Method for determining the probability of failure of at least one cell system

Abstract

The invention relates to a method for determining the probability of failure, in particular at least one state of an increased probability of failure, of at least one cell system (10a), having at least one electrochemical cell (12a), wherein in at least one determining step (14a; 14b; 14c; 14d), at least one degradation parameter of the at least one electrochemical cell (12a) is determined by means of a virtual sensor unit (16a), the degradation paramater in particular allowing a conclusion to be drawn about the degradation state of the at least one electrochemical cell (12a) in a defined state. According to the invention, the probability of failure of the at least one electrochemical cell (12a) is determined as a function of the degradation parameter in a determining step (14a; 14b; 14c; 14d).

Description

[0001] R.414133

[0002] - 1 -

[0003] Description

[0004] title

[0005] Methods for determining the probability of failure of at least one cell system

[0006] State of the art

[0007] A method for determining the probability of failure, in particular at least a state of increased probability of failure, of at least one cell system with at least one electrochemical cell, has already been proposed, wherein in at least one determination step at least one degradation parameter of the at least one electrochemical cell, which in particular gives a conclusion about a degradation state of the at least one electrochemical cell in a defined state, is determined by means of a virtual sensor unit.

[0008] Disclosure of the invention

[0009] The invention relates to a method for determining a probability of failure, in particular at least one state of increased probability of failure, of at least one cell system with at least one electrochemical cell, wherein in at least one determination step at least one degradation parameter of the at least one electrochemical cell, which in particular gives a conclusion about a degradation state of the at least one electrochemical cell in a defined state, is determined by means of a virtual sensor unit.

[0010] It is proposed that, in one step of the investigation, a failure probability of at least one electrochemical cell be determined as a function of the degradation parameter. R.414133

[0011] - 2 -

[0012] Preferably, the cell system comprises only one electrochemical cell, in particular the one already mentioned. Alternatively, it is conceivable that the cell system comprises several electrochemical cells, for example, two, three, or several hundred electrochemical cells. The electrochemical cell is preferably designed as a solid oxide fuel cell. Alternatively, however, it is also conceivable that the electrochemical cell is designed as another type of fuel cell, for example, as a polymer electrolyte fuel cell or the like, as an electrolysis cell, or as a battery cell. For example, it is also conceivable that the electrochemical cell is used in an electrolyzer. Preferably, the electrochemical cells of the cell system comprising several electrochemical cells are grouped into a stack and, in particular, connected in series.It is also conceivable that the cell system comprises several stacks formed from multiple electrochemical cells, preferably grouped into a module or tower. Preferably, the cell system, and in particular the electrochemical cell, is configured to generate electrical energy from a fuel, especially hydrogen, natural gas, biogas, or any other fuel that would be suitable to a person skilled in the art, through a chemical process, or, in electrolysis mode, to produce hydrogen by supplying water and electricity. In particular, it is conceivable that the cell system is configured to operate in both fuel cell mode and electrolysis mode. Preferably, the cell system is configured to operate at a temperature of preferably 400 °C to 1200 °C, more preferably 500 °C to 1100 °C, and most preferably 600 °C to 700 °C.Preferably, in an operating state, a fuel such as hydrogen or hydrocarbons reacts with oxygen, producing electrical energy, heat and water.

[0013] The cell system preferably includes a control unit. The control unit is intended to execute the method for determining the probability of failure of at least one cell system. "Intended" is understood to mean specifically designed, specifically configured, and / or specifically equipped. The fact that an object is intended for a specific function is understood to mean that the object fulfills and / or performs this specific function in at least one application and / or operating state. R.414133

[0014] - 3 -

[0015] The control unit comprises, in particular, at least one processor and one memory element, as well as an operating program stored on the memory element. The memory element is preferably designed as a digital storage medium, for example, a hard drive or the like. The cell system comprises, in particular, a housing in which, preferably, the at least one electrochemical cell or the stack formed by several electrochemical cells is arranged. It is conceivable that the control unit is arranged at least partially on, and in particular at least partially in, the housing. Alternatively or additionally, it is also conceivable that the cell system has an external unit, wherein the external unit comprises at least a part of the control unit. The external unit can, for example, be a cloud, a server, or the like. It is also conceivable that the external unit comprises the memory element of the control unit.Preferably, the cell system includes a virtual sensor unit, in particular the one mentioned above. Preferably, the virtual sensor unit is configured to determine, and in particular calculate, a degradation state of the cell system. Preferably, the virtual sensor unit is stored as a sensor program on the control unit. Preferably, the virtual sensor unit and the control unit are integrated into a single component. Preferably, the sensor program is executed on a processor of the control unit. Preferably, the sensor program includes or is formed by a physical model. The physical model is preferably a simulation model of the cell system, which in particular describes the operation of the cell system in a steady state. Preferably, the physical model is based on the finite element method or the finite volume method.The physical model exhibits, in particular, a unique invertibility with respect to the degradation state, preferably the degradation parameter(s). It is conceivable that the physical model exhibits a unique invertibility with respect to the degradation state, preferably the degradation parameter(s), at least by filtering out physically irrelevant results. The degradation state describes, in particular, an aging state of the cell system. Preferably, the degradation state describes a loss of performance of the cell system, especially the electrochemical cell, over an operating period. For example, R.414133 exhibits this.

[0016] - 4 -

[0017] The degradation parameter has a value of 1 in an unaged state of the cell system and preferably a value greater than 1 for an aged state. Alternatively, the degradation parameter has a value of 1 in an unaged state of the cell system and preferably a value less than 1 for an aged state. Combinations of these are also possible. Preferably, the virtual sensor unit is configured to describe a degradation state using at least one degradation parameter. In particular, it is conceivable that the virtual sensor unit is configured to describe a degradation state using at least two degradation parameters. A degradation parameter is preferably a scalar quantity. In particular, it is conceivable that a degradation state is detected using several scalar degradation parameters or using a vector degradation parameter.Preferably, the degradation state is influenced by chemical degradation, material migration, and / or electrolyte degradation. In chemical degradation, the chemical reactivity of the materials in the electrochemical cell leads to a reduction in active material or phase changes, which affect electrode performance. For example, cathode materials can react with air contaminants, thereby reducing their activity and conductivity. Similarly, carbon deposits can form on the anode if hydrocarbons are used as fuel. With a loss of catalytic activity, the electrochemical reactions at the electrodes can become less efficient over time, as the active surface areas decrease due to contamination, aging, or structural changes.Material migration occurs when materials such as nickel or other metallic components diffuse over extended operating times and spread throughout a cell system, impairing the function of the electrodes and electrolytes. Electrolyte degradation can cause the electrolyte to lose conductivity due to ionic losses or the formation of defects in its structure, increasing electrical resistance and affecting the overall performance of the cell system. During operation, the cell system, and especially the electrochemical cell, exhibits continuous degradation, which can be time- and load-dependent. Additionally, the cell system, particularly the electrochemical cell, can lose all functionality due to "spontaneous failures." Spontaneous failure in a cell system, especially in R.414133.

[0018] - 5 - electrochemical cell, an abrupt and unexpected cell failure occurs. This leads to an immediate loss of functionality, so that the cell can no longer generate or convert electrical energy and / or hydrogen. The failure probability, in particular at least one state of increased failure probability, of at least one cell system, in particular at least one electrochemical cell, preferably describes the probability of spontaneous failure and / or accelerated aging of the cell system, wherein spontaneous failure is understood to mean the sudden and unforeseen failure of the cell system's function, and accelerated aging is understood to mean a state in which the degradation of the cell system progresses faster than would be expected under normal operating and environmental conditions.

[0019] In at least one determination step, at least one degradation parameter of the at least one electrochemical cell is preferably determined using the virtual sensor unit. This parameter, in particular, allows conclusions to be drawn about the degradation state of the at least one electrochemical cell in a defined state. The virtual sensor unit preferably determines a degradation parameter of the electrochemical cell from at least one input variable of the electrochemical cell in at least one determination step. Preferably, the at least one input variable of the virtual sensor unit comprises at least one operating characteristic of the cell system, in particular at least one input characteristic of the cell system, and preferably at least one set operating parameter of the cell system.Preferably, the at least one input variable of the virtual sensor unit comprises at least one further operating characteristic of the cell system, in particular at least one input characteristic of the cell system, preferably at least one determined, in particular measured, operating parameter of the cell system. The operating parameter is, for example, a queried load current, a temperature, in particular a gas temperature, a volume flow rate, an air utilization rate, a fuel utilization rate, a gas composition, a pressure, a pressure difference, or the like. Preferably, the at least one input variable of the virtual sensor unit comprises at least one performance characteristic of the cell system. The performance characteristic is, for example, R.414133.

[0020] - 6 - a cell voltage and / or a quantity of hydrogen or the like. Preferably, in a determination step, the degradation parameter is determined by means of the virtual sensor unit as a function of the input variables, which allows a conclusion to be drawn about the degradation state of the cell system, in particular the electrochemical cell. Preferably, in a determination step, a degradation parameter is calculated at a defined time as a function of the input variables using the sensor unit. Preferably, in a determination step, a degradation parameter is determined at a current time. Preferably, in a determination step, a conclusion is drawn about a degradation state at a current time. In particular, it is conceivable that in a determination step, a degradation parameter of a defined past time window is determined.Preferably, in a determination step, a degradation parameter is determined at defined intervals within a defined past time window. In particular, it is conceivable that a determination step is carried out at defined intervals over the duration of the cell system's operation, whereby a degradation parameter is determined within a predefined time interval. Preferably, the determination step is carried out at regular time intervals. Alternatively, it is conceivable that the determination step is carried out at defined irregular time cycles and / or through operator interaction. Preferably, a determination step is carried out within an operating step. Preferably, an operating step of a cell system is understood to be a specific phase or action within the operating cycle in which certain operationally necessary processes or functions are executed.This includes various states and processes that are crucial for the proper operation, performance, and lifespan of the cell system. Preferably, the electrochemical cell is in a steady state during the determination step.

[0021] Preferably, the determination step is performed during an operational step, during normal operation of the cell system. More preferably, the determination step is performed during regular operation of the cell system. The determination step can, in particular, be carried out continuously or iteratively during the operating step of the cell system. Preferably, in the determination step, a failure probability of at least one electrochemical cell is determined as a function of a degradation parameter. Preferably, in R.414133

[0022] - 7 -

[0023] In this step, at least one state of increased failure probability for at least one electrochemical cell is determined. In particular, it is conceivable that this step determines an increased failure probability for a cell system comprising one electrochemical cell and at least one other electrochemical cell. Preferably, this step determines a degradation parameter that provides information about the degradation state of the electrochemical cell. Preferably, this step provides information about a state of increased failure probability for at least one electrochemical cell, depending on a current degradation parameter. Preferably, a state of increased failure probability is an operating state of the cell system in which the probability of system failure or failure of individual components is significantly increased.This condition is particularly critical because it can impair the reliability, efficiency, and safety of the entire system. Early detection of this condition allows for preventive measures to be taken to avert potential failures and extend the system's service life. This condition can be caused by various factors, such as advanced degradation, unfavorable operating conditions, material fatigue, or the occurrence of anomalies in the electrochemical processes. Preferably, in a preliminary step, a failure probability is determined by comparing the degradation parameter with a reference value.Preferably, in a determination step, a state of increased failure probability occurs when the degradation parameter is less than or equal to, for example, a determined degradation quantity, or greater than or equal to, for example, a degradation rate, than a reference value.

[0024] Preferably, in a determination step for a cell system with multiple stacks of electrochemical cells, the deviation of the degradation parameters, in particular the degradation rates or degradation magnitudes, of the individual stacks is monitored. A stack in a cell system is an arrangement of several electrochemical cells connected in series or parallel to achieve the desired overall power. By stacking cells, the voltage and current of the system can be scaled according to the requirements. Preferably, in a determination step, the R.414133

[0025] - 8 -

[0026] Monitoring of the degradation parameters of individual stacks is performed, and then, for example, the measured operating conditions for each stack are recorded individually. In particular, it is conceivable that, when operating multiple stacks, the difference in degradation parameters is used as a measure of an increased probability of failure. Preferably, it is checked whether the deviation of the degradation parameter, especially the degradation rate or the degradation magnitude, of an individual stack from a mean value of all stacks exceeds a defined warning threshold. In particular, the maximum difference in degradation parameters between two stacks can also serve as a criterion. Preferably, in a determination step, a warning of a state of increased probability of failure is issued as soon as one of the two threshold types is exceeded.Alternatively, a warning for a state of increased failure probability can be triggered only when the specified criterion is met by more than one stack in a cell system, for example, by more than one-fifth of the stacks. In this case, too, monitoring of the degradation parameters is carried out as an investigative step to enable timely responses to potential states of increased failure probability.

[0027] The inventive design of the method for determining the failure probability of at least one cell system provides advantageous characteristics with regard to cost-efficient operation of the cell system. In particular, the aforementioned method can extend the operating life of the cell system, increase its availability, and reduce costs for short-term service interventions or unexpected failures. Advantageous characteristics with regard to the service life of the cell system can be provided. Particularly advantageous characteristics with regard to planning reliability can be achieved through the planning of operator interactions or the adjustment of operating parameters. Advantageous characteristics with regard to user-friendliness can also be provided.Particularly advantageous properties regarding the economic efficiency of the cell system's operation can be provided. R.414133.

[0028] - 9 -

[0029] Furthermore, it is proposed that in at least one evaluation step, all degradation parameters, in particular all degradation rates of the at least one electrochemical cell, are calculated within a reference range. Preferably, in at least one evaluation step, a time window width for the degradation rate is defined. Preferably, in one evaluation step, a calculation method for the degradation rate within a reference range and / or within a current time window is defined. For example, a least-squares fit method for a straight line is used as the calculation method. Preferably, in the case of a least-squares fit method for a straight line, the degradation parameters are fitted to the measured data. The least-squares fit method, also known as the method of least squares, is a mathematical procedure for determining the best possible linear relationship between two variables.A straight line is determined that best corresponds to the given data points by minimizing the sum of the squares of the vertical distances (residuals) between the measured data points and the straight line. Preferably, the slope of a straight line representing the degradation quantity is calculated as a function of time, which correlates with a degradation rate, using the least-squares-fit method. Alternatively and / or additionally, any other calculation method that would be considered appropriate by a person skilled in the art is also conceivable. Preferably, in at least one evaluation step, a reference range is defined within the time window width of the degradation quantity's progression. Preferably, a reference range is defined at the beginning of the cell system's operation or within a defined area.Preferably, in the evaluation step, a degradation parameter, preferably a degradation rate, for at least one electrochemical cell is calculated at regular intervals. Preferably, in an evaluation step, all degradation parameters, in particular all degradation rates of the at least one electrochemical cell, are calculated in a reference range using the control unit. In particular, it is conceivable that in an evaluation step, a degradation parameter, preferably a degradation rate, for at least one electrochemical cell and / or stack is calculated in a reference range. In particular, it is conceivable that in an evaluation step, all degradation parameters calculated in a reference range in a determination step are converted into a degradation rate as a function of time. Preferably, this is described in R.414133.

[0030] - 10 - In an evaluation step, a degradation rate is calculated from degradation values ​​calculated in a reference range within a defined interval. Preferably, all degradation rates within a reference range are calculated in a single evaluation step using a sliding window method. The sliding window method is a procedure in which the total operational data is divided into smaller, consecutive time windows or "slices." This division allows for better detection and analysis of local trends and changes in the data. Preferably, in a single evaluation step, the mean and standard deviation of the reference range are determined from the calculated degradation rates. Alternatively, any other mathematical methods for calculating the degradation rate that would be considered appropriate by a person skilled in the art are also conceivable.This allows for the provision of particularly advantageous properties with regard to the determination of a reference degradation rate. In particular, it allows for the advantageous provision of a reference value for normal operation of the cell system.

[0031] Furthermore, it is proposed that in at least one comparison step, the current degradation parameter, in particular a degradation rate of the at least one electrochemical cell, is compared with a distribution of degradation parameters, in particular the degradation rates of the at least one electrochemical cell, from the reference range. Preferably, the comparison step is performed after an evaluation step. Preferably, in the evaluation step, the current degradation rate, in particular of a last available window, is determined. Preferably, in the evaluation step, the current degradation rate is calculated from a last available time window of the time window width.Preferably, in the evaluation step, the current degradation rate is calculated using the defined calculation method from the last available time window of the time window width, using the degradation values ​​calculated by the virtual sensor unit from the last available time window. Alternatively, in the evaluation step, an average of several degradation rates from a wider window and / or a plurality of windows within a last available time window is calculated to determine the current degradation rate. Preferably, in the comparison step, the current degradation rate calculated in an evaluation step is compared with R.414133.

[0032] - 11 - a distribution of the degradation parameters, in particular the degradation rates of the at least one electrochemical cell, from the reference range is compared. Preferably, in a comparison step, it is checked whether a current degradation rate deviates significantly from one of the degradation rates of the at least one electrochemical cell from the reference range. Preferably, in a comparison step, a warning message is issued as soon as the current degradation rate, in particular at least one standard deviation of the current degradation rate, is greater than a distribution of the degradation rates of the at least one electrochemical cell from the reference range. Preferably, the distribution of the degradation rates of the at least one electrochemical cell from the reference range has a safety factor. A safety factor of 3 is particularly preferred. Alternatively, any other safety factors that appear reasonable to a person skilled in the art are conceivable.For example, a safety factor of 3 means that a warning is issued as soon as the current degradation rate exceeds the distribution of degradation rates of at least one electrochemical cell by at least three standard deviations from the reference range. Alternatively or additionally, it is conceivable that, in at least one comparison step, a trend of the recent degradation rates, particularly the last three, is analyzed. An example is when the degradation rates increase sharply and, in particular, do not follow a linear pattern (e.g., exponentially), which indicates a rapid deterioration of the cell system, especially the electrochemical cell. Such an increase is, in particular, a further indication of an increased probability of failure and cannot always be detected by simple threshold calculations.In particular, it is conceivable that a trend analysis using time series analysis, regression, or a moving average is evaluated in a comparative step. This allows for the identification of particularly advantageous characteristics regarding the assessment of a state with an increased probability of failure. It is especially advantageous for the early detection of a potential failure of the cell system. It is particularly advantageous for ensuring planning reliability.

[0033] It is further proposed that in at least one learning step a prediction algorithm be developed based on an assumption about the course of the degradation parameter in R.414133

[0034] - 12 -

[0035] The system is calibrated based on the set operating conditions of the cell system, particularly of the at least one electrochemical cell, and on degradation parameters determined by the virtual sensor unit. Preferably, a prediction algorithm is trained and / or calibrated in a learning step. This algorithm enables the prediction of the future degradation of a specific parameter over a future period. For example, the prediction is made within a limited window of approximately eight weeks. The prediction algorithm combines values ​​from the virtual sensor unit, such as adjustable or measured operating parameters, up to a current time with a prediction of the degradation parameter's progression.In particular, it is conceivable that by assuming future operating parameters as input parameters for the prediction algorithm, a prediction for a degradation parameter at a defined point in the future or within a defined time window in the future can be determined. Preferably, the prediction algorithm can be specifically trained using a database with operating parameters of the cell system where the respective value of the degradation parameter lies within or outside the critical value range. Preferably, the prediction algorithm is trained by machine learning, in particular to execute a regression procedure for classifying identifiable states with an increased probability of failure of the cell system. The training of the prediction algorithm is based, for example, on a random forest, a support vector machine, a neural network, or the like.Preferably, the prediction algorithm is used in operation of the cell system to detect and, in particular, classify a state of increased failure probability in the cell system based on the operating parameters of the cell system recorded during operation, for example, by assigning at least one degradation parameter to it. In particular, it is conceivable that in the learning step, future operational use of the electrochemical cell is assumed in order to calculate the degradation magnitude in the future. It is also conceivable that in a learning step, the prediction algorithm is calibrated using standard operating conditions under which no significant changes are expected during operation. Alternatively, the average of the R.414133.

[0036] - 13 -

[0037] Operating conditions from a previous time frame are calculated and projected into the future to calibrate the prediction algorithm. Specifically, it is conceivable that, in a training step, the prediction algorithm is calibrated using actual operating data from a previous period by adopting the operating profile from that period for the forecast. Furthermore, the prediction algorithm can be calibrated with extreme values ​​or worst-case scenarios of the electrochemical cells to examine how degradation might develop under the most unfavorable conditions, thus supporting risk assessment and maintenance planning. The prediction algorithm can be trained precisely and efficiently, and conditions with an increased probability of failure in a cell system can be detected particularly easily and reliably.Advantageously, this allows an error in a cell system to be assigned to a degradation parameter.

[0038] Furthermore, it is proposed that in at least one evaluation step, a degradation rate of the at least one electrochemical cell at a defined point in the future, in particular within the last available window in the future, is calculated as a function of a prediction algorithm. Preferably, in at least one evaluation step, a degradation profile is calculated as a function of the prediction algorithm up to a defined point in the future. Preferably, in one evaluation step, a degradation rate is determined as a function of a profile determined by the prediction algorithm at a defined point in time. Particularly preferably, in one evaluation step, a degradation rate is calculated within the last available window of the degradation profile predicted by the prediction algorithm.Alternatively, it is conceivable that in the evaluation step, an average of several degradation rates from a wider window and / or a multitude of windows is calculated within the last available time window of the degradation quantity predicted by the prediction algorithm. Alternatively, it is conceivable that in at least one evaluation step, a future degradation rate at a defined time is calculated as a function of a prediction algorithm. This can enable particularly advantageous properties with regard to predicting a degradation rate at a defined time. R.414133.

[0039] - 14 -

[0040] Furthermore, it is proposed that in at least one comparison step, the degradation rate of the at least one electrochemical cell at the defined time in the future, particularly in a last available window in the future, is compared with the distribution of degradation rates of the at least one electrochemical cell from the reference range. Preferably, in the comparison step, the degradation rate of the at least one electrochemical cell calculated in an evaluation step at the defined time in the future, particularly in a last available window in the future, is compared with a distribution of degradation parameters, particularly the degradation rates of the at least one electrochemical cell, from the reference range.Preferably, a comparison step is performed to check whether the degradation rate of the at least one electrochemical cell at a defined future time, particularly within a last available window in the future, deviates significantly from one of the degradation rates of the at least one electrochemical cell from the reference range. Preferably, a warning is issued in a comparison step as soon as the degradation rate of the at least one electrochemical cell at a defined future time, particularly within a last available window in the future, is greater than a distribution of degradation rates of the at least one electrochemical cell from the reference range by at least one standard deviation of the current degradation rate. Preferably, the distribution of degradation rates of the at least one electrochemical cell from the reference range includes a safety factor.A safety factor of 3 is particularly preferred. Alternatively, any other safety factor that would appear reasonable to a person skilled in the art is conceivable. For example, a safety factor of 3 means that a warning is issued as soon as the degradation rate of the at least one electrochemical cell at the defined time in the future, in particular in a last available window in the future, exceeds the distribution of degradation rates of the at least one electrochemical cell from the reference range by at least three standard deviations. Alternatively or additionally, it is conceivable that in at least one comparison step, a trend of the last degradation rates of the at least one electrochemical cell at the defined time in the reference range is determined.

[0041] - 15 - future, in particular in a last available window in the future, especially the last three degradation rates of the at least one electrochemical cell at the defined time in the future, in particular in a last available window in the future, is analyzed. An example is when the last degradation rates of the at least one electrochemical cell at the defined time in the future, in particular in a last available window in the future, increase sharply and in particular are non-linear (e.g. exponential), which indicates a rapid deterioration of the cell system, in particular the electrochemical cell. Such an increase is in particular an additional indication of an increased probability of failure and cannot always be captured by simple threshold considerations.In particular, it is conceivable that a trend analysis using time series analysis, regression, or a moving average is evaluated in a comparative step. This allows for the identification of particularly advantageous characteristics regarding the assessment of a state with an increased probability of failure. It is especially advantageous for the early detection of a potential failure of the cell system. It is particularly advantageous for ensuring planning reliability.

[0042] It is further proposed that, in at least one evaluation step, a degradation parameter of the electrochemical cell is calculated for a defined point in the future, depending on a prediction algorithm. Preferably, in at least one evaluation step, a trend of the degradation parameter up to a defined point in the future is calculated, depending on the prediction algorithm. This enables particularly advantageous properties with regard to predicting a degradation parameter for a defined point in time.

[0043] Furthermore, it is proposed that in at least one evaluation step, a distribution of the degradation quantity within a defined time window and / or a defined time interval is calculated based on a probabilistic prediction algorithm. Preferably, the distribution of the degradation quantity at a defined time point is calculated using a probabilistic prediction algorithm. A probabilistic prediction algorithm utilizes R.414133.

[0044] - 16 -

[0045] Probabilistic models are used to account for uncertainties in predictions. Instead of providing a single predicted value, they indicate a distribution of possible outcomes, which advantageously allows for more realistic and flexible modeling. Examples of probabilistic methods include Gaussian process regression, Bayesian networking, Monte Carlo simulation, and the hidden Markov model. These methods offer particularly advantageous properties for predicting a degradation quantity at a defined point in time.

[0046] Furthermore, it is proposed that in at least one comparison step, the determined degradation level for a defined point in the future and / or a distribution of the degradation level within a defined time window and / or a defined time period is compared with a previously defined critical degradation level. Preferably, in the comparison step, a degradation level calculated in an evaluation step for a defined point in the future is compared with a previously defined critical degradation level. Preferably, in a comparison step, a warning is issued as soon as the degradation level calculated in an evaluation step for a defined point in the future exceeds or falls below a previously defined critical degradation level.In particular, it is conceivable that the critical degradation value includes a safety factor designed to ensure a safety tolerance to a critical state with an increased probability of failure. In a comparison step, a distribution of the degradation value determined in an evaluation step at a defined time and / or time interval is compared with a previously defined critical degradation value. Preferably, in this embodiment, a warning is issued as soon as at least one value of the distribution of the degradation value at a defined time and / or time interval falls below a previously defined critical degradation value.Furthermore, it is conceivable that an exceedance of the previously defined critical degradation value is also analyzed, whereby a significant exceedance of a previously defined critical degradation value can also trigger an adjustment step. In particular, a critical value range of the critical R.414133 exists for at least one electrochemical cell.

[0047] - 17 -

[0048] A predefined degradation value is stored, for example, on the control unit's memory element. It is conceivable that the critical value range of the critical degradation value can be adjusted, for example, automatically by the control unit or manually by the operator via an input unit of the cell system. For instance, a database containing critical value ranges of the critical degradation value for different operating parameters could be stored on the memory element. In particular, it is conceivable that the prediction algorithm could access the database to utilize extreme values ​​or worst-case scenarios of the electrochemical cell in a learning step. This would allow for the identification of particularly advantageous properties with regard to the assessment of a state with an increased probability of failure.It can be particularly advantageous to detect a potential failure of the cell system at an early stage. It can also be particularly advantageous to ensure planning reliability.

[0049] Furthermore, it is proposed that in at least one learning step, a prediction algorithm is calibrated based on the normal behavior of the degradation parameter, in particular the degradation magnitude. Preferably, in one learning step, the prediction algorithm is calibrated and / or trained using data from units or time intervals exhibiting "normal" behavior of the cell system. Preferably, the "normal behavior" of a cell system, in particular an electrochemical cell, is characterized by a stable, predictable change in the performance parameters, such as occurs within the context of a typical aging process. For example, normal behavior includes a steady decline in performance and a moderate increase in the degradation parameter, in particular the internal resistance, without sudden or unexpected jumps in the measured values.Preferably, the normal behavior is calibrated in an evaluation step based on long-term observation of a reference range and / or a previously completed operating time. In particular, it is conceivable that normal behavior is further refined by specific temperature ranges, charge and discharge cycles, and typical usage patterns that do not cause significant deviations in cell performance. Preferably, in a learning step, a uniform R.414133 is used as the normal behavior.

[0050] - 18 - A moderate, continuous, and in particular linear decrease in the degradation quantity is assumed, resulting in a constant degradation rate. Alternatively, any other function of the degradation quantity that appears meaningful to a person skilled in the art and that correlates with normal behavior can be considered. This allows for particularly advantageous properties with respect to a precise representation of normal degradation behavior, including possible deviations or anomalies in the future.

[0051] It is further proposed that, in at least one comparison step, a current degradation value of the electrochemical cell is compared with a predicted degradation value for the current time, determined in the past by the prediction algorithm. Preferably, in the evaluation step, a future trend of the degradation value is determined based on a prediction algorithm for the normal behavior of a degradation value of the at least one electrochemical cell. In particular, it is conceivable that, in at least one evaluation step, a degradation value is calculated for at least a defined period in the future based on a prediction algorithm for the normal behavior of a degradation value.Preferably, in at least one evaluation step, a smoothing parameter is selected to achieve a balance between the accuracy of the fit to the data and the avoidance of overfitting. For example, the smoothing parameter is a value between 0 and 1. Preferably, the smoothing parameter is applied to the determination of the future degradation value. Preferably, in a comparison step, a warning is issued as soon as the current degradation value falls below or exceeds a previously determined future degradation value at a given time. In particular, it is conceivable that the previously determined future degradation value at a given time includes a safety factor intended to ensure a safety tolerance to a critical state with an increased probability of failure.This allows for particularly advantageous properties regarding an assessment of a state of increased failure probability of a current R.414133.

[0052] - 19 -

[0053] The degradation level can be determined. In particular, it can be advantageous to detect a potential failure of the cell system at a given time.

[0054] Furthermore, it is proposed that, in a comparison step, a current degradation value at a defined point in time and / or a defined time interval is compared with a distribution of the future degradation value determined in the past at a current point in time. For example, the distribution is determined using a probabilistic prediction algorithm. Preferably, in this embodiment, a warning is issued as soon as a current degradation value falls below at least one distribution of the future degradation value determined in the past at a current point in time. In particular, it is conceivable that, in a comparison step, a series of values ​​is compared with a series of future degradation values ​​determined in the past within a current time window, with a warning being issued if a critical number from N is met.Furthermore, it is conceivable that, in a comparison step comparing the current degradation level with a historically determined distribution of future degradation levels at a given time, the probability of a state with an increased failure probability increases with decreasing or increasing degradation levels. This allows for particularly advantageous assessments of a state with an increased failure probability for a given current degradation level. In particular, it can be advantageous to detect a potential failure of the cell system at a given time.

[0055] Furthermore, it is proposed that in at least one adaptation step, at least one operating parameter of the cell system, in particular of the at least one electrochemical cell, is adjusted and / or at least one operator interaction is injected into the cell system, in particular of the at least one electrochemical cell, depending on a warning signal determined in a comparison step. Preferably, an adaptation step is initiated by the control unit. Preferably, operator interaction and / or an adjustment of the operating parameters takes place via the control unit. Preferably R.414133

[0056] - 20 - The adaptation step is initiated depending on a warning signal for a state of increased failure probability determined in a comparison step. Preferably, the warning signal is processed by the control unit in a comparison step, and then an adaptation step is automatically triggered. Preferably, an adjustment of the operating parameters is made in an adaptation step to slow down degradation and, for example, to prevent the cell voltage from falling below a minimum or maximum permissible value. Preferably, an immediate check of the cell system by a technician and / or operator is triggered in an adaptation step during operator interaction. This operator interaction can preferably be carried out either remotely (human-in-the-loop) or on-site (service), depending on the requirements.In particular, a controlled shutdown of the cell system, either automatically or through operator interaction, is conceivable as part of an adjustment step. Preferably, depending on the warning message received during an adjustment step, the replacement of the stack and / or the entire cell system can be scheduled within a given timeframe, for example, within 8 weeks, to avoid premature or delayed replacement, such as in the case of a planned replacement. This can result in particularly advantageous properties regarding the extension of system operation. It can also lead to significant cost savings, for example, by avoiding short-notice service calls or unexpected failures, as well as improved planning reliability.

[0057] Furthermore, a cell system, in particular the one already mentioned, for carrying out a method according to the invention is proposed, comprising at least one electrochemical cell, in particular the one already mentioned, at least one virtual sensor unit, in particular the one already mentioned, and at least one control unit, in particular the one already mentioned. Advantageously, a particularly durable cell system can be provided. Advantageously, a cell system can be provided that enables particularly precise and convenient maintenance planning. Particularly advantageous properties with regard to cost-efficient operation of the cell system can be provided. R.414133

[0058] - 21 -

[0059] The cell system and / or the method according to the invention shall not be limited to the application and embodiment described above. In particular, the cell system and / or the method according to the invention may, to achieve a functionality described herein, comprise a different number of individual elements, components, units, and process steps than that specified herein. Furthermore, values ​​within the specified limits of the value ranges stated in this disclosure shall also be considered disclosed and freely usable.

[0060] drawing

[0061] Further advantages become apparent from the following description of the drawings. The drawings illustrate four exemplary embodiments of the invention. The drawings, the descriptions, and the claims contain numerous features in combination. It is advantageous for those skilled in the art to also consider the features individually and combine them into meaningful further combinations.

[0062] They show:

[0063] Fig. 1 shows a cell system according to the invention, comprising at least one electrochemical cell, in a schematic representation.

[0064] Fig. 2 shows a schematic flowchart of a method according to the invention for determining a failure probability of at least one cell system,

[0065] Fig. 3 shows a schematic diagram of the rate of change of a degradation quantity as a function of time in a schematic representation.

[0066] Fig. 4 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining the probability of failure of at least one cell system.

[0067] Fig. 5 a schematic flow diagram of an alternative embodiment of a degradation quantity as a function of time in a schematic representation, R.414133

[0068] - 22 -

[0069] Fig. 6 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining a failure probability of at least one cell system,

[0070] Fig. 7 shows a schematic flow diagram of an alternative embodiment of a degradation quantity as a function of time in a schematic representation.

[0071] Fig. 8 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining a failure probability of at least one cell system and

[0072] Fig. 9 shows a schematic flow diagram of an alternative embodiment of a degradation quantity as a function of time in a schematic representation.

[0073] Description of the exemplary implementations

[0074] Figure 1 shows a schematic representation of a cell system 10a according to the invention, comprising at least one electrochemical cell. The cell system 10a includes at least one electrochemical cell 12a. Alternatively, the cell system 10a may have several electrochemical cells 12a, for example, two, three, or several hundred electrochemical cells 12a. The electrochemical cell 12a is configured as a solid oxide fuel cell. Alternatively, however, it is also conceivable that the electrochemical cell 12a is configured as another type of fuel cell, for example, as a polymer electrolyte fuel cell or the like, as an electrolysis cell, or as a battery cell. For example, it is also conceivable that the electrochemical cell 12a is used in an electrolyzer.In particular, it is conceivable that the electrochemical cells 12a of the cell system 10a, which comprises several electrochemical cells 12a, are grouped into a stack and, in particular, connected in series. It is also conceivable that the cell system 10a comprises several stacks formed from several electrochemical cells 12a, which are preferably grouped into a module or a tower. The cell system 10a, in particular the electrochemical cell 12a, is R.414133.

[0075] - 23 - is configured to generate electrical energy or, in electrolysis mode, hydrogen by supplying water and electricity from a fuel, in particular hydrogen, natural gas, biogas, or any other fuel that would appear suitable to a person skilled in the art, through a chemical process. In particular, it is conceivable that the cell system 10a is configured to operate in both a fuel cell mode and an electrolysis mode. The cell system 10a is configured to operate at a temperature preferably of 400 °C to 1200 °C, more preferably of 500 °C to 1100 °C, and most preferably of 600 °C to 700 °C. In an operating state, a fuel such as hydrogen or hydrocarbons reacts with oxygen in a cell system 10a, in particular an electrochemical cell 12a, producing electrical energy, heat, and / or water.

[0076] The cell system 10a comprises a control unit 46a. The control unit 46a is designed to perform the method for determining the probability of failure of at least one cell system 10a. The control unit 46a comprises at least one processor and one memory element, as well as an operating program stored on the memory element. The memory element is preferably designed as a digital storage medium, for example, a hard drive or the like. The cell system 10a particularly comprises a housing in which the at least one electrochemical cell 12a or the stack formed by several electrochemical cells 12a is preferably arranged. It is conceivable that the control unit 46a is arranged at least partially on, and in particular at least partially in, the housing.Alternatively or additionally, it is also conceivable that the cell system 10a has an external unit, wherein the external unit comprises at least a part of the control unit 46a. The external unit could, for example, be a cloud, a server, or the like. It is also conceivable that the external unit comprises the memory element of the control unit 46a. The cell system 10a has a virtual sensor unit 16a. The virtual sensor unit 16a is configured to determine, and in particular to calculate, a degradation state of the cell system 10a. The virtual sensor unit 16a is stored as a sensor program on the control unit 46a. The virtual sensor unit 16a and the control unit 46a are integrated into a single unit. The sensor program is executed on a processor of the control unit 46a. The sensor program is designated R.414133.

[0077] - 24 - A physical model is constructed. The physical model is a simulation model of the cell system 10a, which in particular describes the operation of the cell system 10a in a steady state. Preferably, the physical model is based on the finite element method or the finite volume method. The physical model exhibits, in particular, unique invertibility with respect to the degradation state, preferably the degradation parameter(s). It is conceivable that the physical model exhibits unique invertibility with respect to the degradation state, preferably the degradation parameter(s), at least by filtering out physically irrelevant results. The degradation state describes a power loss of the cell system 10a, in particular of the electrochemical cell 12a, over an operating period.Alternatively, the degradation parameter has a value of 1 in an unaged state of the cell system and preferably a value less than 1 for an aged state of the cell system. Combinations of these are also possible. The virtual sensor unit 16a is configured to describe a degradation state using at least one degradation parameter. In particular, it is conceivable that the virtual sensor unit 16a is configured to describe a degradation state using at least two degradation parameters. A degradation parameter is preferably a scalar quantity. In particular, it is conceivable that a degradation state is detected using several scalar degradation parameters or a vector degradation parameter. The degradation state is influenced by chemical degradation, material migration, and / or electrolyte degradation.In an operational setting, the cell system 10a, particularly the electrochemical cell 12a, exhibits continuous degradation, which can be time- and load-dependent. Additionally, the cell system 10a, and especially the electrochemical cell 12a, can lose its entire functionality due to "spontaneous failures".

[0078] Fig. 2 shows a schematic flowchart of a method according to the invention for determining the probability of failure of at least one cell system. In at least one determination step 14a, at least one degradation parameter of the at least one electrochemical cell 12a, which is in particular R.414133, is determined by means of a virtual sensor unit 16a.

[0079] - 25 -

[0080] To draw conclusions about the degradation state of at least one electrochemical cell 12a in a defined state, a failure probability of the at least one electrochemical cell 12a is determined in a determination step 14a as a function of the degradation parameter. The virtual sensor unit 16a preferably determines a degradation parameter of the cell system 10a, in particular of the electrochemical cell 12a, from at least one input variable of the cell system 10a, in particular of the electrochemical cell 12a, in a determination step 14a. The at least one input variable of the virtual sensor unit 16a has at least one set operating parameter of the cell system 10a. The at least one input variable of the virtual sensor unit 16a has at least one determined, in particular measured, operating parameter of the cell system 10a.The operating parameter is, for example, a queried load current, a temperature, in particular a gas temperature, a volume flow rate, an air utilization rate, a fuel utilization rate, a gas composition, a pressure, a pressure difference, or the like. The at least one input variable of the virtual sensor unit 16a has at least one performance characteristic of the cell system 10a. The performance characteristic is, for example, a cell voltage and / or a quantity of hydrogen, or the like. Using the virtual sensor unit 16a, the degradation parameter is determined in a determination step 14a as a function of the input variables, which provides information about the degradation state of the cell system 10a, in particular the electrochemical cell 12a. In a determination step 14a, a degradation parameter is calculated at a defined time as a function of the input variables using the sensor unit 16a.In investigation step 14a, a degradation parameter is determined at a current time 48a. In investigation step 14a, a conclusion is drawn about a degradation state at a current time 48a. In investigation step 14a, a degradation parameter is determined at defined intervals within a defined past time window. In particular, it is conceivable that investigation step 14a is carried out at defined intervals over the duration of the operation of the cell system 10a, whereby a degradation parameter is determined within a predefined time interval. Investigation step 14a is carried out at a regular time interval. Investigation step 14a is carried out during the operation of the cell system 10a. R.414133.

[0081] - 26 -

[0082] Determination step 14a is performed during an operational step, specifically during normal operation of the cell system 10a. In determination step 14a, the probability of failure of at least one electrochemical cell 12a is determined as a function of a degradation parameter. Determination step 14a identifies at least one state of increased failure probability for at least one electrochemical cell 12a. Specifically, it is conceivable that determination step 14a identifies an increased failure probability for a cell system 10a comprising one electrochemical cell 12a and at least one other electrochemical cell. Determination step 14a identifies a degradation parameter that provides information about the degradation state of the electrochemical cell 12a.In step 14a, a conclusion is drawn regarding a state of increased failure probability for at least one electrochemical cell 12a, depending on a current degradation parameter. In step 14a, the failure probability is determined by comparing the degradation parameter with a reference value. In step 14a, a state of increased failure probability occurs when the degradation parameter is less than or equal to, for example, a determined degradation quantity, or greater than or equal to, for example, a degradation rate, a reference value.

[0083] In particular, it is conceivable that in a determination step 14a, for a cell system 10a with several stacks of electrochemical cells 12a, the deviation of the degradation parameters, especially the degradation rates or degradation magnitudes, of the individual stacks is monitored. In determination step 14a, the monitoring of the degradation parameters of the individual stacks is carried out, and then, for example, the measured operating conditions for each stack are recorded individually. In particular, it is conceivable that, when operating several stacks, the difference in the degradation parameters is used as a measure of an increased probability of failure. It is checked whether the deviation of the degradation parameter, especially the degradation rate or degradation magnitude, of an individual stack from a mean value of all stacks exceeds a defined warning threshold. In particular, the maximum difference of the degradation parameters between R.414133 can also be used.

[0084] - 27 - two stacks serve as a criterion. In investigation step 14a, a warning of a state of increased failure probability is issued as soon as one of the two threshold types is exceeded.

[0085] In at least one evaluation step 18a, all degradation parameters, in particular all degradation rates 22a of the at least one electrochemical cell 12a, are calculated within a reference range 20a. In at least one evaluation step 18a, a time window width 52a for the progression of the degradation parameter 62a is defined. The progression of a degradation parameter 62a of the electrochemical cell 12a is visually represented in a trend diagram. The x-axis of the trend diagram represents a time 56a of operation of the electrochemical cell 12a, and the y-axis represents the degradation parameter 62a (see Fig. 3). In evaluation step 18a, a calculation method for the degradation rate 22a within the reference range 20a and / or within a current time window 54a is defined. For example, a least-squares-fit method for a straight line is used as the calculation method.The least-squares-fit method is used to calculate the slope of a straight line representing the degradation parameter 62a as a function of time 56a, which correlates with a degradation rate 22a, 58a. In evaluation step 18a, the reference range 20a is defined within the time window width 52a of the degradation parameter 62a's curve. The reference range 20a is defined at the start of operation of the cell system 10a or within a defined range. In evaluation step 18a, a degradation parameter, preferably a degradation rate 22a, is calculated at regular intervals for at least one electrochemical cell 12a. In evaluation step 18a, all degradation rates 22a within a reference range 20a are calculated using a sliding-window method. In evaluation step 18a, the mean and standard deviation of the reference range 20a are determined from the calculated degradation rates 22a (see Fig. 3).

[0086] In evaluation step 18a, the current degradation rate 58a, specifically of a last available window 54a, is determined. In evaluation step 18a, the current degradation rate 58a is calculated from a last available time window 54a with a time window width 52a. In evaluation step 18a R.414133

[0087] - 28 - The current degradation rate 58a is calculated using the specified calculation rule from a last available time window 54a of the time window width 52a, using the degradation parameters 62a calculated with the virtual sensor unit 16a from the last available time window 54a. A comparison step 24a is performed after an evaluation step 18a. In the comparison step 24a, the current degradation parameter, in particular a degradation rate 58a of the at least one electrochemical cell 12a, is compared with a distribution of the degradation parameters, in particular the degradation rates 22a of the at least one electrochemical cell 12a, from the reference range 20a. In the comparison step 24a, the current degradation rate 58a calculated in the evaluation step 18a is compared with a distribution of the degradation parameters, in particular the degradation rates 22a of the at least one electrochemical cell 12a, from the reference range 20a.In comparison step 24a, it is checked whether a current degradation rate 58a deviates significantly from one of the degradation rates 22a of the at least one electrochemical cell 12a from the reference range 20a. In comparison step 14a, a warning is issued as soon as the current degradation rate 58a, in particular at least one standard deviation of the current degradation rate 58a, is greater than a distribution of the degradation rates 22a of the at least one electrochemical cell 12a from the reference range 20a. The distribution of the degradation rates 22a of the at least one electrochemical cell 12a from the reference range 20a includes a safety factor.

[0088] Following comparison step 24a, an adaptation step 50a is performed depending on the warning message. In adaptation step 50a, at least one operating parameter of the cell system 10a, in particular of the at least one electrochemical cell 12a, is adjusted and / or at least one operator interaction is injected into the cell system 10a, in particular of the at least one electrochemical cell 12a, depending on a warning message determined in comparison step 24a. Adaptation step 50a is initiated by the control unit 46a. Operator interaction and / or an adjustment of the operating parameters is performed via the control unit 46a. Adaptation step 50a is initiated depending on a warning message determined in comparison step 24a for a state of increased failure probability. The warning message is displayed by the control unit 46a in an R.414133

[0089] - 29 -

[0090] Comparison step 24a is processed, and then an adjustment step 50a is automatically triggered. In adjustment step 50a, the operating parameters are changed to slow down degradation and, for example, prevent the cell voltage from falling below a minimum or maximum permissible level. In adjustment step 50a, if operator interaction occurs, an immediate check of the cell system by a technician and / or operator is triggered. This operator interaction can be performed either remotely (human-in-the-loop) or on-site (service), depending on the requirements. In particular, a controlled shutdown of the cell system, either automatically or through operator interaction, is also conceivable in adjustment step 50a.Depending on the warning, in an adjustment step 50a, the replacement of the stack and / or the entire cell system 10a can be planned within a given time window, for example in 8 weeks, in order to avoid a replacement that is too early or too late, for example in the case of a planned replacement.

[0091] Figures 4 to 9 show three further embodiments of the invention. The following descriptions and drawings are essentially limited to the differences between the embodiments, whereby, with regard to identically designated components, particularly those with the same reference numerals, reference may also be made to the drawings and / or the description of the other embodiments, especially Figures 1 to 3. To distinguish the embodiments, the letter "a" is appended to the reference numerals of the embodiment in Figures 1 to 3. In the embodiments of Figures 4 to 9, the letter "a" is replaced by the letters "b" to "d".

[0092] Fig. 4 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining the probability of failure of at least one cell system. The method comprises at least one learning step 28b, one evaluation step 18b, one comparison step 24b, and one adjustment step 50b. The learning step 28b, the evaluation step 18b, the comparison step 24b, and the adjustment step 50b are performed in a determination step 14b of the method. R.414133

[0093] - 30 -

[0094] In learning step 28b, a prediction algorithm 30b is calibrated based on an assumption about the course of the degradation parameter as a function of the set operating conditions of the cell system, in particular of the at least one electrochemical cell, and of degradation parameters determined with the virtual sensor unit. In learning step 28b, a prediction algorithm 30b is trained and / or calibrated, which makes it possible to predict a future degradation value 42b of the electrochemical cell 12b over a period in the future. For example, the prediction is made within a limited window of approximately eight weeks. The prediction algorithm 30b combines values ​​from the virtual sensor unit, for example, adjustable or measured operating parameters, up to a current time 48b with a prediction about the course of a degradation parameter.In particular, it is conceivable that by assuming future operating parameters as input parameters for the prediction algorithm 30b, a prediction for a degradation parameter at a defined point in the future or within a defined time window in the future can be determined. The prediction algorithm 30b can be specifically trained using a database with operating parameters of the cell system where the corresponding value of the degradation parameter lies within or outside the critical value range. The prediction algorithm 30b is trained using machine learning, specifically to execute a regression procedure for classifying identifiable states with an increased probability of failure of the cell system. The training of the prediction algorithm 30b is based, for example, on a random forest, a support vector machine, a neural network, or similar technologies.The prediction algorithm 30b is used in learning step 28b, in an operation of the cell system, to detect and, in particular, classify a state of increased failure probability in a cell system based on the operating parameters of the cell system recorded in an operation, for example by assigning it at least one degradation parameter.

[0095] In evaluation step 18b, a degradation rate 32b of at least one electrochemical cell at a defined point in the future 34b, in particular a last available window in the future, is calculated as a function of a prediction algorithm 30b. In evaluation step 18b, R.414133

[0096] - 31 - Depending on the prediction algorithm 30b, a degradation parameter 62b of the electrochemical cell 12b is calculated up to a defined time in the future 34b. The degradation parameter 62b is visually represented in a trend diagram. The x-axis of the trend diagram represents a time 56b of operation of the electrochemical cell 12b, and the y-axis represents the degradation parameter 62b. In evaluation step 18b, a degradation rate 32b is determined as a function of a trend determined by the prediction algorithm 30b at a defined time 34b. In evaluation step 18b, a degradation rate 32b is calculated in a last available window 54b of the degradation parameter 36b predicted by the prediction algorithm 30b (see Fig. 5).

[0097] In comparison step 24b, the degradation rate 32b of the at least one electrochemical cell at the defined time in the future 34b, in particular in a last available window 54b in the future, is compared with the distribution of at least one degradation rate 22b of the at least one electrochemical cell from a reference range 20b. In comparison step 24b, the degradation rate 32b of the at least one electrochemical cell calculated in evaluation step 18b at the defined time in the future 34b, in particular in a last available window 54b in the future, is compared with a distribution of the degradation parameters, in particular the degradation rates 22b of the at least one electrochemical cell, from the reference range 20b.In comparison step 24b, it is checked whether a degradation rate 32b of the at least one electrochemical cell at the defined time in the future 34b, in particular in a last available window 54b in the future, deviates significantly from one of the degradation rates 22b of the at least one electrochemical cell from the reference range 20b. In comparison step 24b, a warning is issued as soon as the degradation rate 32b of the at least one electrochemical cell at the defined time in the future 34b, in particular in a last available window in the future 54b, is greater than a distribution of the degradation rates 22b of the at least one electrochemical cell from the reference range R.414133 by at least one standard deviation of the degradation rate 32b.

[0098] - 32 -

[0099] 20b The distribution of the degradation rates 22b of the at least one electrochemical cell from the reference range 20b has a safety factor. The safety factor 3 is particularly preferred.

[0100] Fig. 6 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining the failure probability of at least one cell system. The method comprises at least one learning step 28c, one evaluation step 18c, one comparison step 24c, and one adjustment step 50c. The learning step 28c, the evaluation step 18c, the comparison step 24c, and the adjustment step 50c are performed in a determination step 14c of the method. In the evaluation step 18c, a degradation parameter 36c of the electrochemical cell is calculated for a defined time in the future 34c, depending on a prediction algorithm 30c. In the evaluation step 18c, a trend of the degradation parameter 36c up to a defined time in the future 34c is calculated, depending on the prediction algorithm 30c.The degradation value 36c is visually represented in a graph. The x-axis of the graph represents a time 56c of operation of the electrochemical cell 12c, and the y-axis represents the degradation value 62c (see Fig. 7). In evaluation step 18c, a distribution of the degradation value 36c is calculated within a defined time window and / or a defined time interval. For example, a distribution of the degradation value 36c at a defined time point can be calculated using a probabilistic prediction algorithm 60c. A probabilistic prediction algorithm 60c uses probability models to account for uncertainties in the predictions. Examples of probabilistic methods include Gaussian process regression, Bayesian networking, Monte Carlo simulation, and the hidden Markov model (see Fig. 7).

[0101] In comparison step 24c, the determined degradation value 36c of a defined point in the future 34c is compared with a previously defined critical degradation value 38c. In comparison step 24c, a degradation value 36c calculated in evaluation step 18c of a defined point in the future 34c is compared with a previously defined critical degradation value 38c. R.414133

[0102] - 33 - Degradation value 38c is compared. In comparison step 24c, a warning is issued as soon as the degradation value 36c calculated in evaluation step 18c for a defined point in the future exceeds or falls below a previously defined critical degradation value 38c. Alternatively, it is conceivable that in comparison step 24c, a distribution of the degradation value 36c determined in evaluation step 18c for a defined point in time and / or a defined time period is compared with a previously defined critical degradation value 38c. In this alternative embodiment, a warning is issued as soon as at least one value of the distribution of the degradation value 36c for a defined point in time and / or a defined time period falls below a previously defined critical degradation value 38c.In particular, a critical value range for the critical degradation parameter 38c is specified for at least one electrochemical cell, which is stored, for example, on the memory element of a control unit of the cell system 10. It is conceivable that the critical value range of the critical degradation parameter 38c is adjustable, for example, automatically by the control unit or manually by the operator via an input unit of the cell system. For example, it is conceivable that a database is stored on a memory element of the control unit, containing critical value ranges of the critical degradation parameter 38c for different operating parameters. In particular, it is conceivable that the prediction algorithm 30c can access the database to use extreme values ​​or worst-case scenarios of the electrochemical cell in a learning step 28c.

[0103] Fig. 8 shows a schematic flowchart of an alternative embodiment of a method according to the invention for determining the failure probability of at least one cell system. The method comprises at least one learning step 28d, one evaluation step 18d, one comparison step 24d, and one adjustment step 50d. The learning step 28d, the evaluation step 18d, the comparison step 24d, and the adjustment step 50d are performed in a determination step 14d of the method. The course of a degradation parameter 62d of the electrochemical cell 12d is visually represented in a trend diagram. The x-axis of the trend diagram shows a time 56d of operation of the electrochemical cell 12d, and the y-axis of the trend diagram shows the degradation parameter 62d. R.414133

[0104] - 34 -

[0105] In learning step 28d, a prediction algorithm 30d is calibrated based on the normal behavior of the degradation parameter, in particular the degradation quantity 62d. In learning step 28d, the prediction algorithm 30d is calibrated and / or trained using data from units or time intervals in which the cell system exhibits "normal" behavior. This normal behavior is calibrated in an evaluation step 18b based on long-term observation of a reference range and / or previously completed operating time. In particular, it is conceivable that normal behavior is further refined by specific temperature ranges, charge and discharge cycles, and typical usage patterns that do not cause significant deviations in cell performance.In learning step 28d, a uniform, linear decrease of the degradation quantity 62d is assumed as normal behavior, causing a constant degradation rate.

[0106] In evaluation step 18d, a future trend of the degradation parameter 48d is determined based on a prediction algorithm 30d of normal behavior of a degradation parameter 62d of at least one electrochemical cell. In evaluation step 18d, a smoothing parameter is selected to achieve a balance between the accuracy of the fit to the data and the avoidance of overfitting. The smoothing parameter is applied to the determination of the future trend of the degradation parameter 62d. In comparison step 24d, a current degradation parameter 40d of the electrochemical cell is compared with a predicted degradation parameter 42d for the current time 48d, determined by the prediction algorithm 30d in the past 44d.In comparison step 24d, a warning is issued as soon as the current degradation value 40d falls below or exceeds a previously determined future degradation value 42d at a current time. In comparison step 24d, a current degradation value 40d at a defined time and / or time period is compared with a distribution of the future degradation value 42d determined in the past 44d at a current time 48d. For example, the distribution is determined using a probabilistic prediction algorithm 60d. In R.414133.

[0107] - 35 - In this alternative embodiment, a warning is issued as soon as a current degradation quantity 40d falls below at least a distribution of the degradation quantity 42d of the future determined in the past 44d at a current time 48d.

Claims

R.414133 - 36 - Claims 1. Method for determining a probability of failure, in particular at least a state of increased probability of failure, of at least one cell system (10a), with at least one electrochemical cell (12a), wherein in at least one determination step (14a; 14b; 14c; 14d) at least one degradation parameter of the at least one electrochemical cell (12a), which in particular allows conclusions to be drawn about a degradation state of the at least one electrochemical cell (12a) in a defined state, is determined by means of a virtual sensor unit (16a), characterized in that in the at least one determination step (14a; 14b; 14c; 14d) a probability of failure of the at least one electrochemical cell (12a) is determined as a function of the degradation parameter.

2. Method according to claim 1, characterized in that in at least one evaluation step (18a; 18b) all degradation parameters, in particular all degradation rates (22a; 22b) of the at least one electrochemical cell (12a), are calculated in a reference range (20a; 20b).

3. Method according to one of the preceding claims, characterized in that in at least one comparison step (24a) the current degradation parameter, in particular a degradation rate (58a) of the at least one electrochemical cell (12a), is compared with a distribution of the degradation parameters, in particular the degradation rates (22a) of the at least one electrochemical cell (12a), from the reference range (20a). R.414133 - 37 - 4. Method according to claim 1, in particular according to claim 1, characterized in that in at least one learning step (28b; 28c) a prediction algorithm (30b; 30c) is calibrated on the basis of an assumption of the course of the degradation parameter as a function of the set operating conditions of the cell system, in particular of the at least one electrochemical cell, and of degradation parameters determined with the virtual sensor unit.

5. Method according to one of the preceding claims, characterized in that in at least one evaluation step (18b) a degradation rate (32b) of the at least one electrochemical cell at a defined time in the future (34b), in particular a last available window in the future, is calculated as a function of a prediction algorithm (30b).

6. Method according to claim 5, characterized in that in at least one comparison step (24b) the degradation rate (32b) of the at least one electrochemical cell, at the defined time in the future (34b), in particular in the last available window in the future, is compared with the distribution of the degradation rates (22b) of the at least one electrochemical cell, from the reference range (20b).

7. Method according to one of the preceding claims, characterized in that in at least one evaluation step (18c) a degradation quantity (36c) of the electrochemical cell of a defined time in the future (34c) is calculated as a function of a prediction algorithm (30c).

8. Method according to one of the preceding claims, characterized in that in at least one evaluation step (18c) a distribution of the degradation quantity (36c) is calculated in a defined time window and / or a defined time interval depending on a, in particular probabilistic, prediction algorithm (30c). R.414133 - 38 - 9. Method according to one of claims 7 or 8, characterized in that in at least one comparison step (24c) the determined degradation quantity (36c) of a defined point in time in the future (34c) and / or a distribution of the degradation quantity (36c) in a defined time window and / or a defined time period is compared with a previously determined critical degradation quantity (38c).

10. Method according to claim 1 , characterized in that in at least one learning step (28d) a prediction algorithm (30d) is calibrated depending on a normal behavior of the degradation parameter, in particular the degradation quantity (62d).

11. Method according to claim 10, characterized in that in a comparison step (24d) a current degradation quantity (40d) at a defined time point and / or a defined time period is compared with a distribution of the degradation quantity (42d) of the future determined in the past at a current time point.

12. Method according to claim 10, characterized in that in at least one comparison step (24d) a current degradation value (40d) of the electrochemical cell is compared with a predicted degradation value (42d) for the current time determined in the past (44d) by the prediction algorithm (30d).

13. Method according to one of the preceding claims, characterized in that in at least one adaptation step (50a; 50b; 50c; 50d) at least one operating parameter of the cell system (10a), in particular of the at least one electrochemical cell (12a), is adapted and / or at least one operator interaction is injected into the cell system (10a), in particular of the at least one electrochemical cell (12a), depending on a warning indication determined in a comparison step (24a; 24b; 24c; 24d).

14. Cell system (10a) for carrying out a method according to one of the preceding claims, comprising at least one electrochemical cell R.414133 - 39 - (12a), with at least one virtual sensor unit (16a) and with at least one control unit (46a).

Images