Neural-Network-Based Adaptive Model Predictive Control for a Flexure-Based Roll-to-Roll Contact Printing System
Neural-Network-Based Adaptive Model Predictive Control for a Flexure-Based Roll-to-Roll Contact Printing System
复制标题
DOI：
10.1109/tmech.2022.3172949
复制
发表时间：
2022-12
期刊：
影响因子：
--
通讯作者：
Jingyang Yan;Xian Du
中科院分区：
文献类型：
--
作者：
Jingyang Yan;Xian Du
高精度的接触力控制对于通过使用邮票在柔性卷筒纸基材上进行机械接触的连续卷筒接触印刷是必不可少的。印章接触力控制不均匀会导致印花失败，尤其是大面积印花。由于挠曲机构在定位和力控制方面的高精度，已被应用于辊对辊接触印刷系统中，然而，传统的基于物理模型的控制系统无法处理基于挠曲的辊对辊接触印刷系统中存在的非线性效应。为了实现精确的接触力控制，提出了一种基于神经网络的柔性辊对辊接触印刷系统的自适应模型预测控制。利用人工神经网络对柔性机构的非线性进行学习和建模。为了消除模型失配和外部扰动引起的稳态误差，通过更新神经网络模型输出层的偏差，设计了一种在线自适应机制。实验结果表明，在印辊两端有天平的情况下，接触力的均方根误差可以控制在0.075-0.5 N范围内，优于比例-积分-导数控制器、基于神经网络的标准模型预测控制和基于神经网络的鲁棒预测控制。所提出的控制算法已在45-μm宽的金图案的微接触印刷过程中实现，并在88.9 mm宽的柔性基板上实现了不同位置的平均金线宽度的0.3μm的变化。均匀微尺度印刷结果表明，所提出的神经网络自适应模型预测控制在实际印刷过程中是有效的。
High-precision contact force control is essential for continuous roll-to-roll contact printing via mechanical contact on flexible web substrates using stamps. Nonuniformly controlled stamp contact force will cause failures during printing, especially for large-area printing processes. Due to their high precision in positioning and force control, flexure mechanisms have been applied in roll-to-roll contact printing systems; however, conventional physical model-based control systems cannot manage the nonlinear effects that exist in flexure-based roll-to-roll contact printing systems. To achieve precise contact force control, we propose a neural-network-based adaptive model predictive control for a flexure-based roll-to-roll contact printing system. The nonlinearity of the flexure mechanism is learned and modeled by an artificial neural network. To eliminate the steady-state error caused by model mismatches and external disturbances, an online adaptive mechanism is designed via updating the biases of the output layer of the neural network model. Experimental results show that the root-mean-square error of the contact force can be controlled in the range of 0–0.075 N with balances on two ends of the print roller, outperforming a proportional–integral–derivative controller, a neural-network-based standard model predictive control (MPC) controller, and a neural-network-based robust MPC controller. The proposed control algorithm is implemented in a microcontact printing process that prints 45-μm width gold patterns and achieves a variation of 0.3 μm in the average gold line width at different locations on an 88.9-mm width flexible substrate. The uniform microscale printing results have shown the effectiveness of the proposed neural-network-based adaptive model predictive control in the applied printing process.