Reliability Estimation Based on Condition Monitoring Data

• Author / Creator

  Liu, Yang

• Reliability estimation based on condition monitoring data contains two important parts: thresholding and probability density estimation. Thresholding is to determine a critical level of an indicator corresponding to the transition of system states. Probability density estimation is to estimate the probability density function (PDF) of the indicator value over a certain time period. Existing reliability estimation methods usually require prior knowledge such as design criteria and past experiences or event data such as time-to-failure data to determine a threshold for estimating reliability; while the information as such might be costly and impractical to acquire for expensive or highly reliable systems. Therefore, there is a demand on the reliability estimation methods which can determine thresholds and estimate probability density relying on solely condition monitoring data. However, very few studies are reported in this respect. A method falling into this type is recently reported. The reported method jointly uses one-class support vector machine (OC-SVM) solution path for thresholding and kernel density estimation (KDE) for probability density estimation to estimate system reliability based on only condition monitoring data. This thesis studies in-depth the reported method and finds that there are four aspects that are unclear and deficient. These four aspects are thus investigated and the suggestions are provided to address the concerns. The findings of this thesis are listed as follows: (1) The impact of the width parameter of OC-SVM on reliability estimates is investigated and an applicable range for width parameter selection is given to acquire reasonable reliability estimates; (2) The impact of the bandwidth parameter of KDE on reliability estimates is investigated and an applicable range for bandwidth selection is given to acquire reasonable reliability estimates; (3) The impact of the sliding window size of KDE on reliability estimates is investigated. A strategy of variable window size is developed to enable stationary data to be used for probability density estimation. The results show that the proposed strategy can provide not only the reliability estimates comparable to the fixed sliding window size but also the reliability estimates corresponding to the transition of system states; (4) The impact of outliers is investigated and a strategy of removing outliers is developed for KDE to ensure reasonable reliability estimates can be obtained.

• Subjects / Keywords

  • Reliability estimation
  • One-class support vector machine
  • Condition monitoring
  • Kernel density estimation

• Graduation date

  Fall 2016

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/R3CJ87R50

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.