In rapid hot-embossing of microarray products, sensors accuracy drifts, mechanical wears and environmental changes produce the nonlinear relationship between micro-forming accuracy and process parameters. Generally, the process parameters need to be adjusted according to ex-situ detection and on-spot experiences, leading to inefficiency. Therefore, an in-process optic detection of micro-forming heights is proposed to closed-loop control the micro-forming accuracy on macro hot-embossed surface. On the base of ex-situ detection data, the in-process detected data are related to micro-forming heights to adjust hot-embossing parameters by intelligent algorithms. The objective is to resolve the uncertainty during precision micro-forming. First, an optic detection was developed to recognize the micro-forming heights on macroscopic workpiece surface in real-time; then artificial neural networks and Naïve Bayes method were adopted to select the initial process parameters; next, the correction algorithm was modeled to perform fine adjustment instead of on-spot experiences, based on the recognized forming heights; finally, this system was applied to the hot-embossing of microprism arrays on light-guide plates. It is shown that the illuminance ratio is related to the hot-embossed microstructure heights. This may be used to in-process detect the micro-forming heights on macro workpiece surface. For the neural networks trained with process parameters, the RBF eliminates nonlinearity-caused local minimization better than the BP. For ambiguous process data, the Naïve Bayes method updates incomplete process parameter database more precisely and timely than neural networks. As a result, the micro-forming height may be controlled within the allowable error band under unstable hot-embossing situations.

在微阵列产品的快速热压印过程中，传感器精度漂移、机械磨损和环境变化会产生微成型精度与工艺参数之间的非线性关系。一般需要根据异地检测和现场经验调整工艺参数，效率低下。因此，提出了一种微成形高度的过程中光学检测，以闭环控制宏观热压印表面的微成形精度。在异地检测数据的基础上，将在制品检测数据与微成型高度相关联，通过智能算法调整热压参数。目标是解决精密微成形过程中的不确定性。第一的，开发了一种光学检测来实时识别宏观工件表面的微成形高度；然后采用人工神经网络和朴素贝叶斯方法选择初始工艺参数；接下来，基于识别的成型高度，对校正算法进行建模以进行微调，而不是现场经验；最后，将该系统应用于光导板上微棱镜阵列的热压印。结果表明，照度比与热压印显微组织的高度有关。这可用于在过程中检测宏观工件表面上的微成形高度。对于使用过程参数训练的神经网络，RBF 比 BP 更好地消除了非线性引起的局部最小化。对于模糊的过程数据，朴素贝叶斯方法比神经网络更准确和及时地更新不完整的过程参数数据库。因此，在不稳定的热压印情况下，微成形高度可以控制在允许的误差范围内。