基于深度流形表示学习的工业过程多故障识别方法

Industrial Process Multiple Faults Identification Method Based on Deep Manifold Representation Learning

深度流形表示学习对于自动学习系统的本质特征有着重要的作用。论文提出了一种基于深度流形表示学习的多故障识别方法。所提的多故障识别方法可以分为三个阶段:第一,将故障识别问题转化为分类问题,定义正常和故障状态,以及预处理原始数据;第二,利用深度流形表示学习对深度神经网络进行预训练;第三,利用故障标签数据全局训练深度网络。所提出的方法被应用于由一种典型的工业系统生成的两个不同尺寸以及多个故障类型的数据集。测试结果表明,所提方法能够准确预测故障类型,优于其他两种分类方法。此外,由于所提出的方法仅需要数据,因此很容易迁移到其他的工业系统。

Deep manifold representation learning plays an important role in learning essential characteristics of systems.This paper proposes a multiple faults identification method based on deep manifold representation learning.The proposed multiple faults identification method can be divided into three phases.First,this paper transformes the fault identification problem into a classification problem,defines normal and fault states,and preprocesses the original data.Second,deep manifold representation is used to pre-train the deep neural network.Third,the deep network is trained globally using fault tag data.The proposed method is applied to data sets of two different sizes and multiple fault types generated by a typical industrial system.The test results show that the proposed method can accurately predict the fault type and is superior to the other two classification methods.In addition,since the proposed method requires only data,it is easy to migrate to other industrial systems.