Predicting the Remaining Useful Lifetime of a Proton Exchange Membrane Fuel Cell using an Echo State Network
Year: 2014
Type: conference-paper
Abstract: One remaining technological bottleneck to develop industrial Fuel Cell (FC) applications resides in the system limited useful lifetime. Consequently, it's important to develop failure diagnostic and prognostic tools enabling the optimization of FC. The Prognostic and Heath Management (PHM) is a discipline involved in the process of industrial maintenance. The objective, in PHM, is to estimate the Remaining Useful Life (RUL) of a system by predicting its future behavior. The RUL enables to predict the moment when a fault could occur on a system. It also allows identifying the relevant part of the system where a fault could happen. Then, a preventive maintenance could be performed to avoid non-reversible degradations. Three main prognosis approaches can be distinguished: model-based, data-based and hybrid methods. Data-based methods such as Artificial Neural Network (ANN), aim to estimate the ageing behavior of the process without specific knowledges related to the physical system phenomenon. Nevertheless, the deployment of such an approach can be a tedious work, mainly due to the trial and error algorithm method, which represents a real problem for industrial applications where real-time complying algorithms must be developed. Among the various methods of this area, the tool chosen here is called Echo State Network (ESN). An ESN consists in the use of a dynamical neurons reservoir where the training step consists in performing a linear regression. The computation time of this algorithm is thus shorter while keeping the same modeling capability of a Recurrent Neural Network (RNN). Created in 2001 by H. Jaeger, an ESN proposes a better human brain paradigm than traditional ANN, and are based on a reservoir of neurons randomly connected to each other. The aim of this paper is to study the application of ESN as a prognostics system enabling the estimation of the Remaining Useful Life of a Proton Exchange Membrane Fuel Cell using an iterative predictive structure, which is the most common approach performing a one-step prediction. This estimation output value is used in the next step as one of the input regressor and these operations can be repeated until the desired prediction horizon. The results obtained thanks to this method exhibits good prediction and they will be detailed in this paper.
Source: HAL
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3160760839