Cutting Force Prediction in Drilling Operations Using Temporal Convolutional Networks (TCN) and Spindle Speed Data

The prediction of forces in drilling is vital for enhancement of machining process, reduction of tool wear rates, and quality improvement of the final product. The present work introduces a new method for force estimation based on Temporal Convolution Networks (TCNs) and trained with spindle speed data. TCNs take advantage of the ability to process sequential data that perform well in capturing short-term and long-term dependencies, and hence are used in time-series analysis to analyze machining environments that are ever changing. The process is real-time acquisition of spindle speed and cutting force signals, filtering of signals to remove noise, and feature extraction for training of the models. To make correct predictions while making efficient computations, a TCN architecture that has been specifically customized with causal and dilated convolutions is used. The proposed model has MAE = 2.15 N and 98.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> inclusive squared coefficient of determination, which is significantly better than traditional machine learning models like ANN, SVM, and LSTM and faster on inference time. Indeed, all of the present model's validation has been done across different spindle speeds and for different types of materials. Also, the TCN model implementation into real-time monitoring allows optimizing the modelling parameters in proximity to changing conditions, which is a feature of industry 4.0. It also strongly supports the practice of sustainable manufacturing, which in turn improves the efficiency of machining. High value is credited to the potential of TCNs as a transformative tool in intelligent machining systems.