概率神经网络用于机匣振动故障诊断
Casing Vibration Fault Diagnosis based on Probabilistic Neural Networks

对机匣振动加速度信号进行采集与分析是航空发动机振动故障诊断的重要方法。对振动信号进行处理与特征提取后，可以利用神经网络非线性映射的能力，对振动故障实现分类。利用转子-轴承-机匣耦合振动实验台模拟了5种风扇机匣的振动故障，从频谱的初步分析中并未能够实现对故障的准确判断。对振动数据进行了处理，提取了振动波形的频域与幅域参数，采用概率神经网络的方法实现单一故障的分类，并对不同参数所训练的网络进行了比较，检验了该诊断方法对于机匣振动故障的可行性。
The acquisition and analysis of casing's acceleration vibration signal is a significant method to realize aero-engine's fault diagnosis. Neural networks can classify vibration faults with their strong nonlinear mapping ability. 5 different types of fan casing vibration faults are simulated on the rotor-bearing-casing coupled vibration experiment system. Accurate judgment of faults can't gain from the preliminary analysis of signal's frequency spectrum. Single fault is classified from faults by probabilistic neural networks (PNN) using vibration data's frequency and amplitude domain parameters. Methods with different parameters are compared and the feasibility of PNN using in casing vibration fault diagnosis is tested