Preprocessing ultrasonic scanning data with the help of Hopfield-style neural network

Ultrasonic scanning with coherent data treatment is a very promising method for nondestructive inspection of welded pipeline joints. This method requires processing of high-dimensional raw data obtained from ultrasonic probes. In practice, the raw data have a substantial noise level. Besides, acoustic contact between scanning probe and inspected pipe is sometimes lost, which may cause gaps in data, resulting in performance degradation during further analysis. This paper describes a neural network system providing noise removal and gaps filling in ultrasonic data.