Research On Intelligent Fault Diagnosis Technology Of Laser Gyro

The laser gyroscope is an important inertial element in the strapdown inertial navigation system.Based on mechanically dithered ring laser gyroscope,modern signal processing technology and neural network were used to study on the fault diagnosis methods of frequency stabilization circuit and dither control circuit.Firstly,the appropriate test points were selected in the frequency stabilization control circuit and the dither control circuit,and the data was collected through fault injection.Different diagnostic methods were adopted respectively according to the signal characteristics of the two circuits.The essence of fault diagnosis technology is pattern classification and recognition.Extracting stable and effective fault signal characteristics is the key step in the diagnosis process.The signals of the frequency stabilization circuit are mostly direct current except the lamb wave signal which is a sine wave.Through the analysis of the fault signal,it is concluded that the component of the test signals were relatively simple,the fault symptom is usually the change of threshold or the fluctuation of a certain range.Therefore,the average value and the standard deviation were selected as the characteristic parameters,and the characteristic parameters which have good correlation with the fault modes were selected.Then,for the benefit of the powerful learning ability and nonlinear mapping ability,BP neural network was used to learn the data samples.Therefore,the knowledge was stored in the form of numerical value.Thus the complex reasoning process was replaced.In view of the difficulty in determining the initial structure and initial parameters of BP network,genetic algorithm was used to select the initial weights and thresholds of the network.The experimental results show that the neural network optimized by genetic algorithm improves the diagnostic accuracy of the frequency stabilization circuit.The dither control circuit is the detection and feedback control of the dither device.The signal is a sinusoidal wave with random amplitude variation.The signal components are complex.Moreover,under some fault modes,the fault signal features are weak to extract and the signal is unstable.Wavelet packet transform is an advanced signal processing tool,which can simultaneously decompose the high frequency and low frequency parts of the signal,and have good time-frequency resolution in both high and low frequency parts.Therefore,wavelet packet transform can retain the original information of the signal to the maximum extent.Wavelet packet decomposition combined with the signal frequency band energy was used to extract fault feature vectors.The results show that the distribution of signal energy in each frequency band of different fault modes is different,and the energy spectrum is an effective fault feature vector.Then,the RBF neural network was used to study the samples.It is concluded that the result of fault diagnosis based on different test signals is obviously different.The accuracy of fault diagnosis can be improved through the combination of test signals.