Critical Parameter Identification in Power Electronic Inverters Using Optimized Monte Carlo Analysis Under Parameter Uncertainties

Abstract

There is a significant effort in understanding the cumulative impact of tolerance chains on the behavior of a power-electronic device. This step is particularly important to optimize design margins and ensure homologation of a serieslevel development. Nevertheless, the computational cost of a brute-force analysis can be prohibitive when a large number of parameters with complex interactions are involved. To address this issue, this paper presents an efficient Monte Carlo analysis framework for power-electronic circuits under parameter variability. Using Latin hypercube sampling combined with a standard error of the mean based stopping criterion for sampling, the proposed approach minimizes the number of simulations by 84% while maintaining high accuracy. Additionally, sensitivity analysis serves to identify device- and system-level parameters with the greatest impact on the device loss profile, enabling targeted exploration of the most influential variables. Lastly, guidelines for optimal sampling density are provided to efficiently probe the parameter space and avoid excessive computational cost. Applied to LTspice double-pulse tests with realistic MOS-FET bare-die and parasitic models of the power module and DC-link capacitor, the method evaluates system efficiency under device-level variations.