Application of Signal Processing and Machine Learning Techniques for Segmentation and Spatial Registration of Material Property Data

Errors introduced into data acquired for nondestructive evaluation due to suboptimal digitization rates, bandwidth, and signal processing settings can dominate the perceived noise in acquired data, leading to artifacts and erroneous interpretation. Furthermore, the presence of such errors incurred through the data acquisition process can also inhibit post-processing techniques utilized in multimodal data segmentation and registration efforts. This study illustrates the use of advanced signal processing techniques to limit the effects of quantization errors in normal-incidence ultrasonic inspection data, thereby optimizing the signals for further processing while maintaining the integrity of the data. In conjunction with signal processing methods, K-means and Expectation-Maximization algorithms are investigated for applications in automated data segmentation and multimodal spatial registration. Using results from segmented and registered data, techniques in constructing computer aided design (CAD) models are investigated for importing measured material property and flaw information into various modeling software platforms.