Method for increasing the service life of a fuel cell

Abstract

The invention relates to a method for increasing the service life of a fuel cell which is used to provide electrical drive energy in a vehicle, for which purpose, when the vehicle is parked, an anticipated standstill duration is estimated, on the basis of which a decision is made as to whether an air / air start is accepted or cyclical hydrogen replenishment is initiated, wherein the anticipated standstill duration is estimated on the basis of various parameters which take into account legal driving time regulations and standstill times of foreseeable duration. The method according to the invention is characterised in that the parameters, together with personal driver-related data from a specified time period before the vehicle is parked, are used as input parameters for a deep learning model in order to estimate the standstill duration.

Description

[0001] Methods for increasing the lifespan of a fuel cell

[0002] The invention relates to a method for increasing the service life of a fuel cell used to provide electrical drive energy in a vehicle, according to the type defined in more detail in the preamble of claim 1.

[0003] In principle, a crucial factor influencing the lifespan of a fuel cell is how it is restarted after a period of inactivity. The so-called air-to-air start occurs when air is present on both the cathode and anode sides. During startup, the air on the anode side is displaced by incoming hydrogen. The disadvantage of this is that an air / hydrogen front is forced through the anode. Due to the potential difference with the air-filled cathode, the catalyst on the anode side can oxidize along this front. This results in catalyst material loss and reduces the fuel cell's lifespan.

[0004] For this reason, WO 2022 / 214170 A1 describes a procedure in which the vehicle's downtime is estimated in order to determine, based on this estimated downtime, whether such an air-to-air start should be accepted or whether it should be prevented by repeated hydrogen injection. The application is specifically aimed at use in trucks, so that the driver's prescribed rest periods can be taken into account when estimating the downtime.

[0005] The disadvantage of the method described there lies in the relatively coarse estimation of the standby time, which leads to high inaccuracies and thus ultimately to an unnecessarily high number of air-to-air starts. This can negatively affect the lifespan of the fuel cell.

[0006] DE 102009 036 199 A1 describes a method for operating a

[0007] Fuel cell system in a vehicle. This essentially involves a start / stop system, in which hydrogen typically continues to be supplied as long as a longer stop is not expected.

[0008] Furthermore, US 2015 / 0099200 A1 describes a cold start preparation for a fuel cell system, which essentially aims to prevent freezing.

[0009] For further information on the state of the art, reference can also be made to WO 2024 / 017463 A1. This describes the use of a deep learning model which, according to the aforementioned international publication, is essentially used to predict the state of health of the fuel cell.

[0010] Starting from the first-mentioned state of the art, the object of the inventive method is to improve the estimation of the expected downtime in order to achieve an optimized assessment of whether an air-to-air start is acceptable or not.

[0011] According to the invention, this problem is solved by a method with the features in claim 1. Advantageous embodiments and further developments result from the dependent claims.

[0012] The method according to the invention, similar to the prior art mentioned above, uses an estimation of the expected downtime of the vehicle. Like the prior art, the method takes into account various parameters that consider legal driving time regulations and longer downtimes of foreseeable duration.

[0013] According to the invention, in addition to these parameters, personal driver-related data from a past period of time until the vehicle is parked are used, and the parameters and the personal driver-related data are used as input for a deep learning model in order to estimate the standstill time.

[0014] This use of a deep learning model offers a crucial advantage in terms of the accuracy of estimating the expected downtime.

[0015] Particularly advantageous is the use of a deep learning model for pattern recognition, so that, based on characteristic driver behavior patterns, in addition to foreseeable conditions and legal regulations, the length of the expected downtime can be specified more precisely.

[0016] According to a particularly advantageous further development of the inventive method, the deep learning model is trained using fleet data that utilizes parameters from vehicles currently in the field. In addition to the data of the vehicle itself and the driver-related data of its driver, data from other vehicles in the field within a fleet—be it the fleet of a specific manufacturer, a transport company, a transport association, or the like—can also be used to train the deep learning model and thus improve the accuracy of the estimation. The parameters used can include, in particular, engine speed, vehicle speed, engine torque, and / or fuel level.

[0017] Based on these parameters, the operating condition of each vehicle in the fleet can be determined, whereby, according to a very advantageous further development, the length of expected driving and rest times is determined as average values ​​for the vehicles in the fleet based on the determined operating condition and the legal guidelines for driving and working hours and is used for the vehicle under consideration.

[0018] According to a highly beneficial training program, the personal driver-related data can include sensor data from a past period, specifically 14 days, to determine the length of individual breaks and the total driving time within this period. Furthermore, the total driving time can be determined, and the length of past rest periods can be categorized accordingly. This allows for a distinction between weekly rest periods, daily rest periods, and driving time interruptions. All this data can then be provided as input to a deep learning model, enabling a comparatively accurate prediction of the expected downtime.Based on the probabilities with which these standstill durations occur in certain recorded situations of the vehicle, it is then possible to decide simply and efficiently whether an air / air start should be accepted, or whether operational readiness should be maintained via hydrogen dosing, in a known manner, in order to avoid such an air / air start, which is harmful to the fuel cell, in any case.

[0019] Further advantageous embodiments of the method according to the invention can also be seen from the flowchart, which is described in more detail below with reference to the figure.

[0020] The only accompanying figure shows a flowchart for implementing the method according to the invention in a specific embodiment.

[0021] Figure 1 shows several boxes, each describing the acquisition and provision of data and / or the evaluation of data or the training of a deep learning model.

[0022] In the upper right, box number 1 symbolizes the acquisition of data from vehicles currently in the field. Here, relevant parameters are recorded to determine the operating status of each vehicle. These parameters include, in particular, engine speed, vehicle speed, engine torque, and fuel level. The data recorded in box number 1 is then transferred to boxes numbered 2 and 3. Box 2 is used to determine break patterns by identifying characteristic driving profiles based on the vehicle's operating status.In Box 3, based on the determined operating status and taking into account the relevant legal guidelines for driving and working hours—typically the corresponding EU directive in the applicant's area of ​​operation—an estimate of the length of driving and rest periods is calculated, and average values ​​are determined. The data generated from Boxes 2 and 3 are then fed into a deep learning model in Box 4, which uses this aggregated data for training. The appropriate deep learning model is thus trained with this data, either in the vehicle itself or in a data center connected to it via a communication link.

[0023] The system now checks in box number 5 whether the current vehicle, for which the standstill time is to be estimated, is parked. This means that the parking brake is activated and the electric vehicle's drive readiness is deactivated. The tachograph status can also be evaluated so that a change to the pause state can be detected.

[0024] In such a case, the driver's social data for the past 14 days is determined in box 6. This includes the length of individual breaks and the sum of all driving times over the last 14 days. The total driving time is also calculated. The length of the past rest periods is then categorized to differentiate between weekly rest periods, daily rest periods, and breaks in driving time. This data is then provided as input to the trained deep learning model. Box 7 is then supplied with this data from boxes 5 (as a trigger) and 6 to the deep learning model in box 4, which has been trained with the data from boxes 1, 2, and 3. In box 7, the data is processed by the deep learning model, which then provides an estimate of the expected duration of the stop.The deep learning model indicates which values ​​are considered likely based on the input data from boxes 5 and 6, so that ultimately the probabilities of the expected downtime durations can be included in the decision.

[0025] In Box 8, the outputs of the deep learning model from Box 7, taking into account the probability that the length of the interruption is accurate, determine the starting value for the downtime. This predicted or estimated downtime value is then fed into Box 9, which defines the operating strategy for the fuel cell based on the estimated downtime.

[0026] Box 9, based on the length of the estimated downtime (which is highly probable in terms of degradation effects and hydrogen consumption for maintaining operational capability), presents a decision between two methods that extend the service life of the fuel cell system and improve the preservation of the fuel cells, particularly the membranes and catalysts. These two options are then illustrated in boxes 10 and 11.

[0027] Box 10 is intended to symbolize the air-to-air start. If the system is stationary for a predetermined duration of, for example, 18 hours or more, it is completely shut down and all valves for hydrogen supply and air intake / exhaust, if present, are closed. The system is then largely left to its own devices. Depending on the system's tightness, more or less air, and therefore oxygen, enters the anode area, so that upon restarting, an air-to-air start can be assumed. However, maintaining operational readiness for such a long estimated downtime would require a comparatively large amount of hydrogen by constantly refueling, which would increase emissions and negatively impact hydrogen consumption and ultimately the vehicle's range.

[0028] Box 11 represents the alternative hydrogen replenishment strategy. For shorter standby times, for example, 18 hours or less, the cathode side, and thus the valves for air intake and exhaust, are closed. At defined intervals, a specific amount of hydrogen is added to the system, ensuring that the hydrogen atmosphere on the anode side is maintained throughout the entire standby time. This prevents the air-to-air start described earlier, and the fuel cell can be restarted without affecting its lifespan.

Claims

Patent claims 1. A method for increasing the service life of a fuel cell used to provide electrical drive energy in a vehicle, wherein, when the vehicle is switched off, a likely standstill duration is estimated, based on which a decision is made whether an air-to-air start is accepted or a cyclic hydrogen refueling is initiated, wherein the likely standstill duration is estimated based on various parameters which take into account statutory driving time regulations and standstill times of foreseeable duration, characterized in that the parameters together with personal driver-related data from a given time period before the vehicle is switched off are used as input parameters for a deep learning model for estimating the standstill duration.

2. Method according to claim 1, characterized in that pattern recognition is performed via the deep learning model.

3. Method according to claim 1 or 2, characterized in that the deep learning model is trained using fleet data which includes parameters of vehicles currently in the field.

4. Method according to claim 3, characterized in that The parameters used are those of vehicles currently in the field, including their engine speed, vehicle speed, engine torque and / or fuel level.

5. Method according to claim 3 or 4, characterized in that, based on the determined operating conditions of the fleet vehicles and a legal guideline for driving and working hours, the length of driving and rest times are determined as average values, which are then used to train the deep learning model.

6. Method according to one of claims 1 to 5, characterized in that the personal driver-related data includes social data of a past period of time, on the basis of which the length of individual breaks as well as the sum of all driving times in this period together with the total driving time is determined.

7. Method according to claim 6, characterized in that the rest periods recorded in the time period are ordered in order to be able to differentiate between weekly rest period, daily rest period and driving time interruption.

Images