Sensorless Tool Wear Estimation by using the Artificial Intelligence (AI) tools from the currents of motors generating linear motions

Abstract

Timely replacement of cutting tools reduces machining costs and prevents the manufacture of defective products. Many researchers have developed Tool Condition Monitoring (TCM) systems to estimate tool wear using reliable, low-cost instrumentation. This paper proposes estimating tool wear by interpreting motor current signals from the programming logic controllers (PLCs) of CNC machines with artificial intelligence (AI) tools. Experimental data were collected from three cutting tools at four wear states: normal, and with one, two, or three worn cutting edges. Four AI models were used to classify normal and anomalous cases: 1D Convolutional Neural Networks (1D CNN), Long Short-Term Memory (LSTM) networks, a hybrid 1D CNN-LSTM architecture, and an LSTM autoencoder. Hyperparameter tuning was performed to optimize each model. Three approaches were explored to address different production needs: · Approach A - Supervised Learning: 1D CNN showed the best performance, achieving 90% validation accuracy. · Approach B - Semi-Supervised Learning: An LSTM autoencoder trained only on “normal” data achieved 96% validation accuracy but a low F1 score of 0.51, indicating limited anomaly detection capability. · Approach C - Transfer Learning: Using transfer learning, the 1D CNN model reached 85% validation accuracy and an F1 score of 0.84 when tested on a third tool with a different diameter.