基于深度学习的集成化装备故障诊断方法综述

Overview of Integrated Equipment Fault Diagnosis Methods Based on Deep Learning

摘要:

集成化装备的结构和功能日益复杂，传统的故障诊断方法难以进行准确的特征提取，而深度学习具有强大的学习能力，能够有效挖掘特征，适合于集成化装备的故障诊断。为此，首先按照应用领域的不同，分别描述了国内外基于深度学习的故障诊断最新研究进展；其次，简要介绍了三种典型的深度学习故障诊断方法（深度置信网络、堆栈自编码机和卷积神经网络），整理出已经取得的成果和存在的问题并做出总结；而后将基于深度学习实现故障诊断面临的挑战总结为六种类型，并根据前文总结出的研究成果提出了研究思路；最后结合实际应用情况，提出了四种发展方向。

Abstract:

The structure and function of the integrated equipment is increasingly complex, traditional fault diagnosis methods are difficult to extract accurate features. Deep learning has strong learning ability, which can effectively mine features and is suitable for fault diagnosis of integrated equipment. For this purpose, firstly, the latest research progress of fault diagnosis based on deep learning at home and abroad is described according to the different application fields; secondly, three typical fault diagnosis methods based on deep learning (deep belief networks, stacked auto-encoders and convolutional neural networks) are briefly described, the achievements and challenges are sorted out and summarizes are made; then, the challenges of fault diagnosis based on deep learning are summarized into six types, and the research ideas are proposed according to the research results summarized above; finally, combined with the actual application, four development directions are proposed.