Continue
By using our website you accept our cookies policy.Find out more

Visualisation to Help Engineering Decision Making Under Uncertainty

Posted on 06/11/2016

Uncertainty in Industrial Design

Virtual design and simulation provides the opportunity to go beyond experienced-based judgment in engineering design

We are all faced with uncertainty. In our everyday lives we must make decisions where uncertainty causes those decisions to be difficult. With time, however, we learn to make judgments based on accumulated experience and become capable of balancing uncertainties and risks in our heads to make sensible decisions.

When it comes to engineering design decisions, quantifying and dealing with uncertainty is becoming increasingly important. Uncertainty analysis provides a more accurate description of the engineering system; the benefit being that this improved system description might support the identification of better engineering solutions [i] as well as providing confidence to a decision maker at key stages of the design process.

However, as the timescale from concept to commercialisation is eroding, there is decreasing time to accumulate the experience we rely on to make intuitive decisions, and turnover of personnel and the cost of personnel time both mean that it is not practical to pass on all that experience to one’s successor [ii – link].

However, we live in an era of virtual design, and there exists an opportunity to supplement experience-based judgment with computer simulation and handle uncertainty in a systematic way to fully realise the potential of uncertainty driven engineering design.

The Uncertainty Quantification and Management Special Interest Group is working to unite the UK capability in this field, and wishes to switch industry on to the benefits of incorporating uncertainty into their industrial design processes.

 

There are different kinds of uncertainty to consider

Early in the establishment of the Special Interest Group, we appreciated that a major roadblock in the wider industry uptake of uncertainty quantification (UQ) and management was the difficulty in communicating information arising from an UQ analysis to a decision maker – probabilities which output from a UQ analysis are notoriously difficult to communicate effectively.

There are different kinds of uncertainty to consider in this communication, and decision makers may need to take this into account. Broadly, uncertainties classify themselves into two camps (1) aleatory uncertainties, and (2) epistemic uncertainties.

  • Aleatory uncertainties are due to the physical variability in the system being examined. They are not strictly due to lack of knowledge and cannot therefore be reduced. The random nature of some material properties or operating conditions of a physical system typically leads to aleatory uncertainties.
  • Epistemic uncertainties are solely due to a lack of knowledge. They can arise from assumptions in the mathematical model used for the simulation or simplifications made to complex systems. It is clearly possible to reduce the epistemic uncertainty by using data, refinement of models etc.

 

Decision-making under uncertainty

At a recent seminar organised by the Uncertainty Quantification & Management Special Interest Group, Luis Crepso, Senior Research Scientist at NASA discussed the difficulty in handling uncertainties in engineering practices.

Decision-makers are often tasked with choosing a course of action where the data used to make that decision are uncertain. Shown below in Figure 1 is a simple example of this; ‘yay’ or ‘nay’ decision outcomes, this decision will be made based on some Figure of Merit (FOM) and the decision is made with respect to a decision criteria (in blue).
If the FOM is determined to be a scalar, then the decision maker has an easy job – if the FOM lies to the left of the criteria, a ‘nay’ decision is returned. If the FOM lies to the right, a ‘yay’ decision is returned.

However, in a UQ analysis, the decision maker will need to make a decision where the FOM will be determined to lie within an uncertainty range (displayed as the bracket on the x-axis). The size and distribution of the bracket is the role of the UQ analyst to define, and this needs to be done with care. An underestimate of this bracket could result in the wrong decision being made. An overestimate might prevent a decision being made at all as the range is too wide to sensibly be able to say anything.
This bracket can be a probability, but a decision cannot be – a decision can never be half made!

 

Using Visualisation to Help the Decision Maker

Decision makers are often faced with the task of identifying the “best” option from a set. However, as mentioned above, the inclusion of uncertainty makes the task difficult. Fortunately, it is increasingly easy to present data in the form of interactive visualisations and in multiple types of representation that can be adjusted to user needs and capabilities.

Nonetheless, communicating deeper uncertainties resulting from epistemic uncertainty remains a challenge. Perhaps the greatest challenge is to make a visualisation that is attractive and informative, and yet conveys its own contingency and limitations.
Matt Butchers – Knowledge Transfer Manager, Industrial Mathematics