Research On Tool Wear And Its Condition Monitoring Technology During The Cutting Of Nickel-based Superalloys

The huge demand for products such as aero-engine and gas turbines has driven the manufacture of superalloy components.Superalloys are typically difficult materials to machine.The cutting process of superalloy has the characteristics of fast tool wear,low tool life,high cutting temperature,severe work hardening and poor quality of the machined surface of the workpiece.Tool wear is closely related to tool life,dimensional accuracy of superalloy workpieces,machined surface quality of superalloy workpieces,intelligent operation and maintenance of smart manufacturing lines,etc.For many years,tool wear during cutting of superalloys has mainly focused on the study of the tool wear mechanism in the sharp wear stage.These studies mainly focus on various cutting methods,various superalloy materials,various cutting tool substrate and coating materials,different tool geometry parameters,different cooling methods,etc.Related studies neglected the study of tool wear morphology and wear mechanisms in the initial and stable wear stages during the cutting process of superalloys.In recent years,artificial intelligence(machine learning,deep learning,transfer learning,etc.)has been introduced into the field of tool wear condition monitoring and tool life prediction,which has contributed to the development of smart manufacturing.Tool wear is influenced by the cutting parameters,the cutting method(turning,milling,drilling,grinding,etc.),the tool material of cutting tool,the geometrical parameters of the cutting tool and the cooling conditions.Due to the high cost of test workpiece materials and tools,high time and manpower costs,and expensive cutting test equipment,high resolution images of tool wear morphology and wear mechanisms during cutting of superalloys are scarce.So it is a difficulty to construct high-resolution image datasets of tool wear for artificial intelligence learning.In-depth studies of tool wear morphology and wear mechanisms in the initial and stable wear phases of turning and milling processes help to examine the cutting performance of tool coatings and substrate materials.And in-depth exploration of the difficult machining characteristics of superalloy GH4169 provides a reference for the reasonable selection of cutting parameters for turning and milling of superalloy GH4169.This helps to ensure the dimensional accuracy and machined surface quality of high value-added superalloy components during the machining process.High-resolution images of tool wear acquired during cutting of nickel-based superalloys are used to construct an image dataset for the study of tool wear condition monitoring techniques.The main contents of the paper are as follows.(1)A study on tool wear microscopy,wear mechanism,wear curve and cutting force during the turning of nickel-based superalloy GH4169 was carried out.And the process of initial,stable and rapid wear stages of abrasive wear and adhesive wear were revealed by optical microscopy and scanning electron microscopy(SEM).Severe abrasive wear and adhesive wear were found in the initial,steady and rapid wear stages,severe wear of the tool tip and tool base material occurred,the build-up edge was found near the cutting edge.The cutting tool is subjected to severe thermal and mechanical fatigue as well as cutting between hard carbides in superalloy workpiece and cutting tool.These causes lead to severe wear of the tool by abrasive particles.Significant adhesion was found on the flank and rake faces,cutting edges and tool nose,superalloy workpiece materials such as Ni,Fe,Cr,Nb and Mo were detected on the flank and rake tool faces.These workpiece materials were repeatedly adhering and detached from the cutting tool surface,resulting in adhesive wear on the tool surface during cutting.From the tool wear curve,it can be seen that tool wear rate is very fast during turning superalloy GH4169.Under the combined effect of multiple tool wear mechanisms,as the average wear width of the flank tool face increases,the absolute mean value of the cutting force and its increase rate from the largest to the smallest in order of the main cutting force Fc,backward force Fp and feed force Ff.Energy dispersive spectroscopy(EDS),X-ray photoelectron spectroscopy(XPS)and X-ray diffraction(XRD)methods have identified a variety of oxides(Al2O3,etc.)on the tool surface during tool wear,which damage the tool coating and substrate structure and reduce the cutting performance of the tool,leading to oxidative wear of the tool during cutting.The diffusion of workpiece materials such as Ni,Fe and Cr into the tool substrate material during the rapid wear stage was detected by SEM and EDS.This phenomenon destroys the tool substrate structure and reduces the cutting performance of the cutting tool,which leads to diffusion wear of the cutting tool,with the most severe diffusion of workpiece materials along the tool nose compared to diffusion along the rake and flank tool faces.(2)The 45 groups of tool wear milling tests were carried out under different cutting conditions.The tool wear microscopy,wear mechanisms,wear curves,cutting forces,spindle motor currents and the morphology of the machined surface of the workpiece were studied by optical microscopy,SEM,EDS,XRD,force measurement and current sensors.The 45 groups of wear curves during the milling of nickel-based superalloy GH4169 tool wear were obtained,milling speed and axial depth of cut had the most significant effect on tool wear.The variation of cutting force and spindle motor current with the average wear width VB of the tool flank face was obtained.Under the interaction of various tool wear mechanisms,the cutting forces Fx,Fy and Fz increase by more than 30%at 150 μm<VB<300 μm,and tool wear has the greatest effect on Fx and Fy.The spindle motor current increases significantly with tool wear,with the spindle motor current increasing by more than 60%at 100 μm<VB<300 μm.During the milling of the superalloy GH4169,the cutting tool suffered from severe adhesive wear and abrasive wear,and a large number of different microscopy of adhesions,built-up edge and built-up layer were found.Various oxides(Al2O3 etc.)that cause oxidative wear on the cutting tool surface have been identified by EDS and XRD methods.The vibration and shock caused by the periodic intermittent cutting during milling of superalloy GH4169 leads to a very serious breakage of the cutting nose and cutting edge.In the sharp wear stage,different microscopic morphology of chips and adhesions were caused by strong extrusion and stretching.The microstructure of the machined surface of the workpiece in terms of ridges and grooves is closely related to the different stages of tool wear.(3)The "Fine-tune" method of transfer learning based on a lightweight network model was used,a study was carried out to monitor the state of tool wear at different wear stages of the cutting process of superalloy GH4169 by optical microscopy images.The results of the study show that when the EfficientNet-b0 network model is used,the accuracy of the rake and flank tool faces wear image monitoring during turning is 72.43%and 83.10%,respectively,while the accuracy of the rake and flank tool faces wear image monitoring during milling is 88.62%and 92.81%,respectively.Compared with the MobileNet-v2,ShuffleNet and SequeezeNet lightweight network models,the EfficientNet-b0 network model has higher monitoring accuracy.(4)The Inception-ResNet-v2 network model was compressed using the "knowledge distillation" method,and research was conducted on tool wear mechanisms during cutting of superalloy GH4169 by optical microscopy and scanning electron microscopy image condition monitoring techniques.The results of the study show that when the Inception-ResNet-v2 compression network model is used,the image monitoring accuracy of the rake and flank tool faces wear mechanism during turning is 85.39%and 86.29%,respectively,while the image monitoring accuracy of the rake and flank tool faces wear mechanism during milling is 82.22%and 90.66%,respectively.Compared with eight network models such as Inception-ResNet-v2 and its compression model,RestNet50,RestNet101,InceptionV3,Xception,InceptionResNet-v2 compressed network model has 34%of the number of parameters than the uncompressed state,and has higher state monitoring accuracy.