In-line statistical monitoring of machine tool thermal error through latent variable modeling

Because thermally induced errors account for a large percentage of machine tool errors, in-line monitoring of machine thermal stability is a very important issue for part quality. Although system identification (SI) theory has been used in modeling machine tool thermal errors for improving accuracy and robustness, there are several unsolved issues-for example, how to determine the number of thermal sensors that is sufficient for building an SI model, especially when sensor readings are highly correlated; and how to effectively monitor the machine thermal status and predict machine performance when the sensing resource is limited. This paper presents a new concept-in addition to the widely recognized error avoidance and error compensation approaches-of controlling machining thermal effects by monitoring of machine thermal status. Based on an experimental study, an in-line monitoring method based on Latent Variable Modeling (LVM) is proposed to overcome the aforementioned difficulties. The results have shown that the LVM method based in-line monitoring provides a powerful tool for variable selection and is effective in monitoring variations of thermal status when the values of thermal deformation and temperature are highly correlated.