Online fault detection on fuel cell systems

In this project high temperature PEM fuel cells are investigated during normal operation and during abnormal operation, with the aim of developing algorithms to detect when faults occur. By detecting faults the system can change operation parameters to go into normal operation or shutdown the system before components are damaged.

To determine if the fuel cell is in normal operation the fuel cell dynamics or statics needs to be characterized. This is in the literature most widely done by electrochemical impedance spectroscopy (EIS) measurements. The detection methods are then divided into signal based and model based diagnostic algorithms, where the signal based methods handle the EIS measurements directly via statistical or machine learning methods. The model based methods adapt the measurement to an electrical equivalent model (EEM) and compares the EEM parameters to normal operation values. The drawback of the EIS measurements is that the technique is expensive to implement in mass production systems.

In this project an alternative characterization technique have been investigated namely the small amplitude current pulses injection method. The advantage of this method is that it can be implemented in a real system with a transistor and a resistor, and is therefore low cost and suitable for mass production. Based on the time signals the EEM parameters can be determined via online parameter estimation. The EEM estimated is in general simpler than what can be observed by EIS measurements, but could be sufficient for many fuel cell diagnostics methods.

As a case study during this PhD study, detection of CO contamination in the anode gas of a high temperature fuel cell was chosen. A data driven method to detect CO content in the anode gas of a high temperature fuel cell have been designed. The fuel cell characterized by EIS measurements in normal and abnormal operation, and an EEM was fitted to the measurements. The normal operation EEC parameters were mapped as a function of the load current. To detect CO contamination in the anode gas a general likelihood ratio test detection scheme was designed to detect whether an EEM parameters differ from the normal operation. It was proven that the general likelihood ratio test detection scheme, with a very low probability of false alarm, could detect CO content in the anode gas of the fuel cell.

Publications or other dissemination activities:

Conference proceedings:

Diagnosis of CO Pollution in HTPEM Fuel Cell using Statistical Change Detection. Jeppesen, Christian; Blanke, Mogens; Zhou, Fan; Andreasen, Søren Juhl., 9th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes SAFEPROCESS 2015. Vol. 48 21. udg. 2015. s. 547-553 (IFAC-PapersOnLine). 2nd September 2015 Paris, France.

Oral presentations: 

Fuel cell characterization using current pulse injection, Fuel Cells Science and Technology 2016, 13th April 2016 Glasgow, Scotland.

Diagnosis of CO Pollution in HTPEM Fuel Cell using Statistical Change Detection, 9th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes SAFEPROCESS 2015, 2nd September 2015 Paris, France.

Poster presentation: 

Fuel Cell Equivalent Electric Circuit Parameter Mapping, 4th CARISMA conference 2014, Cape Town, South Africa. 


Contact

Christian-Jeppesen
PhD Student
Institute of Energy Technology (AAU)
http://www.4m-centre.dk/Research/christian-jeppesen
25 SEPTEMBER 2017