Machine Learning-Based Uncertainty Quantification of Passive Intermodulation in Aluminum Contact

This paper deals with the development of a surrogate model for the uncertainty quantification and the stochastic analysis of passive intermodulation (PIM) in an Aluminum-Aluminum contact based on the least-squares support vector machine (LS-SVM) regression. Starting from a small set of training pairs collecting the configuration of the un-certain parameters and the corresponding PIM level, the LS-SVM allows to build a closed-form approximation of such non-linear relationship. Such model, can be suitably used within a Monte Carlo (MC) scenario in order to accelerate the simulation process and provide all the statistical quantities of interest. The results show a considerable speed-up on the computational time compared to a plain MC simulation, while achieving an accurate approximation of the PIM probability density function.