Eficient Reliability-Based Design Optimization Of Buildings Using Surrogate Models In Ai

1Mayank Kumar

1Department of Civil Engineering, Graphic Era Hill University, Dehradun, Uttarakhand, India 248002

131 Views
50 Downloads
Abstract:

This approach for reliability-based design optimization (RBDO) of buildings using surrogate models in artificial intelligence (AI). RBDO aims to find the optimal design of structural systems that satisfy performance requirements while considering uncertainties in the design variables and the loads. Traditional RBDO methods often require a large number of computationally expensive simulations, which hinder their practical applicability. The proposed approach utilizes surrogate models, trained using AI algorithms, to approximate the structural response and reliability analysis. This enables a significant reduction in computational cost while maintaining accuracy. The paper outlines the methodology, discusses the construction of surrogate models, describes the RBDO formulation, presents a case study, and provides insights into the efficiency and effectiveness of the proposed approach. Reliability-based design optimization (RBDO) plays a crucial role in enhancing the performance and safety of buildings by considering uncertainties associated with structural design. However, the conventional RBDO methods often suffer from high computational costs, making them impractical for real-world applications. This abstract presents an innovative approach that leverages surrogate models in artificial intelligence (AI) to achieve efficient and reliable design optimization for buildings. The proposed methodology integrates surrogate models, such as neural networks and Gaussian processes, with RBDO techniques to create a computationally efficient framework for optimizing building designs while accounting for uncertainties. Surrogate models serve as approximation functions that mimic the behaviour of complex structural analysis models, enabling rapid evaluations of the structural responses and associated reliability metrics. To establish accurate surrogate models, a set of training samples is generated by exploring the design space using a sampling strategy, such as Latin hypercube sampling or Monte Carlo simulation. These samples are used to train the surrogate models, which can then predict the response and reliability metrics for any given set of design variables, eliminating the need for repetitive and time-consuming evaluations of the full-fledged structural models. The surrogate models are integrated within an RBDO framework, which combines optimization algorithms, reliability analysis methods, and surrogate model-based response surface models. This integration enables the efficient exploration of the design space to identify the optimal design that minimizes cost or maximizes performance while satisfying reliability constraints. By utilizing surrogate models, the proposed approach significantly reduces the computational burden associated with RBDO of buildings, allowing for more extensive exploration of the design space within feasible timeframes. The surrogate-based optimization process achieves near-real-time design evaluations, enabling designers and engineers to make informed decisions promptly. The effectiveness and efficiency of the proposed methodology are demonstrated through case studies involving various building types and design objectives. The results highlight the significant computational savings achieved by surrogate models while maintaining accurate predictions of structural responses and reliability metrics. The integration of surrogate models provides a powerful tool for designers and engineers to enhance the structural performance and safety of buildings while considering uncertainties, paving the way for more advanced and practical design optimization methodologies in the field of civil engineering.

Keywords:

Reliability-Based Design Optimization (RBDO), Surrogate Models, Artificial Intelligence (AI), Simulations, Reliability, Design Optimization, Surrogate Models

Paper Details
Month05
Year2019
Volume23
IssueIssue 3
Pages1203-1211