Estimation of Internal Surface Roughness of Additively Manufactured Components Under Complex Conditions Using Artificial Intelligence and Measurements of Ultrasonic Backscatter

Abstract

Additive Manufacturing (AM) is increasingly being considered for fabrication of components with complex geometries in various industries such as aerospace and healthcare. Control of surface roughness of components is thus a crucial aspect for more widespread adoption of AM techniques. However, estimating the internal (or ‘far-side’) surface roughness of components is a challenge, and often requires sophisticated techniques such as X-ray computed tomography, which are difficult to implement online. Although ultrasound could potentially offer a solution, grain noise and inspection surface conditions complicate the process. This paper studies the feasibility of using Artificial Intelligence (AI) in conjunction with ultrasonic measurements for rapid estimation of internal surface roughness in AM components, using numerical simulations. In the first models reported here, a pulse-echo configuration is assumed, whereby a specimen sample with rough surfaces is insonified with bulk ultrasonic waves and the backscatter is used to generate A-scans. Simulations are carried out for various combinations of the model parameters, yielding a large number of such A-scans. A neural network algorithm is then created and trained on a subset of the datasets so generated using simulations, and later used to predict the roughness from the rest. The results demonstrate the immense potential of this approach in inspection automation for rapid roughness assessments in AM components, based on ultrasonic measurements.