Reliability-based structural optimization using neural networks and Monte Carlo simulation

• Published in: Computer methods in applied mechanics and engineering Vol. 191; no. 32; pp. 3491 - 3507
• Main Authors: Papadrakakis, Manolis, Lagaros, Nikos D
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 07.06.2002; Elsevier
• Subjects: Applied sciences; Buildings. Public works; Computation methods. Tables. Charts; Computational techniques; Evolution strategies; Exact sciences and technology; Fundamental areas of phenomenology (including applications); Inelasticity (thermoplasticity, viscoplasticity...); Mathematical methods in physics; Monte Carlo simulation; Neural networks; Parallel computations; Physics; Reliability analysis; Solid mechanics; Structural analysis. Stresses; Structural and continuum mechanics; Structural optimization; Viscoelasticity, plasticity, viscoplasticity; 79; 26; Structural optimization; Neural networks; Evolution strategies; Parallel computations; 70; Reliability analysis; 61; 62; Monte Carlo simulation; Monte Carlo method; Probabilistic approach; Plastic flow; Frame construction; Rupture; Structural reliability; Numerical method; Neural network; Optimization; Inelasticity; Three dimensional model; Elastoplasticity; Plastic limit; Distributed algorithm; Parallel computation; Structural analysis
• ISSN: 0045-7825; 1879-2138
• DOI: 10.1016/S0045-7825(02)00287-6

This paper examines the application of neural networks (NN) to reliability-based structural optimization of large-scale structural systems. The failure of the structural system is associated with the plastic collapse. The optimization part is performed with evolution strategies, while the reliability analysis is carried out with the Monte Carlo simulation (MCS) method incorporating the importance sampling technique for the reduction of the sample size. In this study two methodologies are examined. In the first one an NN is trained to perform both the deterministic and probabilistic constraints check. In the second one only the elasto-plastic analysis phase, required by the MCS, is replaced by a neural network prediction of the structural behaviour up to collapse. The use of NN is motivated by the approximate concepts inherent in reliability analysis and the time consuming repeated analyses required by MCS.