We wanted to develop a new technique for measuring fuel cell impedance that would empower us to rapidly measure the characteristics of a wide range of frequencies in high precision, rather than just the impedance of a single frequency. To achieve this, we needed to build a system capable of high-speed, high-precision measurement, and then implement a measurement theory verification environment using actual fuel cells.
We used PXI products to build a system capable of high-speed sampling and performing measurements of multiple channels. Using the LabVIEW library, we wrote programs for measurement, as well as programs for fast Fourier transform (FFT) calculations and running median calculations to obtain impedance frequency characteristics.
Each year, fuel economy regulations for vehicles become stricter in countries around the world. Vehicle manufacturers are redoubling efforts to satisfy these regulations to cut CO2 emissions. One such effort involves attempting to migrate power sources from conventional gasoline engines to electric motors, which means demand for migration to electric vehicles is increasing. One kind of electric vehicle is a fuel cell vehicle that utilizes hydrogen energy. The fuel cells generate electric power by using the chemical reaction of hydrogen and oxygen (air). This reaction produces only water—no CO2 is emitted. Accordingly, research and development in this field is progressing as an extremely effective technique for reducing CO2 emissions.
Honda Motor Co., Ltd. has also been focusing on fuel cell vehicles and has conducted research and development since the 1980s. A key achievement of this R&D was the Clarity fuel cell vehicle released in March 2016. The Clarity can be refueled (hydrogen refilling time) in around three minutes, and it achieves faster high-speed refueling times than electric vehicles. Furthermore, its driving range can reach approximately 750 km on a single tank.
Honda has procured fuel cells from external suppliers in the past. However, we currently use fuel cells that have been manufactured in-house. The Clarity is also equipped with Honda’s own proprietary fuel cells. Our group is in charge of developing fuel cell systems. Our mission is to develop systems that achieve a high enough level of precision at a cost low enough to be brought to market. However, we must research and develop a diverse range of subjects to bring a fuel cell system to market. One such topic our group focused on was developing a system that could measure the impedance of fuel cells. We outline our reasons for wanting to measure fuel cell impedance below.
Fuel cells produce water by causing a chemical reaction between hydrogen and oxygen. In a fuel cell stack, the hydrogen and oxygen channels are made extremely narrow to make the stack more compact. If water becomes trapped in these channels, the hydrogen and oxygen cannot pass through and power fails to be generated. Accordingly, we must monitor the volume of water produced during operation of the fuel cells. We use impedance measurement as a means for monitoring this water due to the correlation between the volume of water and the impedance of fuel cells. The impedance of a fuel cell decreases as the volume of water trapped on the electrolyte membrane increases. Therefore, if the level of impedance becomes too low, we can decrease the volume of water by increasing the volume of air or modifying the temperature. Conversely, if the volume of water is too low, the impedance of the fuel cells could increase excessively and lead to other problems. We can use impedance measurement as a means for monitoring the water volume, and then based on the results, control the water volume for balance accordingly.
Impedance-based control techniques are already used in practical applications. The Clarity’s fuel cells also come with a function for measuring impedance. We input a single-frequency signal correlated to the water volume and measure the impedance at the time of input. We then control and modify the volume of water within the fuel cells based on the measurement results.
As previously mentioned, we can obtain information about the volume of water in fuel cells by measuring the impedance of a single frequency correlating to the water volume. Conversely, it could be said that the only information that we can obtain by this measurement is information about the water volume. Naturally, water volume is not the only factor that may affect the performance of fuel cells. In fact, if we also measure the impedance of other frequencies, we can obtain information on a variety of other elements of fuel cell performance. A Nyquist diagram such as the one shown in Figure 1 can help to visualize, for example, an event such as the deterioration of the catalyst, which can be a primary cause of decreased fuel cell performance. If we can measure the impedance of a wide range of frequencies, rather than just a single frequency, we can detect causes of performance deterioration other than the water volume, and then conduct the appropriate controls to enable sustained performance of the fuel cell vehicle accordingly.
When measuring the impedance of a wide range of frequencies rather than a single frequency, we can take impedance measurements repeatedly as the frequency of the input signal is modified. This technique is known as frequency response analysis (FRA). If we need to measure the impedance of only a single high frequency, we can obtain the result in one second or less. However, measuring a wide range of frequencies using the FRA technique takes five to ten minutes, which can be problematic. Clearly, this is not a practical solution.
There is another known method for measuring impedance frequency characteristics. It involves measuring (sampling) voltage/current while inputting m-sequence (maximum length sequence) (quasi random) signals, and then applying fast Fourier transform (FFT) calculations to the data obtained, to find the impedance frequency characteristics. This FFT method means we can perform measurements in just a few seconds. However, it also has its own problem. The FFT method is extremely vulnerable to the effects of noise, rendering it impractical (Figure 2).
Noise is always going to occur in the real world. In fact, we cannot perform high-precision measurements even on stand-alone fuel cells using the FFT method because of the effects of noise. Many electronic units operate when a vehicle runs, which inevitably generates massive amounts of noise. To measure impedance frequency characteristics at a practical level, we would need to develop a new method to suppress the effects of noise while continuing to use the FFT method for completing measurements quickly.
To resolve our noise problems, we adopted a method based on averaging. However, measurement would take too long if we obtained the average value by simply taking repeated measurements using the same conditions. Accordingly, we added the concept of ergodicity to the FFT method.
Ergodicity refers to the nature of time average being equal to the ensemble average. As an example, imagine rolling a die 100 times and then taking the average of all the resulting numbers thrown. If it takes one second to roll the die once, it will take 100 seconds to roll it 100 times, and we can then take the average of the resulting numbers. Now, if we throw 100 dice simultaneously, it would take only one second to obtain the average. Ergodicity implies that the resulting average values will be equal regardless of the method used.
However, if the method requires the preparation of 100 measurement devices to perform the averaging quickly, there is little point to the exercise. We considered how we could simulate this concept of ergodicity. We want to perform measurements on a wide range of frequencies, but imagine that we want to know the results at the 1,000 Hz point. For the measurement method, we can input m-sequence signals as per the usual FFT method and select sampling frequencies that will enable us to obtain sufficiently high frequency resolution within the range of 999 Hz to 1,001 Hz. We can then conduct the sampling and perform FFT calculations on the results obtained and use the average value (mean) of 999 Hz to 1,001 Hz as the measurement results at the 1,000 Hz point. We can repeat this process for multiple other frequency points (running median value) to obtain the frequency characteristics of impedance. If we apply the concept of obtaining the running median, we can suppress the effects of noise without increasing measurement times.
We confirmed the effectiveness of the aforementioned theory in a computer simulation, but we needed to see if we could verify this theory using actual fuel cells. We built a system for conducting the required experiments. To meet our objective, we chose NI products, including the LabVIEW graphical system development platform. Our company already had a proven track record of using NI products and believed LabVIEW would be extremely suitable for our proposal.
Figure 3 shows an outline of the system we built. It uses a PXIe-1082 chassis, a PXIe-8135 controller, and a PXIe-6358 DAQ device. The system injects oxygen and hydrogen into the fuel cell as per regular usage in a fuel cell. In this state, the m-sequence signal is input as the load, and the PXI-based system measures the voltage/current. We perform calculations on the obtained values to find the FFT and running median, then obtain the impedance value for each frequency.
The special feature of this measurement system is that it uses a high sampling frequency to make averaging function effectively. For example, to perform measurement in a frequency range of up to 10 kHz, we select a sampling frequency of 800 kHz. Actual fuel cell stacks are composed of as many as several hundred cells. To take detailed impedance measurements for each cell, the same number of measurement channels must be made available. The PXIe-6358 enables measurement to be performed using up to 16 channels at a maximum sampling frequency of 1.25 MHz. Furthermore, if increasing the number of PXIe-6358 units in order to increase the number of measurement channels to match the number of cells, synchronization between each of the units can be facilitated easily.
We can use LabVIEW to describe all programs needed for measurement or calculation graphically. The signal processing library already contains functions for performing FFT calculations or generating m-sequence signals. In developing the program, we first asked NI to provide us with the most basic of sample programs to use for measuring voltage/current. We then revised that program as needed. We needed various kinds of logic verifications, so we proceeded with our task by making revisions to the program and then conducting measurements, and then repeating the process. Although this seemed a little unfamiliar at first, we managed sufficiently without the need for studying huge volumes, unlike text-based programming languages. Furthermore, we could execute the completed program at extremely high speeds.
We knew that if we used NI platform products, we could select the optimal hardware as required. We also knew that we could graphically develop software for the purpose of controlling the operations of this hardware and to perform signal processing. These special features demonstrated clear advantages for our purpose.
Figure 4 shows an example of the measurement results obtained by the system we developed. The red plot represents the data obtained over a ten-minute period using the FRA method. We have plotted the impedance of frequencies at 43 different points within the range of 0.1 Hz and 10 kHz. Conversely, the blue plot represents the impedance measurement results obtained using our new technique. Although we plotted 256 points, the measurement required only 3.1 seconds to perform. As observed, both sets of results are remarkably similar. We have cumulatively plotted the measurement results approximately 20 times. Although there is some scattering in the lower frequencies, we have achieved high reproducibility.
We based our new technique on the FFT method, with an added process for obtaining the running median value. This technique helped us suppress the effects of noise while still maintaining measurement precision to acquire impedance frequency characteristics in just a few seconds.
In principal, vehicles will evolve in the direction of enhanced functionality while continuing to keep component costs down. We can’t say how commercial-level fuel cell vehicles will evolve in the future. It is unlikely that our new FFT method will be applied immediately due to matters such as cost. Autonomous vehicles will require processors that can perform extremely fast calculations. Once autonomous vehicle functionality is popularized, such processors might make it possible to perform calculations using the new FFT method. If so, the method outlined in this case study could be put to practical use.
Mr. Masahiro EGAMI, PhD
Honda R&D Co., Ltd.