Research on Manufacturing Deviation Vehicle Performance Transmission Mechanism and Intelligent Compensation Control for Automotive Engineering

Manufacturing deviations in multi-stage automotive processes propagate nonlinearly, ultimately reducing vehicle performance. However, practical closed-loop compensation remains limited by model mismatch and uncertainty. We propose a performance deviation digital twin model that combines a changeflow state-space model with a physically informed graph agent and uncertainty quantification, then computes opportunity-constrained MPC actions under actuator constraints. On a dataset of 500 participants containing 120 KPCs and multiple KPIs, the proposed hybrid model reduces the KPI prediction RMSE from 0.86 (SoV) to 0.41 and significantly alters the performance loss distribution (paired t-test p < 1e-16; mean reduction of 0.271, 95% CI [0.263, 0.280]). Conclusion: This framework achieves interpretable, scalable, and statistically reliable compensation, providing a practical approach to linking manufacturing variation control with vehicle-level performance assurance.