PNNL

Abstract

In order to decrease the energy consumption and to improve the utilization rate of raw materials, the fiber length distribution (FLD) is generally employed as one of the important production indices in the refining process. Considering that the conventional mean and variance of fiber length are unable to perfectly characterize the distribution properties that displays non-Gaussian distributional feature, this paper proposes a novel geometric analysis based double closed-loop iterative learning control (ILC) method for output probability density function (PDF) shaping of FLD. Primarily, the radial basis functions (RBFs) neural network (NN) with Gaussian-type is utilized to approximate the square root of output PDF in the inner loop, and RBFs parameters can be tuned between each batch by using ILC method in order to improve the closed-loop performance batch-by-batch. Secondly, on the basis of the optimal RBFs, the weights are extracted by employing the output PDF in one batch and the state-space model between the input variables and the weights vector is established by utilizing the subspace identification method. Then, for the sake of accelerating the convergence rate of the output PDF and improving the robustness of closed-loop system, an improved ILC strategy based on geometric analysis is adopted to obtain the optimal control input so as to quickly achieve the desired output PDF tracking control of the actual output PDF. Finally, both simulation and experiments demonstrate the effectiveness and practicability of the proposed approach.