Prediction of tool wear in milling operations using machine learning techniques

One of the fundamental components of Industry 4.0 is the transformation of manufacturing processes into Smart Manufacturing through Big Data Analytics and Machine Learning Techniques. Deep Learning, considered the latest and most impactful of the Machine Learning techniques, is rapidly gaining popularity in sectors such as automotive, aerospace, healthcare, and finance, leading to revolutionary new products and applications in these sectors. In the field of manufacturing, it is suggested that the most prominent research potential lies in the areas of Advanced Condition Monitoring, Process Optimization, and Predictive Maintenance. Advanced process monitoring relies on the timely collection of machine and system generated data with sufficient accuracy, resolution, and reliability. In this context, vibration, acoustic, and machine power consumption data were collected using an accelerometer, microphone, and energy meter, respectively. At the same time, the value of flank wear occurring on the cutting tool was collected using a microscope, and the surface roughness of the machined part was also collected using a surface roughness measuring device. The aim of the studies conducted within the scope of this thesis is to develop Machine Learning-based models for process control and monitoring parameters, surface roughness of the machined part, consumed energy, and cutting tool wear by applying deep learning techniques to the collected data together with machining processes. A Deep Multi-Layer Perceptron (DMLP) based prediction model has been developed to predict the surface roughness and specific cutting energy (ÖKE) of the machined part. In addition, two different prediction models based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) have been developed to predict the wear occurring on the cutting tool. Using the DMLP-based prediction model developed within the scope of the thesis, it is observed that the error rate between the obtained specific cutting energy and surface roughness results and the target outputs is below 10%. In addition, 8 different experiments were conducted to train the CNN and LSTM-based models developed for the prediction of cutting tool wear. The models were trained using the vibration and acoustic data obtained from the experiments. The experimental data not used during training was used as test data. Here, 5 different scenarios were studied for both prediction models. The aim in the scenarios is to try to predict the wear occurring on the cutting tool for different cutting speeds (m/min) and different cutting tool diameters. When the best results in the scenarios are examined, the average error value of the Wavelet-CNN based prediction model is 0.031, while the average error value of the WLSTM-DMLP based prediction model is 0.004.