Industrial internet model based predictive control for applications using thermal imaging

本提案的总体目标侧重于开展旨在利用工业互联网平台开发智能预测控制方法的研究。这种方法涉及开发用于大容量数据分析的计算密集型算法，以实现连续的过程建模和协作机器学习。两名硕士研究生和两名博士生将进行研究。其中一些已经接受了培训，并将为长期研发和产业整合建立坚实的研究基地。对于许多先进的制造应用，制定一个准确的模型可能需要计算和实验时间。拟议的研究将使用工业互联网作为快速和并行算法计算的机制，这些算法计算将被导出，以便逐步制定这些复杂的学习模型，并继续使用所有可用的数据，以尽可能准确地预测过程动态。第一位博士研究员将使用热成像数据来制定这些先进的模型。结合预测分析和机器学习，该提案旨在重新制定先进的预测控制器，以提高计算效率，以便将它们集成到快速反应系统中，特别是先进制造中的机器人系统和高速运动控制。将考虑各种预测方案，包括重新拟订的最新发展。将研究将标准矩阵代数形式转换为离散空间，并与标准公式进行计算速度的比较。第一位MScE学生将负责这项工作。第二个MScE学生将开始设计各种系统的基于图像的控制，整合由第一个MScE衍生的快速控制例程。该策略是制定控制方案，可以提供最小的跟踪误差的复杂设定值轨迹在这样的系统，包括连续机器人焊接和快速运动制造过程。导出的算法将与第一个博士工作的初始基于互联网的监测例程集成。第二名博士生将把上述项目的关键要素整合到他/她的研究中。这项初始任务将包括通过工业互联网结合过程建模算法的元素，热图像处理例程和快速计算控制例程。由于研究涉及到热能和成像数据的使用，因此需要制定一种新的自适应和空间预测的各种热过程控制律。这项工作将在各种规模工业系统和大型工业系统上实施。这项研究将促进研究的长期愿景的成功。

The overall objective of this proposal focuses on conducting research aimed at developing intelligent predictive control methodologies using the platform of the industrial internet. This approach involves developing computationally intensive algorithms for high volume data analytics to enable continuous process modelling and cooperative machine learning. Two MScEs and 2 PhD students will conduct the research. Some are already being trained and will establish a solid research base for long-term R&D and industrial integration. For many advanced manufacturing applications, formulating an accurate model can be demanding on computations and time consuming in experimentation. The proposed research will use the industrial internet as a mechanism for fast and parallel algorithmic computations that are to be derived in order to progressively formulate these complex learning models, and continue to use all available data to the extent possible for accurate predictive process(es) dynamics. Thermal imaging data will be used in formulating these advanced models by the the first PhD researcher. In conjunction with predictive analytics and machine learning, the proposal aims at reformulating advanced predictive controllers for computational efficiency in order to integrate them on fast reacting systems, specifically robotic systems and high speed motion control in advanced manufacturing. Various predictive schemes will be considered including the most recent developments for reformulation. Research on transforming standard matrix algebraic forms to discrete space will be performed and applied, as well as comparing to the standard formulations for computational speed. The first MScE student will conduct this work. The second MScE student will commence work on designing image-based control of various systems, integrating the fast control routines derived by the first MScE. The strategy is to formulate control schemes that can provide minimum tracking errors of complex setpoint trajectories on such systems, including continuous robotic welding and fast motion manufacturing processes. The derived algorithms will be integrated with initial internet-based monitoring routines from the first PhD’s work. The second PhD student will conduct the integration of key elements of the above projects into her/his research. This initial task will include combining the elements of process modelling algorithms via the industrial internet, thermal image processing routines and the fast computational control routines. As the research involves the use of thermal energy and imaging data, a new adaptive and spatial predictive control law for various thermal processes has to be formulated. The work would be implemented on various scaled industrial systems as well as large-scaled industrial systems. This research will promote the success of the research’s long-term vision.