Imprecise probabilities in engineering analyses
Summary (2 min read)
1. Introduction
- The analysis and reliability assessment of engineered structures and systems involves uncertainty and imprecision in parameters and models of di erent types.
- Information is often not available in the form of precise models and parameter values; it rather appears as imprecise, di use, uctuating, incomplete, fragmentary, vague, ambiguous, dubious, or linguistic.
- If this balance is violated or not achieved, computational results may deviate signi cantly from reality, and the associated decisions may lead to serious consequences.
- Solutions to this problem are discussed in the literature from various perspectives using di erent mathematical concepts.
- Beyond this coarse review, the collection of papers in this Special Issue provides detailed insight into various imprecise probability approaches and high- lights their bene ts in engineering analyses.
2. Facets of epistemic uncertainty
- If subjective probabilistic statements can be formulated on rational grounds and some data of suitable quality are available, then Bayesian updating can play its important role.
- This conceptual understanding together with the classi cation into probabilistic uncertainty and imprecision provides intuitive motivation for imprecise probabilities and their terminology.
- Suppose the dependency between load and de ection is known deterministically, but a load parameter is available in the form of bounds only.
- Fuzzy numbers are a generalization and re nement of intervals for representing imprecise parameters and quantities.
3.1. Emergence in engineering
- A key feature of imprecise probabilities is the identi cation of bounds on probabilities for events of interest; the uncertainty of an event is character- ized with two measure values a lower probability and an upper probability [54].
- The distance between the probability bounds re ects the indeterminacy in model speci cations expressed as imprecision of the models.
- Imprecise probabilities include a large variety of speci c theories and mathematical models associated with an entire class of measures.
- The adoption of imprecise probabilities and related theories for the solution of engineering problems can be traced, in its early stage, with the publications [83, 84, 85, 86, 87].
3.2. Engineering application elds
- From the initial developments imprecise probabilities have emerged into several application elds in engineering with structured approaches.
- In [94] intervals are employed for the description of the imprecision in probabilistic models for a structural reliability assessment.
- The developments in this area have been extended to applicability to larger, realistic and practical problems.
- The proposed interval-valued sensitivity index measures the relative contribution of individual model input variables, in the form of intervals or sets of distribution functions, to the uncertainty in the model output.
4.1. Conceptual categories
- The ideas of imprecise probabilities may be categorized into three basic groups of concepts associated with three di erent technical approaches to construct imprecise probabilistic models.
- In general, these coarse specications may be the best information available, or they may arise from limitations in measurement feasibility.
- This imprecision may arise, for example, when con icting information regarding the distribution type is obtained from statistical tests, that is, when the test results for di erent distributions as well as for compound distributions thereof with any mixing ratio are similar.
- The authors shall see below that interval probabilities can be used to represent this group of concepts.
- But from a practical point of view, this categorization and the associated features of the concepts as elucidated in the subsequent sections can provide the engineer with a good sense for the modeling of a problem.
4.2. Evidence theory
- If the information available possesses some probabilistic or probability-related background, but does not meet the preconditions to be speci ed as a random variable, evidence theory often provides a suitable basis for an appropriate quanti cation and subsequent processing.
- On this basis various speci c uncertainty measures may be derived within a range from plausibility to belief which include traditional probability as a special case.
- The compliance with the traditional de nition of probability is, however, not complete because the focal subsets are not required to be disjoint.
- The subjects included in the discussion concern uncertainty quanti cation, computation of model predictions and system responses, simulation, optimization and design, and decision making.
- A crucial point in the practical application of evidence theory is realizing the basic probability assignment in each particular case.
4.3. Interval probabilities
- Initial applications, although few, already indicate the usefulness of interval probabilities and demonstrate their capabilities.
- Numerically, conditional probability is quite di erent from the probability of the conditional.
4.4. Probability bounds analysis with p-boxes
- Probability bounds analysis [162, 163, 164, 61] is another of the uncertainty quanti cation approaches that are considered part of the theory of imprecise probability [44].
- A p-box represents a class of probability distributions consistent with these constraints.
- There are several other techniques that originated in the eld of interval analysis that are likewise fruitfully extended to probability bounds analysis, three of which the authors mention here.
- Taylor arithmetic can also project uncertainty expressed as p-boxes whether they represent epistemic uncertainty or aleatory uncertainty or both.
- Like Monte Carlo simulation, it sidesteps the problem of repeated uncertain variables to approximate best-possible bounds.
5. Conclusions
- In solving engineering problems, it is extremely important to properly take uncertainty and imprecision into consideration.
- In engineering applications, there are two main sources of uncertainty and imprecision.
- T. Augustin, F. Coolen (Eds.), Special Issue: Imprecise probability in statistical inference and decision making, Vol. 51 of International Journal of Approximate Reasoning, 2010. [73].
Did you find this useful? Give us your feedback
Citations
151 citations
Cites background from "Imprecise probabilities in engineer..."
...A detailed reasoning and discussion in this direction with an overview on available generalized models is provided in [14]....
[...]
108 citations
102 citations
Cites background from "Imprecise probabilities in engineer..."
...In PHM, the formulation of appropriate solutions should also take significant information into account but without introducing unwarranted assumptions to remain applicable and sufficiently general (Beer et al., 2013)....
[...]
101 citations
88 citations
References
[...]
52,705 citations
14,565 citations
"Imprecise probabilities in engineer..." refers background in this paper
...Imprecise probabilities have a close relationship to the theory of random sets [56, 57] and cover, for example, the concept of upper and lower probabilities [58], sets of probability measures [59], distribution envelopes [60], probability bounds analysis using p-boxes [61], interval probabilities [62], Choquet capacities [63] of various orders, and evidence theory (or Dempster-Shafer Theory) [64, 65]...
[...]
...[133] E. Auer, W. Luther, G. Rebner, P. Limbourg, A Veri ed MATLAB Toolbox for the Dempster-Shafer Theory, in: Proceedings of the Workshop on the Theory of Belief Functions, 2010....
[...]
...A mathematical description with details and background information is provided in [65]....
[...]
...Imprecise probabilities have a close relationship to the theory of random sets [56, 57] and cover, for example, the concept of upper and lower probabilities [58], sets of probability measures [59], distribution envelopes [60], probability bounds analysis using p-boxes [61], interval probabilities [62], Choquet capacities [63] of various orders, and evidence theory (or Dempster-Shafer Theory) [64, 65] as a theory of in nitely monotone Choquet capacities [66, 67]....
[...]
7,877 citations
7,826 citations