Intelligent monitoring of drilling tool wear with a sensory tool holder

Abstract

During machining, friction between the tool and a workpiece or shavings causes unavoidable physical or chemical damage to the tool. The amount of damage depends on several factors, including the machining parameters, material hardness, tool material, and cooling. To achieve the online estimation of the tool wear condition, a sensory tool holder (STH) was developed in this study to measure the thrust and torque signals of a drill during machining. Motor current signals from the spindle and feed shaft and the machining center were also measured. The collected current and STH signals were processed using two methods. First, empirical mode decomposition was performed to obtain intrinsic mode functions from the first five iterations as feature inputs for training wear estimation models. Second, third-order discrete wavelet transform (DWT) was performed to obtain approximation coefficients as the feature inputs. Moreover, an image-based tool wear detection system with a charge coupled device camera was installed in the machine to quantify the tool wear after drilling; the wear data were labeled and used as input labels for model training. Three algorithms were used to train various models using the feature inputs obtained from the two aforesaid processing methods, and a 1D convolutional neural network (CNN) model on the STH dataset with DWT was verified to demonstrate the highest accuracy, yielding a mean absolute error of 0.013, root mean square error of 0.018, and coefficient of determination R² of 0.933. The use of this system helps to significantly reduce the preliminary costs of building wear models. Through the combination of near real-time wear image measurement and STH, the time and labor consumption of wear experiments can be reduced. The finally deployed CNN model can achieve real-time drill bit wear prediction during STH machining.