Experimental machine learning approach for building structural health monitoring application

Abstract The research and development field of structural health monitoring is evolving, and as a result, it now provides choices for disaster prevention and life cycle extension. This can help boost structural safety and infrastructures resiliency. Researchers have taken an interest in the use of machine learning techniques for these applications as a result of the promising outcomes that those approaches have delivered. In spite of this, damage detection applications necessitate the use of sophisticated and reliable data-driven enabled technologies. This is because of the complex behavior of structures that is notoriously difficult to track. In this investigation, triaxial accelerometers are used for data acquisition purposes in order to monitor and examine the structural performance of an experimental four-story building frame when subjected to a variety of structural configurations. The effectiveness of supervised machine learning algorithms was analyzed and compared to that of unsupervised learning approaches. In spite of the fact that supervised models have demonstrated good performances for the detection of initially established anomalies, they have only partially exhibited performances for newly produced damage patterns, which restricts their application in real life cases. The results obtained by the unsupervised algorithms, on the other hand, were comparable and satisfying, revealing essential capabilities for damage detection applications.