During normal operation of high-voltage shunt reactor

the vibration of its housing causes the solid structure to be subjected to alternating mechanical stress. The mechanical strength of structural materials is gradually reduced under long-term stress

thus resulting in serious defects in the reactor. In this paper

combined with the multi-physics finite element simulation model and the reactor vibration experiment

the stress distribution of the three-dimensional model of the reactor during operation and the parts prone to mechanical fatigue are obtained firstly. On this basis

the vibration characteristics of the housing wall under different mechanical states are compared by simulating the mechanical defects of the corresponding parts of a 500 kV iron core reactor and the Markov transfer field and convolutional neural network are used to identify different mechanical states. Each signal channel has achieved a high recognition accuracy

which provides a new method for the diagnosis of mechanical defects of reactor.