Tool life prediction integrating cutting power: a comparative analysis of machine learning models

Abstract Monitoring and predicting tool wear remains a critical challenge in hard turning operations because it directly affects process efficiency, surface integrity, and manufacturing sustainability. Although cutting power is readily available in industrial environments, its potential as an indirect indicator of tool condition and an auxiliary predictor of tool life remains limited. This study proposes a Machine Learning (ML) framework that integrates cutting power consumption with machining parameters to improve tool life prediction in the dry turning of hardened AISI 52,100 steel using mixed ceramic inserts. A Central Composite Design with three factors (cutting speed, feed rate, and depth of cut) was employed. Several ML surrogate models optimized through Bayesian Optimization were evaluated under three predictive scenarios: (i) machining parameters predicting tool life ( T ); (ii) machining parameters predicting cutting power ( Pc ); and (iii) machining parameters combined with Pc predicting T. In general, the inclusion of cutting power contributed to improved predictive performance, reducing errors for some of the evaluated models. Gaussian Process Regression achieved the best overall performance for tool life prediction (RMSE = 0.6998) in Scenario 3, whereas Artificial Neural Networks provided the best power prediction (RMSE = 0.1583) in Scenario 2. The results demonstrate that low-cost cutting power can act as a reliable indirect indicator of tool condition, enabling more accurate tool life prediction without additional sensor infrastructure. This approach supports sustainable and efficient machining operations by balancing productivity and resource conservation while avoiding premature tool replacement and the energy waste associated with using worn tools.