Abstract: Inference is carried out through intermediate hypotheses representing physical states of a blast furnace using HG (Heuristic Grade) in order to comprehensively diagnose the state of the furnace to ascertain optimum actions. Specific parameters are monitored to immediately recognize a transition of conditions inside the furnace to additionally execute the inference. Various types of actions such as defensive actions and offensive actions are covered in this inference. A burden distribution estimation model considering collapse of a coke bed is used for calculating distribution inside the furnace to aid in deciding on an optimum action when an action to alter distribution in the furnace is required as the result of the inference. Creation and alteration of a knowledge base for the inference are carried out without interrupting the inference.