Machine learning for plane wave imaging

Abstract

Phased array ultrasonics have enabled the recording of an ever-increasing amount of data from the inspection targets. With the latest advancements in total-focusing method with plane wave imaging the amount of data has increased exponentially when compared to conventional ultrasonic methods. As more data allows more reliable evaluation, the cost of evaluation also increases. Since there is more data for the inspector to evaluate, the inspector’s job becomes more difficult and laborious with the modern technology. Moreover, as phased array techniques evolve to even more sophisticated approaches such as total focusing method and the latest form, plane wave imaging total focusing method (PWI-TFM), reading raw ultrasonic data is too convoluted for human inspectors. As the raw data is pre-calculated for more understandable image, the data can be reconstructed in multiple ways for optimal detection. However, the more presentations are reconstructed from the data, the more time consuming it is for the inspector to evaluate the images. Machine learning powered inspection enables the full use of all the data, while allowing the best possible presentation for the inspector. In this paper we demonstrate PWI-TFM inspection powered by machine learning model. The ML model is used to present the flaw indications to the inspector. Moreover, PWI-TFM image reconstruction is studied from ML performance aspect.