Ultrasonic thickness monitoring of material extrusion parts: machine learning analysis, and erosion testing

This is the authors' abstract. We don't add key points for this paper.

Material extrusion is a cost-effective additive manufacturing method, yet parts can serve in corrosive or abrasive environments where hidden flaws can be critical. Ultrasonic non-destructive testing (NDT) offers in-situ quality evaluation, but layer-wise anisotropy and varied process parameters in additive manufacturing can complicate interpretation. We assess the capability of A-scan ultrasound to measure material extrusion part thickness across different infill patterns and infill densities, nominal part thicknesses, nozzle diameter, presence/absence of top or bottom skins and orientation of the measurements. A Random Forest model links process parameters to ultrasound and caliper data, enabling data-driven thickness prediction. Concentric infill produced time-of-flight readings that agreed best with caliper measurements and were insensitive to transducer orientation. Overall, A-scan reliably quantified thickness within the 2.5–10 mm range of the chosen contact probe. The machine-learning model achieved high R 2 and small errors for thickness and time-of-flight, demonstrating its potential for automated process monitoring. Finally, real-time ultrasound tracking of specimens subjected to simulated slurry erosion captured progressive wall loss, underscoring the feasibility of continuous integrity surveillance in service. These results establish practical guidelines for ultrasonic NDT of material extrusion parts.