Figure 1
The diagram illustrates a workflow for a physics-guided anomaly detection framework. It begins with raw machining data acquisition from a CNC vertical milling machine, monitoring parameters such as spindle speed, depth of cut, feed rate, and tool diameter. The MTConnect multi-sensor monitoring system synchronizes time series data for spindle power, energy consumption, vibration RMS, spindle load, and spindle speed. Interval-wise cutting data extraction identifies key novelty intervals by extracting cutting data and removing idle or non-cutting data. Statistical feature extraction calculates mean, standard deviation, and maximum values for each sensor signal, creating a feature vector. The physics-guided unified damage index (UDI) is formulated using weights for vibration, power, depth of cut, and feed rate. XGBoost-based anomaly detection involves model training with data splits, hyperparameter tuning, and cross-validation to predict healthy or anomalous states.Proposed workflow of the physics-guided anomaly detection framework