提出了一种针对变质量的非传统多调谐阻尼器(MTMD)基于可靠度的优化设计方法.其中主结构部分考虑多阶模态,MTMD则考虑为集中质量.地震动激励采用金井清模型,利用状态空间法提高计算效率.地震危险性则依据地震动衰减关系和Gutenberg-Richter模型模拟.定义危险性、主结构与多调谐阻尼器相关参数均为随机变量,通过考虑结构多维输出及对应限值,并结合激励与结构随机性得到结构绝对失效概率,并进一步以绝对失效概率为目标函数形成优化问题,通过求解变质量MTMD最优解进行设计.本文将该方法应用于带多煤斗的火电厂房煤斗隔震设计.其中,煤斗隔震形成MTMD,而煤斗内部储煤量变化则形成变质量MTMD.采用拉丁超立方抽样法生成样本,研究确定了合适样本数.通过选取结构角柱位移角作为结构响应,采用规范限值计算失效概率.利用基因算法求解了该优化问题并得到了最优设计与对应的失效概率.本文进一步对比了平动隔震体系和摆隔震体系.研究表明,尽管摆体系的频率与质量不相关,其失效概率并未优于平动隔震体系;且摆体系摆动圆弧曲率的变异系数对结构失效概率影响不大,规律不明显.最后,将煤斗与主结构碰撞亦考虑为失效事件,考虑了煤斗碰撞问题.本文所提出的设计方法基于可靠度,直接以降低结构失效概率为目标,可同时考虑结构多维输出以及煤斗与主结构相对位移,并对应地考虑多个响应限值,将多个响应综合为一个失效概率,避免了计算量巨大的多目标优化.“,”A reliability-based optimization design framework for mass uncertain nonconventional multiple tuned mass damper (MTMD) systems was proposed. The structure was modelled as a hybrid-model with multiple tuned mass dampers simplified as lumped masses. Seismic excitation was modelled by the Kanai-Tajimi filtered white noise. State-space representation was adopted to enhance the simulation efficiency. Attenuation relationship and truncated Gutenberg-Richter relationship were used for hazard condition definition. Multiple parameters associated with hazards, the main structure, and multiple tuned mass dampers were considered to be random. The objective function was defined following the uncondi-tional failure probability with multiple limit state bounds, incorporating both structural and excitation uncertainties. A case study based on a thermal power plant with multiple scuttles was carried out to illustrate the framework. The Latin Hypercube Sampling (LHS) method was imple-mented to reduce the sample size. Drifts of corner columns were considered as structural responses and code limits were set to assess structural failures. By using genetic algorithm, an optimum design was obtained. A parametric study was further performed to study the influence of isola-tion system type and seismic gap, along with which the pendulum system and collision problem were investigated. It is found that scuttle isolation via the pendulum system is not a better design as expected since the failure probability is not sensitive to the variance of pendulum curvature. By considering scuttle collision as additional factor, the failure probability with the inclusion of this constraint was obtained. The proposed design method is based on structural reliability, which is capable of considering multi-dimensional outputs and corresponding response limits and integ-rating multiple structural responses into one failure probability.