Innovative tool condition classification: utilizing time–frequency moments as inputs for BiLSTM networks in milling processes

Abstract

Milling is one of the most important processes in the manufacturing industry, and it uses rotating cutting tools to sculpt raw materials into intricate shapes and structures. However, tool wear and breakage present significant challenges influenced by various factors, such as machining parameters and tool fatigue, which directly impact surface quality, dimensional accuracy, and production costs. Therefore, monitoring cutter wear conditions is essential for ensuring milling process efficiency. This study proposes applying BiLSTM networks to classify end mill cutter conditions based on vibration signals. Significant improvements in classification accuracy are achieved by extracting features and employing spectrogram analysis. Specifically, using dual spectral features, instantaneous frequency and spectral entropy, increases the BiLSTM’s average accuracy from 86 to 98.5%, based on a comparative analysis of models trained with raw vibration signals and those trained with extracted spectral features. These findings demonstrate the effectiveness of the proposed method for real-time cutter condition monitoring in milling operations, offering potential benefits for manufacturing processes.