DE102021207048A1 - Battery management system and method for operating a battery management system - Google Patents

Abstract

The invention relates to a battery management system (1), comprising at least one AI-optimized computing device (2), wherein the at least one AI-optimized computing device (2) is configured to have at least one of cells (22-x) of a battery (20) and/or receiving electrical and/or other physical variables (10, 11) detected on modules (21) of the battery (20) and/or on the battery (20), starting from the at least one received electrical and/or other physical variable (10 ,11) to estimate at least one cell status (42-x) and/or at least one module status (41) and/or one battery status (40-x) by means of at least one trained neural network (3). The invention also relates to a method for operating a battery management system (1).

Description

The invention relates to a battery management system and a method for operating a battery management system.

Due to the increasing electrification of vehicles and the increasing use of batteries as stationary energy storage, a battery management system is becoming increasingly important. The battery management system performs tasks such as charge state detection, deep discharge protection, overcharge protection, determining cell health (ageing, residual capacity, internal resistance, etc.) and balancing cells of the battery.

From the U.S. 2018/0086222 A1 systems and methods for monitoring and controlling a battery are known. Systems may include a battery having an output voltage and an output current when delivering power, a load powered by the power delivered by the battery, a battery output voltage circuit, a battery current detection circuit, and a processing circuit coupled to the battery output voltage circuit and the battery current detection circuit. The processing circuitry may implement a recurrent neural network for battery state estimation.

The object of the invention is to improve a battery management system and a method for operating a battery management system.

The object is achieved according to the invention by a battery management system having the features of patent claim 1 and a method having the features of patent claim 7 . Advantageous configurations of the invention result from the dependent claims.

In particular, a battery management system is created, comprising at least one AI-optimized computing device, wherein the at least one AI-optimized computing device is configured to at least one electrical and / or other physical detected on cells of a battery and / or on modules of the battery and / or on the battery Receiving variable, and based on the at least one received electrical and / or other physical variable by means of at least one trained neural network at least one cell state and / or at least one module state and / or a battery state to estimate.

Furthermore, in particular a method for operating a battery management system is provided, with at least one trained neural network being provided by means of a computer-optimized computing device of the battery management system, with at least one electrical and/or other physical variable of cells of a battery and/or of modules of the battery and /or is detected and/or received by the battery, at least one cell state and/or at least one module state and/or one battery condition is estimated.

The battery management system and the method make it possible to better estimate states within the battery. For this purpose, it is provided that an AI-optimized computing device is used, which provides at least one trained neural network. The at least one neural network is trained to calculate at least one cell state and/or at least one module state and/or one Estimate battery health.

One advantage of the battery management system and the method is that the performance of a functionality provided by means of the at least one trained neural network can be increased, since a computing device that is specifically optimized for AI applications is used. This can improve the estimation of the status or statuses. In particular, the at least one (trained) neural network can be made more complex in terms of a structure, so that functionality can be improved, in particular increased.

An electrical quantity is in particular a voltage, a current or a resistance. An electrical variable is recorded in particular by means of a sensor set up for this purpose. A physical quantity is in particular a temperature or a pressure. A temperature can be detected, for example, at locations provided for this purpose on the cells, on modules or on the battery by means of one or more temperature sensors. A pressure can be detected, for example, at points provided for this purpose on the cells, on modules or on the battery by means of one or more pressure sensors.

An AI-optimized computing device is in particular one that is designed to provide functions an artificial intelligence (AI) optimized microprocessor. In particular, the microprocessor is optimized for the provision of arithmetic operations that are required for the functionality of neural networks. Examples of commercially available AI-optimized computing devices are: Nvidia Corp. Jetson Nano, Texas Instruments Inc. TDA4VM, XMOS xcore.ai, NXP Semiconductors i.MX 8M Plus, and Intel Movidius Myriad X.

The AI-optimized computing device is or will be configured in such a way that it provides at least one neural network. In particular, the AI-optimized computing device is or will be configured to map a structure and parameters of the at least one neural network and thereby provide a functionality of the neural network. In particular, the at least one neural network is or will be trained. Training data can be generated and made available in particular on the basis of (known) theoretical models, empirical tests (e.g. destruction tests, overload tests, cyclic charging and discharging tests and/or aging tests, etc.), measurement data recorded in the field and/or simulations. Depending on the functionality, the at least one neural network is trained using methods known per se, such as supervised learning, unsupervised learning (e.g. for finding outliers in a comparison of several cells, modules and/or batteries to be monitored) and reinforcement learning learning). Depending on the embodiment, both regression problems and classification problems can be solved using the trained neural network.

A neural network is in particular a deep neural network with several inner layers. The neural network can be designed, for example, as a convolutional network, generative adversarial network (GAN), autoencoder or recurrent neural network (e.g. LSM, LTSM, GRU), with a structure being selected depending on a desired functionality.

A cell state can in particular include a state of charge (SOC) and/or a state of health (SOH) and/or a cell temperature and/or a cell pressure. A module state can in particular include a state of charge (SOC) and/or a state of health (SOH) and/or at least one module temperature. A battery state can in particular include a state of charge (SOC) and/or a state of health (SOH) and/or at least one battery temperature. However, the states can additionally or alternatively also include an error state of at least one cell, at least one module and/or of the battery.

In particular, it is provided that the estimated at least one cell state and/or the estimated at least one module state and/or the estimated battery state is provided, for example output as an analog or digital signal, for example in the form of a digital data packet. The battery management system is set up accordingly for this. In particular, it is further provided that the estimated at least one cell state and/or the estimated at least one module state and/or a battery state is taken into account by the battery management system when controlling the battery, for example when balancing and/or charging and/or discharging.

In one embodiment, it is provided that the at least one AI-optimized computing device is configured to process time series of the at least one electrical and/or physical variable recorded in a time-resolved manner. As a result, the at least one cell state, the at least one module state and/or the battery state can be determined in an improved manner. For this purpose, the time series recorded in a time-resolved manner is supplied to the at least one trained neural network in the form of individual data points, each of which corresponds to a time step.

In one embodiment, it is provided that the at least one AI-optimized computing device is configured to estimate at least one piece of diagnostic information using the at least one trained neural network. A piece of diagnostic information relates in particular to a current state of the battery. Diagnostic information includes, for example, estimating an error state. For example, diagnostic information can include the following states for the cells, the modules and/or the battery: lithium plating, overcharging, deep discharge and/or unbalanced cells of a module. The at least one trained neural network is or will be trained for this purpose in such a way that it estimates the at least one piece of diagnostic information, for example a state of the lithium plating or a balance between the charge states of the cells, based on the received at least one electrical and/or other physical variable. In other words, the trained neural network carries out a corresponding classification in particular. Training data for training the at least one neural network can be obtained here in particular through empirical tests, measurement data collected in the field with known diagnostic information and/or simulations.

In one embodiment, it is provided that the at least one AI-optimized computing device is configured to estimate at least one piece of prognosis information using the at least one trained neural network. Forecast information relates in particular to a future state of the battery. Forecast information includes, for example, estimating a future error state based on the current state of the battery. For example, forecast information can include the following future states for the cells, the modules and/or the battery: lithium plating, overcharging, deep discharge and/or unbalanced cells within a module. The at least one trained neural network is or will be trained for this purpose in such a way that, based on the received at least one electrical and/or other physical variable, it calculates the at least one piece of prognosis information, for example a future state of the lithium plating and/or a balance of charge states of the cells within of a module, guess. Fixed time periods and/or future points in time can be provided here, for which the at least one item of prognosis information is estimated. Alternatively or additionally, the at least one item of prognosis information can also include an estimated future time at which a specified state will occur, for example an expected time for a short circuit caused by progressive lithium plating, or an overcharge, etc. Training data for training the at least one Neural networks can be obtained here in particular through empirical tests, measurement data collected in the field with known diagnostic and/or prognosis information and/or simulations.

Both the estimation of the at least one item of diagnostic information and the estimation of the at least one item of prognosis information are, in particular, a classification problem in which the at least one neural network is trained or has been trained on the basis of the input data supplied in each case, a classification, for example according to different error patterns , to undertake. In addition, however, numerical values can also be estimated, for example in order to indicate a future point in time at which an error state is expected.

In one embodiment it is provided that the at least one AI-optimized computing device is configured based on the estimated at least one cell state and/or the estimated at least one module state and/or the estimated battery state and/or the estimated at least one item of diagnostic information and/or estimate at least one control action for adapting an operating strategy of the battery and/or a vehicle from the estimated at least one piece of prognosis information by means of at least one further trained neural network. As a result, a control action can be generated and provided directly from the current state of the battery. In particular, it is provided that the estimated at least one control action is carried out by the battery management system, for example by changing the load distribution to individual cells when charging or discharging, by deactivating cells or modules, by opening or closing power relays or by activating or reconfiguring battery cooling etc. With regard to a control action for adapting an operating strategy of the vehicle, a power draw from the battery can be limited (or increased), for example. On the vehicle side, for example, a maximum speed that can be called up and/or a maximum torque that can be called up and/or a maximum acceleration that can be called up can be reduced and/or another power draw can be restricted by limiting or reconfiguring the vehicle (e.g. by reducing a power draw from an air conditioning unit or an infotainment system Etc.).

Furthermore, in particular a vehicle is also created, comprising at least one battery management system according to one of the described embodiments. A vehicle is in particular a motor vehicle, in particular an electric or hybrid vehicle. However, a vehicle can also be any other land, rail, water, air or space vehicle, for example an air taxi or a drone.

Further features for the configuration of the method result from the description of configurations of the battery management system. In this case, the advantages of the method are the same as in the embodiments of the battery management system.

• 1 eine schematische Darstellung einer Ausführungsform des Batteriemanagementsystems;
• 2 eine schematische Darstellung zur Verdeutlichung einer weiteren Ausführungsform des Batteriemanagementsystems;
• 3 eine schematische Darstellung zur Verdeutlichung einer weiteren Ausführungsform des Batteriemanagementsystems;
• 4 ein schematisches Ablaufdiagramm einer Ausführungsform des Verfahrens zum Betreiben eines Batteriemanagementsystems.

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 battery management system;
• 2 a schematic representation to illustrate a further embodiment of the battery management system;
• 3 a schematic representation to illustrate a further embodiment of the battery management system;
• 4 a schematic flowchart of an embodiment of the method for operating a battery management system.

In 1 a schematic representation of an embodiment of the battery management system 1 is shown. The embodiment is shown by way of example together with a module 21 of a battery 20, the module 21 having a plurality of cells 22-x (for the sake of clarity, only a few are provided with their own reference numbers). In principle, however, the battery 20 can also include more than one module 21 . The battery 20 and the battery management system 1 can be arranged in a vehicle 50 in particular.

The battery management system 1 includes an AI-optimized computing device 2. The AI-optimized computing device 2 provides a trained neural network 3. For this purpose, the computing device 2 comprises, for example, one or more of the following microprocessors: Nvidia Corp. Jetson Nano, Texas Instruments Inc. TDA4VM, XMOS xcore.ai, NXP Semiconductors i.MX 8M Plus and/or Intel Movidius Myriad X. The battery management system 1 also includes a battery module controller 4, which controls the module 21 and the cells 22-x. In principle, however, the battery module controller 4 can also not be part of the battery management system 1 and can be embodied separately.

The battery management system 1 also includes, in particular, a control device 5 for controlling (and/or regulating) an operating strategy for the battery 20, the module 21 and the cells 22-x.

Cell voltages 12-x are recorded as an electrical quantity 10 at the cells 22-x. For each detected cell voltage 12-x, the neural network 3 has a corresponding input node 31-x (not all of them are provided with their own reference symbols), to which the respectively detected cell voltage 12-x is supplied as input data.

Furthermore, temperatures 13-1, 13-2 are recorded as a physical quantity 11 at different locations in the battery module 21 and are fed to the input nodes T1, T2 of the neural network 3 as input data.

Based on the received cell voltages 12-x and the received temperatures 13-1, 13-2, the trained neural network 3 estimates cell states 42-x and/or a module state 41 and/or a battery state 40. In particular, the trained neural network 3 estimates a respective state of charge (State of Charge, SOC) and a respective state of health (State of Health, SOH) for the cells 22-x and/or the module 21 and/or the battery 20.

The estimated cell states 42-x and/or the estimated module state 41 and/or the estimated battery state 40 are then supplied to the control device 5 of the battery management system 1, with the control device 5 based on an operating strategy for the battery 20, the module 21 and the cells 22-x from the estimated cell states 42-x and/or the estimated module state 41 and/or the estimated battery state 40.

In addition to the cell voltage 12-x, other electrical variables can be detected at the cells 22-x, at the module 21 and/or at the battery 20, such as a respective current. Proceeding from this, the neural network 3 can be set up and trained to estimate further electrical and/or physical variables, for example an ohmic resistance and/or an impedance and/or a temperature.

Provision can be made for the at least one AI-optimized computing device 2 to be configured to process time series of the at least one electrical variable 10 and/or physical variable 11 recorded in a time-resolved manner. This is shown schematically in the 2 illustrates, which shows a neural network 3 set up accordingly for this purpose, which is provided by the AI-optimized computing device 2. The embodiment is otherwise like that in FIG 1 Formed embodiment shown, the same reference numerals denote the same features and terms. In the embodiment shown, the neural network 3 is supplied with time-resolved time series of a cell voltage 12-1-t (with t=1, 2,..., n; where n is the number of time steps in the time series), with each measured value being a period under consideration here corresponds to an input node 31-1-t (with t=1, 2, . . . , n; where n is the number of input nodes) of the neural network 3. For the sake of clarity, this is only shown for a time series of a single cell, but the procedure is basically analogous for other cells, the module and the battery.

Furthermore, a (dominant) frequency, a detected current of the cell and an impedance of the cell can be fed to the neural network 3 as input data at input nodes F, I, Z provided for this purpose.

The neural network 3 is trained to estimate a cell state 42-1 of the cell under consideration based on the input data. In particular, this cell state 42-1 includes a drawer state (SOC) and a state of health (SOH).

It can be provided that the neural network 3 is also trained to estimate a temperature 43-1 of the cell.

It can be provided that the at least one AI-optimized computing device 2 ( 1 ) is configured to use the at least one trained neural network 3 to generate at least one item of diagnostic information 44 ( 2 ) appreciate. In particular, the trained neural network 3 is trained to estimate the at least one item of diagnostic information 44 . For example, diagnostic information 44 can include the following states for the cells (in other embodiments, alternatively or additionally, also for the modules and/or the battery): lithium plating, overcharging, deep discharge (and/or unbalanced cells in a module).

It can be provided that the at least one AI-optimized computing device 2 ( 1 ) is configured to use the at least one trained neural network 3 to generate at least one piece of prognosis information 45 ( 2 ) appreciate. In particular, the trained neural network 3 is trained to estimate the at least one piece of prognosis information 45 . For example, prognosis information 45 can include the following future states for the cells (in other embodiments, alternatively or additionally also for the modules and/or the battery): lithium plating, overcharging, deep discharge (and/or unbalanced cells in a module).

In 3 A schematic representation of an embodiment of the battery management system is shown. The battery management system is basically like that in the 1 Embodiment shown configured, the same reference numerals denote the same features and terms. the 3 shows the trained neural network 3, which is generated by the computing device 2 ( 1 ) provided.

It is additionally provided that the at least one AI-optimized computing device 2 is configured based on the estimated at least one cell state 42-x and/or the estimated module state 41 and/or the estimated battery state 40 and/or the estimated at least one item of diagnostic information 44 and/or the estimated at least one piece of prognosis information 45 by means of a further trained neural network 7 at least one control action 60 for adapting an operating strategy of the battery 20 ( 1 ) and/or a vehicle 50 ( 1 ) appreciate. In the embodiment shown, the additional neural network 7 is integrated with the neural network 3 .

In 4 a schematic flowchart of an embodiment of the method for operating a battery management system is shown. The battery management system 1 is, for example, like that in FIG 1 configured embodiment shown.

In a measure 100, at least one trained neural network is provided by means of a battery-optimized computing device of the battery management system. In particular, the AI-optimized computing device provides a functionality of the at least one trained neural network.

In a measure 101, at least one electrical and/or other physical variable of cells of a battery and/or of modules of the battery and/or of the battery is recorded and/or received and fed to the at least one trained neural network as input data.

In a measure 102, at least one cell status and/or at least one module status and/or one battery status is estimated on the basis of the at least one recorded and/or received electrical and/or other physical variable by means of the at least one trained neural network.

Provision can be made here for time series of the at least one electrical and/or other physical variable recorded in a time-resolved manner to be processed for the estimation.

It can also be provided in measure 102 that at least one item of diagnostic information and/or at least one item of prognosis information is estimated using the at least one trained neural network.

Furthermore, in a measure 103 it can be provided that starting from the estimated at least one cell state and/or the estimated at least one module state and/or the estimated battery state and/or the estimated at least one piece of diagnostic information and/or the estimated at least one piece of forecast information using at least one further trained neural network at least one control action for adapting an operating strategy of the battery and / or a vehicle is estimated. In a measure 104 it can then be provided that the battery management system carries out the at least one estimated control action by changing an operating strategy.

In particular, it is provided that the method is repeated continuously.

Reference List

1

battery management system

2

computing device

3

trained neural network

4

battery module controller

5

control device

10

electrical size

11

physical size

12-x

cell voltage

12-x-t

cell voltage (time-resolved value)

13-x

temperature

20

battery

21

module

22-x

cell

31-x

input node

31-x-t

Input node (for time-resolved value)

50

vehicle

f

input node

I

input node

T1

input node

T2

input node

Z

input node

40

Battery health (estimated)

41

Module condition (estimated)

42-x

cell condition (estimated)

43-x

temperature (estimated)

44

Diagnostic information (estimated)

45

Forecast information (estimated)

60

Control action (estimated)

100-104

Measures of the procedure

QUOTES INCLUDED IN DESCRIPTION

This list of documents cited by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent Literature Cited

• US20180086222A1[0003]

Claims (10)

Battery management system (1), comprising: at least one AI-optimized computing device (2), wherein the at least one AI-optimized computing device (2) is configured to connect at least one to cells (22-x) of a battery (20) and/or to modules (21) of the battery (20) and/or to the battery (20 ) to receive detected electrical and/or other physical variables (10,11), based on the at least one received electrical and/or other physical quantity (10,11) by means of at least one trained neural network (3) at least one cell state (42-x) and/or at least one module state (41) and/or a battery state ( 40-x) estimate. Battery management system (1) according to claim 1 , characterized in that the at least one AI-optimized computing device (2) is configured to process time series of the at least one electrical and/or physical variable (10, 11) recorded in a time-resolved manner. Battery management system (1) according to claim 1 or 2 , characterized in that the at least one AI-optimized computing device (2) is configured to estimate at least one item of diagnostic information (44) by means of the at least one trained neural network (3). Battery management system (1) according to one of the preceding claims, characterized in that the at least one AI-optimized computing device (2) is configured to estimate at least one piece of forecast information (45) using the at least one trained neural network (3). Battery management system (1) according to one of the preceding claims, characterized in that the at least one AI-optimized computing device (2) is configured, starting from the estimated at least one cell state (42-x) and/or the estimated at least one module state (41 ) and/or the estimated battery state (40) and/or the estimated at least one item of diagnostic information (44) and/or the estimated at least one item of prognosis information (45) by means of at least one further trained neural network (3) at least one control action (60) for adapting to estimate an operating strategy of the battery (20) and/or a vehicle (50). Vehicle (50) comprising at least one battery management system (1) according to one of Claims 1 until 5 . Method for operating a battery management system (1), wherein at least one trained neural network (3) is provided by means of an AI-optimized computing device (2) of the battery management system (1), wherein at least one electrical and/or other physical quantity (10,11) of cells (22-x) of a battery (20) and/or of modules (21) of the battery (20) and/or of the battery (20) is recorded and/or is received, based on the at least one detected and/or received electrical and/or other physical variable (10,11) by means of the at least one trained neural network (3) at least one cell status (42-x) and/or at least one module status (41) and/or a battery condition (40) is estimated. procedure after claim 7 , characterized in that time series of the at least one electrical and/or other physical variable (10, 11) recorded in a time-resolved manner are processed for the estimation. Procedure according to one of Claims 7 or 8th , characterized in that at least one item of diagnostic information (44) and/or at least one item of prognosis information (45) is estimated by means of the at least one trained neural network (3). procedure after a Claims 7 until 9 , characterized in that based on the estimated at least one cell condition (42-x) and/or the estimated at least one module condition (41) and/or the estimated battery condition (40) and/or the estimated at least one item of diagnostic information (44) and/or or at least one control action (60) for adapting an operating strategy of the battery (20) and/or a vehicle (50) is estimated from the estimated at least one piece of prognosis information (45) by means of at least one further trained neural network (3).

Images