Error motion trajectory-driven diagnostics of kinematic and non-kinematic machine tool faults

Error motion trajectory data are routinely collected on multi-axis machine tools to assess their operational state. There is a wealth of literature devoted advances in modelling, identification and correction using such data, as well the collection processing alternative streams for purpose tool condition monitoring. Until recently, there has been minimal focus combining these two related fields. This paper presents general approach identifying both kinematic non-kinematic faults error by framing issue generic pattern recognition problem. Because typically-sparse nature datasets this domain – due infrequent, offline procedures foundation involves training purely simulated dataset, which defines theoretical fault-states observable trajectories. Ensemble methods investigated shown improve generalisation ability when predicting experimental data. Machine often have unique ‘signatures’ can significantly-affect trajectories, largely repeatable, but specific individual machine. As such, experimentally-obtained will not necessarily be easily defined simulation. A transfer learning introduced incorporate trajectories into classifiers were trained primarily simulation domain. The was significantly test set performance, whilst also maintaining all information learned initial, simulation-only phase. ultimate represents viable powerful automated classifier encode with efficacy remain adaptable machine-specific signatures.