METHOD AND DEVICE FOR ROBUST ONLINE CONDITION DETERMINATION OF LARGE BATTERY STORAGE SYSTEMS

Description

[0001] The invention relates to a method for robust online state determination by determining several state variables of large battery storage systems according to claim 1, as well as to a device for carrying out such a method according to claim 5.

[0002] Lithium-ion batteries (Li-ion batteries) are highly efficient electrochemical energy storage devices that, due to their positive properties, play a key role in consumer electronics, mobility, logistics, and stationary storage. A Li-ion battery is generally used within a so-called operating window. This operating window of current, voltage, and temperature is limited in that leaving the safe operating window can pose safety hazards such as fire, explosion, and / or the release of toxic gases. Unlike lead-acid or nickel-based battery cells, the current state of technology does not allow for the suppression or recombination of the resulting reaction products. Therefore, stable, reliable, and temporally high-resolution status monitoring of each individual cell in the Li-ion battery system is essential.The rate of aging is also determined by compliance with the limited operating window.

[0003] To store large amounts of energy in lithium-ion batteries, many individual cells are connected in multiple parallel and / or series circuits to form large, partially distributed battery systems. This involves a broad distribution of the states of the individual cells within the system. The challenge here is to reliably record the battery states for each individual cell to ensure overall safety. The most important state variables are the state of charge and the state of aging.

[0004] Particular challenges arise for batteries used in automotive and industrial traction applications, as well as in stationary applications. These are characterized by load profiles with undefined charges, such as the recovery of braking energy (recuperation), opportunity charging, hybrid storage operation, or operation at intermediate states of charge in stationary applications. Consequently, with such load profiles, both the charging and discharging conditions cannot be specifically defined and are generally unpredictable. Defined charging at chargers can occur in longer cycles or never at all. Therefore, algorithms for state detection must be variable enough to function robustly across a wide range of load profiles.

[0005] Various estimation methods have been established for determining states of charge. These can be divided into direct measurement methods, model-based measurement methods, and data-based measurement methods. Each category encompasses a variety of measurement methods. Each of these measurement methods, in turn, has a variety of advantages and disadvantages.

[0006] For large, distributed battery systems, data-based methods involve immense computational effort, which can only be implemented in outsourced server-based computing units. In practice, direct measurement methods or model-based methods on the battery management system are therefore primarily used. Each individual method involves temporally inconsistent errors, so the state-of-charge measurement is subject to variable inaccuracy.

[0007] EP 3 056 917 B1 discloses a method and a device for estimating the condition of a battery. EP 2 107 385 B1 shows a device for estimating the condition of a battery. EP 3 136 118 B1 discloses a device and a method for determining the voltage of a battery. EP 1 562 048 B1 discloses a battery capacity calculation method, a battery capacity calculation device, and a battery capacity calculation program. EP 2 502 086 B1 discloses a method and a device for error-compensated current measurement of a battery.

[0008] A disadvantage of all the aforementioned publications is that only one estimation method is used for each state variable to be determined. Individual estimation methods only provide sufficiently accurate estimates within specific operating windows. It can be observed that the use of only one estimation method often does not allow for reliable and accurate state measurements with undefined load and / or temperature profiles.

[0009] To reduce the resulting inaccuracy, US Pat. No. 9,960,625 B2 proposes a sequential method in which several battery parameters are determined one after the other using several methods. The method is as follows: A parameter A is estimated by a first model 1. This parameter thus determined serves as the input for further downstream models 2...n to estimate further parameters B...Z.

[0010] A disadvantage of the latter method is that an error remains unknown during use, so sufficient plausibility and reliability cannot be guaranteed. US 2017 / 225584 A1, EP 3 663 780 A1, DE 10 2012 207817 A1, and EP 3 564 693 A1 disclose further prior art examples of approaches for determining the state of charge and aging of batteries.

[0011] In contrast, the invention is based on the object of creating a method and a device that ensure state estimations of battery systems with sufficient reliability.

[0012] This object is achieved by a method having the features of patent claim 1 and by a device having the features of patent claim 5. Further advantageous embodiments of the invention are described in the dependent patent claims.

[0013] The disadvantages resulting from the state of the art are overcome by a parallel application of several estimation methods in a higher-level battery management system.

[0014] Multiple estimation methods are used for each state variable to be determined. These multiple estimation methods are applied in parallel, with both the multiple estimation methods for each state variable running in parallel, and the additional estimation methods for the other state variables running in parallel.

[0015] All estimation methods are used in parallel for each individual battery cell of the battery system.

[0016] After a first, initial state variable determination using one or more estimation methods for one and / or more of the state variables, the resulting state variable(s) form the basis for further estimation methods based on an error-based selection. The method is applied cyclically, meaning that the estimation methods are repeatedly performed in parallel, with a state variable determined in the previous process run forming the basis for the next subsequent process run.

[0017] It is particularly preferred if the state variables are composed of at least one, preferably several and / or the entirety of the following list: A state of charge of the battery system, which, like the states mentioned below, can also be determined for subgroups at a battery cell level or a cell assembly composed of multiple battery cells. The state of charge is also referred to as the "state of charge" (SoC). An aging state, also referred to as the "state of health" (SoH), a safety state, known as the "state of safety" (SoS), a usability state, "state of function" (SoF), a state of charge distribution, also referred to as the "state of balance" (SoB), and an internal cell temperature, where the internal cell temperature refers to the temperature inside a battery cell.

[0018] Both continuous and discontinuous estimation methods are used for the state variable "state of charge." The former can calculate a new state variable, including the error, almost in real time, without preconditions in each high-resolution computing cycle. Examples of the state variable "state of charge" include Coulomb counting, all forms of the Kalman filter, and data-based estimation methods. In contrast, discontinuous methods are subject to preconditions and therefore not always available. Examples of the state variable "state of charge" include rest voltage measurement, incremental capacity analysis (ICA), and load step.

[0019] In each calculation cycle, the errors determined individually decide which of the values ​​of a state variable determined in parallel using several estimation methods is further favored for the state determination of the respective battery cell or the respective cell assembly.

[0020] The required computing power is adapted to the estimation methods used, the cycle frequency requirements and the required accuracy.

[0021] A device according to the invention for carrying out a method described above has a plurality of temperature sensors and a plurality of voltage sensors. These are each operatively connected to battery cells of the battery system, with one temperature sensor and one voltage sensor being connected to each battery cell. In other words, the device according to the invention has exactly as many temperature sensors as voltage sensors and battery cells. Thus, both the temperature and the voltage of each battery cell of the battery system can be measured via the sensors, enabling maximum data density, which in turn enables high accuracy of at least some of the state variables to be determined.

[0022] In addition, a current measuring device is provided on a cell assembly formed from a plurality of battery cells.

[0023] If data is acquired by a plurality of sensors, then advantageously several computing units are provided, arranged centrally or decentrally. These are designed to control the method according to the invention and process the data acquired by the sensors. Advantageously, the computing units are designed to process the resulting data almost in real time.

[0024] In addition to the sensor data collected, the processing units also record the underlying errors. Furthermore, the processing units are designed to compare the collected data with each other and / or with data stored in a database and to evaluate the acquired data. Furthermore, the processing unit features an interface that enables the transmission of the processed and evaluated data.

[0025] A higher-level master processing unit is assigned to the processing units. This master processing unit is designed to receive the data transmitted by the processing units, bundle it in a predetermined manner, and transmit it to a higher-level battery management system. By dividing the system into a master processing unit and several sub-processing units, also known as slave processing units, the necessary computing power can be optimized for the respective processes. This allows for particularly fast and efficient processing of the acquired data.

[0026] For measurement accuracy reasons, it is advisable for the voltage sensors to be designed with a measurement accuracy of 0.5 to 2 mV, preferably 1 to 1.5 mV, especially 1.2 mV. The resolution of the voltage sensors is 0.05 to 0.2 mV, particularly preferably 0.1 mV.

[0027] For measurement accuracy reasons, it is advisable for the temperature sensors to be designed with a measurement accuracy of 0.5 to 2 K, preferably 0.8 to 1.5 K, especially 1.2 K. The resolution of the temperature sensors is 0.05 to 0.2 K, particularly preferably 0.1 K.

[0028] For measurement accuracy reasons, it is advisable for the current measuring device to be designed with a measurement accuracy of <1.5 A, preferably <1 A. The resolution of the current measuring device is 0.5 to 2 mA, particularly preferably 0.8 to 1.5 mA, especially 1 mA.

[0029] In a particularly preferred embodiment, the device is decoupled from the battery system via redundant, galvanically decoupled bus systems. The device's power supply is independent of the battery system and is preferably provided via a system superordinate to the battery system, into which the battery system can be integrated.

[0030] An embodiment of the invention is shown in the figures.

[0031] It shows Figure 1 a flow chart showing the basic sequence of a process cycle using the example of the state variable state of charge (SoC), Figure 2 a schematic flow diagram for the evaluation of an estimation method using the example of the state variable state of charge (SoC), Figure 3 three related diagrams showing the time sequence of determining the state of charge (SoC) and Figure 4 an overview of state variables and the applicable estimation methods.

[0032] Figure 1shows a flowchart of the basic sequence of a process cycle using the state of charge (SoC) as an example. This process cycle is also applicable to the state variables state of aging (SoH), state of safety (SoS), state of usability (SoF), state of charge distribution (SoB) and the internal cell temperature. In an ongoing measurement cycle, measurement data is already available, which in a first step S1 forms the basis for the subsequent determination of the state variables. In a subsequent step S2, several algorithms 1 to n are applied to determine the SoC, each of which outputs a state variable and an error associated with this state variable as a result S3. From this multitude of state variables and errors, the SoC with the lowest error is selected in a fourth step S4.From this selection, a current SoC and a current error are determined in the following step S5 through an evaluation, which serve as the basis for the next cycle. The selection is made from a function based on the SoC and the error from steps S1 and S4.

[0033] This evaluation is shown in the flow chart in Figure 2 The SoC selected in S4 and the corresponding error are compared with the SoC and the corresponding error of the previous cycle in step S6. If the selected SoC and the SoC from the previous cycle differ, this distinction is considered a jump. Jumps in the SoC are evaluated with regard to their size. In the case of an implausibly large jump, the state variables, in this case the SoC, are weighted according to the following formula 1 before and after the jump based on their respective co-determined errors. SoC mean = SoC favorite ⋅ Error LastCycle Error favorite + Error LastCycle + SoC LastCycle ⋅ Error favorite Error favorite + Error LastCycle Error mean = Error favorite + Error favorite Error favorite + Error LastCycle ⋅ Error LastCycle − Error favorite

[0034] Consequently, two options are conceivable in the subsequent step S7. Either the above formula is applied, or the differences between the two SoCs and the corresponding errors are below a corresponding tolerance threshold. The result of S7 is then a new SoC and a corresponding error, which is presented as the final step 8. These values ​​form the basis for the next cycle. The method is described in more detail below using a specific example.

[0035] As an example of the state of charge (SoC) variable, the following describes a method and device used to determine the state of charge in a battery system consisting of six cell arrays, each consisting of 216 battery cells. Each cell array is grouped into six modules. Examples of estimation methods used include Incremental Capacity Analysis (ICA), open-circuit voltage balancing, load step, and Coulomb Counting.

[0036] The device is designed so that current measurements of current flows in the battery system are recorded during each process cycle and are available at the beginning of a cycle via internal communication to the master processing unit. A cycle interval typically lasts 10 to 100 ms.

[0037] The Coulomb counting method is a continuous estimation method, while the other three are called discontinuous estimation methods.

[0038] The ICA estimation method takes advantage of the fact that, within a voltage operating window, very different amounts of charge can be charged into or removed from battery cells in different ranges. This is due to different electrode reactions, which have different standard electrode potentials. The mathematical derivation of the charged / removed charge quantities with respect to the battery cell voltage shows peaks for the voltage ranges in which particularly large amounts of charge can be charged / removed. Depending on the cell chemistry, several of these peaks can be detected, the position and amplitude of which sometimes depend on aging, sometimes occur independently of an aging state at certain charge states. If such a peak maximum is detected during the charging and / or discharging processes, the state of charge and / or aging state can be read directly from it.The prerequisites for the ICA are, on the one hand, a corresponding current flow and, on the other hand, that the state of charge covers exactly the area in which a peak occurs according to the cell chemistry.

[0039] In the state of charge estimation method using open-circuit voltage adjustment, the respective battery cell voltage is measured when a certain current flow has been undershot for a certain period of time. It is assumed that the battery cell voltage corresponds to a resting voltage, which is an erroneous assumption. The time factor here largely determines the size of the error. Due to the asymptotic relaxation behavior of the battery cell to its resting voltage, the error becomes smaller and smaller over time. The resting voltage characteristic curve (SoC over resting voltage) is stored in the battery management system, for example, using an assignment table. Due to the temperature dependence, this table is stored for several temperatures and it is also taken into account that this characteristic curve changes over the course of battery cell ageing. For example, a resting voltage characteristic curve averaged over the usual service life is stored for each temperature.In this way, the ageing-related error is minimized without maintaining a critical circular relationship between the state of charge determination and the ageing state determination and the state of charge determination.

[0040] In the state of charge estimation method using Coulomb counting, the amount of charge charged / discharged is continuously counted during each process cycle to determine the current capacity of each battery cell. This estimation method also requires the currently available total capacity of all battery cells, based on their aging status, to be known. The quotient of the current capacity and the currently available total capacity corresponds to the battery cell's state of charge.

[0041] In the state-of-charge estimation method using a load step, after detecting a rapid change in the battery's output / consumption (typically about 50% of the total power per second), the impedance of each battery cell is determined from the currents and battery cell voltages before and after the load step. A resting voltage is calculated from the voltage, current, and impedance.

[0042] An example of the state of charge and its errors as well as the state of charge and its errors determined from the estimation methods described above are shown in Figure 3Coulomb counting (CC) is performed continuously, and the estimated SoC value is adjusted accordingly. At two points in time, the error of other estimation methods is smaller than that of the CC estimation method ("SoC-CC error"): at the beginning (first measurement point) and after 120 ms. At these points in time, the SoC is updated according to the more accurate estimation method, using either the so-called OCV method or the previously discussed ICA method. At two further points in time, alternative estimates of the CC were performed (load step after 60 ms and OCV after 270 ms), but the calculated error of the estimated value was greater than the parallel error determined using CC.

[0043] Further estimation methods are described in more detail below and illustrated using an exemplary embodiment of the state of aging (SoH) variable. The exemplary embodiment described above uses a battery system consisting of 6*216 battery cells.

[0044] In an estimation method based on capacity determination, the current total capacity is determined based on two consecutive state-of-charge determinations. The two state-of-charge determinations are separated by an intermediate phase consisting of charging, discharging, or a combination of both (typically: load profile with recuperation), and differ by at least 50% state of charge. Coulomb counting (CC) must be performed during the intermediate phase. The relationship between this determined total capacity and the total capacity determined at the beginning of the service life represents the state of health (SoH). The error in this SoH determination results from Gaussian error propagation from the co-determined SoC errors and the CC error, as well as a capacity error determined at the beginning of the service life.

[0045] The ICA estimation method uses the ICA estimation method described above to determine the state of health (SoH). This method takes advantage of the fact that the derivative of capacity versus battery cell voltage exhibits one or more age-dependent peaks depending on the battery cell chemistry. This estimation method has already been described in detail above, so please refer to the relevant sections of the text.

[0046] An estimation method based on impedance determination takes advantage of the fact that the cell impedance of lithium-ion batteries exhibits a specific aging-related increase. In the simplest case, which occurs with a large number of battery cells of different chemistries, this increase is linear with the state of aging (SoH). By determining the cell impedances online using the previously described load step, the aging state is determined based on the specific impedance-aging curve, which is stored in a mapping table in the device.

[0047] The estimation method for determining aging by determining irreversible thickness growth involves measuring the battery cell thickness and correlating it with the aging state. Thickness determination can be carried out using both measurement methods where the measuring device can be housed within battery cell modules due to limited space requirements and those where the device is positioned outside the cell modules and where the measurement is carried out using non-contact measuring methods (optical, acoustic, magnetic, or electromagnetic). The specific thickness growth curve over the permissible aging range is stored in a mapping table. Thickness determination is carried out within a defined state-of-charge range to minimize errors caused by state-of-charge-dependent, reversible thickness growth. After successful thickness determination of selected or all battery cells, the aging state is estimated using the mapping table.

[0048] For battery cells with flexible cell casings ("pouch cells"), although they do not increase in thickness when firmly clamped in a module casing, the internal cell pressure increases due to aging. For such battery cells, the internal pressure can be measured externally. For example, the irreversible pressure increase is determined using thin absolute pressure sensors between two battery cells or between a battery cell and a solid surface adjacent to the battery cell. Analogous to the thickness determination described above, the pressure increase occurs within a defined state-of-charge range to minimize errors caused by reversible processes. The correlation between irreversible internal cell pressure and aging is also stored in a correlation table.

[0049] In an estimation method using the rainflow counting algorithm, complex load profiles are broken down into microcycles. For this purpose, the profile is converted into a sequence of reversal points classified according to SoC, and closed hystereses are then counted. This makes it possible to determine the average state of charge and the depth of discharge (DoD). Based on previous laboratory tests or field data, damage can be assigned to each completed (micro) cycle and summed up. The damage describes the loss of capacity and / or the increase in impedance. The accuracy of the rainflow counting algorithm depends significantly on the state of charge estimate.

[0050] Analogous to the described state variables of charge state and aging state, the safety state ("SoS"), the usability state ("SoF"), the charge state distribution ("SoB"), the internal cell temperatures and their respective errors are estimated by applying various estimation methods in parallel. An overview of the applicable estimation methods is provided in Figure 4 The SoF and SoB are higher-level state variables that can be determined from the SoC and SoH state variables, or from SoC and internal temperature, respectively. The applicable methods have already been discussed above, so a repetition here seems obsolete.

[0051] A device according to the invention is described below with reference to the above exemplary embodiment: The device according to the invention comprises a plurality of temperature and voltage sensors on each of the 1296 battery cells installed in the battery system, a current measuring device on each of the six cell clusters, and one or more computing units that centrally or distributedly control the temporal sequence of the method according to the invention, acquire the sensor data, and calculate and compare the state variables and their errors, and then evaluate the changes. To cope with the comparatively immense computational effort, the calculation of the state variables is distributed across subunits (master-slave system). The high-resolution determination and evaluation of the state variables is carried out for each battery cell by a computing unit assigned to it.The determined state variables are communicated from these processing units to a central, higher-level master processing unit. This unit consolidates the information and communicates it to the higher-level energy management system or to a battery management system (BMS) (e.g., the vehicle computer for traction batteries).

[0052] The BMS enables synchronous recording of internal battery currents and individual cell voltages and their processing of at least ten data points per second, which is particularly necessary for the load step algorithm described above.

[0053] The BMS monitors the proper functioning of the sensor and actuator units installed in the battery as well as the data processing, on the one hand through self-monitoring functions of the computing units and their peripherals, on the other hand through redundant design of the processing units and further through plausibility checks of the data recorded in the sensor units.

[0054] The BMS is not installed in the high-voltage battery, but is connected to the subunits via redundant, galvanically isolated bus systems. The power supply, at least for the BMS master processing unit, is provided, for example, by the vehicle's electrical system, which houses the BMS and the battery system. This ensures that in the event of a failure of the battery system and / or small subassemblies of the battery system, the BMS remains functional and can trigger the actuators.

[0055] Due to the relatively complex algorithms, microcontrollers with a flash memory size of at least 2.5 MB and a powerful multi-core processor architecture are used in the BMS. Despite the increased performance, the BMS's own power consumption in standby mode is a maximum of 100 µA.

[0056] In addition to the functions mentioned above, thermal management is provided via the BMS, which includes temperature detection and control of the actuators.

[0057] To prevent manipulation of the algorithms by third parties, so-called crypto algorithms are used.

[0058] Disclosed is a method and a device for robust online state determination of large battery storage systems, wherein several state variables and their errors can be determined using several parallel estimation methods. List of reference symbols

[0059] S1Step 1 S2Step 2 S3Step 3 S4Step 4 S5Step 5 S6Step 6 S7Step 7 S8Step 8

Claims

1. Method for cyclic determination of the state of battery system including a plurality of battery cells, in which, in one cycle per battery cell, a plurality of different state variables is determined by a plurality of different estimation processes per state variable to be determined is applied in parallel, wherein the estimation processes output the respective state variable and an error assigned to the state variable, in which, for each state variable to be determined, the state variable with the lowest error is selected from the plurality of state variables output by the estimation methods and in which the selected state variable and the error associated with this state variable are used as the current state variable and current error in the next process cycle.

2. Method according to claim 1, characterized in that a state of charge, a state of aging, a state of safety, a state of usability, a state-of-charge distribution and / or a battery cell internal temperature represent a selection and / or a combination of the state variables to be determined.

3. Method according to claim 1 or 2, characterized in that Incremental Capacity Analysis (ICA) open-circuit voltage adjustment, step change in load or Coulomb Counting are used as estimation methods.

4. Method according to claim 1 or 2, in which an estimation procedure is carried out by capacity determination, impedance determination, determination of an irreversible thickness increase or by means of a Rainflow counting algorithm.

5. Device adapted for carrying our a method according to any one of the preceding claims 1 to 4, comprising a plurality of temperature and voltage sensors operationally connected to the battery cells, wherein a temperature sensor and a voltage sensor are respectively applied to a battery cell, wherein a current measurement device is respectively provided at a cell bond, wherein several computing units are provided which are designed for controlling the method and which process the data collected by the sensors, wherein the computing units are configured to not only determine the data collected but also its errors, compare the data and the errors with each other and carry out an evaluation, wherein several computing units are assigned a higher-level master computing unit that is designed to receive the data recorded by the computing units, to bundle it and to transmit it to a higher-level battery management system.

6. Device according to claim 5, wherein the voltage sensors are designed to be operated with a measuring accuracy of 0.5 to 2 mV, preferably 1 to 1.5 mV, preferably 1.2 mV, and a resolution of 0.05 to 0.2 mV, preferably 0.1 mV.

7. Device according to claim 5, wherein the temperature sensors are designed to be operated with a measuring accuracy of 0.5 to 2 K, preferably 0.8 to 1.5 K, in particular 1 K, and a resolution of 1 / 20 to 1 / 5 K, preferably 1 / 10 K.

8. Device according to claim 5, wherein the current measurement device is designed to be operated with a measuring accuracy of <1,5 A, preferably < 1 A, and a resolution of 0.5 to 2 mA, preferably 0.8 to 1.5 mA, preferably 1 mA.

9. Device according to claim 5, wherein the device is designed with redundant, galvanically decoupled bus systems remote from the battery system and is connected to a voltage supply that is independent of the battery system.