RESEARCH ARTICLE Unsupervised real‐time SHM technique based on novelty indexes Rharã de Almeida Cardoso 1 | Alexandre Cury 2 | Flavio Barbosa 2 | Carmelo Gentile 3 1 Department of Civil Engineering, University of Ouro Preto, Campus Universitário, Ouro Preto, Brazil 2 Department of Applied Computational Mechanics, University of Juiz de Fora, Rua José Lourenço Kelmer, Campus Universitário, Juiz de Fora, Brazil 3 Department of Architecture, Built environment and Construction engineering (DABC), Politecnico di Milano, Milan, Italy Correspondence Alexandre Cury, Department of Applied Computational Mechanics, University of Juiz de Fora, Rua José Lourenço Kelmer, Campus Universitário, Juiz de Fora, Minas Gerais 36036‐900, Brazil. Email: alexandre.cury@ufjf.edu.br Funding information CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior); CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico), Grant/Award Numbers: 303361/2016‐6 and 311576/ 2018‐4; FAPEMIG (Fundação de Amparo à Pesquisa do Estado de Minas Gerais); IFSTTAR (former LCPC—Laboratoire Central des Ponts et Chaussées); SNCF (Société Nationale des Chemins de fer Français), Grant/Award Number: 01V0527 RGCU Summary Structural health monitoring programs play an essential role in the field of civil engineering, especially for assessing safety conditions involving large structures such as viaducts, bridges, tall buildings, towers, and old historical buildings. Mostly, an SHM process needs to be based on a trustful strategy for detecting structural novelties or abnormal behaviors. Usually, such an approach is complemented with human inspection and structural instrumentation rou- tines, where the latter requires proper hardware equipment and software tools. Recently, many advances were achieved regarding the hardware resources, such as wireless communication, remotely configurable sensors, and other data management devices. On the other hand, the software counterpart still is in its early developments. Several researches are in progress to fill this gap. In this context, this paper presents a novel online SHM identification method suitable to unsupervised real‐time detection of abnormal structural behaviors. The pro- posed methodology includes the use of an original representation of raw dynamic signals, that is, in situ measured accelerations. To assess the proposed approach, numerical simulations and two experimental applications are stud- ied: a railway viaduct, PK 075+317 in France and an old masonry tower in Italy. The results suggest that the proposed detection indexes are suitable for a wide range of SHM applications. KEYWORDS novelty detection, real‐time monitoring, structural health monitoring, symbolic data analysis, unsupervised statistical learning 1 | INTRODUCTION Large structures under continuous operation, such as viaducts, bridges, tall buildings, historical towers, among others, need to have their structural condition monitored permanently. Usually, this task is performed through visual inspec- tion and/or other manual analyses. However, to enhance the structure's safety assessment, it is necessary to develop a continuous monitoring program based on computations capable of looking after the structure 24/7, uninterruptedly. The primary purpose of such an approach is to automatically detect damage and/or other structural novelties immedi- ately after their occurrence, ideally at their initial stages. Recent technological advances allowed the production of reliable, accurate, and efficient monitoring equipment that can be remotely configured and operated. That fact encouraged the installation of dynamic monitoring systems in a Received: 17 April 2018 Revised: 17 December 2018 Accepted: 17 March 2019 DOI: 10.1002/stc.2364 Struct Control Health Monit. 2019;e2364. https://doi.org/10.1002/stc.2364 © 2019 John Wiley & Sons, Ltd. wileyonlinelibrary.com/journal/stc 1 of 18