Uncertainty Analysis Methods for Multi-dimensional Data Fusion in Flight Tests

The development of modern aviation technology has made flight test multi-dimensional sensor data complex, highdimensional, and uncertain, making it difficult for traditional methods to cope. This paper proposes a multi-dimensional data fusion uncertainty analysis method based on machine learning. Through intelligent preprocessing, adaptive feature selection, and integrated classification models, it can accurately identify six typical fault modes and construct a multi-category system containing normal states and five types of faults. To address data imbalance, an improved SMOTE oversampling and random undersampling combination strategy was adopted, optimizing the ratio from 10.0:1 to 2.8:1; 160 key features were selected to achieve 60% dimensionality reduction; and random forest was determined to be the optimal classification algorithm. Our experiments show that this method improves the classification accuracy from 35.3% to 72.4%, and it achieves an F1 score of 71.8% with a training time of only 2.3 seconds-traits that demonstrate its excellent real-time performance in practical scenarios. When we break down the results by category, we find the F1 scores for normal conditions and engine faults reach 87.5% and 79.3%, respectively. This method ensures high-precision identification of major fault types while maintaining acceptable detection capabilities for minor faults, and these combined strengths offer technical support for both flight safety monitoring and predictive maintenance.