Proceedings of the
World Congress on Micro and Nano Manufacturing (WCMNM 2022 )
19–22 September 2022, Lueven, Belgium
doi:10.3850/978-981-18-5180-3_RP54-0053
Prediction of Milling-µEDM Tool Wear: The Benefit of Process Monitoring and Machine Learning Model
Micro -& Precision Engineering Group, Department of Mechanical Engineering, KU Leuven, Belgium
Members Flanders Make (https://www.flandersmake.be/en), Leuven, Belgium
ABSTRACT
Tool wear in the milling-µEDM process always adversely affects the fabrication accuracy of micro-components and thus requests an advanced prediction technique before a compensation strategy. Traditional efforts in analytical models are limited in prediction accuracy considering the complexity of the tool wear phenomenon. As a representative of the complex discharge condition, the process indicators are effective in predicting the tool wear in real-time. However, empirical indicators require a subtle calculation specific to the generator types and are often sensitive to the preset thresholds. This paper presents the application of statistical indicators that are computed from the discrete wavelet transform (DWT). The DWT indicators are proven to effectively encode the essential information about the single pulse and pulse trains in the temporal and frequency domain. A machine learning model receiving these DWT indicators is trained to predict the layer-wise tool wear. The prediction accuracy shows that the DWT features outperform the empirical indicators and demonstrate their potential as generalized indicators for tool wear monitoring.
Keywords: Machine Learning, Process Monitoring, Micro-EDM, Milling Tool Wear.
Micro -& Precision Engineering Group, Department of Mechanical Engineering, KU Leuven, Belgium
Members Flanders Make (https://www.flandersmake.be/en), Leuven, Belgium
ABSTRACT
Tool wear in the milling-µEDM process always adversely affects the fabrication accuracy of micro-components and thus requests an advanced prediction technique before a compensation strategy. Traditional efforts in analytical models are limited in prediction accuracy considering the complexity of the tool wear phenomenon. As a representative of the complex discharge condition, the process indicators are effective in predicting the tool wear in real-time. However, empirical indicators require a subtle calculation specific to the generator types and are often sensitive to the preset thresholds. This paper presents the application of statistical indicators that are computed from the discrete wavelet transform (DWT). The DWT indicators are proven to effectively encode the essential information about the single pulse and pulse trains in the temporal and frequency domain. A machine learning model receiving these DWT indicators is trained to predict the layer-wise tool wear. The prediction accuracy shows that the DWT features outperform the empirical indicators and demonstrate their potential as generalized indicators for tool wear monitoring.
Keywords: Machine Learning, Process Monitoring, Micro-EDM, Milling Tool Wear.