Intelligent-based structural damage detection model: Fid-Bau Portal

Intelligent-based structural damage detection model

Lee, Eric Wai Ming / Lam, Heung Fai

Structural health monitoring using ANN (artificial neural network) has attracted much attention in the last decade. However, only very little publications address the problem of ANN design. The lack of a practical and robust ANN design method is one of the main reasons for the delay in the applications of ANN in the structural damage detection industry. A new structural damage detection method, which follows the pattern matching approach, is presented in this article. With the help of the damage signature and ANN techniques, the proposed damage detection method solves many problems of the original pattern matching approach, such as the problem of a large number of damage patterns and the lack of systematic matching method. Apart from the damage detection method, a general ANN design method is also presented in this article. Since the actual structural measurements are usually noise contaminated, the performance of the ANN model working in a noisy environment was examined by the introduction ot the Gaussian noise into the training and validation samples for network training. The trained ANN model was applied to the noise-free testing samples to evaluate the prediction error. Since the randomization processes (e.g., data extraction, noise generation and data presentation order, etc.) were involved in the course or the network training, bootstrapping technique was employed to mitgigate these randomization effects and to obtain a statistically justified result. The proposed structural damage diagnosis methodology consists of the new structural damage detection and ANN design methods. One of the objectives of this article is to study the performance of the GRNNFA (general regression neural network fuzzy adaptive resonance theory), a fusion of the Fuzzy Adaptive Resonance Theory (FA) model and the General Regression Neural Network (GRNN) model, in structural damage diagnosis following the pattern matching approach. The GRNNFA model was developed for noisy data regression. It has been proven to be effective in the tasks of regression and classification in a noisy environment. In the numerical example, the proposed structural damage diagnosis methodology is tested with different noise levels. The results from different noise levels are consistent, and it can be concluded that the GRNNFA prediction is not sensitive to the noise level. It is one of the outstanding advantages of the improposed methodology. Furthermore, the narrow width of the 95% confidence intervals of the bootstrap means demonstrates the stable performance of the proposed methodology.