A number of investigators have evaluated the suitability and capabilities of neural networks for damage detection purposes. Szewezy and Hajela[1] presented a neural network approach based on mapping the static equilibrium requirement for a structure in a finite-element formulation, with the assumption that structural damage is reflected in terms of stiffness reduction. Studies by Bani-Hani et al.[2] and Masri et al. [3] complement the work of other investigators by concentrating on a class of problems where knowledge of the failure states is not available. In this research, two neural networks are proposed to help in determining the health state of fixed platform offshore structures. Natural Health Monitoring Neural Network (NHMNN) and Environmental Health Monitoring Neural Network (EHMNN). NHMNN relates the structural geometry, jacket and bracing cross-section, and defected story to the natural frequency of the structure. EHMNN requires additional information about the environmental loading, and provides damage assessment based on the deck drift. In the absence of sufficient experimental data, it was not possible to provide the neural network with a set of data that allows both training and testing. The finite element method was used to simulate such data, by providing "computed" information on natural frequency and deck drift for the intact model and the "damaged" model, instead of measured data. The numerical model for the analysis of offshore structures is performed using the finite element program NASTRAN [4]. Three basic models for rigid steel platforms were analyzed; eight, ten, and twelve stories high. The analysis and evaluation of offshore structures included free vibration parameters (natural frequency and mode shapes). The analysis for environmental conditions accounted for wind, water depth and sea level variations (including tides), sea states and currents. The effect of damage was represented by decreasing the diameter cross-section area along each element. This was done at different locations, one at a time. In each case, cross sectional area for all members of 1st, 4th, or 7th story members, are reduced by 10%, and then by 50% to simulate damage.

For each network, several parameters such as learning rate, momentum rate, and number of nodes in hidden layers, were carefully selected and optimised using the root mean square error. Full details of the design and training of the networks may be found in [5]. EHMNN provides four output nodes representing the four health monitoring statements (safe, slight damage, moderate damage, and severe damage), which correspond to (do nothing, routine maintenance, repair and/or reduce exposure and replace structure). The networks are trained and tested with the simulated (numerical) data. The test samples included platforms with nine, and eleven story heights, and although these story heights were not included in the training samples, results were satisfactory.

The two neural network classifiers proposed proved to be optimal with regard to classification error. They are also adaptable by retraining, and efficient. A health monitoring system for offshore structures is planned based on the networks The system will issue a warning signal if damage exists, and confirm diagnosis with site investigation, then add the signal/diagnosis to the training set. If no damage exists, the neural network will update the base line and add the updated base to the training set. In both cases, the neural network will be retrained with the training set to perform the next diagnostic cycle.

1

Szewczyk, P. Z., Hajela, P., (1994). "Damage Detection in Structures based on Feature-Sensitive Neural Networks." Journal of Comp. in Civil. Eng., ASCE, 8(2), Paper No. 3967. doi:10.1061/(ASCE)0887-3801(1994)8:2(163)

2

Bani-Hani, K., and Ghaboussi, J. (1998). "Nonlinear Structural Control using Neural Networks." Journal of Eng. Mech., ASCE, Vol. 124, No. 3, pp. 319- 327. doi:10.1061/(ASCE)0733-9399(1998)124:3(319)

3

Masri, S.F., Smyth, A.W., Chassiakos, A.G., Caughey, T.K., and Hunter, N.F. (2000). "Application of Neural Networks for Detection of Changes in Nonlinear Systems." Journal of Eng. Mech., ASCE, Vol. 126, No. 7, pp. 666- 676. doi:10.1061/(ASCE)0733-9399(2000)126:7(666)

4

MSC/NASTRAN (1998) MSC.Software Corporation, Santa Ana, CA92707, U.S.A.

5

Battisha, A.F. (2001). "Structural Health Monitoring of Offshore Structures using Artificial Intelligence." A Dissertation submitted to the Faculty of Engineering, Cairo University, in partial fulfilment of the requirements of Ph.D. in Civil Engineering.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)