An accurate detection of tool wear type in drilling process by applying PCA and one-hot encoding to SSA-BLSTM model

Tool condition monitoring (TCM) is significant in advanced manufacturing systems for achieving high productivity in the manufacturing industries. The main objective of the research study is to design a high-quality tool wear detection system. This methodology was experimentally executed by installing the dynamometer to the CNC drilling machine to perform drilling operations using the normal tool, crater wear tool, chisel wear tool, flank wear tool, and outer corner wear tool. The removal of noise from the raw force signal data and extraction of features was done by using the singular spectrum analysis (SSA) algorithm. The algorithm based on the technique of principal component analysis (PCA) is designed for dimensionality reduction and for improving performance, accuracy, and efficiency by avoiding the overfitting of the model. Early stopping and dropout algorithms are designed to effectively overcome the overfitting problem. Both techniques have made the training process efficient by automatic selection of the most suitable number of epochs. The textual form of target variables of the model was converted into binary numerical form by using one-hot encoding as the deep learning algorithm can read numerical data only. The model would determine whether the test tool is worn or not and would also predict the type of tool wear and achieved an accuracy of 97.94%.