Fault Diagnosis Model for UAV with Small Sample Size and Imbalanced Data

With the development of Unmanned Aerial Vehicles (UAV), their application scenarios and task requirements have become more complex, demanding higher standards for subsequent fault diagnosis and equipment maintenance. The actual flight data of UAV usually exhibits small sample sizes and imbalanced characteristics. To achieve more efficient fault prediction for UAV, this paper establishes a SMOTE-IWOA-RF fault diagnosis model. This model optimizes the Whale Optimization Algorithm (WOA) using chaotic initialization, nonlinear convergence factors, and Lévy flight strategies, resulting in better convergence accuracy and global search capability, as well as faster convergence speed. The model is compared with other classification models on data with different imbalance ratios, demonstrating that it is more efficient and accurate in UAV fault diagnosis, with the F1 score improving by about 10% compared to the pre-optimization state. Furthermore, the IWOA-RF model is combined with other data preprocessing methods, showing that SMOTE-IWOA-RF performs better on UAV operational data, indicating its practical value.