In this paper the iterative learning control scheme is considered from the standpoint of minimizing the special case of quadratic cost criterion for linear systems. First, an iterative learning control scheme based on the gradient method is proposed. It can be applied to any plant if the full information of the adjoint system is available. Second, an iterative learning control scheme for unknown plants is proposed. Since the convergent condition of the scheme is equivalent to the input-output condition of the second block of cascade-connected adjoint systems, the applicable range of the scheme is determined by the amount of the available information of system models. Consequently, it is proved that the problem of designing iterative learning control schemes is equivalent to the problem of decomposing adjoint systems.
In this paper, a discrete-time preview-repetitive control method using an ARMAX model is presented and experimental results are reported. At each sampling time, a control input is modified so as to minimize the cost function on predicted future tracking errors for a prediction horizon. The sufficient conditions for the stability of the control system are derived. The preview-repetitive control system includes a phase lead compensator and the amount of phase lead is proportional to the length of the prediction horizon. This method was applied to a position control of a DC servomotor. A simple model was used for predictions. However, the stability conditions were satisfied by extending the prediction horizon. Experimental results showed that errors were reduced very quickly and did not increase after the convergence. In the 1000th period, a one-pulse error occurred one time only through the period. A stable high-precision tracking property was achieved.
In recent years, there have been a lot of efforts in solving scheduling problems by using the techniques of artificial intelligence. But, through development of a variety of AI-based systems, it became well-known that eliciting effective problem-solving knowledge from human experts is an arduous job, and human schedulers typically lack the knowledge of solving a large and complicated scheduling problem in the sophisticated manner. In this paper our case-based approach, implemented in the system called CABINS, is presented for capturing human expert's preferencial criteria about schedule quality and search control knowledge to speed up problem solving. By iterative schedule repair, CABINS improves the quality of sub-optimal schedules, and during the process CABINS utilizes past scheduling experiences for (1) repair tactic selection and (2) repair result evaluation. In the paper, it was empirically demonstrated that CABINS could improve the efficiency of repair process while preserving the quality of a resultant schedule.
A dynamics model suitable for applying the system control theory to pulverizers has been developed. In this report, the dynamics model is discussed through the following order : (1) describing pulverizers, in which particle size distributions have the non-Gaussian properties, with the state space form as stochastic systems, (2) reducing the order of the state space by the proposed parameterizing method using the moments of the particle size distributions, (3) deriving the modeling algorithms for the proposed method, and (4) validations of the model through analysis on a commercial plant and dynamics response data of a pilot plant.
This paper proposes a method for measuring the position and heading of a vehicle moving on a curved course. The method uses laser fan beam transmitters and photo detectors mounted on the vehicle and an array of corner cubes on the course side as retro-reflecting targets. The laser beams have different direction angles and offset distances with each other and, therefore, return to the vehicle at different time according to the position and heading of the vehicle. The on-board computer calculates the position and heading from the time data that indicate the instants of detecting the reflections from the target corner cubes. The measured data are used for correcting the dead reckoning system.