涡扇发动机气路故障定量诊断的BP网络研究

FAULT QUANTITATIVE DIAGNOSIS OF TURBOFAN GAS PATH COMPONENT BY BPN

为了克服ＢＰ算法收敛速度慢的问题，提出了一种基于混合学习规则的ＢＰ算法，并采用模归一化方法，成功地定量组织了故障的学习样本，建立了能够定量分析发动机气路部件故障的人工神经网络（ＢＰＮ）。通过分析测量系统随机误差的影响和实际试车数据的效验结果，表明该网络具有较强的推广能力及适应性，能基本满足故障定量诊断的要求，并具有较好的工程实用性。

In order to overcome the convergent difficulty of BP algorithm, a new learning rule of BP algorithm named hybrid rule is proposed.Normalizing the neural net input by its modulus, a fault library with fault magnitude and a BPN which can diagnosis the turbofan gas path component faults quantitatively are built successfully. A validation of the data of noisy measurements and the real engine ground test is made. The diagnostic results show that the BPN can quantify the magnitude of deterioration of the various engine components, detect the multiple faults and has robust adaptation of random error in measurement system.