Predictive Model Evaluation for PHM

Chunsheng Yang, Yanni Zou, Jie Liu, and Kyle R Mulligan
Publication Target: 
Publication Issue: 
Submission Type: 
Full Paper
ijphm_14_019.pdf476.82 KBJanuary 28, 2015 - 2:39pm

In the past decades, machine learning techniques or algorithms, particularly, classifiers have been widely applied to various real-world applications such as PHM. In developing high-performance classifiers, or machine learning-based models, i.e. predictive model for PHM, the predictive model evaluation remains a challenge. Generic methods such as accuracy may not fully meet the needs of models evaluation for prognostic applications. This paper addresses this issue from the point of view of PHM systems. Generic methods are first reviewed while outlining their limitations or deficiencies with respect to PHM. Then, two approaches developed for evaluating predictive models are presented with emphasis on specificities and requirements of PHM. A case of real prognostic application is studied to demonstrate the usefulness of two proposed methods for predictive model evaluation. We argue that predictive models for PHM must be evaluated not only using generic methods, but also domain-oriented approaches in order to deploy the models in real-world applications.

Publication Year: 
Publication Volume: 
Publication Control Number: 
Page Count: 
Submission Keywords: 
Machine Learning Algorithms
Generic Methods
binary classifier
prognostics and health management (PHM)
Time to Failure
Predictive Models Evaluation
Submission Topic Areas: 
Data-driven methods for fault detection, diagnosis, and prognosis
Submitted by: 

follow us

PHM Society on Facebook Follow PHM Society on Twitter PHM Society on LinkedIn PHM Society RSS News Feed