Building Surrogate Models in JMP: Combining Statistical and Physical Modelling Techniques (2026-EU-30MP-2776)

This presentation highlights the application of statistical modelling to perform sensitivity analysis and optimisation of complex physical models more efficiently and effectively.  

Physical models can often be complex, as it is the reality that they are representing, which requires significant resources for performing sensitivity analysis and makes their optimisation very challenging, or even not possible. Sensitivity analysis involves exploring different scenarios, after which models are used to determine optimal values for the input parameters. Often this process is not straightforward, as it requires running the model across many different parameter combinations.

Recent projects have highlighted the benefits of using statistical techniques, such as design of experiments, to select representative data to build a surrogate model, based on the physical model outputs. The Prediction Profiler in JMP is then utilised to perform the sensitivity analysis and optimisation of the surrogate model instead. This capability enables focused exploration of the experimental space and helps to explain the relationships between different parameters in a relatively short timescale. This approach has been successfully applied to both atomistic and engineering models.

This presentation showcases the use of statistical tools in JMP to maximise the value obtained from complex physical computational models.