Skip to main content
RISE logo

Machine Learning for Uncertainty Propagation (MLUP)

Machine learning (ML) is an efficient tool to mimic complex processes. Numerical simulations, such as computational fluid dynamics (CFD) is often extremely time demanding and may occupy considerable computer resources for an extended period of time resulting in a very limited series of result. A proper uncertainty propagation study is in most cases not possible to perform, however a novel use of ML may give a good estimate of the propagation of input uncertainties and how they affect the simulation results. We call it MLUP.

Velocity
Photo: Olle Penttinen
A comparison of model and predicted deviation of the streamwise velocity field.

A simple example of a complex process is that of air flowing in a channel with e.g. variable temperature, viscosity inlet streamwise velocity and boundary conditions.  A testcase considering the air flow over a backward facing step is considered to show the methodology for MLUP. Uncertainty quantification deals with quantitative characterization and limitations of uncertainties in both calculations and experimental evaluations. The basic problem is to determine how credible a result is if certain aspects of the system are not exactly known. The methodology is tested in the following way, two separate simulation series with known input distributions are created using full factorial design. The artificial neural network (ANN) is trained on the first set of data and is then used to predict the spread in the data (standard deviations) on the second set of data using just the centre case of the second study as input.

The predicted results are compared with the output of the full factorial design (targets) of the second set. Even though training and test datasets are not overlapping, the results are correlated. It should be stressed that it would be possible to perform MLUP in a wide range of applications, such as different process and system simulations, LCA and coupled technoeconomic applications. Uncertainties are a key parameter to take into consideration regarding quality assessment, decision making and safety precautions. Don’t hesitate to contact us if you are interested in further information.


More information

 

Contact person

Johan Anderson

Forskare

+46 10 516 59 26
johan.anderson@ri.se

Read more about Johan