Development of “AI-VISION” for fluidized-bed incinerator

The paper presents “AI-VISION” which is an intelligent combustion control equipment for a fluidized-bed incinerator (FBI). Since thermal plants are MIMO systems, it is important to grasp the characteristic of the plant by sensors for precise combustion control. Furthermore, in a refuse incineration plant (RIP), the fuel property is unstable and minimization of exhaust emission is required. Thus, optimization from an overall standpoint is required in a RIP with consideration of sensor and control technology. Particularly, in a FBI, the combustion time is short in comparison with other incinerator types. The CO generation can suddenly increase when the refuse property and quantity changes. In this paper, we realize a low CO concentration combustion at an operating FBI, and report the development of “AI-VISION” which consists of a combustion image processing unit, neural networks which discriminate the combustion state by combustion images, and an online learning method which timely selects an optimized neural network. The combustion control system can use the “AI-VISION” output to operate the amount of manipulated value in a real plant