In-situ Milling Cutter Condition Monitoring Based On Machine Vision

Cutting tool is the direct performer of the cutting process.Good tool wear condition is one of the important factors to ensure the safety,reliability,and stability of the machining process.The tool wear condition monitoring method based on machine vision is not affected by cutting conditions and workpiece materials.It has higher accuracy and reliability in measuring the actual geometric changes of tool wear,which is convenient to understand the tool wear mechanism.Moreover,it is helpful for optimization of machining parameters,prediction of tool life,and evaluation of machining quality.This paper will focus on the key technologies of machine vision based in-situ milling cutter condition monitoring,and the influencing factors that limit the industrial application of the direct method.On this basis,the identification and evaluation methods of tool wear conditions will be researched.The research results are of great significance to promote the intelligent and automatic development of tool condition monitoring based on machine vision.The main research work is as follows:1.Aiming at the problem of long installation time of machine vision system with small field and high precision,and how to deploy tool condition monitoring system to capture high quality tool image,hardware selection of machine vision monitoring system under wide field of view was carried out,and polynomial regression model based on improved information entropy was proposed.Firstly,the machine vision hardware system is selected according to the coordination relationship between the tool wear detection accuracy and the wide field of view,as well as the dynamic detection requirements.The tool condition monitoring system is guaranteed to be fixed on the machine tool table by switching the working mode.In addition,the evaluation metrics of information entropy influenced by the background model has a main effect relationship with the working distance and exposure time,which accords with the objective variation law of the deployment parameters of the tool condition monitoring system.2.Regarding the issue that a single image cannot fully reflect the tool wear characteristics and capturing a single image will bring downtime of the machine tool,the concept of tool condition image sequence(TCIS)and an adaptive capture method are proposed.Based on the wear mechanism,imaging principle and experimental verification method,the concept of the TCIS is proposed for the first time,which explains which images can comprehensively reflect tool wear characteristics in dynamic images.Then,the adaptive identification of the TCIS is proposed based on oriented gradient histograms,image coding,and logistic regression model.Finally,the tool wear area in TCIS is located by balancing wear measurement elements and wear measurement benchmarks,and the accurate tracking of subsequent tool wear area is realized based on motion model and local search method.3.Regarding the issue that traditional high-complexity neural network models are difficult to deploy in industrial applications and deep learning requires high-quality and quantitative data sets,a light weight network based on data augmentation and multiple activation functions is proposed.Firstly,the mechanism of imaging quality change caused by complex working conditions in industrial environment is analyzed.On this basis,data augmentation is proposed to solve the problem of data scale on the premise of ensuring data quality and richness.Then,an adaptive activation function and an h-swish activation function are introduced in the front end and back end of the network respectively to avoid information loss and reduce the cost caused by the activation function.Finally,a light weight tool wear classification network model based on edge-cloud collaboration is constructed.The model is optimized iteratively in the cloud and deployed on edge embedded devices.4.Regarding the issue that the tool image captured by the high-resolution camera in the wide field of view occupies a large amount of memory,a high-precision tool wear detection method is proposed,including location,segmentation and measurement.Firstly,the homomorphic filter and histogram contrast method are used to roughly locate the tool wear area.On this basis,the Grab Cut algorithm is improved to achieve accurate segmentation of the tool wear area.Finally,the least square method is used to fit the main cutting edge as the measurement benchmark,so as to realize the division and measurement of the flank wear area.5.Regarding the issue that the existing tool wear evaluation metrics cannot distinguish different flank wear forms because of ignoring the structural characteristics of the tool wear area,a novel flank wear evaluation metric based on the wear distance dispersion is proposed.Firstly,a normal distribution model is constructed with prior knowledge to describe the probability distribution of the wear distance in the indirect tool-chip contact area in a wide field of view.The confidence interval is calculated by the probability distribution threshold to realize the extraction of the indirect tool-chip contact area.Then,a novel metric based on the wear distance dispersion in the indirect tool-chip contact is proposed to describe the flank wear forms in more detail.Finally,the change rate of distance dispersion with machining time is constructed to identify the time nodes when three kinds of flank wear forms change in the tool life cycle.