ASSGCN Model Based Vibration Signal Reconstruction in Motor Vibration Testing

For modern electric powertrain applications
(wind, electric vehicles/ships/aircrafts,…), the vibration
analysis of the electric motor is one of the most important
tasks. Normally, a large number of vibration sensors are
placed evenly around the stator of the prototype to sample the acceleration and vibration signals. To decrease the
vibration testing cost and time, in this article, an attentionbased spatial-spectral graph convolutional network (ASSGCN) model is proposed to reduce the number of sensors
to reconstruct the vibration signal of the motor. Three spectral features of the vibration signal are modeled separately,
and the correlation of the operating condition force, acceleration and vibro-impedance matrices are investigated and
analyzed in the spatial dimension. Via dynamic correlation
analysis of spatial configuration and spectral response,
the proposed ASSGCN model predicts vibration signals at
different sensor sampling points. A 21 kw integrated permanent magnet synchronous motor testing rig with Brüel
and Kjær’s vibration sensing equipment is employed to test
the proposed ASSGCN model and the proposed method
successfully reconstructs the vibration source signal and
achieves well performance.