A cointegration-based approach for automatic anomalies detection in large-scale structures

In recent years, the development of structural health monitoring (SHM) solutions for the automatic evaluation of the health state of engineering structures is continuously growing. However, when considering real-world applications, structures are highly influenced by meteorological variations or human activities (like temperature, wind and traffic loading) which can overwhelm the changes induced by a damage. Thanks to its ability to remove the long-term trends from a set of variables of the same process, cointegration, a technique born in the field of econometrics, has been introduced about ten years ago in SHM applications as a valid method to project out the confounding influences, such as environmental and operational variations. Because of the few examples of implementation currently available, this paper provides an in-depth review of all the relevant aspects to consider when cointegration is used as damage detection strategy and data are acquired from real-world structures of large dimensions. The methodology is applied for the first time on a complex structure of a singular nature, i.e. the steel roof of the G. Meazza stadium in Milan, which consists of multiple modular elements referred to as rafts. The time series which measures the rotations of the rafts are used as input data for the development of the cointegration-based method. Then, Johansen procedure is adopted to create a unique feature from the multivariate dataset, namely cointegrating residual, in which the effects of environmental and operational variables are suppressed, while the effects due to damage remain evident. The obtained residual is therefore used for novelty detection by means of a control chart, demonstrating its effectiveness into identifying the presence of anomalies or modifications in the structure in a clear and timely manner.