Model selection for degradation modeling and prognosis with health monitoring data

This paper aims to address the problem of degradation model selection including goals, procedure and evaluation criteria

Khanh T.P. Nguyen; Mitra Fouladirad; Antoine Grall

2017

Scholarcy highlights

  • Degradation modeling in the presence of health monitoring data is extremely important for lifetime prognosis and maintenance planning
  • The selection criteria are classified into two groups: classical statistical criteria that are based on the discrepancy between observation degradation data and the values expected under the considered model and prognostic criteria that are based on the relevance between failure time and its expected distribution under the considered model
  • The advantages and disadvantages of these criteria are considered through numerous numerical examples for model selection between solutions of stochastic differential equations and Gamma processes
  • As the model complexity is more strongly penalized by Minimum Description Length criterion with the hypothesis of universal priors, MDL criterion frequently favors a simple model
  • The degradation process generated by another model can be close to real data during a short period of time
  • It may be counterproductive to increase the observation time for a model selection based on the prognostic accuracy criterion criterion
  • The same degradation data are considered for classical model selection and prognostic criteria

Need more features? Save interactive summary cards to your Scholarcy Library.