[8A5] An automatic non-destructive inspection based on the combination of multiple phased array probes and convolutional neural network
Bojie Sheng1, Istvan Szabo1, Sergio Malo1, Jamil Kanfoud1 and Channa Nageswaran2
1Brunel University London, UK
2TWI Ltd, UK
Currently ultrasonic phased array technology has been applied wildly as a non-destructive inspection technique to assess flaws in lined pipes. However due to the complexity of pipeline structure, such as a pipeline made of a steel layer and multiple cladding layers, defects in different layers can not be detected by a single phased array probe. Thus multiple different phased array probes are required for the inspection, which results in the complicated measured data. Consequently, data analysis for this application in industry is still implemented by phased array experts.
In order to develop an automatic non-destructive inspection technique able to automatically assess flaws in lined pipes, this paper proposed a technique based on ultrasonic phased array method and convolution neural network (CNN) method. At first, a combination of different phased array probes, such as Transmit-Receive Longitudinal (TRL) probe and linear array probe, are applied to scan pipeline surface which generate sets of B-scan and S-scan data. Then, all the B-scan and S-scan data are combined in one image for each measurement. Finally CNN method is applied to process the image data for automatic flaw detection. In the paper a case study has shown a good performance of the method.
In order to develop an automatic non-destructive inspection technique able to automatically assess flaws in lined pipes, this paper proposed a technique based on ultrasonic phased array method and convolution neural network (CNN) method. At first, a combination of different phased array probes, such as Transmit-Receive Longitudinal (TRL) probe and linear array probe, are applied to scan pipeline surface which generate sets of B-scan and S-scan data. Then, all the B-scan and S-scan data are combined in one image for each measurement. Finally CNN method is applied to process the image data for automatic flaw detection. In the paper a case study has shown a good performance of the method.
