Method for determining a state of health and / or an internal resistance of a battery

Abstract

The present invention relates to a method for determining a state of health (SOH) and / or an internal resistance (R0) of a battery, comprising the following steps: stimulating the battery with current, measuring a resulting voltage and a resulting current due to the stimulation of the battery, and measuring a temperature (T) of the battery and a state of charge (SOC) of the battery, applying an FFT algorithm to the resulting voltage to determine a plurality of frequency components of the resulting voltage, and applying the FFT algorithm to the resulting current to determine a plurality of frequency components of the resulting current, calculating a plurality of frequency components of the impedance from the plurality of frequency components of the resulting voltage and the plurality of frequency components of the resulting current, and determining the state of health (SOH) and / or the internal resistance (R0) of the battery by means of a trained neural network (NN), wherein the plurality of frequency components of the impedance, the temperature (T) of the battery, and the state of charge (SOC) of the battery are used as input data (70) for the trained neural network (NN).

Description

[0001] Method for determining an aging state and / or an internal resistance of a battery

[0002] The present invention relates to a method for determining an aging state and / or an internal resistance of a battery, a determining device for determining an aging state and / or an internal resistance of a battery and a computer program product for carrying out such a method for determining an aging state and / or an internal resistance of a battery.

[0003] Electrochemical impedance spectroscopy (EIS) is well known in the art. Electrochemical impedance spectroscopy (EIS) is a powerful analytical technique used to investigate the behavior of electrochemical systems, such as batteries. In particular, it can also be used to measure lithium-ion batteries. Electrochemical impedance spectroscopy can be used to determine the state of health (SOH) and the internal resistance of a battery. In electrochemical impedance spectroscopy, a small sinusoidal voltage is applied to the battery, and the resulting current (which is sinusoidal) is measured. This process is repeated for different frequencies of the applied voltage. At each frequency, the voltage is divided by the current to obtain the impedance for that frequency.By analyzing the battery's impedance behavior, insights into its electrochemical properties and performance can be gained. This is known, for example, from WO 2023 / 041727 A1, where a non-invasive measurement of the battery's condition is achieved by injecting a sinusoidal alternating current into a battery.

[0004] However, electrochemical impedance spectroscopy can only be measured with special laboratory instruments. Therefore, these measurements are very complex and cannot be performed everywhere and in every environment. Against this background, one object of the present invention is to at least partially remedy the disadvantages described above. In particular, the object of the present invention is to be able to easily and universally determine the aging state and internal resistance of a battery.

[0005] The above object is achieved by a method having the features of claim 1, a determination device having the features of claim 14, and a computer program product having the features of claim 15. Further features and details of the invention emerge from the subclaims, the description, and the drawings. Features and details described in connection with the method according to the invention, the determination device according to the invention, and the computer program product according to the invention apply, and vice versa, so that reciprocal reference is or can always be made to the individual aspects of the invention with regard to the disclosure.

[0006] Accordingly, a method for determining the aging state and / or internal resistance of a battery is described. The method comprises the following steps:

[0007] Stimulating the battery with current,

[0008] Measuring a resulting voltage and a resulting current due to the stimulation of the battery, as well as measuring a temperature of the battery and a state of charge of the battery,

[0009] Applying an FFT algorithm to the resulting voltage to determine a plurality of frequency components of the resulting voltage and applying the FFT algorithm to the resulting current to determine a plurality of frequency components of the resulting current,

[0010] Calculating multiple frequency components of the impedance from the multiple frequency components of the resulting voltage and the multiple frequency components of the resulting current, and

[0011] Determining the aging state and / or the internal resistance of the battery by means of a trained neural network, wherein the plurality of frequency components of the impedance, the temperature of the battery and the state of charge of the battery are used as input data for the trained neural network.

[0012] By using a trained neural network, the battery's aging state and / or internal resistance can be determined without the need for complex electrochemical impedance spectroscopy measurements. Advantageously, the complex electrochemical impedance spectroscopy measurements can be performed beforehand when creating the dataset for training the neural network.

[0013] A method according to the invention can also be understood as a combination of a direct and an indirect measurement method. Thus, the method combines direct measurement steps, in which, for example, the temperature and state of charge of a real battery are recorded directly. This involves an actual, physical measurement of these parameters. Based on these direct measurement results, a series of indirect measurement steps are then carried out indirectly. These use the measured parameters as a starting point and, using the FFT algorithm, determine several frequency components of the resulting current. This allows the aging state to be further determined via the frequency components of the impedance and the resulting voltage.The aging state output in this way now corresponds to a real value in a specific operating situation of the battery, which would otherwise not be directly measurable or would only be possible with increased effort. This method can therefore be used, for example, as part of a control device in a battery management system (BMS). The determined aging state values ​​can be made available for monitoring connected consumers, such as a vehicle's electric drive motor.

[0014] A frequency component of a current, within the meaning of the invention, corresponds to a non-zero value for a specific frequency after a fast Fourier transform (FFT) of the current. Thus, multiple frequency components of a resulting current have multiple non-zero values ​​for different frequencies. The same applies to the frequency component of the resulting voltage and the frequency component of the impedance. Thus, multiple frequency components of the impedance have multiple non-zero values ​​for different frequencies.

[0015] For each frequency component of the resulting voltage and current, a frequency component of the impedance is calculated. Each calculated frequency component of the impedance has a real part, an imaginary part, and a frequency. Each frequency component of the impedance is calculated by dividing a frequency component of the resulting voltage by a frequency component of the resulting current. The number of multiple frequency components of the impedance is determined by the number of multiple frequency components of the resulting voltage or the number of multiple frequency components of the resulting current, where the number of multiple frequency components of the resulting voltage and the number of multiple frequency components of the resulting current are equal.

[0016] The frequency of a frequency component of the impedance can range from 0.5 Hz to 1000 Hz. The number of frequency components of the impedance can range from 3 to 100, from 5 to 50, or from 10 to 30.

[0017] The FFT algorithm (Fast Fourier Transformation algorithm) is an algorithm that can decompose a signal into individual spectral components. Here, the resulting voltage is decomposed into individual frequency components of the resulting voltage, and the resulting current is decomposed into individual frequency components of the resulting current.

[0018] The battery can, in particular, be a lithium-ion battery. The battery's aging state is referred to as its "state of health" (SOH). The trained neural network is created and trained as described below.

[0019] According to one embodiment of the method, the battery is stimulated with current pulses. The battery can be stimulated with current pulses using a charging and / or discharging current with a specific pulse width as a stimulus signal. A current pulse, within the meaning of the invention, is understood to be a time-limited electrical current flow whose frequency spectrum, in a fast Fourier transformation (FFT), has multiple values ​​other than zero for different frequencies. In particular, the current pulse according to the invention does not have a sinusoidal signal waveform or multiple superimposed sinusoidal signal waveforms. In contrast, a sinusoidal signal waveform has a frequency spectrum with a dominant frequency component.

[0020] According to a further embodiment of the method, the resulting voltage, the resulting current, the temperature, and / or the state of charge are measured for a specified period of time. The specified period of time can be in the range of 1 to 100 seconds, in the range of 20 to 80 seconds, or in the range of 30 to 60 seconds.

[0021] According to a further embodiment of the method, the FFT algorithm decomposes the resulting voltage into a plurality of frequency components of the resulting voltage and the resulting current into a plurality of frequency components of the resulting current, each frequency component being a sinusoidal signal with an amplitude and a phase shift relative to the resulting voltage or the resulting current. Each frequency component of the resulting voltage and each frequency component of the resulting current thus has a sinusoidal signal with an amplitude and a phase.

[0022] This has the advantage that, for example, no special hardware needs to be implemented to generate various sinusoidal signals that sample a frequency range. This advantageously saves a considerable amount of time compared to stepping through all frequencies with special hardware and leads to efficient acquisition of the entire frequency spectrum. According to a further embodiment of the method, each frequency component of the impedance from the plurality of frequency components of the impedance is calculated by dividing a frequency component of the resulting voltage from the plurality of frequency components of the resulting voltage by a frequency component of the resulting current from the plurality of frequency components of the resulting current.For 1 to n different frequency components, to determine the first frequency component of the impedance, the first frequency component of the resulting voltage is divided by the first frequency component of the resulting current, to determine the second frequency component of the impedance, the second frequency component of the resulting voltage is divided by the second frequency component of the resulting current, and so on, until finally, to determine the nth frequency component of the impedance, the nth frequency component of the resulting voltage is divided by the nth frequency component of the resulting current.

[0023] According to a further embodiment of the method, the trained neural network was trained using data from electrochemical impedance spectroscopy measurements. This advantageously allows the advantages of electrochemical impedance spectroscopy measurements to be retained. In particular, the frequency components of the impedance determined in the electrochemical impedance spectroscopy measurements are partly the same as those determined using the FFT algorithm.

[0024] Furthermore, the neural network is trained for use in a method according to the invention for determining the aging state and / or internal resistance of a battery. The neural network is trained through the following steps:

[0025] Providing at least one battery in different states,

[0026] Performing an electrochemical impedance spectroscopy measurement at a number of different frequencies at each of the different states of the battery,

[0027] Creating a data set from all electrochemical impedance spectroscopy measurements performed in all different states of the battery, and

[0028] Training the neural network with the dataset.

[0029] Advantageously, the neural network trained in this way can be used in a method according to the invention for determining an aging state and / or an internal resistance of a battery. In particular, the at least one battery used during training of the neural network is identical to the battery to which the method according to the invention for determining an aging state and / or an internal resistance of a battery is applied.

[0030] Alternatively, instead of a single battery, multiple batteries can be measured and used to train the neural network. The state of the at least one battery is determined by various parameters of the at least one battery. Providing at least one battery in different states therefore means changing several parameters of the at least one battery, so that the at least one battery can be measured in many different states.

[0031] The first three steps of training can be performed alternately. This means that the battery is first placed in a specific state, then an electrochemical impedance spectroscopy measurement is performed at a specific number of frequencies, and the results are then stored in the dataset. The battery is then placed in the next specific state, and the electrochemical impedance spectroscopy measurement is repeated for all frequencies. This process is repeated with additional specific battery states until the dataset is complete.

[0032] The number of frequencies in an electrochemical impedance spectroscopy measurement can range from 3 to 100, from 20 to 80, or from 40 to 60. The frequencies can be in a frequency range from 0.2 Hz to 1434 Hz.

[0033] According to one embodiment of the neural network training, the various battery states are determined by a battery charge level, a battery temperature, and / or a battery aging level. Advantageously, all relevant battery states can be represented by varying the parameters battery charge level, battery temperature, and / or battery aging level.

[0034] The battery charge level parameter can be in the range 0% to 100%, 10% to 90%, or 20% to 80%. For example, 1 to 10, 2 to 8, or 4 to 6 charge levels can be selected. Specifically, the charge levels available are 0%, 10%, 40%, 70%, and 100%.

[0035] The battery temperature parameter can be set within the range of 1°C to 60°C, 5°C to 50°C, or 10°C to 45°C. For example, 1 to 5, 2 to 4, or 3 temperature values ​​can be selected. Specifically, the temperatures 10°C, 25°C, and 45°C can be selected.

[0036] Furthermore, the battery aging state parameter can be in the range from 70% to 100%, from 75% to 100%, or from 80% to 100%. For example, 1 to 10, 2 to 8, or 4 to 6 battery aging states can be selected. Specifically, the aging states 80%, 85%, 90%, 95%, and 100% can be selected.

[0037] According to a further embodiment of the training of the neural network, the following steps are performed when performing an electrochemical impedance spectroscopy measurement:

[0038] Applying a sinusoidal voltage with a specific frequency to the at least one battery,

[0039] Measuring the resulting sinusoidal current, and

[0040] Calculate a frequency component of the impedance by dividing the sinusoidal voltage by the sinusoidal current, repeating the steps according to the number of different frequencies.

[0041] If all frequency components of the impedance are plotted in polar coordinates for a specific frequency range, the result is a Nyquist diagram. The specific frequency is changed repeatedly until all different frequencies have been used, according to the number of different frequencies.

[0042] According to a further embodiment of the training of the neural network, the data set has the following data for each individual entry: real part of the impedance, imaginary part of the impedance, frequency, state of charge of the at least one battery, temperature of the at least one battery, aging state of the at least one battery, and internal resistance of the at least one battery.

[0043] In addition, the data set can also contain the frequency points, ie the number of the frequency used when the frequencies are numbered consecutively.

[0044] The aging state can be known from initialization. The internal resistance of at least one battery can be calculated directly from the high-frequency impedance of the electrochemical impedance spectroscopy measurement.

[0045] According to a further embodiment of the training of the neural network, input data for the neural network and output data for the neural network are determined from the data set for training the neural network, wherein various frequency components of an impedance, an associated state of charge of the battery and an associated temperature of the battery are used as input data of the neural network and an associated aging state and an associated internal resistance of the battery are used as output data of the neural network.

[0046] In particular, the different frequency components of an impedance are the same for each electrochemical impedance spectroscopy measurement.

[0047] According to another embodiment of neural network training, the neural network is an artificial neural network. The artificial neural network is a type of machine learning algorithm built according to the principles of neural organization. The artificial neural network is based on a collection of interconnected nodes. The nodes can be grouped into layers, with the different layers performing different transformations on the input data. Each node has different adjustable weights for learning and making predictions based on this input data. The first layer is called the input layer, the last layer is the output layer, and the layers in between are called hidden layers. The data signals pass through the artificial neural network from input to output.The input layer receives the input data. The output layer outputs the output data. Based on the input data, the neural network uses the hidden layers for various calculations, makes predictions, and feeds the results to the output layer.

[0048] According to another embodiment of training the neural network, a backpropagation algorithm is used to train the neural network. Advantageously, the backpropagation algorithm can be used to correct the weights of the neural network.

[0049] Furthermore, a determination device for determining an aging state and / or an internal resistance of a battery is described. The determination device comprises the following: a stimulation module for stimulating the battery with current, a measuring module for measuring a resulting voltage and a resulting current due to the stimulation of the battery, as well as for measuring a temperature of the battery and a state of charge of the battery, an application module for applying an FFT algorithm to the resulting voltage to determine several frequency components of the resulting voltage and for applying the FFT algorithm to the resulting current to determine several frequency components of the resulting current, a calculation module for calculating several frequency components of the impedance from the several frequency components of the resulting voltage and the several frequency components of the resulting current,and a determination module for determining the aging state and / or the internal resistance of the battery by means of a trained neural network, wherein the plurality of frequency components of the impedance, the temperature of the battery, and the state of charge of the battery can be used as input data for the trained neural network, wherein the stimulation module, the measuring module, the application module, the calculation module, and / or the determination module are designed in particular to carry out a method according to the invention for determining an aging state and / or an internal resistance of a battery.

[0050] Thus, the determination device provides the same advantages as those described in detail with reference to a method according to the invention for determining an aging state and / or an internal resistance of a battery.

[0051] Furthermore, the present invention provides a computer program product comprising instructions which, when executed by a determination device according to the invention, cause the device to perform the steps of a method according to the invention for determining an aging state and / or an internal resistance of a battery. Thus, a computer program product according to the invention also provides the same advantages as those explained in detail with reference to a method according to the invention for determining an aging state and / or an internal resistance of a battery.

[0052] Further possible implementations of the invention also include combinations of features described above or below with respect to the exemplary embodiments that are not explicitly mentioned. In this case, the person skilled in the art will also add individual aspects as improvements or additions to the respective basic form of the invention.

[0053] Further advantages, features, and details of the invention will become apparent from the following description, in which exemplary embodiments are described in detail with reference to the drawings. They show:

[0054] Fig. 1 is a schematic view of a determining device for determining an aging state and / or an internal resistance of a battery;

[0055] Fig. 2 is a flowchart showing the inputs and outputs of a trained neural network;

[0056] Fig. 3 is a flowchart of a neural network training process; Fig. 4 is a graphical representation of the influence of a battery's aging state on the impedance;

[0057] Fig. 5 is a schematic diagram for training a neural network; and

[0058] Fig. 6 shows a flowchart for training a neural network.

[0059] Fig. 1 shows a schematic view of a determination device 10 for determining an aging state SOH and / or an internal resistance R0 of a battery. The determination device 10 comprises a stimulation module 20, a measuring module 30, an application module 40, a calculation module 50, and a determination module 60. The determination device 10 is designed in particular to carry out a method according to the invention for determining an aging state SOH and / or an internal resistance R0 of a battery. The method comprises the steps described below.

[0060] In a first step, the battery is stimulated with current. The battery is stimulated using current pulses. This step can be performed using the stimulation module 20.

[0061] In a second step, a resulting voltage and current are measured due to the battery stimulation. A battery temperature T and a battery state of charge SOC are also measured. This step can be performed using the measuring module 30.

[0062] In a third step, an FFT algorithm is applied to the resulting voltage to determine several frequency components of the resulting voltage. Furthermore, the FFT algorithm is applied to the resulting current to determine several frequency components of the resulting current. This step can be performed using application module 40.

[0063] In a fourth step, several frequency components of the impedance are calculated from the several frequency components of the resulting voltage and the several frequency components of the resulting current. This step can be performed using the calculation module 50. In a fifth step, the aging state SOH and / or the internal resistance RO of the battery are determined using a trained neural network NN. The several frequency components of the impedance, the temperature T of the battery, and the state of charge SOC of the battery are used as input data 70 for the trained neural network NN. This step can be performed using the determination module 60.

[0064] The trained neural network (NN) can be trained using data from electrochemical impedance spectroscopy measurements. Accordingly, the aging state and internal resistance of the battery can be determined easily and reliably.

[0065] Furthermore, the resulting voltage, resulting current, temperature T, and / or state of charge SOC can be measured for a specified period of time. This ensures that the data is recorded correctly.

[0066] Fig. 2 shows a flowchart of the inputs and outputs of a trained neural network. It shows the input data 70 that enters the neural network NN, the neural network NN, and the output data 72 that is output by the neural network. The input data 70 are the frequency components Z1, Z2, ... Zn of the impedance, the measured temperature T of the battery, and the measured state of charge SOC of the battery. The output data 72 are the aging state SOH and / or the internal resistance RO of the battery.

[0067] The FFT algorithm decomposes the resulting voltage into multiple frequency components of the resulting voltage and the resulting current into multiple frequency components of the resulting current. Each frequency component of the resulting voltage and each frequency component of the resulting current comprises a sinusoidal signal with an amplitude and a phase. Furthermore, each frequency component of the resulting voltage exhibits a phase shift with respect to the resulting voltage. Likewise, each frequency component of the resulting current exhibits a phase shift with respect to the resulting current.Each frequency component of the impedance among the multiple frequency components of the impedance is calculated by dividing a frequency component of the resulting voltage among the multiple frequency components of the resulting voltage by a frequency component of the resulting current among the multiple frequency components of the resulting current. Specifically, the nth frequency component of the impedance Zn is calculated by dividing the nth frequency component of the resulting voltage Iln by the nth frequency component of the resulting current In.

[0068] Fig. 3 shows a flow chart of a training of the neural network NN .

[0069] In a first step S1, at least one battery is provided in different states. The different states of the battery can be determined, for example, by a battery state of charge (SOC), a battery temperature (T), and / or a battery aging state (SOH).

[0070] In a second step S2, an electrochemical impedance spectroscopy measurement is performed at a number of different frequencies for each of the different states of the battery. In an electrochemical impedance spectroscopy measurement, in particular, a sinusoidal voltage with a specific frequency is applied to the at least one battery. The resulting sinusoidal current is then measured. The frequency component of the impedance is calculated by dividing the sinusoidal voltage by the sinusoidal current. These steps of the electrochemical impedance spectroscopy measurement are repeated for all different frequencies.

[0071] In a third step S3, a dataset is created from all electrochemical impedance spectroscopy measurements performed in all different battery states. Such a dataset can contain the following data for each individual entry: real part of the impedance, imaginary part of the impedance, frequency, state of charge (SOC) of the at least one battery, temperature (T) of the at least one battery, state of aging (SOH) of the at least one battery, and internal resistance (RO) of the at least one battery. In a fourth step S4, the neural network (NN) is trained with the dataset.

[0072] The described steps S1 to S4 can be carried out alternately.

[0073] Fig. 4 shows a graphical representation of the influence of the aging state (SOH) of a battery on the impedance. The impedance is shown as a function of frequency in a Nyquist diagram. The imaginary part of the impedance Z is plotted against the real part of the impedance Z. This example shows how the battery's impedance changes as the battery ages. The battery's state changes when the aging state (SOH) changes. All electrochemical impedance spectroscopy measurements shown in Fig. 4 were performed at a specific state of charge (SOC) and a specific battery temperature (T).

[0074] Fig. 5 shows a schematic diagram for training a neural network. To train the neural network (NN), input data 80 for the neural network (NN) and output data 82 for the neural network (NN) are determined from the data set created using the electrochemical impedance spectroscopy measurements. Various frequency components of an impedance Z1 to Zn, a corresponding state of charge (SOC) of the battery, and a corresponding temperature T of the battery are used as input data 80 of the neural network (NN). A corresponding aging state (SOH) and a corresponding internal resistance (R0) of the battery are used as output data 82 of the neural network (NN).

[0075] The neural network NN can in particular be an artificial neural network ANN.

[0076] Fig. 6 shows a flowchart for training a neural network (NN). The neural network (NN) can be trained using a backpropagation algorithm. The neural network (NN) is trained by processing example data sets with known input data 80 and output data 82 from the neural network (NN). During this processing, the weights within the neural network (NN) are adjusted. Such a backpropagation algorithm, for example, has the following steps. In step S10, the weights for all nodes and for all layers in the neural network are initialized. The weights can be randomly selected.

[0077] In a step S20, all outputs for each node from the input layer through the hidden layers to the output layer are calculated.

[0078] In a step S30, the output error between the output data 82 and the calculated output of the output layer is calculated.

[0079] In a step S40, it is compared whether the output error is below a specified value.

[0080] Step S50 is executed if the condition from step 40 is met, i.e., the output error is smaller than the specified value. In this case, the neural network is trained.

[0081] Step S60 is executed if the condition from step 40 is not met, i.e., the output error is greater than the specified value. In this case, the weights of the output layer are adjusted.

[0082] In a step S70, the weights of the hidden layers are further adjusted.

[0083] In step S80, the weights of all layers are adjusted. Step 20 is then repeated.

[0084] List of reference symbols

[0085] 10 Determination device

[0086] 20 Stimulation module

[0087] 30 measuring module

[0088] 40 Application module

[0089] 50 Calculation module

[0090] 60 Determination module

[0091] 70 input data

[0092] 72 initial data

[0093] 80 input data

[0094] 82 output data

[0095] SOH state of health

[0096] SOC state of charge

[0097] R0 internal resistance

[0098] T Temperature

[0099] NN neural network

[0100] ANN artificial neural network

[0101] Z Impedance

Claims

Patent claims 1 . Method for determining a state of aging (SOH) and / or an internal resistance (RO) of a battery, characterized by the steps: Stimulating the battery with current, Measuring a resulting voltage and a resulting current due to the stimulation of the battery, as well as measuring a temperature (T) of the battery and a state of charge (SOC) of the battery, Applying an FFT algorithm to the resulting voltage to determine a plurality of frequency components of the resulting voltage and applying the FFT algorithm to the resulting current to determine a plurality of frequency components of the resulting current, Calculating multiple frequency components of the impedance from the multiple frequency components of the resulting voltage and the multiple frequency components of the resulting current, and Determining the state of aging (SOH) and / or the internal resistance (RO) of the battery by means of a trained neural network (NN), wherein the plurality of frequency components of the impedance, the temperature (T) of the battery and the state of charge (SOC) of the battery are used as input data (70) for the trained neural network (NN).

2. Method according to claim 1, characterized in that the stimulation of the battery is carried out with current pulses.

3. Method according to claim 1 or 2, characterized in that the resulting voltage, the resulting current, the temperature (T) and / or the state of charge (SOC) are measured for a specified period of time.

4. Method according to one of the preceding claims, characterized in that the FFT algorithm decomposes the resulting voltage into several frequency components of the resulting voltage and the resulting current into several frequency components of the resulting current, each frequency component being a sinusoidal signal with an amplitude and a phase shift with respect to the resulting voltage or current.

5. Method according to one of the preceding claims, characterized in that each frequency component of the impedance of the plurality of frequency components of the impedance is calculated by dividing a frequency component of the resulting voltage of the plurality of frequency components of the resulting voltage by a frequency component of the resulting current of the plurality of frequency components of the resulting current.

6. Method according to one of the preceding claims, characterized in that the trained neural network (NN) was trained using data from electrochemical impedance spectroscopy measurements.

7. Method according to one of claims 1 to 6, characterized in that the neural network (NN) is trained by the following steps: Providing at least one battery in different states, Performing an electrochemical impedance spectroscopy measurement at a number of different frequencies at each of the different states of the battery, Creating a data set from all electrochemical impedance spectroscopy measurements performed in all different states of the battery, and Training the neural network (NN) with the dataset.

8. The method according to claim 7, characterized in that the different states of the battery are determined by a state of charge of the battery (SOC), by a temperature (T) of the battery and / or by an aging state (SOH) of the battery.

9. Method according to claim 7 or 8, characterized in that when carrying out an electrochemical impedance spectroscopy measurement, the following steps are carried out: Applying a sinusoidal voltage with a specific frequency to the at least one battery, Measuring the resulting sinusoidal current, and Calculate a frequency component of the impedance by dividing the sinusoidal voltage by the sinusoidal current, repeating the steps according to the number of different frequencies.

10. Method according to one of claims 7 to 9, characterized in that the data record for each individual entry has the following data: Real part of the impedance, Imaginary part of the impedance, Frequency, State of charge (SOC) of at least one battery, Temperature (T) of at least one battery, State of aging (SOH) of at least one battery, and Internal resistance (R0) of at least one battery.

11. Method according to one of claims 7 to 10, characterized in that for training the neural network (NN) input data (80) for the neural network (NN) and output data (82) for the neural network (NN) are determined from the data set, wherein as input data (80) of the neural network (NN) various frequency components of an impedance, an associated state of charge (SOC) of the battery and an associated temperature (T) of the battery are used and as output data (82) of the neural network (NN) an associated state of aging (SOH) and an associated internal resistance (R0) of the battery are used.

12. Method according to one of claims 7 to 11, characterized in that the neural network (NN) is an artificial neural network (ANN).

13. Method according to one of claims 7 to 12, characterized in that a backpropagation algorithm is used to train the neural network (NN).

14. Determination device (10) for determining an aging state (SOH) and / or an internal resistance (R0) of a battery, characterized by a stimulation module (20) for stimulating the battery with current, a measuring module (30) for measuring a resulting voltage and a resulting current due to the stimulation of the battery, as well as for measuring a temperature (T) of the battery and a state of charge (SOC) of the battery, an application module (40) for applying an FFT algorithm to the resulting voltage to determine several frequency components of the resulting voltage and for applying the FFT algorithm to the resulting current to determine several frequency components of the resulting current, a calculation module (50) for calculating several frequency components of the impedance from the several frequency components of the resulting voltage and the several frequency components of the resulting current,and a determination module (60) for determining the state of aging (SOH) and / or the internal resistance (R0) of the battery by means of a trained neural network (NN), wherein the plurality of frequency components of the impedance, the temperature (T) of the battery and the state of charge (SOC) of the battery can be used as input data (70) for the trained neural network (NN), wherein the stimulation module (20), the measuring module (30), the application module (40), the calculation module (50) and / or the determination module (60) are designed in particular to carry out a method having the features of one of claims 1 to 6.

15. A computer program product comprising instructions which, when executed by a determining device (10) having the features of claim 14, cause the device (10) to carry out the steps of a method having the features of any one of claims 1 to 6.

Images