During the normal operation of the oil-immersed iron core reactor

its vibration is caused by the winding stressing

the magnetostrictive effect of the iron core and the Maxwell force between the iron cores. The axial and radial vibration response distributions of the reactor under three loads are analyzed by constructing a three-dimensional model of the magnetic-solid coupling of the reactor. On this basis

a reactor vibration test platform is set up to obtain the vibration signal of the reactor housing. The vibration signal of the housing is collected by adjusting the axial compression force of the iron core. It is known by extracting the vibration spectrum characteristic under various mechanical states that with the reduction of the axial compression force of the reactor core

the vibration signal spectrum of the housing is changed. Three kinds of Bayesian optimized network classifier models are adopted to identify the mechanical state under different compression forces

and the accuracy rate has reached 94.5%

94.2% and 95.0% respectively

and the high efficiency of the spectrum feature in combination with the optimization algorithm for the looseness judgement of the reactor core is verified further.