Research On Intelligent Monitoring Technology For Tool Wear And Breakage Conditions During Milling Process

Milling is widely used in the manufacturing process of key parts in aviation,energy,automobile,and other fields,where interrupted cutting with multi-tooth tools removes material rapidly from the metal surface and obtains the workpiece’s required geometry and dimensional accuracy.As the most active element during milling processes,tool wear and breakage conditions directly affect workpiece quality and productivity.When faced with difficult-to-machine materials,the extremely high mechanical stresses,thermal stresses,and shocks make the tool failure problems particularly prominent.How to monitor tool wear and breakage conditions in a timely and accurate manner has become a critical technical problem to be solved for the intellectualization of the milling process.The complex and variable milling environment and the variations in tool performance between individual tools can lead to a significant difference between the actual and theoretical tool life.In addition,compared with progressive tool wear,tool breakage usually occurs instantaneously and randomly without early warning,posing a significant challenge for online monitoring of tool conditions.This paper takes the high-feed milling process of508-III steel as the primary research object and carries out research on intelligent monitoring technology for tool wear and breakage conditions,aiming to provide theoretical reference and technical support for the accurate control of tool conditions and remaining useful life in the actual machining process and avoid unnecessary economic losses caused by tool failure.The specific research contents are as follows:The tool failure forms and related mechanisms during high-feed milling of508-III steel are studied.The performance of different sensors in the tool condition monitoring task is comprehensively evaluated to determine the signal acquisition scheme for tool wear and breakage monitoring.The experimental platform for tool condition monitoring based on multi-sensor signal fusion is built,and the condition monitoring signals of milling tools and the corresponding flank wear are collected to provide data support for subsequent research.Aiming at the recognition accuracy of milling tool wear condition,a novel method based on radar map feature fusion and signal singularity analysis is proposed.The feature extraction and correlation analysis of multi-sensor signals are carried out,and the health indicator of tool wear evolution is obtained by radar map feature fusion.An ensemble learning model of adaptive gradient boosting decision tree is established to realize online recognition of tool wear conditions.The effectiveness and versatility of the proposed method are verified by milling public datasets and high-feed milling experiments.The signal singularity analysis theory is introduced for the error recognition samples in the transition region of different wear stages to establish the mapping relationship between the mean value of the Holder exponent,the number of singular points,and the tool wear condition.The misclassified samples are modified by finding the transition points between different wear stages to further improve the recognition accuracy of tool wear conditions.Aiming at the reliability problem of the milling tool remaining useful life(RUL)prediction,a data-model linkage tool RUL prediction method considering the wear physical process is proposed.In the data-driven module,a convolutional stacked bi-directional long short-term memory network with a temporal-spatial attention mechanism is developed to learn the nonlinear mapping relationship between multi-channel monitoring signal features and tool wear,and then predict the tool wear.In terms of physical modeling,a three-stage RUL prediction model based on the Wiener process is developed considering the physical process of tool wear and prediction uncertainties,and the probability density distribution of RUL corresponding to different wear stages is calculated.The weight-optimized particle filter algorithm under the Bayesian framework is used to dynamically update the physical model parameters,thus realizing a data-model linkage tool RUL prediction.The effectiveness of the proposed method is verified by tool life milling experiments under single and multiple conditions.For the problem of sparse and imbalanced tool breakage data in the milling process,a tool breakage monitoring method based on an auxiliary classier Wasserstein generative adversarial network with gradient penalty term is proposed from the perspective of data generation.The one-dimensional convolutional neural network is used to construct the generator and discriminator,and the loss function of the network training process is improved to realize the generation of multi-mode breakage fault samples of milling cutter,including chipping,one tooth broken,two teeth broken,etc.A sample filter based on multiple statistical indicators is designed to further ensure the quality of the generated data.The qualified samples after quality assessment are added to the original imbalanced dataset to improve the performance of the tool breakage classifier.Finally,the feasibility of the proposed method is verified through artificially controlled tool breakage monitoring experiments.