Applying Extreme Value Theory for alarm and warning levels setting under variable operating conditions

Alarm configuration is one of the main challenges of power generation and associated industries. The configuration challenge is compounded by machines being operated under variable conditions as a change in operational condition i.e. speed or torque affects the vibration response. Thus, if the data used to determine the alarm and warning threshold levels characterises only limited range of operational conditions a false alarm may be triggered indicating onset of a fault while only the operational regimes have changed. Another possibility is the fault to be masked by change in the operational condition which leads to misdetection. Central to determining the alarm and warning threshold levels is establishing the type of the data distribution. The distributions are usually assumed to be Gaussian even though a number of possible distributions should be considered in the search of the best fit. Incorrect distribution fit may result in sub-optimal alarm configuration. In the present paper instead of considering the whole data set only maxima will be taken into account as likely to reveal an outlier. The Generalised Extreme Value distribution is proposed as a possible limit distribution for the maxima. In order to take into account the effect of the variable speed, Extreme Value Theory for non-stationary processes will be applied. The suggested approach is validated on data from an experimental gearbox.

[1]  S. Roberts.  Novelty detection using extreme value statistics , 1999 .

[2]  Robert B. Randall,et al.  Vibration-based Condition Monitoring: Industrial, Aerospace and Automotive Applications , 2011 .

[3]  B. Gnedenko.  Sur La Distribution Limite Du Terme Maximum D'Une Serie Aleatoire , 1943 .

[4]  Branislav Titurus,et al.  Gearbox condition monitoring under variable operational conditions , 2014 .

[5]  Hoon Sohn,et al.  VIBRATION-BASED DAMAGE DETECTION USING STATISTICAL PROCESS CONTROL , 2001 .

[6]  L. Tarassenko,et al.  Bayesian Extreme Value Statistics for Novelty Detection in Gas-Turbine Engines , 2008, 2008 IEEE Aerospace Conference.

[7]  David A. Clifton,et al.  Novelty Detection with Multivariate Extreme Value Statistics , 2011, J. Signal Process. Syst..

[8]  Yi-Qing Ni,et al.  Structural damage alarming using auto-associative neural network technique : exploration of environment-tolerant capacity and setup of alarming threshold , 2011 .

[9]  Charles R. Farrar,et al.  Structural Health Monitoring: A Machine Learning Perspective , 2012 .

[10]  Eric P. Smith,et al.  An Introduction to Statistical Modeling of Extreme Values , 2002, Technometrics.

[11]  E. Bechhoefer,et al.  Setting HUMS condition indicator thresholds by modeling aircraft and torque band variance , 2004, 2004 IEEE Aerospace Conference Proceedings (IEEE Cat. No.04TH8720).

[12]  R. Fisher,et al.  Limiting forms of the frequency distribution of the largest or smallest member of a sample , 1928, Mathematical Proceedings of the Cambridge Philosophical Society.

[13]  Hoon Sohn,et al.  Damage detection in mechanical structures using extreme value statistics , 2002, SPIE Smart Structures and Materials + Nondestructive Evaluation and Health Monitoring.

[14]  E. Bechhoefer,et al.  A Generalized Process for Optimal Threshold Setting in HUMS , 2007, 2007 IEEE Aerospace Conference.

[15]  David A. Clifton,et al.  An Extreme Function Theory for Novelty Detection , 2013, IEEE Journal of Selected Topics in Signal Processing.

[16]  C. Cempel,et al.  Limit value in the practice of machine vibration diagnostics , 1990 .

[17]  Tomasz Barszcz,et al.  Modeling of probability distribution functions for automatic threshold calculation in condition monitoring systems , 2013 .