Optimal tuning of particle filtering random noise for monotonic degradation processes

Matteo Corbetta, Claudio Sbarufatti, and Marco Giglio
Submission Type: 
Full Paper
AttachmentSizeTimestamp
phmec_16_014.pdf5.56 MBJune 21, 2016 - 7:31am

Particle filtering is a model-based, Bayesian filtering algorithm widely employed in many scientific and engineering fields. It has been recently applied many times for the diagnosis and prognosis of engineering systems. Within the prognostics and health management scenario, particle filtering stood out as an effective and robust algorithm to predict the system's remaining useful life because of its capability in tracking nonlinear/non-Gaussian systems. One of the fundamental equations underneath particle filtering is the evolution equation describing the system's dynamics. This equation composes of a deterministic model and an artificially-added random process or random noise, thus making the evolution equation a stochastic equation. The selection of random noise is up to the algorithm's designer discretion and may vary on a case-by-case basis. Concentrating on the field of structural degradation processes, many studies on particle filtering-based prognostics have shown encouraging results. Though, they did not provide detailed discussions in supporting the appropriateness of the selected random noises altering the degradation model. An improper choice may cause the algorithm to be inefficient, moving the projected trajectories outside the state-space domain of the system, sometimes also introducing a bias in the stochastic evolution model. Therefore, this work examines the evolution equation with the aim of creating an optimal prognostic framework for monotonic degradation phenomena. The paper gives special emphasis to structural degradation caused by fatigue, which is a widespread monotonic damage progression process. Some of the existing works are reviewed, discussing the effectiveness of the random noises embedded in the algorithm. Eventually, the paper presents an unbiased, optimal random noise for monotonic damage progression, pointing out the strengths of the proposed solution against formulations suggested in literature. The presented particle filtering-based prognostic algorithm is applied to experimental crack growth observations on an aeronautical stiffened structure and the prediction performance is validated using dedicated prognostic metrics.

Publication Year: 
2016
Publication Volume: 
7
Publication Control Number: 
014
Page Count: 
11
Submission Keywords: 
particle filtering; fatigue prognosis; structural health monitoring; remaining useful life prediction
Submission Topic Areas: 
Model-based methods for fault detection, diagnostics, and prognosis
Modeling and simulation
Structural health monitoring
Submitted by: 
  
 
 
 

follow us

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