Session: 01-05-01: Applications of Artificial Intelligence/Machine Learning for Aerospace Structures

Paper Number: 121466

121466 - OC-DICAM: One Class Defect Identification in Composite Aerostructure Material 

Nondestructive evaluation (NDE) techniques are used in many industries to evaluate the properties of components and inspect for flaws and anomalies in structures without altering the part’s integrity or causing damage to the component being tested. The aerospace industry uses a range of NDE techniques in the evaluation of its components with ultrasonic testing (UT) being a popular method for composite material. These composites are chosen due to their good tensile strength and resistance to compression. Composites are routinely inspected for common flaws including porosity, delaminations, voids, foreign object debris, and other inclusions that may compromise the structural integrity of the part. Current inspection methods rely heavily on human experience and are extremely time consuming. Therefore, there is a need for the development of techniques to reduce the manual inspection time.

Artificial Intelligence (AI) and Machine Learning (ML) have revolutionized the field of image processing thanks to convolutional neural networks (CNNs) and have garnered mainstream attention in language processing in part to ChatGPT. Applied to NDE, these tools can help automate data collection and analyses, provide new insights, and potentially improve detection performance in a quick and low effort manner with great cost savings. Previous researchers have shown that ML is well-suited to NDE. However, most works consider small lab generated composites with artificially created defects that allow for easy model development and evaluation but rarely extend to a production environment. In practice, naturally occurring defects rarely happen resulting in heavily imbalanced datasets. Furthermore, defects are labeled visually using a bounding box by inspectors. Inspectors are concerned with fully encapsulating the defect resulting in large boxes that may contain non-defect material. This labelling process generates an overlap between the defect and non-defect classes. These two issues make it challenging for ML models to learn good defect characteristics which hampers model performance.

Out-of-distribution (OOD) detection focuses on ensuring the reliability and safety of ML systems. OOD detection aims to detect test samples that are drawn from a distribution that is different from the training distribution. OOD detection has found success in autonomous driving, medical image classification, and industrial inspection. Inspired by these works and motivated by the challenges above, we present One Class Defect Identification in Composite Aerostructure Material (OC-DICAM). Applied to NDE, we assume non-defects as in-distribution while defects as out-of-distribution samples. Therefore, OC-DICAM only requires non-defect samples during training, eliminating both the dataset imbalance and label noise issues. Our approach utilizes the A-scan or signal waveform generated through the UT scanning process. To ensure model convergence without a validation dataset of defects, we use a recently developed automated stopping algorithm that uses loss entropy to internally evaluate the model performance during training. After training, we use model confidence to score test signals with those below a dynamically set threshold flagged as defects. OC-DICAM is tested on composites generated in a production environment with real defects. Our preliminary results show that our approach can detect multiple types of defects with minimal false positives.

Presenting Author: Austin Yunker Argonne National Laboratory

Presenting Author Biography: Predoctoral appointee at Argonne National Laboratory in the Data Science and Learning Division. Past work has focused on the development of AI/ML systems for science problems at scale using HPC resources including CT restoration and reconstruction. Current work focuses on developing an artificial intelligence system for improving non-destructive inspection methods used in evaluating state-of-the-art aircraft.