基于进化算法优化GAN的轴承故障诊断

Bearing Fault Diagnosis Based on Generative Adversarial Nets Optimized by Evolutionary Conditions

针对滚动轴承故障诊断故障样本类别不平衡的问题,提出一种基于进化算法优化的条件生成对抗网络(evolutionary conditional generative adversarial nets,简称ECGAN)故障诊断方法。首先,利用进化算法优化条件生成对抗网络中的生成器,使其在不同的损失函数下生成与原始样本分布相似的新样本,扩充数据集;其次,将生成的样本和原始样本输入判别器,提取出样本中有效的数据特征,判断输入样本的真假和类别;最后,通过对抗学习机制优化生成器和判别器,提高网络的故障识别能力。实验结果表明,在轴承故障样本数据类别不平衡的情况下,ECGAN模型具有较好的故障诊断性能。

Aimed at the problem of unbalanced fault samples in rolling bearing fault diagnosis,a fault diagnosis method based on conditional generative adversarial nets optimized by evolutionary algorithm(ECGAN)is proposed.Firstly,the generator in conditional generative adversarial nets is optimized by evolutionary algorithm to generate new samples similar to the original sample distribution under different loss functions so as to expand the data set.Then,the generated sample and the original sample are input to the discriminator to extract the valid data features of the sample and judge the authenticity and category of the input sample.Finally,the generator and discriminator are optimized by counter learning mechanism to improve the fault recognition ability of the nets.The experimental results show that the ECGAN method has better performance in fault diagnosis in the case of unbalanced sample data types.