Acoustic Emission Characterization of Natural Fiber Reinforced Plastic Composite Machining Using a Random Forest Machine Learning Model

Abstract

Natural fiber reinforced plastic (NFRP) composites are eliciting an increased interest across industrial sectors, as they combine a high degree of biodegradability and recyclability with unique structural properties. These materials are machined to create components that meet the dimensional and surface finish tolerance specifications for various industrial applications. The heterogeneous structure of these materials—resulting from different fiber orientations and their complex multiscale structure—introduces a distinct set of material removal mechanisms that inherently vary over time. This structure has an adverse effect on the surface integrity of machined NFRPs. Therefore, a real-time monitoring approach is desirable for timely intervention for quality assurance. Acoustic emission (AE) sensors that capture the elastic waves generated from the plastic deformation and fracture mechanisms have potential to characterize these abrupt variations in the material removal mechanisms. However, the relationship connecting AE waveform patterns with these NFRP material removal mechanisms is not currently understood. This paper reports an experimental investigation into how the time–frequency patterns of AE signals connote the various cutting mechanisms under different cutting speeds and fiber orientations. Extensive orthogonal cutting experiments on unidirectional flax fiber NFRP samples with various fiber orientations were conducted. The experimental setup was instrumented with a multisensor data acquisition system for synchronous collection of AE and vibration signals during NFRP cutting. A random forest machine learning approach was employed to quantitatively relate the AE energy over specific frequency bands to machining conditions and hence the process microdynamics, specifically, the phenomena of fiber fracture and debonding that are peculiar to NFRP machining. Results from this experimental study suggest that the AE energy over these frequency bands can correctly predict the cutting conditions to ∼95% accuracies, as well as the underlying material removal regimes.