摘要
针对传统智能识别需要复杂的特征提取过程,增加了操作的难度和不确定性,采用深度信念网络(Deep Belief Network,DBN)直接从原始数据对故障智能识别的方法。该方法避免了人工特征提取过程,增强了识别的智能性。将以原始数据为输入的DBN应用于轴承故障识别,接近100%正确识别率的实验结果表明:DBN可以直接通过原始数据对轴承故障进行高效识别。
Feature extraction is a crucial step affecting the performance of traditional intelligent identification with difficulty and uncertainty for manual operation.Starting directly from the raw data, the paper proposes a novel method for bearing fault identification based on deep belief network.The method can get the particular features and avoid manual operation for feature extraction, enhancing intelligence of diagnostic process.When DBN is applied to bearing fault identification with raw data, the experimental results with nearly 100% correct recognition rateshows that DBN can achieveefficiently fault identificationdirectly by the raw data.
出处
《电子设计工程》
2016年第4期67-71,共5页
Electronic Design Engineering
关键词
特征提取
受限玻尔兹曼机
DBN
深度学习
故障识别
feature extraction
restricted boltzmann machine
DBN
deep learning
fault identification
