Research On Test Signal Analysis Based On The Manifold Learning For Printing Press

Signal noise reduction and feature extraction are important parts in the mechanical fault diagnosis and the machine running state monitoring. High speed printer is a complex mechanical system, so the measured signals from which have the nonlinear, strong coupling characteristic information and noise interference characteristics. The manifold learning based on the theory of differential geometry and topology manifolds has the advantages in the process of disposing the nonlinear data, and has a wide range of applications in the image processing, information retrieval and pattern recognition, which is also the hot issues in the test signal noise reduction and feature extraction in recent years. The test signal noise reduction and feature extraction method of the high speed printing machine are studied in the paper. This paper’s main works are as following:1. In terms of test signal noise reduction, the characteristics of the empirical mode decomposition(EMD) method and the wavelet threshold method are combined, the EMD wavelet threshold de-noising method is studied. The EMD wavelet threshold de-noising method combines the advantages of EMD’s strong adaptability and wavelet threshold de-noising method’s simple and good noise reduction effect, which can effectively reduce the nonlinear signal noise. The EMD wavelet threshold de-noising method has been verified as a good method in nonlinear signal processing by simulation signal de-noising analysis.2. In terms of test signal feature extraction, the manifold learning method studied the method based on time domain index. Combining the characteristics of the empirical mode decomposition(EMD), sample entropy and the locally linear embedding(LLE) algorithm, the based on EMD sample entropy Hessian LLE algorithm was proposed. The EMD sample entropy is an effective nonlinear feature extraction methods, through the typical fault rotor experiment, verified the EMD sample entropy is better than the effect of time domain index in nonlinear signal feature extraction. Comparing LLE method and Hessian LLE method, concluding that the Hessian LLE method based on the EMD sample entropy has a good recognition effect than LLE on single fault identification.3. In order to improve the manifold learning method in the application of printing machine fault diagnosis, developed the manifold learning platform for the test signal analysis system, the system has the characteristics of a fusion of multiple signal processing method, simple operation and clear steps.4. The manifold learning method was initially applied to the printer test signal analysis. For the issue of the second overlay deviation large group of the high speed color printer, made the impression cylinder bearing vibration test of all color group and found a large acceleration in the second color group impression cylinder bearing. Using the EMD wavelet threshold de-noising method and the Hessian LLE method based on EMD sample entropy, successively making test signal noise reduction processing and feature extraction. Compared with the typical fault characteristics of the rotor, considering the main reason of causing a large acceleration in the second color group impression cylinder is bearing looseness.Through this paper’ research, manifold learning method was introduced into the printer test signal noise reduction and feature extraction, and verified the validity and practicability of this method by the experiment. Due to the structural complexity of the printing press, and a large number of destructive testing can’t be made, it remains to be further research in the printing machine fault features and pattern recognition.