Defect cause estimation device, defect cause estimation method, learning method for defect cause estimation model, operation condition determination method, and method of manufacturing steel product

A defect cause estimation method comprising:

• a step of calculating a depth position at a slab stage of a surface defect in a product by a defect occurrence depth calculation unit included in a computer;
• a step of calculating a defect occurrence rate for each slab depth based on the depth position by a defect occurrence rate calculation unit included in the computer;
• a step of calculating an inclusion adhesion ratio for each distance from a solidification start position during slab casting based on the defect occurrence rate for each slab depth by an inclusion adhesion ratio calculation unit included in the computer;
• and
• a step of estimating a defect cause by selecting a part of manufacturing conditions using an estimation model learned in advance with manufacturing conditions for the product set as input data and with the inclusion adhesion ratio set as output data and by estimating an inclusion adhesion ratio in a case in which the selected manufacturing conditions are changed stepwise by a defect cause estimation unit included in the computer.

A learning method for a defect cause estimation model, comprising:

• a step of calculating a depth position at a slab stage of a surface defect in a product by a defect occurrence depth calculation unit included in a computer;
• a step of calculating a defect occurrence rate for each slab depth based on the depth position by a defect occurrence rate calculation unit included in the computer;
• a step of calculating an inclusion adhesion ratio for each distance from a solidification start position during slab casting based on the defect occurrence rate for each slab depth by an inclusion adhesion ratio calculation unit included in the computer;
• and
• a step of learning an estimation model for estimating the inclusion adhesion ratio with manufacturing conditions for the product set as input data and with the inclusion adhesion ratio set as output data by a model learning unit included in the computer.