Advanced Machine Learning for Non-Destructive Testing

The Army depends on radiographic inspection for non-destructive testing (NDT) of munitions and armament system components to identify flaws and defects before these products reach the warfighter. Small defects, such as gaps, voids, cracks, and foreign materials, in the energetic material can lead to unforeseen reactions that have lessened or increased the potency of the energetic material, that would lead to costly, hazardous, or even life-threatening misfires. Highly skilled and experienced radiographers must interpret an x-ray of every munition for critical defects. Incorrect classification may result in rejection of a part in good condition or acceptance of a part with defects exceeding the limit defined by relevant standards. In many cases, this process is not only costly, laborious, and time-consuming, but also prone to error and inconsistencies. Therefore, the Army has a critical need for accurate and reliable detection of defects to eliminate defective parts from leaving the production floor, thereby reducing the potential for hazardous incidents in the field.   During the Phase II, Sky Park Labs (SPL) developed a software prototype, known as DeepNDT (Deep Learning for Non-Destructive Testing), that employed artificial intelligence (AI) algorithms to assist a radiographer in the interpretation of radiographic inspection results. DeepNDT uses unsupervised and supervised deep learning algorithms for automatically detecting anomalies and highlights any deviations as potential defects. Then, the system classifies the type of defect (i.e., cavities, cracks, porosity, etc.) and characterizes its geometry and size for a radiographer to assess and make a final decision.  During Phase II, we demonstrated these capabilities on a variety of publicly available industrial datasets as well as sponsor-provided data, achieving sub-second inference time and accuracy, precision, and recall rates of over 95% for detection and classification tasks. The Sequential Phase II effort will focus on advanced development and thorough evaluation of the Phase II prototype in preparation for a system integration with inspection equipment at government ammunition production facilities. Transitioning DeepNDT to ammunition production facilities will significantly enhance the advanced manufacturing processes for the U.S. Army. Incorporating DeepNDT into the inspection workflow can potentially reduce the cognitive load on radiographers and thereby increase the efficiency and accuracy of assessment of munitions for defects, resulting in safer products for the warfighter.