The Research On Noise Cancellation For Metal Debris Detection System

The number, size and other indicators of metal debris in lubricating oil system are commonly used in the evaluation of the extent and rate of engine wear. Metal debris detection system is an on-line metal debris testing device installed in the lubricant system. When the metal debris goes through the sensor, the system will output a similar single-cycle sine wave signal, which contains metal debris's size and flow rate information. Because of the noise, we can't accurately obtain this information. The signal generated by the small metal debris is even drowned in the noise. Therefore, it is necessary for the output signal of the system to denoise.The noise contained in the system output signal is considered as Gaussian white noise, so in this paper, a variety of white noise elimination and suppression algorithms are deeply studied, and two improved methods are proposed. The following work has been done in this paper:First, briefly introduce the composition and principle of the system which is used to detect the metal debris in the lubricant and give a description of the physical meaning of the actual output signal and noise. Then simulate the output signal (including the useful signal and noise signal) of the actual system and introduce the simulation parameters. At last this paper details the experimental program results and standards for the assessing algorithm.Second, three methods for eliminating the white noise are introduced in this paper, including:singular value decomposition (SVD), empirical mode decomposition (EMD) and wavelet thresholding. After compare the each denoising algorithm'performance with different parameters, the optimal parameters are figured out. With the conditions of optimal parameters, the three methods' denoising effect is compared in different SNR.Third, two improved algorithms are proposed for eliminating the noise in this signal:" white noise elimination method based on static wavelet decomposition and kurtosis " and " white noise elimination method based on independent component analysis and prefix signal." The first method adopts a static wavelet decomposition, whose advantage is the ability to retain the statistical properties of the wavelet coefficients in each layer, so we can determine which layers contained the useful signal most by the kurtosis values of the wavelet coefficients. According to the results, wavelet coefficients are processed separately. The denoising effect has been significantly improved over the previous three methods. The second method needs two-channel output signal and considers the two-channel noisy signals as three independent source signals. Linear mixing ICA method combined with the prefix signal is used to remove noise. This method not only solves the underdetermined problem, but also corrects the results amplitude and phase in ICA separation. The error between denoised signal and the useful signal is within 0.1%.Finally, a summary of the research and work outline of the next step are introduced.