Cold Plat-shaped Signal Analysis And Pattern Recognition Based On Intelligence Methods

Flatness pattern recognition is an important part of flatness control system. The keytechnology includes the choice of flatness basic model, the choice of the number offlatness parameters and pattern recognition algorithms. With the funds from national"Eleventh Five-Year" Technology Support Program and the National Natural ScienceFoundation project, using Artificial neural network that based on non-contact Sheet ShapeMeasurement, author research on flatness pattern recognition and control program,achieving some research findings.The flatness signal from non-contact Sheet Shape Measurement is not directly signal,it requeses high-lever signal processing technique, and then deals with the noiseelimination using wavelet analysis.In terms of selection of flatness basic models,considering a variety of methods of flatness control of modern mills, a new flatnesspattern recognition method including the cubic flatness is first applied in this paper, whichcan improve accuracy of flatness pattern recognition. The model processes uses the linear,quadratic, cubic, and biquadratic Legendre polynomials as basic patterns. In terms ofpattern recognition algorithms, author chooses self-adaptive chaotic PSO algorithm. Thisalgorithm is used to optimize the structure, the weights and the threshold ofback-propagation(BP) network, which can improve the accuracy and convergence rate ofthe neural networks. Simulation results show that the anti-interference ability andself-learning ability of the ACPSO-BP Network is better. It provides a reliable basis for thedevelopment of flatness control strategy.For example of the HC mills, developed the feedback control strategy, which is basedon the measured result of ACPSO-BP network. The simulation platform is built based onC++ software. Users can set parameters through visual parametric interface, then thissystem can identify the defect shape automatically and get the I value of the defect shapeand the feedback control strategy clearly. The results show that the system has highaccuracy and can be used for online loop feedback control of flatness after analysis ofexperimental.