Recently, there are increasing needs for structural vibration control. Design of structural controller has, however, many problems such as modeling error, actuator saturation, handling of nonlinearity, and etc. Modeling error is one of the most important practical problems. It is a challenging task to accurately model any structure for controller design. If the model is not available, it is impossible to design any control system. Even when one has a mathematical model that is considered to be close to a real structure, some possible error in the model may cause significant deterioration in the performance of the control system. To solve this modeling problem, artificial intelligences appeared as promising tools. Neural network is one of the artificial intelligences and has been used for many engineering problems such as damage detection, optimization, and controller. This paper shows control algoriuun using neural network. Three main ideas for structural control using neural network are proposed in this study. The first one is to propose a sensitivity evaluation algorithm that replaces emulator neural network. Emulator neural network is used to evaluate the sensitivity of structural response to control signal in conventional methods. To use the emulator, it should be trained to predict the dynamic response of the structure. Much of time is usually spent on training of emulator. In proposed algorithm, however, it takes only one sampling time to obtain the sensitivity. Therefore, taining time for emulator is eliminated. The second idea is that a cost function is proposed for the training criterion of neural network. In conventional methods, error function is defined by the squared sum of the deviation of the actual response from the desired response. Then, the error function is used for training criterion. Therefure, the desired response should be found in some way. However, the cost function proposed in the proposed algorism requires only the actual responses. Therefure, the desired response is not needed any more. The third one is to apply CMAC (Cerebellar model articulation controller) to the structural control. CMAC is a kind of neural network that converges very quickly in training. The structure, computation, and training algorioun of CMAC is presented. The training time for CMAC is very short compared with MLP (Multilayer perceptron). Therefore, it can be said that a real time controller is possible with CMAC. In numerical example, three-story building structure under ground motion is controlled by trained neural. The actuator dynamics and control time delay are considered in simulation. Numerical examples show that the proposed control algoricn is valid in structural control.
