The current research aims to develop and implement a robust and reliable automated defect recognition system for defects characterizing in nondestructive evaluation (NDE) applications by developing a novel ultrasound imaging technique and AI-based algorithms. During the manufacturing process, the flaws are proven to form in terms of impurities, such as porosity, slag inclusion, blowholes, and pinholes. During service conditions of the structures, these impurities will grow in size and form a group together to create a more extensive defect, such as cracks in the structures. These structures have to undergo inspection at regular intervals of time to prevent catastrophic failures. The NDE methodologies are widely used across industries to examine and certify the structures to fit for intended service. If the flaws exist, they must be characterized concerning location, size, orientation, shape, and nature to determine their acceptability under operating conditions. An NDE expert is required to resolve these flaws.


In this research work, we have developed a safer and faster ultrasound methodology for automated defect characterization by developing a simulation-assisted AI-based system that is efficient and reliable. However, training an AI system requires vast training datasets, but obtaining such datasets is challenging in the NDE domain. Therefore, to overcome the scarcity of annotated dataset generation, we have used three approaches: (1) The experiments are carried out by developing a novel ultrasonic imaging technique using an arbitrary virtual array source aperture (AVASA) by phased array excitation, which reduces the inspection time and improves the resolution of the output image. (2) Modeling the Finite Element (FE) simulation based on the influence of real-time experimentation, such as expected defect morphologies, nature of the defect, and the sensitivity of the instruments. (3) AI algorithms are employed for the first, automating the process of creating large synthetic ultrasound NDE imaging datasets, and the second for obtaining the ultrasound NDE simulations for phased array techniques by training the smaller set of FE datasets. These generated datasets are used for building a robust AI system for automated defect recognition to improve the structures' quality and life and reduce human intervention.