Least Angle Regression for early-stage identification of earthquake-induced damage in a monumental masonry palace: Palazzo dei Consoli

Abstract

This paper presents a novel methodology for earthquake-induced damage identification of historical constructions through sparse multivariate regression. The proposed methodology comprises a first data cleansing stage using the minimum covariance determinant (MCD) method to mitigate the adverse effects related to the existence of outliers in the training feature dataset. Afterwards, a sparse multiple linear regression model (SMLR) is trained using the least-angle regression (LAR) model to eliminate the influence of environmental effects upon the selected features set. The proposed SMLR model allows to identify the optimal set of predictors in a fully automated way, minimizing the need for expert judgement in the process. The effectiveness of the proposed approach is demonstrated with an application case study of a monumental masonry palace, the Consoli Palace in Gubbio (Italy). The palace has been monitored with an aggregated static/dynamic/environmental SHM system since July 14th 2020. A recent seismic sequence of small intensity hit the palace on May 15th 2021 with a main earthquake of magnitude Mw 4.0. The epicentres of the main seismic event and the following aftershocks were located at a distance of 2–3 km far from the palace, making this case study a prominent example of a monumental construction subjected to near-field ground motion. The presented results demonstrate that a new damage condition arises in the Consoli Palace after the seismic sequence, although its severity remains at an early stage not detectable by visual inspections.