Sensitivity Analysis Lab Kit: Sobol Indices Calculator + Case Dataset

Sobol sensitivity analysis is a variance-based global method that decomposes the output variance of a computationally expensive model into contributions from individual inputs and their higher-order interactions. Unlike local methods (e.g., derivatives), it operates over the full joint input distribution—typically uniform or Gaussian—and yields dimensionless, normalized indices ranging from 0 to 1. The lab kit implements efficient quasi-Monte Carlo sampling (e.g., Sobol sequences) and leverages estimator algorithms (e.g., Saltelli’s extension) to compute both first-order (Si) and total-order (STi) indices with reduced sample requirements and improved convergence. The included case dataset comprises a curated set of benchmark process simulation scenarios—such as a continuous stirred-tank reactor (CSTR) with kinetic uncertainty, a distillation column with thermodynamic property variability, and a battery thermal management digital twin—each accompanied by ground-truth parameter distributions, reference outputs, and metadata describing model fidelity, runtime constraints, and validation metrics. Integration is facilitated through containerized Python modules (with NumPy/SciPy backend), JupyterLab notebooks for pedagogical walkthroughs, and API wrappers compatible with Modelica-based simulators and industrial digital twin platforms (e.g., Siemens XHQ, AspenTech DMC3). This enables engineers and data scientists to rapidly assess parameter influence without reimplementing core sensitivity logic, thereby accelerating model calibration, uncertainty quantification, and digital twin trustworthiness evaluation.