Research data and code for "Reliability-based robust design of a nonlinear automotive assembly under manufacturing tolerances: a chance-constrained surrogate framework"

<p dir="ltr">This repository contains the dataset and Python code supporting the above study on reliability-based robust design of an automotive inner tie rod (ITR) caulking assembly.</p><p dir="ltr">The dataset (251202_rawdata_doe_fea_results_1.csv) comprises 81 high-fidelity finite-element analysis (FEA) cases from a three-level full-factorial design of experiments over four design factors (caulking stroke, housing lip length, machining volume, seating angle), together with the six resulting multi-physics responses (mass, caulking load, contact pressure, tilting torque, rotational torque, pull-out load).</p><p dir="ltr">The Jupyter notebooks reproduce the full analysis pipeline:</p><p dir="ltr">Surrogate modeling and validation: surrogate_comparison.ipynb (GPR vs. deep ensemble / RF / XGBoost / SVR), A_gpr_train_and_uq.ipynb (final GPR training and uncertainty calibration), gpr_parity_plot.ipynb, convergence_per_target.ipynb.</p><p dir="ltr">Interpretability: shap_beeswarm.ipynb, shap_interaction.ipynb, shap_interaction_heatmap.ipynb.</p><p dir="ltr">Optimization and diagnosis: optimizer_stats_v2.ipynb (BO/DE/PSO/SLSQP comparison with statistical tests), A_robustness_gpr.ipynb (local stability and weight sensitivity), diagnose_extrapolation.ipynb, SLSQP_only.ipynb.</p><p dir="ltr">Chance-constrained reformulation: chance_constrained_DE.ipynb, chance_constrained_slsqp.ipynb, chance_constrained_slsqp_CRN.ipynb.</p><p dir="ltr">Supplementary reliability analyses: reexp1_mc_violation.ipynb (Monte Carlo violation-probability estimation with confidence bounds), reexp2_largebudget.ipynb (budget-independence of the boundary collapse).</p><p dir="ltr">Requirements: Python 3.10, with numpy, pandas, scikit-learn, scipy, xgboost, shap, pyswarms, and bayesian-optimization.

The trained GPR models and scalers are generated by running A_gpr_train_and_uq.ipynb first; the remaining notebooks load these artifacts.</p>