Aviation-based nondestructive evaluation data analytics

icaf2023 Tracking Number 67

Recent connectivity and digitization of aviation maintenance equipment has increased the potential of developing an Internet of Things 4.0 approach to enhance aircraft availability. Typically, these systems generate more data which nucleates interest in using analytical methods, such as artificial intelligence and machine learning (AI/ML), to increase the effectiveness of current aviation maintenance practices. It is important to recall that AI/ML methods are based on statistical regression and classification techniques. However, before such algorithms can be applied, considerations must be given to the quantity and quality (precision, accuracy, and noise) of the data to enable AI/ML. Several case studies are presented that explore these questions and indicate a careful assessment of the data is required to understand the accuracy and the distribution of the results from such analysis. The potential for the use of AI/ML is explored further using nondestructive evaluation (NDE) data. A significant challenge for these analytical methods is the limited amount of data captured for the features of interest, such as fatigue cracks and corrosion. Recall that trends in fleets lead to replacement or modification initiatives before an extensive amount of flaws are present. To mitigate this limitation, the Air Force Research Laboratory (AFRL) and collaborators have explored and implemented alternative methods to assist in the analysis of NDE data that integrates at least two of the following: heuristics, model-based, and data-derived analysis techniques. In addition, success has occurred when retaining the expertise of inspectors, i.e. humans-in-the-loop, to ensure the quality of the decision-making process. AFRL calls this approach Intelligence Augmentation (IA). The USAF has a rich history of using IA to analyze large NDE data sets, typically acquired from inspections that use automated scanning to acquire data. Several representative examples that include at least two of the three analysis methods are discussed, including the implementation process. These examples illustrate the benefit of integrating all resources to enable accelerated decisions with data limitations and the value of retaining humans-in-the-loop. Future opportunities include improved integration of models, especially as a function of their maturity through validation.