In the era of data analytics, the use of predictive models for time-consuming simulations is widely accepted in the industry. But many believe predictive models are not useful for simulations that run quickly. This is a common but incorrect belief. With examples from COMSOL Multiphysics, this webinar will illustrate the benefits of building a predictive model for a fast-running simulation.
A simulation model is a black box function. Even with a large number of evaluations, the simulation model does not provide sufficient information to form an understanding of the input and output relationship. A predictive model opens up the black box of the simulation model. Once a predictive model is trained using machine learning algorithms to learn the input-output relationship of the simulation based on a training data set, it can be used conveniently to plot and study the relationship.
Even with quick simulation models, performing advanced analytics tasks like design space exploration, sensitivity analysis, uncertainty propagation, and optimization are time consuming and not practical. But a predictive model can rapidly predict the outputs for input configurations not contained in the training data, making it an excellent option for performing advanced analytics tasks that require a number of function evaluations. Moreover, predictive models are necessary to improve a simulation model’s accuracy and estimate the distribution of inputs by using statistical calibration and inverse analysis techniques, respectively. All these methodologies will be discussed in this webinar.
