For accurately predicting the failure response of composite structures, it is crucial to consider the impacts of process-induced residual stresses and defects. However, given the complex multi-scale nature of both process-induced defects and composite failure, as well as the trade-off between simulation fidelity and computational speed, this aspect is often neglected. To address this challenge, we have developed a novel approach that utilizes both low- and high-fidelity finite element (FE) simulations, as well as surrogate machine learning (ML) models, for integrated process and failure analyses. Our approach begins with macroscale process simulation of thermo–chemical–mechanical responses. This is followed by micro-scale analysis of residual stresses and failure mechanisms, utilizing two distinct modeling fidelities. By generating large datasets from low-fidelity simulations and enriching them with smaller, high-fidelity datasets, we then train surrogate ML models. These models achieve the precision of high-fidelity simulations but operate at speeds surpassing even those of low-fidelity simulations. Benchmark studies, using the HEXCEL AS4/8552 material system, demonstrate our approach’s ability to significantly accelerate the simulation speed of high-fidelity models while enhancing accuracy by integrating process and failure analyses.