GOALI: Turnkey Model Predictive Control: automated design, model identification, tuning, and monitoring

本研究的目标是开发一个集成的框架，用于设计、模型识别、调整和监测基于工业模型的控制系统。新技术将被开发出来，从过程测量中识别模型，使从业者能够调整控制系统以适应感兴趣的应用。所提出的方法被称为交钥匙，因为过程识别实验和随后的控制系统整定参数计算将是自动化的。交钥匙方法的另一个好处是消除了当前控制系统供应商产品中引入的巨大可变性，这些可变性需要用户体验和仿真研究来选择这些调谐参数。所提出的控制技术可以很容易地监测到过程中干扰的变化，并提出干预策略。这种控制系统的自动监控在目前使用的工业方法中是不存在的，这为改善许多工业部门的业务绩效提供了重要的机会。在与该项目的工业合作伙伴伊士曼化学公司的合作下，拟议的方法将在一个全面的商业工业化学过程中进行演示。该方法还将在廉价的大学控制实验室实验中进行演示，以便本科生可以接触到最先进的控制系统。基于模型预测的在线重复优化的直观概念，作为设计自动反馈控制系统的一种手段，现在已经在化学过程工业（例如，模型预测控制- MPC）以及许多其他工业部门（包括机器人运动控制，飞行自动驾驶系统和陆地车辆制导控制）中应用的最先进的控制技术中占据了主导地位。由于控制器部署的困难和耗时因素是获得模型，因此所提出的研究的智力价值在于提高了识别线性执行器-传感器模型以及MPC所需的积分干扰模型的技术水平。这种模型的组合在所有工业应用中都是必需的，但是没有集成的理论可用于这项任务。提供一个交钥匙系统，直接从数据转换为所有调优参数的值，并在具有挑战性的工业过程中展示其性能，这将增强潜在的基本控制理论，并将该技术转移到复杂的工业制造设施中。虽然针对传统和新型化学过程控制应用，但本研究中开发的建模、设计和监测方法足够普遍，可以应用于任何具有生产目标和材料、工作流程和库存限制的制造设施中出现的自动化制造问题。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The goal of this research is to develop an integrated framework for design, model identification, tuning, and monitoring of industrial model-based control systems. New techniques will be developed to identify models from process measurements that enable practitioners to tune the control system to the application of interest. The proposed methodology is termed turnkey because process identification experiments and subsequent control system tuning parameter calculations will be automated. An added benefit of the turnkey approach is to remove the large variability introduced in current control system vendor products that require user experience and simulation studies to select these tuning parameters. The proposed control technology can be readily monitored to detect changes in the disturbances to the process and suggest intervention strategies. This automated monitoring of the control system is absent in industrial approaches in use today, providing a significant opportunity for improved business performance across many industrial sectors. In collaboration with the project’s industrial partner, Eastman Chemical, the proposed approach will be demonstrated on a full-scale, commercial, industrial chemical process. The approach also will be demonstrated in inexpensive university control laboratory experiments so that undergraduate students can be exposed to state-of-the-art control systems.The intuitive notion of online, repeated optimization of a model-based forecast as a means to design an automatic feedback control system has now taken hold in most advanced control technologies applied in the chemical process industries (e.g., model predictive control - MPC) as well as many other industrial sectors that include robotic motion control, flight autopilot systems, and land vehicle guidance control. Since the difficult and time-consuming element of controller deployment is obtaining models, the intellectual merit of the proposed research is to advance the state of the art in identifying linear actuator-to-sensor models plus the integrating disturbance models required for MPC. This combination of models is required in essentially all industrial applications, but no integrated theory is available for this task. Providing a turnkey system to move directly from data to values for all tuning parameters and demonstrating its performance on a challenging industrial process will enhance both the underlying fundamental control theory as well as the transfer of this technology to complex industrial manufacturing facilities. Although targeted to both traditional and new classes of chemical process control applications, the modeling, design, and monitoring methods developed in this research are sufficiently general to be applied to automated manufacturing problems arising in any manufacturing facility having production targets and constraints on materials, workflows, and inventories.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.