Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-08-0270
Tool Condition Prediction by Process Monitoring using Semi-Supervised Learning
1Singapore Institute of Manufacturing Technology, 2 Fusionopolis Way, #08-04, Innovis, 138634, Singapore
ABSTRACT
In precision machining process, part fabrication quality often has a great deal to do with tool condition. For aerospace MRO (Maintenance, Repair and Overhaul) shopfloor, tool condition is particularly critical when the part-to-repair is of substantial size and machining cycle is prolonged, especially during finishing phase when tool change is not allowed due to elevated precision requirement. It is our endeavor to announce a tool change request only when change is possible, right before fabrication quality is to be compromised. Our approach predicts tool condition progression aiming to remind the operator of the best moment for a tool change in achieving both part quality assurance and tool cost saving. The work starts with process data collection in realtime including vibration, current, spindle load, feed rate, etc., and the actual tool wear (flank wear) condition measured from a microscope. Since the tool condition can only be inspected after prolonged machining cycles, the amount of labeled tool wear samples is limited. Hence, a 2-views semi-supervised approach is chosen to utilize information from both labeled and unlabeled datasets. Experimental results support the capabilities of the proposed model in predicting tool condition for a turning process in MRO shopfloor.
Keywords: MRO shopfloor, Precision machining, Process signals, Tool wear prediction, Semi-supervised learning
1Singapore Institute of Manufacturing Technology, 2 Fusionopolis Way, #08-04, Innovis, 138634, Singapore
ABSTRACT
In precision machining process, part fabrication quality often has a great deal to do with tool condition. For aerospace MRO (Maintenance, Repair and Overhaul) shopfloor, tool condition is particularly critical when the part-to-repair is of substantial size and machining cycle is prolonged, especially during finishing phase when tool change is not allowed due to elevated precision requirement. It is our endeavor to announce a tool change request only when change is possible, right before fabrication quality is to be compromised. Our approach predicts tool condition progression aiming to remind the operator of the best moment for a tool change in achieving both part quality assurance and tool cost saving. The work starts with process data collection in realtime including vibration, current, spindle load, feed rate, etc., and the actual tool wear (flank wear) condition measured from a microscope. Since the tool condition can only be inspected after prolonged machining cycles, the amount of labeled tool wear samples is limited. Hence, a 2-views semi-supervised approach is chosen to utilize information from both labeled and unlabeled datasets. Experimental results support the capabilities of the proposed model in predicting tool condition for a turning process in MRO shopfloor.
Keywords: MRO shopfloor, Precision machining, Process signals, Tool wear prediction, Semi-supervised learning
