Rail inspections for internal discontinuities and flaws in the railroad industry primarily use the ultrasonic testing (UT) nondestructive evaluation (NDE) method. Automated ultrasonic systems in rail detector cars or vehicles utilize advanced software systems that primarily use B-scans for data presentation, visualization, and detection of internal anomalies. Given their large data size and storage constraints, A-scans are typically not saved and discarded. A-scans contain the amplitude of reflected sound waves (echoes) over time, reflecting signal strength at varying depths within a material. In contrast, B-scans employ "cross-sectional imaging" to record the maximum signal amplitude values within specific gates established in the A-scans. While A-scans include numerous features and details that may not be readily visible to the naked eye, they provide valuable insights for trending and can enhance data integration with other inspection modalities.

This research discusses MxV Rail's artificial intelligence and machine learning (AI/ML) efforts on the raw A-scan signals without the need for feature extractions. The convolutional neural networks (CNN) model developed in the AWS platform was deployed on an edge computing device for rail UT using A-scans. This work aims to increase the accuracy and speed of ultrasonic rail flaw inspection and open the possibility of using ultrasonic A-scans for future defect trending and data fusion work. This paper will also highlight the technical approach and edge device implementation, which includes edge hardware selection, optimization of the CNN model for edge deployment, data processing pipelines, and evaluation of the deployed system's performance. MxV Rail's work is not aimed at developing new products; rather, it is centered on demonstrating the feasibility of refining and optimizing existing rail UT technology using edge processing with A-scans to open future possibilities for innovation and collaboration to drive rail safety, reliability, and efficiency.