Detection of cell-to-cell variability in water holdup using pattern recognition techniques
View Patent ↗A system and method for preventing low performing cells in a fuel cell stack. The method includes periodically providing a pulse of the cathode input airflow at low stack current densities, and comparing the current density output of each cell in response to the pulse. Those cells that do not have significant water accumulation will provide one voltage signature and those cells that do have a significant water accumulation will provide another voltage signature. If one or more of the cells exhibit the voltage signature for water accumulation, then the cathode inlet airflow pulses can be provided more often to prevent the cells from failing.
1. A method for determining whether flow channels of a fuel cell in a fuel cell stack are flooded, indicating cell instability, said method comprising:
providing pulses of a cathode reactant gas flow to cathode flow channels within the fuel cell stack;
collecting data points of the voltage of each fuel cell in the fuel cell stack in response to the pulse of cathode reactant gas flow for a predetermined period of time;
using a pattern recognition algorithm to determine whether the collected data points indicate a voltage signature for each fuel cell identifying flooded cathode flow channels or non-flooded cathode flow channels; and
increasing how often the cathode air flow pulses are provided if the pattern recognition algorithm determines that one or more of the cells has flooded cathode flow channels.
2. The method according to claim 1 wherein determining whether any of the fuel cells have flooded cathode flow channels includes only determining whether any of the fuel cells have flooded flow channels if the current density of the stack is below a predetermined value.
3. The method according to claim 2 wherein the predetermined value of the current density of the stack is about 0.2 A/cm 2 .
4. The method according to claim 1 wherein using a pattern recognition algorithm includes using a Spearman rank correlation pattern recognition algorithm.
5. The method according to claim 4 wherein using a pattern recognition algorithm includes correlating the data points to a median response voltage signature for all of the fuel cells in the fuel cell stack.
6. The method according to claim 1 wherein using a pattern recognition algorithm includes using a principle component analysis pattern recognition algorithm.
7. The method according to claim 6 wherein using a pattern recognition algorithm includes normalizing the data points, transforming the normalized data points to principle components and calculating a Euclidean norm of the principle components.
8. The method according to claim 7 wherein using a pattern recognition algorithm includes normalizing the data points by subtracting each data point mean and dividing by a standard deviation.
9. The method according to claim 7 wherein using a pattern recognition algorithm includes transforming the normalized data points using a predetermined characteristic transformation matrix.
10. A fuel cell system comprising:
a fuel cell stack including cathode flow channels;
a compressor for providing cathode inlet air to the cathode flow channels in the fuel cell stack; and
a controller for controlling how often cathode air flow pulses are provided to the cathode flow channels to remove water from the cathode flow channels, said controller collecting data points of the voltage output of each fuel cell in the fuel cell stack during a cathode air flow pulse and using a pattern recognition algorithm to determine whether the collected cell voltages indicate a voltage signature for flooded cathode flow channels or non-flooded cathode flow channels, said controller increasing how often the cathode air flow pulses from the compressor are provided if the pattern recognition algorithm determines that one or more of the cells has flooded cathode flow channels.
11. The system according to claim 10 wherein the controller determines whether any of the fuel cells have a flooded cathode flow channels only if the stack current density output is below a predetermined value.
12. The system according to claim 10 wherein the controller uses a Spearman rank correlation pattern recognition algorithm.
13. The system according to claim 12 wherein the Spearman rank correlation pattern recognition algorithm correlates the cell voltages to a median response voltage signature for all of the fuel cells in the fuel cell stack.
14. The system according to claim 10 wherein the controller uses a principle component analysis pattern recognition algorithm.
15. The system according to claim 14 wherein the controller normalizes the collected cell voltage outputs, transforms the normalized cell voltages to principle components and calculates a Euclidean norm of the principle components.
16. The system according to claim 15 wherein the controller normalizes the data points by subtracting each data point mean and dividing by a standard deviation.
17. The system according to claim 15 wherein the controller transforms the normalized data points using a predetermined characteristic transformation matrix.
18. A fuel cell system comprising:
a fuel cell stack including cathode flow channels;
a compressor for providing cathode inlet air flow pulses to the cathode flow channels in the fuel cell stack; and
a controller for determining if any of the cathode flow channels are flooded using pattern recognition of cell voltages wherein the cell voltages are collected during the cathode inlet air pulses.
19. The system according to claim 18 wherein the controller uses Spearman rank correlation for the pattern recognition.
20. The system according to claim 18 wherein the controller uses principle component analysis for the pattern recognition.