Non-probabilistic Reliability Analysis Based On Particle Swarm Optimization And Its Application In Slope Engineering

Reliability analysis is based on uncertainty and the traditional probabilistic reliability has obtained more development. However, the sufficient sample data to definite the probability distribution and the subjection function of the variables is difficult to meet in engineering, which leads to the limitation of the probabilistic model. Therefore, non-probabilistic reliability model, which doesn't involve the concept of probability, is proposed as a complement and deepen of the probabilistic reliability model. The non-probabilistic reliability model, only needing to know the ranges of uncertain variables, breaks the limitation of traditional probability model in analyzing reliability. This paper mainly researches the development and the solutions about non-probabilistic reliability indexes.(1) This paper systematically summarized the development of non-probabilistic reliability indexes. In the meantime, it comparatively studied the applicable scopes, advantages and disadvantages of those indexes. Among all indexes, the convex set scaling factor index, the most economic, reasonable and widely applicable index in the current field of non-probabilistic reliability, can be applied to any kinds of uncertain models about convex sets and limit state functions. Therefore, all analysis on non-probabilistic reliability in this paper is based on this index.(2) The Particle Swarm Optimization (PSO), Simulated Annealing-Particle Swarm Optimization (SA-PSO) and Particle Swarm Optimization based on Bacterial (PSOBC), all proposed in recent years, are introduced to solve the problems of non-probability reliability with explicit limit state functions. The accuracy and feasibility of above methods are verified by examples.(3) The second response surface method based on weighted regression is proposed to solve the problems of non-probability reliability with implicit limit state functions and verified by examples. Aiming at determining the scheme of the design of experiments (DOE) in the response surface method, this paper also analyses some examples by using several common methods as the basis. Then the proposed method is applied to analyze the slope stability and the analysis steps are expounded. According to a strip iron mine in Guangdong Province, the feasibility and practicability of above method is verified by setting up the slope model, simulating safety coefficients, and analyzing non-probability reliability through the FLAC3D software.The probability model and non-probability model about reliability are proposed for different situations, so they are not contradictory but complementary and confluent. Although many countries have putted in a great deal of fund to study non-probability reliability, and achieved preliminary results, the theoretical system of non-probability reliability is not complete. Therefore, further researches need to develop urgently.