Intelligent Fault Diagnosis of Reciprocating Compressor Using Deep Learning Methods
- Publication Type:
- Thesis
- Issue Date:
- 2020
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Fault diagnosis of reciprocating compressors (RCs) based on vibration signals plays a vital role in guaranteeing a high operating reliability in RCs. Conventional maintenance schemes, which are carried out on a regular basis, can lead to unnecessary maintenance and shutdowns. Online health monitoring can monitor the working conditions of RCs continuously and provide more specific information, thus allowing the RC to be maintained as needed.
This PhD research focuses on the development of effective fault diagnosis methods using deep learning methods, thereby greatly advancing traditional health condition monitoring methods. Most traditional data-driven methods analyze the operating conditions using shallow models, which are incompetent at obtaining more confident results. To overcome this problem, a novel scheme based on deep learning models is proposed and applied to RC fault diagnosis. Traditional fault diagnosis methods select and extract features of raw signal with expertise and fuse them with shallow models. However, these methods cannot analyze the characteristics of signal in depth and thus degrade the performance of health monitoring. Deep learning methods are introduced in this research to calculate more representative features self-adaptive from the RC signals to improve fault diagnosis performance. As most fault diagnosis methods are based on vibration signals being the single information source, they cannot reflect the RC operating condition comprehensively. In this research, multi-source signals are collected and analysed for fault diagnosis. A scheme fusing multi-source information is proposed, as well as an auto-denoising network for denoising RC signals self-adaptively.
This PhD thesis consists of seven chapters. Chapter 1 provides research background. Chapter 2 presents a literature review. Chapter 3 proposes a method using intrinsic vibration feature fusion and a Grassmann manifold-based similarity. Chapter 4 introduces the method of RC fault diagnosis using mode isolation-convolutional deep belief networks. Chapter 5 presents the intelligent fault diagnosis method using an optimized convolutional deep belief network. Chapter 6 proposes a novel ensemble empirical mode decomposition-convolutional deep belief network for RC fault diagnosis, and chapter 7 presents the conclusion and discusses future research in this area.
Please use this identifier to cite or link to this item: