Tool wear monitoring for robotic milling based on multi-dimensional stacked sparse autoencoders and bidirectional LSTM networks with singularity features

Abstract

Abstract This study addresses the challenges posed by the vibration-induced wear and breakage of milling cutters during the machining large parts using industrial robots with six degrees of freedom. The proposed tool wear monitoring method (TWM) relies on a sophisticated framework that integrates a multi-dimensional stacked sparse autoencoders (MD-SSAEs) network and bidirectional long short-term memory networks (BiLSTM) incorporating singularity features. The method begins with a singularity analysis (SA) approach, which is employed to extract local features and eliminate the impact of irregular fluctuations. Following this, MD-SSAEs are strategically designed to conduct dimension reduction of SA features and facilitate the deep fusion of multiple features. Subsequently, BiLSTM is employed to map the deep-fused features and model the relationship between continuous tool wear progression. Finally, two milling experiments with full wear cycle were carried out on a self-made robot milling platform to verify the effectiveness of the proposed method. The experimental results affirm that the established method demonstrates exceptional prediction accuracy and robust adaptability to variations in cutting parameters. Leveraging this approach, a TWM system is developed, providing an effective tool replacement guide for real-world manufacturing scenarios.