[Paper Review] Deep Learning and Its Applications to Machine Health Monitoring: A Survey

Rui Zhao, Ruqiang Yan+4 more|arXiv (Cornell University)|Dec 16, 2016

Anomaly Detection Techniques and ApplicationsComputer Science64 references149 citations

TL;DR

This survey reviews how deep learning models (AE, RBM/DBN/DBM, CNN, RNN) are applied to machine health monitoring, highlighting end-to-end learning and data-driven advantages over traditional methods.

ABSTRACT

Since 2006, deep learning (DL) has become a rapidly growing research direction, redefining state-of-the-art performances in a wide range of areas such as object recognition, image segmentation, speech recognition and machine translation. In modern manufacturing systems, data-driven machine health monitoring is gaining in popularity due to the widespread deployment of low-cost sensors and their connection to the Internet. Meanwhile, deep learning provides useful tools for processing and analyzing these big machinery data. The main purpose of this paper is to review and summarize the emerging research work of deep learning on machine health monitoring. After the brief introduction of deep learning techniques, the applications of deep learning in machine health monitoring systems are reviewed mainly from the following aspects: Auto-encoder (AE) and its variants, Restricted Boltzmann Machines and its variants including Deep Belief Network (DBN) and Deep Boltzmann Machines (DBM), Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). Finally, some new trends of DL-based machine health monitoring methods are discussed.

Motivation & Objective

• Motivate the shift from physics-based and hand-crafted feature methods to data-driven, DL-based MHMS in modern manufacturing systems.
• Summarize DL architectures used for MHMS and how they learn hierarchical representations from raw data.
• Compare conventional data-driven MHMS with DL-based MHMS and discuss end-to-end training advantages.
• Provide guidance on applying DL models to fault diagnosis and prognosis in MHMS.

Proposed method

• Describe DL architectures applicable to MHMS, including Auto-encoders and their variants (SAE, DAE, SDA) and pre-training benefits.
• Explain RBM, DBN, and DBM as multi-layer probabilistic models and their role in representation learning.
• Outline CNNs for 1D/2D data in MHMS and detail convolution, pooling, and feature extraction mechanisms.
• Describe RNNs and variants (LSTM, GRU) for sequential data and their memory capabilities.
• Discuss end-to-end DL-based MHMS pipelines where all layers are trained jointly without hand-crafted features.
• Highlight how DL models can switch from diagnosis to prognosis by adapting the top layer.

Experimental results

Research questions

• RQ1 What DL architectures are effective for extracting hierarchical representations from machinery data for health monitoring?
• RQ2 How do DL-based MHMS compare to conventional data-driven approaches in handling large-scale, noisy sensor data?
• RQ3 Can end-to-end DL systems reduce the need for hand-crafted feature design in fault diagnosis and prognosis?
• RQ4 What are the roles of pre-training and variants (AE, RBM/DBN/DBM, CNN, RNN) in improving MHMS performance?

Key findings

• DL-based MHMS enable end-to-end learning of representations from raw data, reducing dependence on hand-crafted features.
• Auto-encoders and their variants can learn meaningful representations and can be used for pre-training deep networks in MHMS.
• RBM-based models (DBN, DBM) and their variants have been applied to fault diagnosis and remaining useful life prediction in MHMS.
• CNNs and RNNs are effectively applied to time-series and time-frequency data for condition recognition and RUL estimation.
• DL-based MHMS demonstrate improved handling of large-scale data and offer flexibility across machines and domains.