Fuzzy Logic Controller and Cerebellar Model Arithmetic Computer Neural Networks for Active Vibration Suppression

本文的主要目的是研究一种新型的人工神经网络体系结构，它协同结合了小脑模型算术计算机(CMAC)神经网络和模糊逻辑控制器(因此我们称之为模糊-CMAC神经网络)的优点，并尝试将这种新型的模糊-CMAC神经网络应用于主动振动抑制(AVS)的实时学习控制系统中。在典型的AVS系统中，被控对象通常是建模较差、复杂、高度非线性和不确定的机械系统。Robotronics公司认为，神经网络的研究进展，特别是本文提出的新型模糊-CMAC神经网络结构，可能为实现自学习非线性控制提供一种理想的方法，从而显著提高AVS系统的性能。作为两种独立的控制方案，CMAC神经网络和模糊逻辑控制器都为智能控制系统提供了有价值的设计。然而，到目前为止，还没有研究活动将这两种方法之间的联系联系起来，并将其应用于AVS。这项拟议的工作试图从建筑角度建立这些模型之间的关系。比较了两种模型的各种特点，提出了一种结合了两种模型各自特点的模糊CMAC模型。讨论了一种将模糊CMAC应用于振动主动抑制的方法。初步仿真结果表明，该控制器具有很好的学习和控制性能。在第一阶段的工作中，将对模糊-CMAC神经网络体系结构的理论问题进行研究。所提出的模糊-小脑模型主动减振方法将通过非线性仿真进行验证，并将使用实时多个DSP控制硬件进行实验。将对仿真和实时实验的结果进行评估，并与现有控制器的结果进行比较。

The primary objective of this proposed work is to study a novel artificial neural network architecture which synergistically combines the preferred features of the CMAC (cerebellar model arithmetic computer) neural networks and the Fuzzy Logic Controller (therefore we called it the Fuzzy-CMAC Neural Network), and attempt to utilize this novel Fuzzy-CMAC neural network in the implementation of the real-time learning control systems for active vibration suppression (AVS) applications. In typical AVS systems, the plants to be controlled are usually poorly-modeled, complex, highly nonlinear, and non-deterministic mechanical systems. Robotronics, Inc. believes that the advances in study of neural networks, especially the novel Fuzzy-CMAC neural network architecture proposed here, may offer an ideal approach to implement self-learning nonlinear control resulting in a significant performance improvement in AVS systems. As two separate control schemes, both CMAC neural networks and fuzzy logic controllers provide valuable designs for intelligent control systems. However, there has been no research activity reported so far which bridge the connection between these two approaches and apply it to the AVS. This proposed work attempts to establish the relationship between these models from an architectural viewpoint. Various features of both models are compared and a Fuzzy CMAC model which combines the desirable features of both models will be developed. An approach to apply the Fuzzy CMAC to active vibration suppression is discussed. Preliminary simulation results show very good learning and control performance. During Phase 1 work, theoretic issues of the Fuzzy-CMAC neural network architecture will be conducted. The proposed Fuzzy-CMAC active vibration suppression approach will be verified through nonlinear simulations and will be implemented using real-time multiple DSP control hardware for experiments. The results of simulations and real-time experiments to be performed will be evaluated and compared with those of existing controllers.