Study On The Extraction Of Engine's Abnormal Working Sound Signals Based On Chaotic Theory

Using sound information to diagnose fault is the method that had continuing to use in the past. But it did not get fast development because it had not combine with the modern signal processing technology very well, and it just stayed the subjected judgement or the contact complex trial judgement by human ears. If we get the engine's signal by the non-touching sensors, extract the abnormal signals of measurable engine's signals using reasonable methods, and then we can get the fault reasons of engine using the system identification. So the studying of extraction of engine's abnormal signals is very important to invent fault diagnosis of engine by non-touch.In fieldwork noise detection of engine's acoustic signals, because of complication of the real acoustic environment we often face with complex spatial acoustic field that covers large area, or is unsteady or has multi-sound source. Especially when the engine to be inspected is out of order, the engine's abnormal signal, the steady normal working signal and background noise mix together. The signal detected in this condition is actually a mixture of multi sources and their reflections. When certain engine is out of order, the sound that represents characteristic of the fault may be canceled or covered totally by signals from other sources, accuracy and reliability of fault diagnosis are degraded drastically. This greatly limits the usage of audio fault diagnosis in real world. It is obvious that how to acquire real acoustic signal of the engines to be inspected is key to improve feasibility of audio sound diagnosis. Because of lacking in effective method to separate signals, nowadays there is no much progress in the study and application of engine's noise fault diagnosis. So there needs an algorithm urgently which can separate the fault sound-source signal to improve the realization of audio engine's fault diagnosis.Most of engine abnormal signals are due to the part abrasion, crack,flexible of touching surface or parts of engine, so the engine's abnormal signal is the acoustic signals by strike. Because these striking course finish shortly, engine's fault signals are stochastic vibrating attenuating impulse signals, which often reflect significant engine's fault information. Moreover the normal acoustic signals of engine are multi-source, complexity, and have lots of noises, we just extract the high signal-to-noise using the methods of traditional signal process. We find that the engine's normal acoustic signals contain chaotic signal through study, so this thesis bring forward the extracting methods of engine's abnormal signals using chaotic background. The normal engine's acoustic signals are produced by mutual-effect of multi-complex, regular, assured parts of engine, so the normal engine's acoustic signals represent like-stochastic, inner regular affirmative, which accord with the affirmation, like-stochasticity, like-noise of chaotic signal. So, the signals we get contain normal engine's acoustic signal (chaotic signal), noise signals, engine's abnormal signals. Because the normal engine's acoustic signal and noise signals are background signal, which decide the signal-to-noise of extraction of engine's abnormal signals. The methods of this thesis are that, using nonlinear forecast model to forecast and extract engine's normal acoustic chaotic signal, and using accustomed methods to remove noises from surplus signals, finally we can get the pure abnormal signals, which reach the aim of extraction of engine's abnormal signals.The thesis analyzes the similarities and differences of engine's acoustic signal, and find the frequency of engine's abnormal signals are entirety submerged with engine's normal signal, which decides that it is incapable to extract small abnormal signals with general methods of frequency spectrum. We study on the chaotic characteristic of engine's normal acoustic signal using the methods of PCA and Lyapunov. We simulate the principal component spectrum graph of engine's acoustic signal and ideal white noise, and find that the principal component spectrum graph of engine's acoustic signal is approximately straight line with negative slope, but the principal component spectrum graph of ideal white noise is a straight line with zero slope. So weknow that it is essential difference between engine's normal signal and noise, although the engine's normal signal shows no sequence in time series, and we just say that it is like-noise, in fact is chaotic. Moreover we figure out the most Lyapunov exponent of engine's acoustic signal, which is 0.0464. This reflect the chaotic characteristic of engine's acoustic signal.In this thesis, all the algorithms are based on Takens Theorem. In the theorem, there are interior relationships among variables during the evolution of the dynamic system, and arbitrary element is determined by reciprocity between itself and others. Meanwhile, each one also concludes the system evolution information. Thus, in order to reconfigure an equivalent state space, only an element's change should be considered. As usual, delayed fixed point data are picked out to form a one-dimensional scale with delay method, which means single variable is mapped into multi-dimension as a vector, hence a phase space is reconfigured by single variables. However, vectors in the reconfiguration still possess the characters of original phase space.This thesis study the non-linear Volterra progression, and the adaptive methods of Least Mean Square(LMS) and Normalized Least Mean Square(NLMS) about the time-varying weighted coefficient of Volterra model, and found the Volterra adaptive filter model using the NLMS rule. The simulation of ideal chaotic signal produced by Lorenz equation validate the forecast and fitting ability of chaotic signal of the model, furthermore we finished the simulation of engine's acoustic signal measured. In ideal state, we can extract the abnormal signals with -18.1374db signal-to-noise, but in practice, we just extract the abnormal signals with -8.8280db signal-to-noise. Although the effect of this model is not very well, the feasibility of extraction of measured engine's acoustic signal using arithmetic is good.This thesis's emphases is the idea of denoise through dimension reduction using local tangent space alignment(LTSA) of manifold learning, and found the adaptive local tangent space alignment model basing on Volterra model, using the Volterra adaptive filter model and singular value decomposition(SVD). In simulated trials, firstly we validate the model's chaotic signal forecast, fittingability using Lorenz ideal chaotic signal, secondly we practise the extracted trial of measured engine's signal. The effect is good about the extraction of abnormal signal for Lorenz chaotic background signal, and we can get the intact abnormal signals from the -79.0526db signal-to-noise mix-signal. Although the effect is not so good as ideal state about the extraction of abnormal signal for measured engine's acoustic background signal, we can get the intact abnormal signals from the -36.7868db signal-to-noise mix-signal, which contain many measured noises or background noises. But we can wipe off these noises using general adaptive denoising method, so get the more pure abnormal signals, which establish the foundation of abnormal signal's identification for latter project.