Machine learning based approach for process supervision to predict tool wear during machining

Abstract Tool wear prediction during machining is a challenging problem. Traditional approaches are available to use the process parameters which influence tool wear but there are certain parameters which are very specific to the machining process and available prediction models fail. Present work discusses a Machine Learning based process supervisory system to predict the tool wear. To illustrate the approach an application for the prediction of tool wear while machining is selected as a case study. The analysis was performed on a machining dataset consisting of certain experiments of different levels of input parameters and for each experiment several sensor logged physical parameters (features). From a chosen training set of experiments the features that best describe the state of tool wear (unworn or worn) along with the input parameters were chosen to build a model. Several models employing logistic regression were built and the best one was chosen. The model obtained had good accuracy and interpretability. The results obtained from the test set show the system’s suitability and potential for industrial application. The presented supervisory model can be utilized to predict tool wear in real time and prior to the tool getting worn before a set number of operations, thus cause a reduction in the delay due to the change over required to an unworn tool.