Data Fusion and Machine Learning for Non-destructive Crack Evaluation

点击翻译按钮获取中文摘要

Failure to detect cracks in rail tracks, pipelines, and airplanes in the early stage can result in fatal accidents. While many crack evaluation methods focus on new sensor technology, this collaborative project aims to conduct novel research on crack detection using multiple non-destructive evaluation (NDE) techniques based on machine learning and data fusion to combine data from different technologies to properly characterize the crack. Deep learning data fusion will be used to combine multiple NDE techniques into a single crack profile that is close to the ideal laser performance. First, data driven crack detection models will be developed for single NDE and multiple NDE sensors using deep learning. Second, multi-modal deep learning crack detectors with uncertainty measure will be developed to provide optimal or close to optimal fusion of multiple NDE sensors. Third, deep learning style transfer will be employed to develop sensor transform for individual NDE and multiple NDE sensors to achieve the laser scanned performance. Developed fusion techniques will be evaluated on the real data provided by the industrial partner. Pipeline and other industries involved in instrumentation maintenance will benefit from more accurate crack detection methods which result in making better decisions on repairs and integrity decisions, increasing the safety of their pipeline systems.