Structural damage detection and localization using decision tree ensemble and vibration data

This paper explores the capabilities of decision tree ensembles (DTEs) for detecting and localizing damage in structural health monitoring (SHM). Unlike research on many other learning models, the goal of this study is to identify damage with a localization accuracy down to the single structural element, rather than limiting the evaluation to the story scale. The SHM methodology herein discussed, denoted as D-2-DTE, is based on decision trees ensemble and belongs to the class of vibration-based approaches, being the health assessment of the structure obtained by analyzing dynamic properties of the structural system, namely, mode shapes and natural frequencies. The proposed damage detection method is validated for three different test cases, including both numerical simulations and experimentally recorded data, which consider a wide array of damage configurations, including single and multiple damages; different damage types and severities; and the presence of random noise levels associated with dynamic properties acquisition. The performances of the D-2-DTE are evaluated in terms of accuracy, confidence of probabilistic predictions, and measurements of physical distances in localization errors. Additionally, two of the investigated test cases are based on available benchmarks, thus allowing a direct comparison with a state-of-the-art methodology. This comparative analysis evidences competitive performances of the DTE learning method.