Phased array ultrasonic imaging and characterization of adhesive bonding between thermoplastic composites aided by machine learning

The testing and evaluation of adhesive bonding quality between thermoplastics are crucial for structural integrity. This article presents the use of phased array ultrasonic testing (PAUT) method to characterise the adhesive interface between thermoplastic composites. Samples with three different bond conditions: control, bad and mid-level were fabricated and tested using PAUT. A damage index (DI) based classification framework aided by machine learning (ML) algorithm is proposed to classify different adhesion conditions. A set of 18 physics-based damage indices were extracted from each PAUT image for quantitative characterisation. ML algorithms were developed to build a non-linear mapping that correlates the input DIs with the output sample types to address the classification problem. The experimental results show that support vector machine (SVM) performs better than other ML algorithms with classification accuracy greater than 95%, and the defined DIs can differentiate among bad, mid-level, and control samples.

Automated summary

This article presents the use of phased array ultrasonic testing method to characterize the adhesive interface between thermoplastic composites. A deep learning algorithm is proposed to classify different adhesion conditions based on a set of physics-based damage indices extracted from the ultrasonic images. The experimental results show that support vector machine performs better than other machine learning algorithms, achieving a classification accuracy of over 95%.