Automated Fault Diagnosis with Calibrated Artefact Probing

Modern five-axis machining centres are able to manufacture parts with extreme pre- cision; they are, however, not free from error, and error affecting the machine kinematics in particular is observed to change with time. To maintain tolerance requirements, machine operators will routinely verify the kinematic health and take action where necessary. One common method for verifying rotary-axis health is calibrated artefact probing, the results of which can be analysed to identify specific kinematic fault states and inform maintenance actions. This analysis requires specialist knowledge, so in order to ensure fault diagnosis is accessible to all users, an automated inference system is desirable. A Residual Neural Network was trained with artificially-generated data, to identify seven common kinematic fault states. The classifiers performed favourably when applied to a test set acquired from operational machine tools, indicating that the approach is viable. However, some faults were found to be easier to identify than others, so further work on developing the training set is required before the method can be deployed in industry.