Abstract: As the core control unit of aero-engines,the fuel metering unit (FMU) is also one of the components with the highest failure rates.To address the challenges of limited training samples and severe sample homogeneity when developing machine-learning-based fault diagnosis models,this paper proposes a small-sample data augmentation method that integrates an optimized Time-series Generative Adversarial Network (TimeGAN) with an Isolation Forest (iForest).The proposed approach employs TimeGAN to learn the temporal correlations of fault sequences generated by the AMESim model and to synthesize sufficient multivariate fault samples.A Particle Swarm Optimization (PSO) algorithm is introduced to optimize the hyperparameters of TimeGAN,enhancing its capability of feature learning across different fault types.Based on the optimized TimeGAN-generated data,iForest is further applied to detect and remove anomalous synthesized samples,thereby improving the overall quality of the augmented dataset.Validation using limited fault data obtained from the AMESim model of a certain aero-engine FMU demonstrates that the proposed method significantly improves data diversity,representativeness,and effective coverage compared with traditional approaches.The enhanced dataset effectively mitigates insufficient feature learning under small-sample conditions and provides stronger data support for intelligent operation,maintenance,and fault identification of aero-engine fuel metering systems.