PHM of fuel cell systems
Axis moderator : Rafael Gouriveau, MCF
Motivations and object
Fuel Cell Systems (FCS) appear nowadays to be a promising and alternative energy source to face economic and environmental challenges of modern society. However, even if this technology is close to being competitive, it is not yet ready to be considered for large scale industrial deployment: FCS still must be optimized, particularly by increasing their limited lifespan. This involves not only a better understanding but also requires emulating the behavior of the whole system. Additionally, a new area of science and technology emerges: prognostic of FCS is a field of scientific and industrial developments that should be increased. This is the aim of this axis in which we propose to develop intelligent Prognostics and Health Management (PHM) methods in order
- to assess the health state of a FCS (State of Health – SoH),
- to predict its remaining useful life (RUL),
- to decide from mitigation/control actions for mission achievement.
Problems to be adressed
Fuel-cell systems are complex multi-physical (electric, fluidic, electrochemical, thermal and mechanical) and multi-scale (time and space) systems. The modeling of their degradation is a very difficult task as a consequence of the nonlinear and complex nature of these systems, the non-reversibility of their reactions and the interactions between their multiple subsystems. Also, constancy of manufacturing processes is still not enough to enable a statistical characterization of FCS’ behavior. From an other point of view, useful data to build prognostics approaches are not clearly identified, and observability of relevant wearing phenomena still remains an open problem. In addition, experiments use to be cost expensive and the amount of available data is thereby reduced. The information framework is finally poor: missing data, imbalanced data between normal and degraded states, etc. According to all this, PHM axis must enable to ensure developments to cope with the following problems (figure 1).
- Generate experimental data for research developments
- Observe ageing
- Model the behavior
- Assess the current state of health of the FCS (SoH)
- Predict future health states of the FCS (prognostics – RUL)
- Test, optimize and validate the approaches developed
- Disseminate results and prepare industrial transfer
Taxonomy of developments
According to literature (in others fields than FCS), numerous approaches can a priori be used to perform prognostics. That said, prognostics can not be considered as a single task. Firstly, failure avoidance can only be reached if degradation phenomena are well tracked (data acquisition, feature extraction, detection, diagnostic). Secondly, a RUL estimates is not an objective in itself: predicting the remaining life should enable to mitigate wearing in order to ensure the accomplishment of the mission (decision support). Following that, various activities (modules), ranging from data collection through the recommendation of specific maintenance/control actions, must be carried out. The whole can be summarized as a PHM (Prognostics and Health Management) system. An illustration is given in figure 2 that depicts the taxonomy of developments of PHM axis from FCLAB.
1) Data acquisition. This module provides the PHM application with digitized sensor or transducer data. Instrumentation must be as non-intrusive as possible, but is however required to correctly track the behavior of the Fuel Cell.
2) Data processing. This module aims at filtering noise, and at extracting/selecting features that enable to characterize the functioning of the FCS. Within other techniques, EIS is explored as well as wavelet analysis techniques.
3) Condition assessment. This module compares on-line data with expected values of system’s parameters in order to detect abnormal behavior. Developments led in this area take into account uncertainty and the variability of functioning conditions.
4) Diagnostic. This module aims at determining if the monitored system, sub-system, or component has degraded, and at providing a diagnostic record and suggesting fault causes. From this point of view, interaction between components is the most challenging problem.
5) Prognostics. This modules aims at predicting the future health states of the system, and at estimating its remaining useful life. FCLAB developments emphasize on reliability of prognostics, by using data-driven approaches like neural networks and neuro-fuzzy systems.
6) Decision support. This module aims at providing recommended maintenance/actions actions on how to run the system, subsystem, or component until the mission is completed. Interactions with « FC system optimization » axis are obviously required.
7) Presentation. This module is generally built into a regular human-machine interface in order to ensure communication between modules. It doesn’t constitute a research area from FCLAB at this moment.