Material recognition for fault diagnosis in machine tools using improved Mel Frequency Cepstral Coefficients

Identification of material type in structural cracks is crucial for fault diagnosis and structural integrity assessment in multi-component operation systems. In the current literature, acoustic emission (AE) signal analysis has been extensively explored as an effective method for crack monitoring and fatigue damage assessment. Most existing research on crack diagnosis is primarily focused on determining the crack formation in structures using the AE signals corresponding to known materials. However, the material recognition at any given cracking conditions using AE signals has lagged behind, hindering fault detection and damage repair in complex and multi-material structures. This research investigates the feasibility of the synergistic use of Mel Frequency Cepstral Coefficients (MFCC) and K-nearest neighbor (KNN) algorithms for identifying the material type in any arbitrary cracking processes. In the proposed method, MFCC and first-order differential MFCC parameters are extracted from the real-time AE signals. Material-specific cracking characteristics are identified and tuned for material recognition using a KNN classifier. The proposed method is validated using the AE signals of different types of materials acquired under uniaxial tension, and its effectiveness in identifying tool breakage is evaluated through case studies in the milling process. The result indicates that the recognition accuracy of up to 96.21 % can be achieved in tool breakage detection in milling operations by extracting material-specific fracture characteristics.