Deep learning based intelligent CNC machining for error prediction using multi-sensor data

Precision machining is critical in modern industry, where minimizing errors ensures product quality and reduces defects. Traditional methods for Computer Numerical Control (CNC) machining estimation rely on either manual experience, physical models, or objective data support, which results in premature tool replacement or excessive tool wear, ultimately increasing the manufacturing cost. This study proposes a Deep Learning (DL)-based framework that uses multi-sensor time-series data for CNC machining error prediction. The dataset was collected from a CNC lathe by using vibrational and acoustic sensors along with the machine controller signals for rotational speed and the current of the spindle and turret. Four DL architectures, i.e., 1D CNN, ResNet, DeepConvLSTM, and InceptionTime, were evaluated to identify the most effective model for CNC machining error prediction. InceptionTime achieved minimum Mean Squared Error (MSE) and Mean Absolute Error (MAE), outperforming other DL models. This enables real-time error estimation in CNC machining, enhancing precision, productivity, and optimization.