带有偏差神经元的内回归神经网络在旋转机械故障诊断中的应用

Application of the Internally Recurrent Net Withneuron Units to Fault Diagnosis of Rotary Machines

针对传统的神经网络在机械故障诊断方面的不足,利用偏差神经元改进了基于BP神经网络算法的内回归神经网络(IRN)算法,加快收敛速度,提高运算质量,并将其应用于旋转机械的振动故障诊断与识别。实例结果表明,该算法学习收敛较快,误差曲线平稳,对复合故障的识别性能好。

Due to the disadvantage of the traditional network to fault diagnosis of machines, the Internally Recurrent Net (IRN) algorithms based on BP algorithms are improved by bias neuro units, which quicken operation speed and improve operation quality. The algorithms are applied successfully to fault diagnosis of rotary machines. The simulation results show that these methods have rapider convergence, smoother error curve, and better learning and diagnostic properties to complex faults.