Application of Neuro-GA Hybrids in Sensor Optimization for Structural Health Monitoring

This paper presents an application of Neuro-GA methodology to tailor an Optimal Sensor Placement (OSP) strategy for effective Structural Health Monitoring. In recent years, OSP has become a prominent research topic for its capability to ensure effective experimental design with minimal resources. Literature suggests that Genetic Algorithms (GAs) have the potential to give promising results in OSP problem. However, in some cases, these results can be consequences of repeated computational effort in a large number of generations leading to a slower convergence. Generally, various operators in GAs were used to overcome these situations. But, the performance of those operators largely depends on the quality of the randomly generated population sets. Thus, to solve these problems, this paper illustrates an approach to improve GA with some populations of better fitness through Artificial Neural Network (ANN) and guide the GA in search of the global optimum. Two back-propagation neural networks with two fast converging training algorithms were used to individually combine with GA and later, both the hybrids were applied to determine the number and location of sensors in a benchmark bridge structure. The computational result demonstrated that proposed hybrid algorithms could sufficiently detect locations with better convergence.