DE102020212193A1 - Method and device for monitoring at least one battery cell of a battery - Google Patents

Abstract

The invention relates to a method for monitoring at least one battery cell (21-x) of a battery (20), a voltage and a current being recorded in a time-resolved manner at the at least one battery cell (21-x) by means of at least one sensor system (2-x), wherein a battery cell condition (22-x) is estimated from the detected voltage and the detected current by means of a battery cell model (10-x), wherein the estimated battery cell condition (22-x) is monitored, and wherein at least the provision of the battery cell model (10-x ) and the battery cell state (22-x) of the at least one battery cell (21-x) is estimated by means of at least one field-programmable gate array (4). The invention also relates to a device (1) for monitoring at least one battery cell (21-x) of a battery (20).

Description

The invention relates to a method and a device for monitoring at least one battery cell of a battery.

As electromobility progresses and the number of electric vehicles and hybrid vehicles increases, more and more batteries, especially high-voltage batteries, are being used. The batteries usually include multiple battery cells. Battery cells are subject to cyclic and calendar aging. Battery operation is characterized by continuous discharging and charging. During this cyclization, chemical components at boundary layers interact with one another in the course of charge exchange. These interactions are expressed in decomposition reactions that can change the boundary layers of the cell, but also the electrolyte and reduce the number of conductive ions. Decomposition reactions can also occur independently of a cyclization, which show their effects over the course of time, i.e. on a calendar basis. As a result, these decomposition reactions lead to an aged battery cell, which has changed electrical time constants with regard to a terminal behavior.

A particularly serious malfunction of a battery is thermal runaway. The thermal runaway of battery cells manifests itself in a strong exothermic cell reaction, which results in very high heat generation and can damage an entire battery pack and the parts of a vehicle surrounding the battery. Accurate prediction of thermal runaway is therefore desirable.

Field-programmable gate arrays are known, which are also referred to as programmable logic gates (Field Programmable Gate Array, FPGA). The field-programmable gate arrays are characterized in that a logical array of gates can be programmed in the field. In other words, a logic circuit can be programmed in hardware. Due to a parallel structure, the programmable gate arrays can process large amounts of data in a short time in a very energy-efficient manner. Field-programmable gate arrays, which can also be partially reconfigured during runtime, have recently been available.

The invention is based on the object of creating a method and a device for monitoring at least one battery cell of a battery, with which in particular a battery cell state can be better estimated and monitored.

The object is achieved according to the invention by a method having the features of patent claim 1 and a device having the features of patent claim 10 . Advantageous configurations of the invention result from the dependent claims.

In particular, a method for monitoring at least one battery cell of a battery is provided, a voltage and a current being detected in a time-resolved manner at the at least one battery cell using at least one sensor system, a battery cell state being estimated from the detected voltage and the detected current using a battery cell model , wherein the estimated battery cell condition is monitored, and wherein at least the providing of the battery cell model and the estimation of the battery cell condition of the at least one battery cell is performed by means of at least one field-programmable gate array.

Furthermore, in particular a device for monitoring at least one battery cell of a battery is created, comprising at least one sensor system set up for the time-resolved detection of a voltage and a current at the at least one battery cell, a computing device with a memory, and at least one field-programmable gate array, the at least one field-programmable gate array is set up to estimate a battery cell state of the at least one battery cell using a battery cell model from the detected voltage and the detected current, and wherein the at least one field-programmable gate array and/or the computing device are set up to monitor the estimated battery cell state.

The method and the device make it possible to estimate and monitor a battery cell state of at least one battery cell of a battery in a time-resolved manner. In this case, the battery cell state is estimated by means of at least one field-programmable gate array. For this purpose, the at least one field-programmable gate array uses or provides a battery cell model of the battery cell to be monitored. In particular, the at least one field-programmable gate array is configured to provide the battery cell model and to estimate the battery cell state based on a voltage detected by a sensor and a detected current using the battery cell model. In particular, the battery cell model represents a digital image of the battery cell, in which the behavior of the battery cell over time can be mapped and monitored. The provision of the battery cell model includes in particular the provision and application of a functionality of the Bat tery cell model, so that a battery cell state can be estimated by means of the field-programmable gate array based on the time-resolved detected voltage and the time-resolved current detected.

An advantage of using a field-programmable gate array is that a signal evaluation, ie an evaluation of the detected voltage and the detected current, can be carried out without a processing-related delay, in particular in real time, with a high sampling rate (MHz). This makes it possible to estimate the battery cell condition without delay, especially in real time. As a result, monitoring of the battery cell can be carried out in an improved manner. In particular, it is then possible to react particularly quickly to possible faults in the battery cell.

Due to the parallel signal processing of a field-programmable gate field, this can also process high-frequency (-MHz) sampled alternating variables, such as those that occur at the terminals of the battery cells in a vehicle due to various consumers (e.g. drive converters, low-voltage DC converters, etc.). Due to the high-frequency sampling, sampling or aliasing effects can be reduced as disturbance variables in the signal. Any noise components of the sensors can be digitally filtered in the field-programmable gate array, so that a very high signal-to-noise ratio can be provided for further digital processing. In comparison, a microcontroller (e.g. DSP) is significantly less suitable for processing high-frequency signals, since the recorded values of the DSP have to be transported and processed by the internal analog-digital converter (ADC) to the processor core, which leads to delay-related and preprocessing-related delays in evaluation. With a field-programmable gate array, the sensed values of the ADC are instead available to the logic units immediately, that is, in real time. Inexpensive ADCs that can generate a high-frequency digital data stream (bit stream), e.g. delta-sigma converters, can also be evaluated with field-programmable gate arrays without further measures.

In particular, the field-programmable gate array is a run-time partially (i.e., partially) reconfigurable field-programmable gate array. The reconfiguration takes place in particular by way of partial reconfiguration, in which only part of the gate array is reconfigured or reprogrammed at runtime, but other parts of the gate array retain their respective configuration or programming. The reconfiguration is carried out in particular by means of a configuration control which carries out and monitors the reconfiguration. The configuration control can also be provided by means of the computing device of the device. Methods for automatic code generation can be used to reconfigure the at least one field-programmable gate field, such as the “Xilinx System Generator for DSP”, which can automatically carry out both source code modeling and a functional analysis. For example, the "System Generator for DSP" enables the modeling of FPGA functions in a Matlab Simulink environment so that the properties can be tested against other Simulink function blocks. The automatic code generation ensures that the field-programmable gate array behaves identically to Simulink modeling.

A configuration of a field-programmable gate array includes, in particular, instructions with which the field-programmable gate array can be programmed in such a way that a specific functionality for data processing can be provided using the programmed (or configured or reconfigured) part of the field-programmable gate array.

The monitoring of the battery cell state includes in particular a comparison of the battery cell state with a target battery cell state. If the comparison shows that the target battery cell state is not (or no longer) present, an error reaction can take place. An error reaction can include, for example, issuing a warning message and/or switching off the affected battery cell. In principle, other error reactions can also take place, for example a power consumption can be reduced. In particular, after an error has been detected, the control device generates at least one corresponding control and/or warning signal and makes it available or outputs it.

As part of the estimation of the battery cell state, time constants of the battery cell can be estimated on the basis of the battery cell model. Since several physical and chemical processes take place in parallel with different time constants in a battery cell, estimating the time constants allows the condition of a battery cell and its aging to be determined. If power is drawn from a battery, the boundary layers are first depleted (smaller time constant), and only later do diffusion processes begin to play a role (larger time constant). The estimated time constants can be compared to target time constants and an error message can be generated in the event of a deviation generated and output, or another error reaction can take place. In particular, the time constants form parameters of the battery cell model and at the same time represent in particular a battery cell state. For estimating the time constants, provision can be made, for example, for estimating the time constants by evaluating prediction errors. With the help of the battery cell model, predictions for the values of the voltages and currents at the battery cell are calculated. This is done based on values of current sampling steps for subsequent sampling steps. The calculated predictions are continuously compared with the voltages and currents recorded on the battery cell, from which a prediction error can be determined as the difference between the calculated and recorded values. The time constants can then be continuously determined, for example, with the aid of a state observer known from control engineering, which simulates the battery cell model as a reference system, and the gradient descent method, by minimizing a prediction error for the voltage and the current. The time constants estimated by the observer can then be compared, for example, with predetermined setpoint time constants or with time constants for known battery cell states for monitoring purposes.

A battery cell status can include electrical parameters (capacitance, resistance, inductance, current, voltage) as well as other physical or chemical parameters (e.g. temperature).

The monitoring can be performed both in the computing device and in the field-programmable gate array. In particular, however, it is provided that the estimation is carried out in the field-programmable gate array, while the monitoring is carried out by means of the computing device, ie the computing device monitors the battery cell state(s) estimated by the field-programmable gate array.

Parts of the device, in particular the computing device, can be designed individually or combined as a combination of hardware and software, for example as program code that is executed on a microcontroller or microprocessor.

One embodiment provides that the battery cell model is designed as an equivalent circuit diagram, with at least values for capacitances and/or resistances of the equivalent circuit diagram being estimated and monitored as the battery cell state. The equivalent circuit is in particular an RC equivalent circuit. In the simplest case, the battery cell can be represented as a capacitance with a resistance connected in parallel. By evaluating a time course of the detected voltage and the detected current, this capacitance and this resistance can be estimated using the equivalent circuit diagram, for example using adaptation and/or estimation methods that are known per se. In particular, the values for the capacitance and the resistance can be adjusted iteratively until the equivalent circuit diagram simulates a behavior of voltage and current observed at the battery cell. The estimated capacitance and resistance values can then be compared to target values and/or target ranges and/or threshold values. If there is a deviation from a target value and/or a target range is left and/or a threshold value is exceeded or not reached, an error message is generated and output and/or the battery cell is switched off. In principle, other error reactions can also take place, for example a power consumption can be reduced. In particular, the equivalent circuit diagram can also include one or more resonant circuits with different time constants. The time constants correspond to the time constants on which (different) chemical processes take place within the battery cell. Estimating the capacitances and resistances (and possibly other electrical variables, e.g. inductances) of the oscillating circuits then enables the associated time constants to be estimated.

In one embodiment it is provided that the battery cell model is provided by means of a machine learning model. As a result, the behavior of the battery cell can be mapped in an improved manner. In particular, the machine learning model also allows the mapping of rare behaviors, provided they are taken into account during training. The machine learning model is in particular a trained or fully parameterized machine learning model. The machine learning model is then embodied in the at least one field-programmable gate array. The machine learning model can be designed in particular as an (artificial) neural network. The battery cell model is provided by means of the machine learning model. The machine learning model can be used, for example, to estimate one or more capacitances and one or more resistances of an equivalent circuit diagram of the battery cell represented by the machine learning model, based on the voltage recorded in a time-resolved manner and the current recorded in a time-resolved manner. The machine learning model is trained or parameterized in particular using training data for which a basic truth is known. In particular, the training data form both an error-free (normal) behavior as well as incorrect behavior of the battery cell.

A further embodiment provides that the machine learning model can be and/or is adapted during operation by at least partially reconfiguring the field-programmable gate array. As a result, the battery cell model can be retrained over time and/or made more complex. This enables an improved estimation of the battery cell condition as well as an adjustment of the battery cell model as the battery cell ages. Furthermore, the battery cell model can also be brought up to date at a later point in time (“update”), for example, in order to take into account new findings and/or other, in particular expanded, training data.

In one embodiment, it is provided that a time behavior of the estimated battery cell state of the at least one battery cell and/or the detected voltage and/or the detected current is evaluated and monitored for monitoring. As a result, an error behavior that manifests itself over time can be recognized and made usable for monitoring. Provision can be made for spectroscopic techniques to be used here, which enable analysis and/or monitoring of individual frequencies and/or frequency ranges. This makes it possible, in particular, to also determine a type of error, so that operating strategies and/or error reactions differentiated according to an error type can be implemented as a reaction to an error that has been identified.

In a further embodiment, it is provided that the time behavior of the battery cell state of the at least one battery cell and/or the detected voltage and/or the detected current is compared to empirically determined error curves, with an error being recognized if a match is found. As a result, an empirically determined wealth of experience can be accessed, which can be used for fault monitoring. If, for example, it has been determined empirically which behavior or which battery states a battery cell shows in the time immediately before a thermal runaway occurs, a future battery cell state (e.g. "thermal runaway in 40 seconds") can also be estimated if such a behavior occurs in the monitored battery cell . This enables the battery cell to be switched off and/or further measures to be taken before the expected error has occurred. Operation of the battery cell or the battery can be made more reliable and safer as a result. In particular, this enables a predictive operating strategy and/or maintenance of battery cells and the battery.

If, for example, time constants are estimated as part of the estimation of the battery cell state, thermal runaway can also be detected early and prevented based on the estimated time constants, in particular based on an observed change in the estimated time constant over time.

In one embodiment it is provided that the estimated battery cell state is compared with at least one estimated battery cell state of at least one other battery cell in order to monitor the battery cell state of the at least one battery cell. In this way, the behavior of individual battery cells in a network can be monitored and/or checked against one another. This is based on the idea that it can generally be assumed that the (similar) battery cells behave in the same way. The battery cell states of the individual battery cells are compared with one another accordingly. For example, the capacities and resistances of the equivalent circuit diagrams of the individual battery cells estimated within the scope of the disclosed method can be compared with one another. If the battery cells are functioning properly, the capacitances and resistances will have the same values (which may be within a specified tolerance range). On the other hand, if the values of, for example, one of the battery cells in the group deviate from the values of the other battery cells, there is a high probability that this battery cell is working incorrectly. As a reaction to the result, this battery cell can be switched off, for example, and/or a warning message can be output. Furthermore, a vehicle in which the battery with the faulty battery cell is installed can also react. For example, a power consumption can be reduced.

In one embodiment, it is provided that the battery cell states of several battery cells of the battery are determined alternately individually or in subgroups and compared with one another for monitoring. In particular, the field-programmable gate array is reconfigured each time to the battery cell to be monitored or to the subgroup of battery cells to be monitored by the field-programmable gate array being reconfigured to provide and use the associated battery cell models. As a result, computing power can be saved and the field-programmable gate array can be smaller and therefore more cost-effective be educated. The procedure is otherwise analogous to the procedure described in the paragraph directly above.

In one embodiment it is provided that the sensor system detects the voltage and the current with a sampling rate of at least 300 kHz, preferably at least 1 MHz. This enables a quick reaction, in particular in real time or almost real time, to incorrect behavior of the at least one battery cell.

Further features for the configuration of the device result from the description of configurations of the method. The advantages of the device are in each case the same as in the embodiments of the method.

The method and the device can be used in particular in a vehicle, in particular in a motor vehicle. In particular, therefore, a vehicle is also created, comprising at least one device according to one of the described embodiments.

If reconfiguration is not desired or necessary, the functionality of the at least one field-programmable gate array can also be provided by means of an appropriately configured application-specific integrated circuit (ASIC).

• 1 eine schematische Darstellung einer Ausführungsform der Vorrichtung zum Überwachen mindestens einer Batteriezelle einer Batterie;
• 2 eine schematische Darstellung eines Ersatzschaltbildes einer Batteriezelle, die mit einem Serienwiderstand und zwei Schwingkreisen mit jeweils unterschiedlichen Zeitkonstanten modelliert ist.

The invention is explained in more detail below on the basis of preferred exemplary embodiments with reference to the figures. Here show:

• 1 a schematic representation of an embodiment of the device for monitoring at least one battery cell of a battery;
• 2 a schematic representation of an equivalent circuit diagram of a battery cell, which is modeled with a series resistor and two resonant circuits, each with different time constants.

In 1 a schematic representation of an embodiment of the device 1 for monitoring at least one battery cell 21-x of a battery 20 is shown. The device 1 performs the method described in this disclosure for monitoring at least one battery cell 21 -x of a battery 20 .

The device 1 comprises a sensor system 2-x for each of the battery cells 21-x for time-resolved detection of a voltage and a current of the battery cells 21-x. To detect the current, the sensors 2-x can have Rogowski coils, for example.

Furthermore, the device 1 comprises a computing device 3 with a memory (not shown) and a field-programmable gate array 4. The computing device 3 is designed as a microprocessor, for example as a Digital Signal Processor (DSP). The field-programmable gate array 4 can be partially reconfigured, particularly at runtime. The computing device 3 can reconfigure the field-programmable gate array 4 .

The sensor system 2-x detects the voltage and the current, in particular with a sampling rate of at least 300 kHz, preferably at least 1 MHz. As a result, processes and changes occurring very quickly in the battery cells 21-x can also be detected and evaluated for monitoring. The analog variables detected by the sensor system 2-x are digitized by means of an analog/digital converter and fed to the field-programmable gate array 4 . This is done, for example, using a sigma-delta converter that provides a bit stream.

The field-programmable gate array 4 is set up to estimate a respective battery cell state 22-x of the battery cells 21-x using a respective battery cell model 10-x for each of the battery cells 21-x from the respectively detected voltage and the respectively detected current. A corresponding area in the field-programmable gate array 4 is configured in each case to provide the battery cell models 10-x. As a result, the battery cells 21-x can all be monitored at the same time. Alternatively, it can be provided that the field-programmable gate array 4 is at least partially reconfigured in order to be able to alternately provide and use one or more of the battery cell models 21-x.

The respectively estimated battery cell state 22-x is monitored. For this purpose, the battery cell state 22-x can be compared, for example, with a target battery cell state or with a target range for the battery cell state 22-x. An error and/or warning message can then be generated if the estimated battery cell state 22-x leaves the target battery cell state and/or the target range. After an error has been detected, an operating strategy of the affected battery cell 21-x and/or the battery 20 can also be changed. The computing device 2 then generates warning and/or control signals designed accordingly for this purpose and outputs them, for example to a battery controller (not shown). Furthermore, power drawn from the battery cell 21-x or the battery 20 can also be reduced.

The monitoring can be carried out in the field-programmable gate array 4, for example in a monitoring module 5 configured for this purpose.

However, as an alternative or in addition, provision can be made for the estimated battery cell states 22 - x to be transmitted to the computing device 3 and for the monitoring to be carried out using the computing device 3 .

It can be provided that the battery cell models 10-x are designed as equivalent circuit diagrams 11-x, with at least values for capacitances C-x and/or resistances R-x of the respective equivalent circuit diagram 11-x being estimated and monitored as the battery cell state 22-x. Based on this, time constants can also be estimated and monitored.

the 2 shows a schematic representation of an equivalent circuit diagram 12 of a battery cell 21-x. The battery cell 21-x has an open circuit voltage V_oc, which is modeled in the equivalent circuit diagram 12 as a voltage source. The equivalent circuit diagram 12 includes two resonant circuits 11-x with different time constants, which simulate chemical processes in the battery cell 21-x occurring on different time scales. For example, the oscillating circuit 11-1 can simulate the depletion of boundary layers of the battery cell 21-x (smaller time constant) and the oscillating circuit 11-2 can simulate diffusion processes taking place within the battery cell 21-x (larger time constant). In this case, each resonant circuit 11-x is modeled via a capacitance Cx and a resistance Rx connected in parallel. Furthermore, the equivalent circuit diagram 12 of the battery cell 21-x also has a series resistance Rs. The entire battery cell 21 - x can be modeled using the equivalent circuit diagram 12 . By taking into account the series resistance Rs and the resonant circuits 11-x, a battery cell voltage V_Batt provided by the battery cell 21-x results from the open-circuit voltage V_oc. To estimate the state of the battery cells, the capacitance Cx and the resistance Rx and the respective time constants are estimated for each resonant circuit 11-x. This can be done, for example, using an observer and the gradient descent method. The estimated capacitances Cx and resistances Rx and/or the estimated time constants can then be compared to predetermined threshold values and/or empirically determined values for monitoring the battery cell 21-x.

That in the 2 The equivalent circuit diagram 12 of the battery cell 21-x shown is only chosen as an example. In particular, it can be provided that additional resonant circuits with other time constants are part of the equivalent circuit diagram 12 in order to take into account additional processes within the battery cell 21 - x.

For a battery 20 like this one in the 1 is shown, the equivalent circuit diagram 12 would correspondingly have five battery cells 21-x connected in parallel via a busbar.

Provision can be made for the battery cell models 10-x to be provided using machine learning models. For example, the capacitances C-x and the resistances R-x can be estimated using a neural network that has been trained for this.

In a further development, it can be provided that the machine learning model can be adapted and/or is adapted during operation by at least partially reconfiguring the field-programmable gate array 4 . In particular, continuous retraining can take place, for example by means of continuous backpropagation. Furthermore, an observer provided by means of the field-programmable gate array 4 can also be adjusted continuously.

Provision can be made for a time behavior of the estimated battery cell states 22-x of the battery cells 21-x and/or the detected voltages and/or the detected currents to be evaluated and monitored for monitoring purposes. If, for example, the time constants of the oscillating circuits 11-x described above are estimated as the battery cell state 22-x, a time profile of the time constants can be monitored.

In a further development, provision can be made for the time behavior of the estimated battery cell states 22-x and/or the detected voltages and/or the detected currents to be compared with empirically determined error curves, with an error being recognized if a match is established. In particular, a future thermal runaway of a battery cell 21-x can be detected and prevented even before the thermal runaway occurs if a time behavior of the battery cell states 22-x was empirically determined immediately, i.e. a few seconds to minutes, before the thermal runaway, for example in a laboratory and therefore known. If such a time behavior of the battery cell state 22-x occurs during operation in the field (e.g. with an estimated capacitance Cx and/or an estimated resistance Rx and/or a time constant of an oscillating circuit 11-x derived therefrom), an immediate reaction can be in that the associated battery cell 21-x is switched off.

When recognizing a time behavior that is characteristic of errors, machine learning methods, for example trained neural networks, can be used to advantage. In particular, these can be formed in the field-programmable gate array 4 so that rapid, in particular high-frequency (eg in the MHz range) processing of the detected voltages and currents is possible. In particular, it can be provided that a trained Harmonically Activated Neural Network (HANN) is used for monitoring.

Provision can be made for the respectively estimated battery cell state 22-x to be compared with at least one estimated battery cell state 22-x of at least one other battery cell 21-x in order to monitor the battery cell state 22-x of the battery cells 21-x. As a result, behaviors in the estimated battery states 22-x that deviate from a normal behavior of the battery cells 21-x can be recognized without prior knowledge in the form of empirically determined behaviors being necessary.

It can be provided that the battery statuses 22-x of several battery cells 21-x of the battery 20 are alternately estimated individually or in subgroups and compared with one another for monitoring. As a result, the resources required in the field-programmable gate array 4 can be saved or reduced. The field-programmable gate array 4 is at least partially reconfigured by the computing device 3 for each change to a different battery cell 21-x and/or to a different group of battery cells 21-x by the field-programmable gate array 4 responding to the provision of the required battery cell model or models 10 times reconfigured. The associated configurations are stored, for example, in the memory of the computing device 3 and are retrieved from this as required and used to reconfigure the field-programmable gate array 4 .

Reference List

1

contraption

2x

sensors

3

computing device

4

field-programmable gate array

5

monitoring module

10-x

battery cell model

11-x

resonant circuit

12

Equivalent circuit (battery cell)

20

battery

21-x

battery cell

22-x

battery cell condition

C-x

capacity

R-x

resistance

Rs

series resistance

V_Batt

battery cell voltage

V_oc

open circuit voltage

Claims (10)

Method for monitoring at least one battery cell (21-x) of a battery (20), wherein a voltage and a current are recorded in a time-resolved manner at the at least one battery cell (21-x) by means of at least one sensor system (2-x), a battery cell state (22-x) being estimated from the detected voltage and the detected current by means of a battery cell model (10-x), wherein the estimated battery cell condition (22-x) is monitored, and wherein at least the provision of the battery cell model (10-x) and the estimation of the battery cell state (22-x) of the at least one battery cell (21-x) is carried out by means of at least one field-programmable gate array (4). procedure after claim 1 , characterized in that the battery cell model (10-x) is designed as an equivalent circuit diagram (11-x), with at least values for capacitances (Cx) and/or resistances (Rx) of the equivalent circuit diagram (11-x ) are estimated and monitored. procedure after claim 1 or 2 , characterized in that the battery cell model (10-x) is provided by means of a machine learning model. procedure after claim 3 , characterized in that the machine learning model can be and/or is adapted during operation by at least partially reconfiguring the field-programmable gate array (4). Method according to one of the preceding claims, characterized in that a time behavior of the estimated battery cell state (22-x) of the at least one battery cell (21-x) and/or the detected voltage and/or the detected current is evaluated and monitored for monitoring. procedure after claim 5 , characterized in that the time behavior of the estimated battery cell condition (22-x) of the min at least one battery cell (21-x) and/or the detected voltage and/or the detected current is compared with empirically determined error curves, an error being recognized if a match is found. Method according to one of the preceding claims, characterized in that for monitoring the battery cell state (22-x) of the at least one battery cell (22-x) the estimated battery cell state (22-x) with at least one estimated battery cell state (22-x) at least one other Battery cell (21-x) is compared. Method according to one of the preceding claims, characterized in that the battery cell states (22-x) of several battery cells (21-x) of the battery (20) are determined alternately individually or in subgroups and compared with one another for monitoring. Method according to one of the preceding claims, characterized in that the sensor system (2-x) detects the voltage and the current with a sampling rate of at least 300 kHz, preferably at least 1 MHz. Device (1) for monitoring at least one battery cell (21-x) of a battery (20), comprising: at least one sensor system (2-x), set up for the time-resolved detection of a voltage and a current at the at least one battery cell (21-x), a computing device (3) with a memory and at least one field-programmable gate array (4), wherein the at least one field-programmable gate array (4) is set up to estimate a battery cell state (22-x) of the at least one battery cell (21-x) using a battery cell model (10-x) from the detected voltage and the detected current, and wherein the at least one field-programmable gate array (4) and/or the computing device (2) are set up to monitor the estimated battery cell state (22-x).

Images