← Back to Use Cases

Climb-Cruise Engine Matching

Name Affiliation Phone Number Email Address
Sanjiv Sharma sanjiv.sharma@airbus.com
Industrial Sectors:



  Consider a use case in the context of a 24-hour operation aircraft; a key implication is that aircraft can operate within noise curfews. Then, using a Set-Based Design approach, explore a multitude aircraft configurations with respect to their climb out Noise Levels, the Cruise Performances and Gaseous Emissions, under uncertainty (Figure 1). Screen Shot 2015-06-20 at 23.11.02 The basic concept is to generate data for a set of representative single aisle aircraft configurations by combining a multitude of airframes with a multitude of engines. Specify the noise measurement criteria in terms of the location of the measurement system and the type of noise level to be deduced. Then, using coupled analyses “plug-ins”, derive the performance models that enable an architect to explore the sensitivities amongst three exemplar measures of aptness. For this use case these are:

  1. Noise Levels (lower is better); conversely, Noise Level Margin (higher is better)
  2. Cruise Fuel Consumptions (lower is better)
  3. Gaseous Emissions (lower is better); conversely Gaseous Emissions Margin (higher is better)
This provides a multi-dimensional challenge for determining, visualising and acting on the uncertainties that propagate through the analyses to these measures of aptness. The main objective is to narrow the set of possible aircraft configurations to a set of feasible ones using uncertain, multi-dimensional decision criteria. The process is then repeated, using analyses models closer to the laws of nature, to narrow the set of feasible aircraft configurations to a set that provides competitive advantages (Figure 2). Screen Shot 2015-06-20 at 23.12.52 Hence, the UQ&M analyses will be used for:
  • robust design-making to narrow a set of possible aircraft configurations
  • discovering the parameters that strongly contribute to the variations in the measures of aptness
  • managing key parameters to drive reliably towards the desired properties and behaviours

2.1 Process Inputs

In the case where historical data or test data is not available (for example: novel airframe or engine configurations) then the uncertainty problem definition may require inputs from domain specialists. These inputs are envisaged to be elicited through the application of Structured Expert Judgement (SEJ) [1]. The results of the analyses would require special treatment and Probabilistic Design Ranking is proposed because it would:

  • account for lack of detailed design knowledge when comparing alternative aircraft configurations
  • account for alternative stakeholder biases and preferences for design metrics
  • provide a measure of confidence in the relative ranking of the aircraft configurations Specifically, the process inputs are clustered into two groups: Design Variables (Table 1) and Uncertainty Variables (Table 2).
Screen Shot 2015-06-20 at 23.14.05 The design variables are treated as being deterministic. However, in the general case a design variable may itself be uncertain, in which case it is decomposed into a dyadic comprising a deterministic design variable and an uncertain variable. The decomposition depends on the nature of the problem and is typically additive or multiplicative. In Table 2, the naming of the variable is indicative of the composition method applied to the decomposed variable: e.g. the words factor or efficiency imply that it they are multiplicative; whereas margin indicates additive composition. Screen Shot 2015-06-20 at 23.14.51 The statistical properties of the uncertainty variables, Table 2, may be assigned in many ways, depending on what is known about them. For example, these statistical properties could be expressed as a range of values, enumerated values, probability distribution functions, or tabulated probability distribution values. Functional dependencies Functional dependencies amongst the variables are not assumed a priori, instead surrogate models are developed using domain-specific models to represent these dependencies. This is specifically needed for novel aircraft configurations where such information may not exist. Discovering these functional dependencies then provides information for deciding how to propagate the uncertainties through the system of analyses. These surrogate models become assets for reuse during future assessments of similar aircraft configurations.  

2.2 Propagation

Whilst the process is multi-organizational and multi-disciplinary, for the purposes of this “use case”, it has been simplified to a chain of coupled analyses. Therefore, it is necessary to propagate the uncertainties stochastically across this chain of coupled analyses. The UQ&M treatment must be a non-intrusive approach around the process because the models are used for other purposes; the application of the UQ&M analyses in this use- case is to reuse these models. These models may be domain-specific models and are at various degrees of representations; from the abstract to the physical (including experimental data). The architect must be able to choose heterogeneous degrees of representation depending on the decisions that need to be made. The intention is that, using the architect’s selection of the models, a series of “design of experiments” and monte-carlo simulations approaches are used to generate domain- specific data. Then, this data is used to create appropriate surrogate models to represent the functional relationships amongst the variables. The surrogate models enable the rapid analyses of the uncertainty propagation and identification of those parameters that significantly contribute to the uncertainties. No prerequisite constrains are applied to the underlying design space; they may be linear or highly non-linear. However, further research is needed for cases where the design space may be discontinuous and where it may contain folds that are important to capture. The surrogates models are created semi-automatically, using multi-layer perception type feed-forward artificial neural networks. In this case, the analysis steps in the process are functionally interdependent; the process consist of tightly coupled analyses models, as shown in Figure 3. Screen Shot 2015-06-20 at 23.16.31   Inverse mapping is an important requirement because we want to be able to identify the uncertainties in the inputs that contribute the most to uncertainties in the output, thereby forming a strategy for robust design decision-making.

2.3 Interpretation and Communication of Results

Probabilistic Design Ranking provides a way of accounting for the lack of detailed design knowledge for comparing alternative conceptual designs; balances the biases and preferences of alternative stakeholders; and provides a measure of confidence in the relative ranking of conceptual design options. Some examples of the types of information to be visualised is given in the following figures: Screen Shot 2015-06-20 at 23.17.19 Screen Shot 2015-06-20 at 23.17.58 Screen Shot 2015-06-20 at 23.18.36 Screen Shot 2015-06-20 at 23.19.21  


  The proof of concept was executed during the period of performance of the of the Innovate UK funded project CONGA (Configuration Optimisation of the Next Generation Aircraft). The methods used were experimental with a view of understanding how to  

  • Propagate uncertainty through tightly-coupled analyses models
  • Discover the significant parameters that give rise to uncertainty
  • Quantify uncertainty when data is not available
  • Visualise and present uncertainty to enable inclusive decision-making (i.e. by stakeholders who do not have a background in UQ&M)
  • Use visual analytics as an interactive knowledge acquisitions approach for uncertainty propagation
  The tools used for generating the data, conducting uncertainty analyses and visualising the results were:
  • Engineering analyses (Environmental Design Space: An engine/airframe modelling and simulation framework [2])
  •  Mathwork’s Matlab® for creating multi-dimensional surrogate models
  • SAS’s JMP® for conducting both batch and interactive statistical analyses[3]


Cooke, R. & Goossens, L. (2000), “Procedures Guide for Structural Expert Judgement in Accident Consequence Modelling”, Radiation Protection Dosimetry 90(3), 303-309.

Kirby, M.R. and Mavris, D.N., “The Environmental Design Space”, 26th International Congress of the Aeronautical Sciences, Anchorage, AK, ICAS-2008-4.7.3, 2008.