Demonstrating Probability of Detection of indication in a Heat Exchanger with Assisted Analysis using Artifical Intelligence

Abstract

Traditional Non-Destructive Testing (NDT) methods for heat exchangers inspections struggle with both efficiency and accuracy. These methods are time-consuming, prone to errors in data interpretation, and heavily reliant on skilled technicians, who are becoming a scarce resource. Artificial intelligence (AI) tools offer a promising alternative, by providing a faster, more automated solution. This study assesses an AI system's performance using Probability of Detection (POD), a key NDT metric. Due to the cost of experimental POD, we leverage numerical simulation with CIVA software to generate large datasets. POD curves are created to evaluate the AI's capabilities across various heat exchangers conditions. Simulations demonstrate the AI's ability to detect indications even with significant noise affecting the signal. The findings are corroborated by real-world testing, which achieved a 97.7% POD for a selected set of indications deemed critical. This study highlights the potential of AI and simulation to improve NDT for critical infrastructure.