Application of a decision tree for choosing appropriate methods to manage uncertainty in industrial Eco-design

Uncertainty is an unavoidable issue in most data-based studies (e.g., life cycle assessment, risk assessment), since they all require amounts of data of good quality or well-known modelling mechanism. As decisions making are... [ view full abstract ]

Uncertainty is an unavoidable issue in most data-based studies (e.g., life cycle assessment, risk assessment), since they all require amounts of data of good quality or well-known modelling mechanism. As decisions making are often constraint by lack of information and ignoring such problem may introduce costly error or bias, so an appropriate management of unknown information helps us to improve the credibility of decisions under uncertainty. Probabilistic methods are mainstream way to deal with data related uncertainty, while alternative methods (e.g., interval or fuzzy interval, evidence theory) have also been developed, since the former still requires information to determine the probability distribution, which may not be available. Therefore, our previous study developed a general decision tree to help choosing appropriate methods to describe the real state of knowledge. This study aims to apply the proposed methodology to deal with uncertainties in an industrial case study for Eco-design, and discuss how uncertain information is described for decision making support.

In this study, an eco-audit model is built to estimate energy consumption and CO₂ emissions during the whole life period of an electronic tablet. The selection of materials and manufacturing processes, as well as their associated data, are determined based on a tool for materials information management. Although this matrix-based model uses single value for input variables, they do give information about uncertainty as a range of value in our database. Therefore, we are going to make two scenarios to deal with uncertainties for illustrative purpose. In the first scenario (S1), we assume probability distribution function for each uncertain input variable as well as scenario uncertainty, even though such information may not be given. In the second scenario (S2), we assume that probability distribution is unknown for a part of uncertain input variables, but their intervals or set of intervals are given. So non-probabilistic method is applied according to our decision tree and a combined propagation method is performed to calculate the overall uncertainty in outputs.

As result, S2 generates graphically two bounds (lower and upper) of probability distributions, rather than a single one from S1 result. The width of two bounds indicates intuitionally potential space to reduce epistemic uncertainty by adding more information. The statistics of interest (e.g., mean, confidence interval) are formed as interval which indicates the unknown about the shape of distribution, while the classic probabilistic method does not consider this source of uncertainty, but ignore it by assumption. Therefore, S2 provides conservative, but more realistic information than that from S1. In conclusion, this decision tree offers several practical methods for the users to manage uncertainty given available information with different levels of precision. Moreover, it guides users to combine different sources of uncertainty in the same framework and evaluate them quantitatively. For example, scenario uncertainty is also integrated by probabilistic method, so that the “hot-spots” uncertainties can be identified via sensitivity analysis that will help decision makers to fill data gaps more efficiently. Although it covers commonly used methods, the decision tree is open for critiques and extensions.