Developing neural-fuzzy adaptive controls with stability margins

*我的研究计划旨在开发结合神经网络和模糊逻辑的机器学习算法，以提高控制系统的性能并保证其稳定性。我目前正在努力寻找一种通用的方法来替代传统的PID反馈控制--工业和机器人自动化系统中普遍存在的回路控制(PID增加了比例、积分和误差的导数)。虽然ID提供了简单、强大的控制，为我们提供了一个世纪以来的良好服务，但它们不能满足未来许多重要系统的系统要求。在保证稳定性不会变差的同时，替代的PID应该在性能方面做得更好(更少的错误和努力)。我正在申请NSERC Discovery的资金，以便在这个问题上取得一些重大进展，特别是开发机器学习算法，将神经网络和模糊逻辑结合在一起，用于一些对加拿大有利的应用。*这个问题的一个有前途的研究方向涉及修改小脑模型关节控制器(CMAC)类型的模糊神经网络。CMAC能够在困难的情况下获得高性能，学习如何补偿非线性、不确定性和周期性干扰。我目前正在修改CMAC控制体系结构，以便在不增加不稳定风险的情况下，始终具有更好性能的潜力。*作为保持加拿大在空间机器人领域领先地位的努力的一部分，我的研究小组正在为国际空间站拟议的手术机器人进行控制系统设计。由于通信信号的理论往返时延高达1秒，实际上高达7秒，我们设想了一种安全的手术模式，即外科医生直接与地球上本地的3D打印宇航员模型互动，而太空中的远程机器人则在一小段时间后安全跟随。通过建议的NSERC发现拨款，我们将寻找低级别反馈控制回路的高级解决方案，以取代本地和远程系统的PID，这些系统将实现(字面上的)手术精度，同时仍满足满足航天机构要求的定量稳定性保证类型。设计中的主要困难是确保机器人的微小灵活性不会在微重力下引起过度振动；我在解决这个问题方面处于有利地位，因为我已经开发出使用CMAC控制柔性关节机器人的专门知识。*在以前的工作中，我还研究了在废水处理、发电焚烧炉和有机朗肯循环(将废热转化为电力)等工业工厂中用CMAC取代PID，并将努力用类似的应用来验证我的方法。因此，这项研究有可能造福于环境和碳减排努力。

***My research program aims to develop machine learning algorithms, incorporating neural networks and fuzzy logic, that will improve the performance of control systems and guarantee their stability. I am currently endeavouring to find a universal replacement for the traditional PID feedback-loop controls ubiquitous in industrial and robotic automation systems (PID adds a Proportion, Integral, and Derivative of error). Although PIDs provide simple, robust controls that have served us well for a century, they will not meet the system requirements for many important systems of the future. A PID replacement should do better in terms of performance (less error and effort) while being guaranteed to do no worse in terms of stability. I am requesting NSERC Discovery funding in order to make some significant progress on this problem, specifically developing machine learning algorithms that incorporate neural network and fuzzy logic for some applications that will be of benefit to Canada.******A promising research direction for this problem involves modifications to the Cerebellar Model Articulation Controller (CMAC) type of fuzzy neural network. The CMAC is capable of high performance in difficult situations, learning to compensate for nonlinearities, uncertainties, and periodic disturbances. I am currently working on modifying the CMAC control architecture so that it always has potential for much better performance without ever increasing the risk of instability.******As part of the effort to maintain Canada's leadership in space robotics, my research group is working on the control system designs for a proposed surgical robot for the International Space Station. Since communication signals experience a theoretical round-trip time delay of up to one second, and in practice up to 7 seconds, we envision a safe model of operation where the surgeon interacts directly with a local 3D-printed model of the astronaut on Earth while the remote robot in space follows along safely a small time later. With the proposed NSERC Discovery grant, we would be looking at advanced solutions for the low-level feedback control loops to replace PID for both local and remote systems that would achieve (literally) surgical precision while still meeting the type of quantitative stability guarantees that would satisfy space-agency requirements. The main difficulty in the design is ensuring the small flexibilities in the robot do not cause excessive vibration in microgravity; I am in a strong position to tackle this problem as I have already developed a particular expertise controlling flexible-joint robots using CMAC.******In previous work I also investigated replacing PID with CMAC in industrial plants like wastewater treatment, power-generating incinerators, and organic Rankine cycles (for turning waste heat into electricity) and will endeavour to validate my methods with similar applications. Thus, the research has the potential to benefit the environment and carbon-reduction efforts.