Self-organizing feature map is often limited by long training time

low diagnosis accuracy and bad classification visualization when it is applied in gear fault diagnosis.A novel method of semi-supervised LDA-GNSOM（Linear Discriminative Analysis Grouping Neuron Self-organized Map） is proposed for fault diagnosis.Firstly

the fault of original feature space is reduced by using LDA

and then the reduced feature subsets are input into the semi-supervised GNSOM network for classification and visualization.The effectiveness of this method is verified by using Iris dataset.And the gear incipient fault experiments are also conducted to demonstrate that the new approach is effective for gear incipient pitting detection.