Multidimensional Adaptive Sim2Real Transfer Learning for Quadcopter Fault Diagnosis

The reliability of unmanned aerial vehicles (UAVs) fundamentally governs their deployment scalability, where advanced fault diagnosis methodologies play a critical role. This study develops a novel fault diagnosis framework for quadcopters, which reduces the dependence on costly and scarce real flight fault data by employing an advanced Simulation-to-Reality (Sim2Real) transfer learning strategy. The proposed method not only improves cost-effectiveness but also introduces novel technical features. The core innovation lies in a novel multidimensional adaptive TrAdaBoost (MAT) framework. MAT integrates both a Maximum Mean Discrepancy-based feature weight allocation mechanism and a BN-layer feature-based instance transfer learning technology, facilitating multidimensional transfer learning analysis. This integrated approach effectively enhances the model’s ability to learn valuable feature elements, ultimately enhancing its adaptability during domain adaptation. In addition, the proposed algorithm optimizes the traditional TrAdaBoost algorithm framework, particularly in cases with small samples, which addresses the issues of overfitting and base classifier selection, and further enhances the model’s generalization ability. Through a series of experiments, the effectiveness of the proposed framework is verified, and the support relationship between simulation data and real flight data is quantified, thereby establishing a new benchmark for the subsequent quality assessment of simulation data acquisition and the optimization of domain-adaptive algorithms. This study not only provides a new perspective in theory but also demonstrates its potential in practical applications, contributing a new solution to the UAV fault diagnosis.