基于AdaBoost算法的故障诊断仿真研究

AdaBoost算法是提高预测学习系统预测能力的有效工具。提出一种基于AdaBoost算法的神经网络故障诊断方法，利用多层前向神经网络作为故障弱分类器，实现了对多类故障的诊断。为了克服AdaBoost对数据噪声比较敏感的不足，通过降低错分样本的权重改进了算法。针对一个涡轮喷气发动机气路部件故障的仿真实验表明，这种方法提高了最终故障分类器的泛化能力，改善了其噪声鲁棒性，便于工程应用。

Simulation research on fault diagnosis using AdaBoost algorithm

AdaBoost is one of the most efficient toots to improve the predictive accuracy of any given learning algorithm. A multi-class fault diagnosis method was developed, which used three-layer perceptions as weak classifiers and combined them to create an aggregate hypothesis with AdaBoost iteration. The ordinary AdaBoost algorithm was modified to overcome noise sensitivity through reducing the weights of misclassified samples. A simulation experiment for the gas path components of a turbojet engine was conducted to demonstrate the effect of the method. 24 groups of testing data from the turbojet engine were correctly classified into 5 fault classes. The simulation results show that the properties of the final fault classifier are enhanced in both generalization and robustness to noise.