This project will investigate the use of CMAC neural networks for the control of vibrations. Vibration control problems can be large scale, non-linear and difficult to model mathematically. Traditional control methods depend on assumptions (models) about the system structure and generally involve extensive real-time calculations. In contrast, CMAC develops the control signal from an experience based of measured system data. One analogy is the sailor who develops `sea legs`. the sailor uses previous experiences to compensate for boat motion, not modeling or real-time calculation. The UNH Robotics Lab has reported many applications of CMAC including robot control, process control, pattern recognition and signal processing. The real-time operational speed and reduced memory requirements make CMAC an attractive alternative to there neural networks. Other neural networks do not offer the bandwidths required for typical vibration control problems. The CMAC approach offer the ability to work with multiple frequency disturbances. The investigation will be both theoretical and applied. The theoretical investigation will be to optimize the CMAC weight patterns improved dynamic range, and better functional approximation (and hence better vibration cancellation) characteristics than the traditional CMAC. the Theoretical investigation of CMAC network are expected to extend the range of applications of CMAC control to include approximate dynamic programming, the adaptive critic, and non-linear optimization problems. The application in this study will be the reduction of hull noise for submarines subject to internal vibration sources. A fully instrumented reduced scale submarine at the underwater test facility at the University of New Hampshire will be used for real-time testing of the CMAC control concepts. the submarine application are has many of the same characteristics as other vibration control problems such as noise reduction in automobile and aircraft cabin interiors and hence, this investigation will impact a variety of applications.
