A learned model generation method involves: outputting, when it is determined that the sequence of the positions of a first end surface of a first sheet material indicated by first data intersects with the sequence of the positions of a second end surface of a second sheet material indicated by second data, information indicating that a roll is defective and that the defect is caused by a first feed reel or a first affixing roller; and generating a facility state diagnosis model by using the first data and the second data read before preservation of the first feed reel when it is determined that a first difference between a first defect rate of the roll before the preservation of the first feed reel and a second defect rate of the roll after the preservation of the first feed reel is equal to or larger than a prescribed value.