A deployable digital twin framework for bolt-torque specification compression in EV chassis assembly

Bolted-joint tightening affects both structural integrity and line performance in EV chassis assembly, yet production torque tables are often conservative to accommodate variability in joint mechanics and shop-floor conditions. This study uses a deployable digital-twin workflow to reduce torque variants while jointly considering mechanics constraints, AGV-related logistics, and cost/OEE objectives. The workflow couples (i) a torque–preload joint model with vibration and fatigue checks, (ii) an uncertainty-calibrated, value-aware LSTM for quality-risk prediction from torque–angle signatures and context, and (iii) a multi-objective PSO variant for mixed discrete–continuous search with feasibility-first handling. Candidate torque tables are admitted only after verification-twin re-evaluation, which serves as a release gate. In an industrial study covering 5,524 vehicles (Feb 2024–Jan 2025), torque specifications are reduced from 23 variants to 8 (− 65.2%) while keeping the observed reject rate within a ≤ 0.05% cap and meeting Goodman safety ≥ 1.5. Standardization reduces end-effector changeovers by 31% and AGV idle time by 14% at a 42 s takt without increasing fleet size. Cross-platform transfer is evaluated on three EV platforms, and ablations are reported with statistical tests and effect sizes alongside V&V metrics. Reproducibility is supported by a complete nomenclature, fully specified algorithms, and a shareable synthetic dataset with scripts that regenerate all figures and tables.