A Study Of System Reliability Optimization Using Genetic Algorithm

During the phase of product design, designers always apply the reliability optimization approach in selecting the most appropriate system elements, components and the number of redundancies to improve the life of products and structure configuration of systems so as to meet the demand of product reliability from customers. Reliability system optimization has already had a great success in both theory and application and the values of applying in engineering have gradually increased during the past decades of years.This thesis mainly does the research on the motifs of the application of genetic algorithm in multi-objective optimization of system reliability and multi-state system reliability optimization. It proposes a NPMOGA algorithm based on non-penalty parameter constraints handling method to resolve the difficulties in obtaining Pareto solutions and maintaining the varieties of solutions in the process of multi-objective optimization of system reliability. It also introduces the universal generating function into multi-state system reliability optimization to resolve the difficulties in obtaining the reliability functions of multi-state system, gives the reliability indices based on the universal generating function, and applies the genetic algorithm as its optimization method.The NPMOGA algorithm based on non-penalty parameter constraints solving method is composed of the niched Pareto genetic algorithm and non-penalty parameter constraints solving method. The niched Pareto genetic algorithm proposes a comparing approach that the domination ability is determined by the comparison between an individual and a comparison set, and the winner is selected to be involved in Pareto solutions. The non-penalty parameter constraints solving method searches for the feasible solutions by binary tournament and the penalty parameters which create bad affection caused by the inappropriate setting are not needed. This thesis integrates the two mentioned above into a new algorithm, gives its criteria and parameters' setting schemes, and adds the equivalent class sharing approach into its criteria to maintain the various Pareto solutions.Because the dimensions of multi-state system reliability optimization problems are high, the traditional methods are hard to apply. The universal generating function can express the relation between system performance level and state probability, and use the u-functions of elements to obtain the u-function of system. This thesis defines the reliability indices of multi-state system by u-function and gives the method of getting u-function of system,according to different system structures and performance measures, and the normal procedure of multi-state system reliability optimization using genetic algorithms.The NPMOGA algorithm based on non-penalty parameter constraints solving method is applied in the design of reliability optimization for an Over-speed Protection System. The goals are maximizing reliability and minimizing cost, the constraints are the weights and volumes of system, and the obtained results are a set of fine Pareto solutions with a good variety. The method based on universal generating function is applied in the design of reliability of optimization for a power station coal feeding system, and obtains an optimal system configuration by using genetic algorithm optimization.