Structural System Reliability Analysis Based On High Performance Parallel Computing

Uncertainties of loads,material properties and structural dimensions are widely found in engineering structures,and reliability analysis is an effective tool to deal with such problems.Structural systems might fail in the terms of different failure modes,which are usually interrelated with each other and form a series-parallel system.For a large complex structural system,the number of possible failure modes is significantly huge.Obviously,it is unrealistic to enumerate all of its possible failure modes.At present,the most commonly used structural system reliability analysis methods are built on the basis of dominant failure modes,which ignores a large number of secondary failure modes that impose little influence on the reliability of structural systems,and analyzes the reliability of structural systems through the dominant failure modes.There are three widely used methods to identify the dominant failure modes:the ?-unzipping method,the incremental load method and the branch and bound method.All of these three methods filter the secondary failure modes by setting reasonable bounding parameters that directly determine the accuracy and efficiency of the algorithm.It is difficult for these methods to reconcile the contradiction between accuracy and efficiency,especially for a large complex structural system.One way to deal with this contradiction is to use parallel computing.Unfortunately,the above methods are not suitable for parallel computing because of its intricate parallel algorithm design and low parallel efficiency.Naturally,Monte Carlo simulation(MCS)is an effectively alternative method,which is labeled "embarrassingly parallelizable”due to its high parallelism and excellent parallel performance.Therefore,this article studies the parallel computing of structural system reliability analysis based on Monte Carlo method.The main work is as follows:(1)Review four commonly used dominant failure mode identification method:the?-unzipping method,the incremental load method,the branch and bound method and the Monte Carlo method,introduce their specific calculation processes,and analyze in detail their parallelism.(2)For single failure mode problem,a structural reliability analysis method based on parallel Monte Carlo simulation is proposed.Since the samples are independent of each other and have basically the same calculation process,they can be divided into several parts on average and computed separately by each processor core.The samples are generated in parallel to keep them completely independent.A seed is set up for each process,and each process generates a random number sequence separately.Finally,the failure probability of the structure is obtained by a reduction operation.(3)A structural system reliability analysis method based on parallel Monte Carlo simulation is proposed to deal with the problem of large computation cost in structural system reliability design.A master-slave model is established,in which the processes are divided into master process and slave processes.The master process is used to search and count the failure modes.The slave processes are used to perform finite element analysis,generate the failure modes,and pass them to the master process.At the end,the failure probability of structural system and the dominant failure modes are calculated by the main process.This method can accurately compute the failure probability of structural system and obtain the dominant failure modes,which can provide reference for structural design,inspection and maintenance.