Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SECOND INTERNATIONAL CONFERENCE ON SOFT COMPUTING TECHNOLOGY IN CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING
Edited by: Y. Tsompanakis, B.H.V. Topping
A Rough Set-Based Revised Counter-Propagation Network Model for Structural Damage Identification
S.F. Jiang1, C. Fu1,2 and J. Lin1
1College of Civil Engineering, Fuzhou University, China
S.F. Jiang, C. Fu, J. Lin, "A Rough Set-Based Revised Counter-Propagation Network Model for Structural Damage Identification", in Y. Tsompanakis, B.H.V. Topping, (Editors), "Proceedings of the Second International Conference on Soft Computing Technology in Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 16, 2011. doi:10.4203/ccp.97.16
Keywords: rough set, revised counter-propagation network, damage identification, attribute reduction, identification accuracy, data processing method.
This paper presents a new damage identification method based on integrating a revised counter-propagation network (RCPN) with a rough set, named RSRCPN. This proposed method is used for damage detection and identification, particularly for cases where the measurement data has many uncertainties.
The proposed structural damage identification method mainly consists of six moduli, namely, data preprocessing, feature extraction, feature parameters reduction, the least input vector, RCPN and results output. In the phase of data preprocessing, threshold method, average method and so on are used to preprocess the raw data. Here the normalized frequency change ratio (NFCR) and the normalized damage signature index (NDSI) are employed in feature extraction phase. Then a rough set is used to reduce the attributes and the spatial dimensions of the input vectors for RCPN, finally the proposed RCPN is used to identify the structural damage. It is noted that the learning rule or learning process of the traditional CPN is revised so as to improve its capabilities of processing uncertainties, classification and running effectiveness.
To validate the method proposed, six damage patterns from a seven-storey steel frame were identified. Two damage magnitudes and three damage locations were defined. The damage magnitude was classified as small and large while the location was designated with the corresponding storey number. The magnitude of the damage was 4.1% for small damage and 16.7% for large damage. In addition, as the noise is inevitable, each set of the analytical computed modal parameters for healthy and damage scenarios were then added by a random sequence to simulate the measured data.
In order to illuminate the advantages of the RSRCPN, we compared the RSRCPN performance with different noise levels. It was found that the recognition accuracy and classification capability of the RSRCPN decreases with an increase of the noise level. A comparison of the RSRCPN with the probabilistic neural network (PNN), probabilistic neural network based on rough set (RSPNN) and RCPN was also made. The results show that, different neural networks have a greater influence on identification accuracy, and the identification accuracy of the RCPN is higher than the PNN obviously at each noise level. For example, when the noise level is 0.1%, 0.2% and 0.3%, respectively, the average identification accuracy of the RCPN was higher than the PNN at 9.33%, 5.00% and 5.50%, respectively. Especially, when the noise level is 0.1%, the identification accuracy of the RCPN is even higher than the PNN at 51% in pattern 1. Similar conclusions are also drawn in the comparison of the RSRCPN and the RSPNN.
In summary, the presented method not only has good damage detection and noise-resistant capabilities, but also significantly reduces the memory requirements for data storage and saves runtime as a consequence of the rough set and RCPN model processing.
purchase the full-text of this paper (price £20)