基于认知计算的机电产品故障诊断技术研究—— 一种深度学习与迁移学习的方法

Human cognitive process has recently attracted increasing attention. Based on its principle, brain cognition can be divided into such two stages as autonomous cognition by extract abstract features layer by layer, transfer knowledge cognition for reinforced autonomously cognition among similar things, which are introduced into the field of mechanical and electronic products fault diagnosis. In this study, first, deep learning and transfer learning theory, the leading edge of the cognitive computing, are utilized for fault feature self-learning from the raw data and reinforced autonomous cognition of the fault modes for fault data deficient equipment. Thus, a new solution for the problem of the uncertainty caused by artificial fault feature extraction and the high costs, as well as the difficulty for building accurate diagnosis model caused by fault data deficient, is proposed. The related theoretical method is verified based on the data collecting from two typical electrical products of rotating machinery component and closed loop feedback control system.

通过借鉴人类认知过程的原理，将脑认知的两个阶段：逐层抽象特征提取的自主认知、迁移应用相似事物认知的强化自主学习，引入到机电产品故障诊断研究领域中；采用认知计算中最前沿的深度学习与迁移学习理论，依次研究基于原始底层数据的故障特征自主认知，以及样本数据缺失设备的故障模式强化自主认知方法。进而，为解决人工故障特征提取的不确定性和高成本，以及故障数据缺失导致精确的诊断模型难以构建等问题提供一种新的解决思路。本项目拟结合旋转机械部件、闭环反馈控制系统两类典型机电产品，对所研究的相关理论方法进行验证。

通过借鉴人类认知过程的原理，将脑认知的两个阶段：逐层抽象特征提取的自主认知、迁移应用相似事物认知的强化自主学习，引入到机电产品故障诊断研究领域中；采用认知计算中最前沿的深度学习与迁移学习理论，依次研究基于原始底层数据的故障特征自主认知，以及样本数据缺失设备的故障模式强化自主认知方法。进而，为解决人工故障特征提取的不确定性和高成本，以及故障数据缺失导致精确的诊断模型难以构建等问题提供一种新的解决思路。本项目结合旋转机械部件、闭环反馈控制系统两类典型机电产品，对所研究的相关理论方法进行验证。