Feedback control for polymer extrusion processes based on virtual die melt temperature profile predictions

目前用于聚合物挤出过程监测的硬件传感器存在许多挑战，如缺乏耐用性、干扰熔体流动、测量延迟等。熔体热均匀性是聚合物挤出过程中一个关键的质量参数，它决定着最终产品的质量。然而，现有的硬件传感器由于其局限性，无法实时测量熔体流动的温度分布，这阻碍了实时质量控制策略的实施。软传感器或基于软件的传感器是一种很有前途的替代方案，可用于取代硬件传感器，以实现实时过程监控。然而，文献中报道的现有软传感器存在各种局限性。基于物理的软测量模型无法捕捉挤压过程的真实动态。此外，它们不能适应不同的工艺条件，因此当工艺条件改变时，它们的性能会下降。这些模型只经过了模拟测试，还没有在真实的工业挤出机上进行过测试。由于缺乏实时过程监控技术，现有的控制机制只是试图将工艺参数保持在预先定义的公差范围内，而不是保持所需的质量水平，这导致了产品质量差。本研究涉及开发一种软传感器来实时预测聚合物挤出过程中熔体的温度分布，这种软传感器应该能够适应不同的工艺条件，如不同的材料、机器和模具设计。此外，将开发一种反馈控制机制，该机制将软传感器估计的值作为输入来调整工艺参数（即螺杆速度和料筒设定温度），以保持最终产品质量在期望的范围内。首先，将开发一种软传感器，根据工艺参数（即螺杆速度和料筒设定温度）实时预测熔体温度分布。然后，模型将配备自适应能力，使模型能够适应不同的工艺条件，如不同的聚合物材料，机器和模具设计，而不会降低性能。在开发自适应软传感器时，将结合过程知识和基于人工智能的数据驱动技术。最后，开发的软传感器将集成到智能反馈控制机制中，以适当调整工艺参数（即螺杆速度和料筒设定温度），以最大限度地减少整个熔体流动中的熔体温度变化。这项研究涉及多个工程学科，包括聚合物加工、聚合物物理、统计学、人工智能、机器学习和控制系统工程。

The hardware sensors currently used for process monitoring in polymer extrusion processes present numerous challenges such as lack of durability, disturbance to the melt flow, measurement delays and so forth. The melt thermal homogeneity is a key quality parameter in polymer extrusion which defines the final product quality. However, the existing hardware sensors cannot measure the melt temperature profile across the melt flow in real-time due to their limitations and this inhibits the implementation of real-time quality control strategies. Soft sensors or software-based sensors are a promising alternative which can be used to replace hardware sensors, to enable real-time process monitoring. However, the existing soft sensors reported in the literature suffer from various limitations. The physics based soft sensor models fail to capture the real process dynamics in the extrusion processes. Moreover, they are not able to adapt to varying process conditions and hence their performance deteriorate when the process conditions change. These models have only been tested using simulation and they have not been tested on real industrial extruders. The existing control mechanisms only attempt to maintain the process parameters within a pre-defined tolerance rather than maintaining the desired quality level due to the absence of real-time process monitoring techniques, and this leads to products with poor quality.This research involves the development of a soft sensor to predict the die melt temperature profile across the melt flow of polymer extrusion processes in real-time, and this soft sensor should be able to adapt to varying process conditions such as different materials, machines and die designs. Moreover, a feedback control mechanism will be developed, which will take the values estimated by the soft sensor as inputs to adjust the process parameters (i.e., screw speed and barrel set temperatures) in order to maintain the final product quality within the desired range. Firstly, a soft sensor will be developed to predict the melt temperature profile in real-time, based on process parameters (i.e., screw speed and barrel set temperatures). Then, the model will be equipped with adaptive capabilities so that the model can adapt to varying process conditions such as different polymeric materials, machines and die designs without deteriorating the performance. Both the process knowledge and artificial intelligence based data driven techniques will be incorporated when developing the adaptive soft sensor. Finally, the developed soft sensor will be incorporated in an intelligent feedback control mechanism to appropriately adjust the process parameters (i.e., screw speed and barrel set temperatures), in order to minimize the melt temperature variations across the melt flow. This research involves multiple engineering disciplines including polymer processing, polymer physics, statistics, artificial intelligence, machine learning, and control systems engineering.