Mechanisms for Integrated Feature Normalization and Remaining Useful Life Estimation Using LSTMs Applied to Hard-Disks

Sanchita Basak, Saptarshi Sengupta, Abhishek Dubey

IEEE International Conference on Smart Computing, SMARTCOMP 2019, Washington, DC, USA 2019

Best Paper Award

PDF DOI

Why This Matters

Hard disk failures are critical concerns in computing infrastructure, and accurate remaining useful life prediction enables proactive maintenance planning. This work is innovative because it addresses device-specific variations in failure patterns through careful feature normalization and selection, enabling models trained on one device to generalize to other devices from the same manufacturer. The approach improves prediction accuracy through integration of multiple machine learning techniques.

What We Did

This paper develops data-driven methods for remaining useful life estimation of hard disk drives using LSTM networks with feature normalization techniques. The approach addresses the challenge that different devices have varying failure characteristics and minimum/maximum feature values. The work implements two LSTM network layers and employs careful feature selection through correlation analysis and Fisher scores.

Key Results

The LSTM-based approach achieved excellent prediction performance with an average precision of 0.8435 and F1 score of 0.72 when predicting whether a disk would fail in the next ten days. The system demonstrated the ability to predict RUL near the critical point of device approach with an average precision of 0.8435. The normalized features successfully enabled generalization across different hard disk models from the same manufacturer.

Full Abstract

Cite This Paper

@inproceedings{Basak2019a,
  author = {Basak, Sanchita and Sengupta, Saptarshi and Dubey, Abhishek},
  booktitle = {IEEE} International Conference on Smart Computing, {SMARTCOMP} 2019, Washington, DC, USA},
  title = {Mechanisms for Integrated Feature Normalization and Remaining Useful Life Estimation Using LSTMs Applied to Hard-Disks},
  year = {2019},
  month = {jun},
  acceptance = {29},
  note = {Best Paper Award},
  pages = {208--216},
  abstract = {In this paper we focus on application of data-driven methods for remaining useful life estimation in components where past failure data is not uniform across devices, i.e. there is a high variance in the minimum and maximum value of the key parameters. The system under study is the hard disks used in computing cluster. The data used for analysis is provided by Backblaze as discussed later. In the article, we discuss the architecture of of the long short term neural network used and describe the mechanisms to choose the various hyper-parameters. Further, we describe the challenges faced in extracting effective training sets from highly unorganized and class-imbalanced big data and establish methods for online predictions with extensive data pre-processing, feature extraction and validation through online simulation sets with unknown remaining useful lives of the hard disks. Our algorithm performs especially well in predicting RUL near the critical zone of a device approaching failure. With the proposed approach we are able to predict whether a disk is going to fail in next ten days with an average precision of 0.8435. We also show that the architecture trained on a particular model is generalizable and transferable as it can be used to predict RUL for devices in other models from same manufacturer.},
  bibsource = {dblp computer science bibliography, https://dblp.org},
  biburl = {https://dblp.org/rec/bib/conf/smartcomp/BasakSD19},
  category = {selectiveconference},
  contribution = {lead},
  doi = {10.1109/SMARTCOMP.2019.00055},
  file = {:Basak2019a-Mechanisms_for_Integrated_Feature_Normalization_and_Remaining_Useful_Life_Estimation_Using_LSTMs_Applied_to_Hard-Disks.pdf:PDF},
  keywords = {remaining useful life, hard disk drives, LSTM networks, predictive maintenance, feature normalization, data-driven methods},
  project = {cps-reliability},
  tag = {ai4cps},
  timestamp = {Wed, 16 Oct 2019 14:14:54 +0200},
  url = {https://doi.org/10.1109/SMARTCOMP.2019.00055},
  month_numeric = {6}
}
Quick Info
Year 2019
DOI 10.1109/SMARTCOMP.2019.00055
Keywords
remaining useful life hard disk drives LSTM networks predictive maintenance feature normalization data-driven methods
Research Areas
CPS ML for CPS
Search Tags

Mechanisms, Integrated, Feature, Normalization, Remaining, Useful, Life, Estimation, LSTMs, Applied, Hard, Disks, remaining useful life, hard disk drives, LSTM networks, predictive maintenance, feature normalization, data-driven methods, CPS, ML for CPS, 2019, Basak, Sengupta, Dubey