Optimal sensor placement using improved simulated annealing algorithm

Abstract

Structural health monitoring based on vibration data has significant potential for improving the safety of structures through observation of long-term deterioration and post-disaster damage. However, the reliability of this technology is limited by uncertainties in the measurement of its sensors. Since many degrees of freedom are required, measurements of large civil structures are particularly difficult to obtain despite a limited number of sensors. The locations of sensors have been determined to play a key role in enhancing vibration data. A review of the literature indicates a shifting trend in the optimisation of the configuration of sensors, from direct ranking and simple iteration methods to combinatorial optimisation methods that employ heuristic algorithms. These heuristic algorithms are capable of providing a global optimal solution. However, their main disadvantage is the computational effort required for implementation. The primary objective of this research is to develop an efficient optimal sensor placement algorithm because a highly effective optimisation method is a critical requirement. An algorithm based on the Simulated Annealing optimisation method with a sensor location coordinate system was developed to search solutions through additional dimensions, enhance random search performance and minimise computational effort. The effectiveness of the proposed method was demonstrated by three numerical models and one experimental specimen. An assessment of objective functions, whose superior quality was identified through the characterisation using mode shape errors, was conducted. Application of the developed algorithm to damaged structures was explored based on simulated and experimental models. The effect of uncertainty in mode shapes on the locations of optimal sensors was examined using a proposed statistical approach. Comparisons of the methods indicate that the proposed method respectively outperforms its conventional method and the Genetic Algorithm method by approximately 10% and 5%.