摘要
为解决小样本和噪声干扰下滚动轴承剩余寿命(RUL)预测准确率低的问题,提出一种基于信息最小二乘生成对抗网络(information least squares generative adversarial network,InfoLSGAN)和行动者-评论家(actor-critic,AC)算法的滚动轴承剩余寿命预测方法。将堆叠降噪自动编码器、信息生成对抗网络和最小二乘生成对抗网络相结合,构建InfoLSGAN,自动地从噪声数据中提取可解释的鲁棒特征,解决梯度消失问题;采用基于AC的训练算法训练InfoLSGAN,减少训练时间,加快收敛速度;根据训练后的InfoLSGAN,利用softmax分类器预测测试样本中滚动轴承的剩余寿命。通过滚动轴承加速疲劳寿命试验验证该方法的有效性。试验结果证明,当信噪比等于0时,该方法对滚动轴承测试样本的寿命预测准确率至少提高了10%。在小样本情况下,滚动轴承剩余寿命预测的平均准确率达95.84%。
In order to solve the problem of low remaining useful life(RUL) accuracy of rolling bearings under small samples and noise interference,a RUL prediction method of rolling bearings using information least squares generative adversarial network(InfoLSGAN) and actor-critic(AC) algorithm was proposed. Stacked denoising autoencoder,information generative adversarial network and least squares generative adversarial network were integrated to construct InfoLSGAN,which can automatically extract interpretable and robust features from noisy data,and solve the problem of vanishing gradients. The training algorithm based on AC was utilized to train the InfoLSGAN to reduce the training time and accelerate the convergence. According to the InfoLSGAN after training,a softmax classifier was used to predict the rolling bearing RUL in test samples. The effectiveness of the proposed method was validated through an accelerated fatigue life experiment of rolling bearings. The experimental results demonstrated that when the signal-to-noise ratio was equal to 0,the proposed method increased the RUL prediction accuracy of rolling bearing test samples by at least 10%. In the case of small samples,the average accuracy of the RUL prediction of rolling bearings was 95.84%.
作者
于广滨
卓识
于军
刘可
YU Guangbin;ZHUO Shi;YU Jun;LIU Ke(Key Laboratory of Advanced Manufacturing and Intelligent Technology,Ministry of Education,Harbin University of Science and Technology,Harbin 150080,China;School of Mechanical and Power Engineering,Harbin University of Science and Technology,Harbin 150080,China;School of Automation,Harbin University of Science and Technology,Harbin 150080,China)
出处
《航空动力学报》
EI
CAS
CSCD
北大核心
2020年第6期1212-1221,共10页
Journal of Aerospace Power
基金
国家重点基础研究发展计划(2019YFB2006400)
黑龙江省“百千万”工程科技重大专项(2019ZX03A02)
黑龙江省杰出青年基金(JC2014020)。
关键词
滚动轴承
剩余寿命预测
信息最小二乘
生成对抗网络
行动者-评论家算法
堆叠降噪自动编码器
rolling bearing
remaining useful life prediction
information least squares
generative adversarial network
actor-critic algorithm
stacked denoising autoencoder
