To address the issue of fatigue cracking in engine exhaust manifolds under high and low-temperature loads and to bridge the gap between traditional bench tests and actual user operation, an accelerated test and reliability assessment method were proposed based on a combined analysis of damage modeling and big data. A temperature-driven quantitative damage model for the exhaust manifold was established based on the Coffin-Manson fatigue theory and Miners linear cumulative damage rule. Then a virtual sensor model for estimating the manifold surface temperature was developed. The operational data from vehicle users were collected and extrapolated to determine a target pseudo-damage value covering 95% of user damage levels. And the accelerated thermal shock tests were subsequently conducted on the engine test bench for comparative validation. The results demonstrate that the virtual sensor model achieves a calculation error of less than 5%, meeting engineering application requirements. The accuracy of damage model was verified through both bench tests and vehicle usage data, with predictions indicating that the optimized exhaust manifold could withstand over 600 000 kilometers of service under general user conditions. The proposed damage calculation model provides a reliable basis for the durability assessment and life prediction of engine exhaust manifolds.