Statistical Aspects in Neural Network for the Purpose of Prognostics

Dawn An, Nam-Ho Kim, and Joo-Ho Choi
Submission Type: 
Full Paper
AttachmentSizeTimestamp
phmce_14_070.pdf464.44 KBMay 30, 2014 - 5:10am

Neural network (NN) is a representative data-driven method, which is one of prognostics approaches that is to predict future damage/degradation and the remaining useful life of in-service systems based on the damage data measured at previous usage conditions. Even though NN has a wide range of applications, there are a relatively small number of literature on prognostics compared to the usage in other fields such as diagnostics and pattern recognition. Especially, it is difficult to find studies on statistical aspects of NN for the purpose of prognostics. Therefore, this paper presents the aspects of statistical characteristics of NN that are presumable in practical usages, which arise from measurement data, weight parameters related to the neural network model, and loading conditions. The Bayesian framework and Johnson distribution are employed to handle uncertainties, and crack growth problem is addressed as an example.

Publication Year: 
2014
Publication Volume: 
5
Publication Control Number: 
070
Page Count: 
7
Submission Keywords: 
Data-driven prognostics
neural network
Statistical uncertainties
Bayesian framework
Johnson distribution
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