of manufacturing industry reform,the monitoring data collected by manufacturing enterprises is increasing rapidly with the widespread application of Internet of Things technology and industrial Internet technology.The tool is an important part of the machining process.By using large amounts of monitoring data,tool remaining useful life(RUL)can be predicted.On the one hand,it can avoid the tool damage and reduce the probability of quality problems in parts processing.On the other hand,it can ensure that the tool runs under high reliability conditions while avoiding the waste caused by excessive maintenance.The traditional methods applied to predict the tool RUL need to extract features first.However,the extraction and selection of the features are highly dependent on specific experience and specialized knowledge.In addition,when the machining condition is changed,the evolution law of tool wear will also change significantly,and the prediction model under the original machining condition is no longer applicable.To solve the above problems,this paper applies deep learning and deep transfer learning methods to predict the tool RUL.The research is carried out under single machining condition and multiple machining conditions respectively.And on this basis,a tool RUL prediction system is developed.Firstly,aimed at predicting the tool RUL under single machining condition,a prediction process based on hybrid approach is proposed.Starting from the concerns in the actual manufacturing process,the piecewise life is introduced.Based on the condition monitoring data,the Residual Net(ResNet)is used to predict the tool wear,and then the ARIMA model belonging to the time series method is used for multi-step ahead tool wear forecast.Combined with the preset wear threshold,the tool RUL at the current moment can be predicted.Secondly,the tool RUL prediction under multiple machining conditions is researched.Under the same processing material,the cutting parameters information and monitoring data are integrated by adding the cutting parameters channel,so that the tool RUL prediction under various cutting parameters can be realized by a single model.Afterwards,for the situation of changing the processing material,the monitoring data under the original processing material is used as the source domain,while that under the new processing material is used as the target domain.The wear prediction model of the source domain is transferred to the target domain through fine-tuning and edge distribution adaptation respectively.And the RUL prediction accuracy of the target domain is improved by transfer learning.Finally,after completing the related theories and algorithm design,the tool RUL prediction system is developed.A big data platform based on Hadoop cluster is built,and large amounts of tool monitoring data are stored in the distributed file system HDFS.The tool database is established to store the tool related information of cutting parameters,machine tool,workpieces,RUL,etc.The system includes the development of user management module,data query module and RUL prediction module,which makes a preliminary exploration for the application of tool RUL prediction in actual manufacturing process.