讲座题目:工程中认知不确定性的模拟与分析 Modeling and Analysiswith Epistemic Uncertainties inEngineering
讲座时间:11月8日(周五)14:15
讲座地点:华南理工大学广州国际校区G2-101
主讲人:Michael Beer教授(欧洲安全与可靠性协会主席)
主讲人简介:
Michael Beer是德国汉诺威大学教授和风险与可靠性研究所所长。他也是利物浦大学的兼职教授,以及同济大学和清华大学的客座教授。他在德国德累斯顿技术大学获得博士学位,曾在美国莱斯大学、新加坡国立大学和利物浦大学工作。Beer教授的研究重点是工程中的不确定性量化,侧重于不精确概率。Beer教授是ASCE-ASME工程系统中的风险和不确定性杂志的主编。他也是地震工程百科全书的主编(联合)和信息科学的副主编。他曾获得多个奖项,包括美国土木工程学会(ASCE)的Alfredo Ang土木基础设施风险分析与管理奖。他是欧洲安全和可靠性协会(ESRA)主席,也是ASCE基础设施韧性部门(IRD)以及风险和韧性测量委员会(RRMC)的联合主席。他是国际安全与可靠性协会的执行委员会成员。他是洪堡基金会的成员,也是ASCE、ASME、IASSAR、CERRA、IACM、ESRA、EASD和GACM的成员。
Michael Beeris Professor and Head of theInstitute for Risk and Reliability, LeibnizUniversität Hannover, Germany. He isalso part time Professor at the Universityof Liverpool and guest Professor atTongji University and Tsinghua University,China. He obtained a doctoral degreefrom Technical University Dresden,Germany, and worked for Rice University,National University of Singapore, andthe University of Liverpool. Dr. Beer’sresearch is focused on uncertaintyquantification in engineering with emphasison imprecise probabilities. Dr.Beer is Editor in Chief of the ASCE-ASME Journalof Risk and Uncertainty inEngineering Systems, Part A Civil Engineering andPart B MechanicalEngineering. He is also Editor in Chief (joint) of theEncyclopedia ofEarthquake Engineering, and Associate Editor of InformationSciences. He haswon several awards including the Alfredo Ang Award on RiskAnalysis andManagement of Civil Infrastructure of ASCE. Dr. Beer is theChairman of theEuropean Safety and Reliability Association (ESRA) and aCo-Chair of Risk andResilience Measurements Committee (RRMC), InfrastructureResilience Division(IRD), ASCE. He is serving on the Executive Board of theInternational Safetyand Reliability Association. He is a Fellow of theAlexander vonHumboldt-Foundation and a Member of ASCE, ASME, IASSAR, CERRA,IACM, ESRA,EASD, and GACM.
讲座摘要:
认知不确定性在所有工程领域中都存在相当大的程度。尽管它们通常可以被现象学和定性地描述,但它们阻碍了严格定量描述的形成,而这种定量描述是现实风险评估的基础。在存在认知不确定性的情况下,指定概率模型和相关的风险分析会导致假设性的结果,这些结果假设了一些直观的猜测来捕捉认知不确定性的影响。也就是说,我们基于代表假设而非事实的条件来量化风险。这样的结果可能会严重误导。因此,尽可能真实地量化认知不确定性至关重要。这种量化既不应引入未经授权的信息,也不应忽视信息。基于此,有一个明确的共识,即需要考虑认知不确定性以实现风险和可靠性的现实评估。然而,并没有一个明确定义的程序来掌握这一挑战。相反,有各种各样的概念和方法可供工程师选择来处理认知不确定性。这种选择之所以困难,是因为人们感觉到现有的概念是相互竞争和对立的,而不是互补和兼容的。显然,首先应该考虑的是概率建模,自然通过主观概率,这表达了专家的信念,并且可以通过贝叶斯方法以一种连贯的方式整合到一个完全的概率框架中。虽然这条路径被广泛接受并被认为是非常强大的,但集合理论方法和不精确概率的潜力只被利用了一小部分。然而,在可用信息不足以有意义地指定主观概率分布的情况下,这些方法越来越受到关注。演讲将展示认知不确定性的模型,并强调它们在工程分析和设计中的能力和附加值。使用示例来解释各自的特征。对模型的讨论通过展示一种强大的数值技术来补充,即使在非常复杂和非线性的工程分析中,也可以处理认知不确定性。这项技术不仅可以用于可靠性分析,还可以用于敏感性分析、设计、模型更新等。
Epistemicuncertainties appear across allengineering fields to quite some significantextent. Although they can often be describedphenomenologically andqualitatively, they counteract a rigorous quantitativedescription, which isneeded as a basis for a realistic risk assessment. In thepresence of epistemicuncertainties the specification of a probabilistic modeland the associatedrisk analysis lead to hypothetical results presuming someintuitive guess tocapture the influence of the epistemic uncertainty. That is,we quantify riskbased on conditions that represent assumptions rather thanfacts. Such resultscan be significantly misleading. It is thus of paramountimportance to quantifyepistemic uncertainties most realistically. Thisquantification should neitherintroduce unwarranted information nor should itneglect information. On thisbasis there is a clear consensus that epistemicuncertainties need to be takeninto account for a realistic assessment of riskand reliability. However, thereis no clearly defined procedure to master thischallenge. There are rather avariety of concepts and approaches available todeal with epistemicuncertainties, from which the engineer can chose. Thischoice is made difficultby the perception that the available concepts arecompeting and opposed to oneanother rather than being complementary andcompatible. Clearly, the firstconsideration should be devoted to aprobabilistic modelling, naturally throughsubjective probabilities, whichexpress a belief of the expert and can beintegrated into a fully probabilisticframework in a coherent manner via aBayesian approach. While this pathway iswidely accepted and recognized asbeing very powerful, the potential ofset-theoretical approaches and impreciseprobabilities has only been utilized tosome minor extent. Those approaches,however, attract increasing attention incases when available information isnot rich enough to meaningfully specifysubjective probability distributions.The presentation will feature models forepistemic uncertainties, and it willhighlight their capabilities and addedvalue when used for engineering analysisand design. Illustrative examples areused to explain the respective features.The discussion on the models iscomplemented by presenting a powerful numericaltechnology for processingepistemic uncertainties even in very complex andnonlinear engineering analyses.This technology can be used not only forreliability analysis, but also forsensitivity analysis, design, model updatingand more.


