Monitoring centerless grinding processes for workpiece roundness prediction using internal machine sensors and Bayesian sparse regression models

Grinding is widely used to produce high-precision components with excellent surface quality. In centerless grinding, guaranteeing consistent quality under variable process conditions typically requires systematic roundness measurements that are significantly slower than the process itself, creating a gap between production speed and quality verification. This study aims to develop an intelligent monitoring system capable of detecting potentially non-conforming parts in centerless grinding using only information available from the machine Numerical Control. The proposed approach exploits process parameters and signals directly accessible from the machine controller, demonstrating that they contain sufficient information to compensate for missing information such as the actual raw workpiece geometry. Experiments were conducted on an industrial centerless grinding machine and verified by roundness measurements of all finished workpieces, resulting in a labelled dataset made publicly available. Features were extracted from the signals in the time, frequency, and time–frequency domains. Several modelling approaches were evaluated, including Bayesian sparse regressions with Least Absolute Shrinkage and Selection Operator regularization and Stochastic Search Variable Selection with conjugate priors, as well as Gaussian Processes and Random Forests. A physics-guided analysis was also performed to investigate the impact of process parameters and design additional informative features. The proposed models achieved a Root Mean Squared Error of about 1 µm and a relative Hybrid Error of about 10% on the test dataset. A Bayesian-informed regression modelling pipeline has been proposed, which achieves the best predictive accuracy and enabled identification of key features, while ridge regression models with correlation-guided feature selection yielded compact models suitable for real-time implementation. Machine-controller signals alone can support accurate monitoring of workpiece roundness in centerless grinding, enabling fast detection of potentially non-conforming parts while improving interpretability and robustness for industrial deployment.