Multiple input CNN architecture for tool state recognition in the milling process based on time series signals

The study presents a tailored application of a multiple-input convolutional neural network (CNN) for tool state recognition in the milling process. Our approach uniquely applies an 11-input CNN to classify tool wear in chipboard milling, utilizing scalogram images derived from time-series signals. The primary objective was to categorize tool wear into three classes: green, yellow, and red, signifying the progression of wear. The study involved 75 samples (25 samples per class), each comprising 11 signals transformed into scalograms via continuous wavelet transform. The dataset of 825 scalogram images enabled the development of a CNN-based diagnostic model, achieving a notable accuracy of 96.00%, which is an improvement over a previous methods (93.33%).