Statistical process control procedures for online damage detection of a monumental masonry palace: The consoli palace in Gubbio, Italy

This paper proposes a new semi-supervised two-class pattern classification method for early damage detection of structures. The proposed approach comprises five consecutive steps, including the (i) collection of data samples representing the healthy condition of the structure (training period); (ii) clustering analysis; (iii) outliers elimination; (iv) pattern recognition using local Principal Components Analysis; and (v) pattern classification through control charts. The effectiveness of the proposed approach is demonstrated through two different case studies. These include a benchmark beam simulation model, and the Consoli Palace in Gubbio (Italy). The latter has been monitored for two years using a long-term vibration-based monitoring system, and represents an extraordinary example of a stiff masonry building with complex environmental effects. Analyses and discussion are presented covering the effects of differing structural/environmental behaviour regimes, the presence of outliers in the training period, and the effectiveness of different control charts.