Argonne National Laboratory has developed SMART-NDI, a multi-tasking, self-supervised deep learning framework designed to accelerate quality assurance in advanced manufacturing. Traditional inspection methods often rely on extensive manual reviews, creating bottlenecks. This AI-powered inspection tool quickly and accurately identifies defects, streamlines workflow, and reduces energy consumption. A key innovation is the ability to identify anomalies using only non-defect data, eliminating the need for time-consuming, curated defect libraries typically required to train AI inspection systems. This adaptability allows manufacturers to meet increasing production demands with higher efficiency and reliability, and can be applied across sectors such as aerospace, automotive, and microelectronics. “SMART-NDI: Scalable Manufacturing Evaluation and Real-Time Testing for Non-Destructive Inspection with Artificial Intelligence” details the capabilities of this technology.
Self-supervised deep learning for anomaly detection in NDI
A surprising element of Argonne National Laboratory’s SMART-NDI system is its ability to accurately identify manufacturing defects without relying on pre-labeled examples of defects. This avoids a common hurdle in artificial intelligence-driven inspection: building comprehensive defect libraries is costly and time-consuming. At the core of SMART-NDI is a multitasking, self-supervised deep learning framework that enables AI to learn patterns from defect-free data and then recognize deviations that indicate anomalies. This process significantly reduces the need for human intervention during the initial training stage, especially considering that traditional inspection methods are often lengthy and rely heavily on manual reviews, creating bottlenecks that hinder manufacturing efficiency and increase operational costs. The modular design of the framework extends its adaptability and allows migration to different inspection methods and materials. Users can tailor SMART-NDI to specific sectors and applications by adapting the core framework to relevant datasets.
The system has been validated in real industrial environments, including aerospace inspection, and has a measurable impact on key performance indicators, achieving high accuracy in defect detection while reducing inspection time. Beyond speed and accuracy, the system’s scalability enables wide adoption across industries such as automotive, critical materials, and microelectronics to support compliance with stringent safety and performance standards. The researchers presented their findings on the technology at the International Conference on Machine Learning and Applications, specifically detailing “DOC-DICAM: Domain-Aware One-Class Defect Identification in Composite Aeronautical Structural Materials,” further cementing its potential to revolutionize non-destructive inspection processes.
DOC-DICAM: Domain-Aware One-Class Defect Identification in Composite Aeronautical Structural Materials,” International Conference on Machine Learning and Applications.
International Conference on Machine Learning and Applications
SMART-NDI saves time, energy and costs in manufacturing
Current manufacturing quality assurance protocols often result in significant delays and energy demands due to reliance on manual inspection processes, creating bottlenecks that prevent efficient production scale-up, especially for advanced materials and complex components. Argonne National Laboratory’s SMART-NDI addresses this challenge with a multitasking, self-supervised deep learning framework designed to automate and accelerate nondestructive testing. Unlike traditional AI inspection systems, SMART-NDI uses only non-defect data to uniquely identify anomalies. This is a strategy that avoids the traditional tedious process of compiling extensive defect libraries for training purposes. This modular design allows users to adapt SMART-NDI to specific sectors and use cases by integrating relevant datasets, increasing its versatility across diverse manufacturing environments. The framework’s functionality extends beyond simple defect detection. This clearly reduces inspection time and is an important factor in increasing throughput and reducing labor costs. SMART-NDI is highly scalable and applicable to aerospace, automotive, microelectronics, and infrastructure, making it a versatile solution for industries with complex manufacturing workflows. The system’s adaptability is enhanced by its ability to function as a standalone tool or seamlessly integrate into existing inspection processes, minimizing disruption during implementation.
SMART-NDI uses a multi-task, self-supervised deep learning framework and the necessary training data provided by the user to identify defects with high accuracy.
Argonne National Laboratory
