To precisely detect abnormality of machining even if characteristics of a driving and mechanism part vary by identifying a driving and mechanism part model by a successive method of least squares when an in-specific-operation signal is inputted from a specific operation decision part and outputting the result as a model correction signal.

When tool replacing operation is started according to a previously specified machining program, a specific operation decision part 19 outputs the in-specific-operation signal 20. A model correction part 21 when inputting the in-specific-operation signal 20 from the specific operation decision part 19 identifies the driving and mechanism part model 12 by a successive method of least squares by using all or some of a command signal 4, an output signal 2a, and an output signal 5 and outputs the result as a model correction signal 22. The driving and mechanism part model 12 is corrected with the model correction pa signal 22 and replaced with a better model. Consequently, an abnormality detection part 18 can detect machining abnormality with precision.