Pattern recognition-based supervision of indirect adaptation for better disturbance handling

An advanced pattern recognition-based supervision algorithm for an indirect adaptive controller is proposed. The aim is to improve performance under certain conditions that are common in the industrial environment, in which indirect adaptive controllers with simple supervision are known to perform poorly or unreliably. Specifically, the problem of large invasive unmeasured disturbances of short or longer duration is addressed. The supervisor is designed to recognize such events as quickly as possible by analysis of recent control signals, without additional measurements. It applies appropriate strategies to prevent model degradation by learning from misleading data and to maintain acceptable performance under unfavorable conditions. As an illustration, it has been applied to the control of a model of a semi-cleanroom HVAC installation subsystem.