| In process control system, collection of reliable and accurate data is vital. However, due to measurement, instrumental, computational, and even human errors, the measured process data always mix with noise ineluctably. De-noising with the purpose of extracting desired information from measured data has proven to be a crucial technique in process data analysis.Due to the requirement of safe and efficient manufacture in industrial process, the quantity of measured process data has exploded with the development of computer and sensor technology. Process data is a kind of information resource and can be used in tasks of process control. Magnanimous process data should been stored. Compression with the purpose of reducing storage memory and preserving the feature of process data is an item of significant research.Wavelets are a very interesting class of functions because of their special properties. The orthonormal bases can be constructed by translation and dilation of a mother wavelet. Wavelets have local property in time domain and frequency domain, so we can extract information from signals using wavelets. Wavelet analysis theory emerges as a new powerful mathematical tool in signal de-noise and data compression of process data analysis.This paper applies wavelet theory to process data analysis, mostly focusing on signal de-noise and data compression.The main contributions of this thesis are as follows:1)As discussion and experiment of de-noise method for removing white, noise from pulp thickness signal, a method was developed by utilizing the different characters of evolution of the wavelet transform maximum across scale of efficient signal and noise. The information of singularity points can be reserved well and the de-noised signal is a good estimation of the original signal.2)Proposed a de-noise method based on wavelet packet coefficient shrinkage.The resolution of the method is finer than that of the method based on wavelet shrinkage in time domain and frequency domain. The de-noise results of different threshold and different threshold function were compared for pulp thickness signal. Appropriate threshold and threshold function should be selected to fit the signal process goal when the method is applied.3)Developed a de-noise method based on translation invariance wavelet transform. The method performed the cycle-spinning for the signal to be de-noised. And then, the soft/hard threshold was used to shrink the wavelet coefficient of the signal and reconstruct the signal. The method can suppress pseudo-Gibbs phenomena on the singularity points of signal produced by de-noise algorithm based on wavelet shrinkage.4)A sparse matrix based on multi-resolution analysis for signal decomposition was introduced. A compression method for arbitrarily long data on chemical process was developed then. The relative error of decompression data is still acceptably small enough and the compression ratio is large enough when the method was applied to the compression decompression manipulation of historical data on chemical process.5)A set partitioning in hierarchical trees encoder for one dimension data was developed. A novel compression method was proposed by combining the encoder with integer wavelet transform. The simulation results showed that features of the novel method were low computation complexity, fast compression rate, high compression ratio and small reconstruction difference when the method was applied to the compression decompression process of temperature signals from a paper mill.6)Probed into the feasibility to apply wavelet networks to signal processing in process control system. A compression algorithm based on wavelet networks was proposed and the features of algorithm were fast parameter approximation rate and high compression ratio.7)Designed a process system of temperature and humidity signal. De-noise and compression algorithm can be tested on the platform.Finally, we point out a few existing problems to be solved after summarizing the works done in the...
|
PDF Full Text Request