Structural damage detection using neural network with learning rate improvement

In this research, we explore the structural damage detection using frequency response functions (FRFs) as input data to the back-propagation neural network (BPNN). Such method is non-model based and thus could have advantage in many practical applications. Neural network based damage detection generally consists of a training phase and a recognition phase. Error back-propagation algorithm incorporating gradient method can be applied to train the neural network, whereas the training efficiency heavily depends on the learning rate. While various training algorithms, such as the dynamic steepest descent (DSD) algorithm and the fuzzy steepest descent (FSD) algorithm, have shown promising features (such as improving the learning convergence speed), their performance is hinged upon the proper selection of certain control parameters and control strategy. In this paper, a tunable steepest descent (TSD) algorithm using heuristics approach, which improves the convergence speed significantly without sacrificing the algorithm simplicity and the computational effort, is investigated. A series of numerical examples demonstrate that the proposed algorithm outperforms both the DSD and FSD algorithms. With this as basis, we implement the neural network to the FRF based structural damage detection. The analysis results on a cantilevered beam show that, in all considered damage cases (i.e., trained damage cases and unseen damage cases, single damage cases and multiple-damage cases), the neural network can assess damage conditions with very good accuracy.