Efficacy of Statistical Formulations on Acoustic Emission Signals for Tool Wear Predictions

Acoustic emission (AE) signals obtained during machining processes can be used to detect, locate and assess flaws in structures made of metal, concrete or composites. This paper aims to characterize AE signals using derived parameters from raw signatures along with statistical feature extractions to correlate with tool wear readings. Missing tool wear values are imputed using domain knowledge rules and compared to AE signals using machine learning models. The amount of effect on tool wear is formulated using Bayesian Inferences on derived parameters such as areas under the raw signal curve in addition to comparisons with the supervised models for predictions. Using the constructed models and formulation, the presented study also includes a trace-back pseudo-algorithm for determining the stage in process where tool wear values begin to approach the wear limits.