Development of an Artificial Intelligence (AI) for the Prediction of Fatigue Failure in Naval Structures: A Digital Twin Application

Abstract

Many of the high speed and high-performance naval vessels use aluminum as a primary structural material. Welded aluminum stiffened structures are very common in these vehicles, but even with much research, performance of these structures under fatigue is not completely known. Many experimental investigations on similar structures were performed over the past two decades. Meanwhile, the desire for a comprehensive naval structural health monitoring (SHM) system is receiving greater interest. Recently, with re-advent of machine learning and artificial intelligence (AI) for structural digital twin, older experimental sensor dataset from crack growth in welded aluminum structure could be better utilized and exploited for crack predictions with few interventions from the sensor data. In this study, piezoelectric wafer active sensors are utilized for SHM of a welded aluminum structure and sensor data were collected at multiple frequencies from multiple specimens. Crack initiation to growth pattern were also recorded. Through machine learning, sensor data from four specimens were exploited to develop an AI algorithm for predicting the crack growth. It is shown that ML and AI frameworks are suitable for ship structure digital twin applications.