Battery Performance Prediction

Abstract

A test system (110) is proposed for determining battery performance during the development of battery constructs in a test environment. The test system (110) includes at least one communication interface (114) and at least one processing device (122). The test system (110) is configured to receive operation data indicating at least one test protocol via the communication interface (114). The test system (110) is configured to receive battery performance input data via the communication interface (114). The processing device (122) is configured to determine at least one predicted time series of at least one state variable indicating battery performance based on the battery performance input data and the operation data using at least one data-driven model. The test system (110) is configured to provide at least a portion of the predicted time series of the state variable.

Description

Technical Field

[0001] The present invention relates to a test system for determining battery performance during the development of battery configurations in a test environment, and a test bench configured to perform at least one battery performance test on at least one battery based on at least one test protocol. The present invention further relates to a method for determining at least one data-driven model for determining battery performance during the development of battery configurations in a test environment, a computer-implemented method for determining battery performance during the development of battery configurations in a test environment, and a computer program for determining battery performance during the development of battery configurations in a test environment. The battery under test can be used in the automotive industry such as for electric vehicles, in consumer devices such as smart phones, laptops, etc., and in energy storage devices for storing energy obtained from renewable energy sources, for example. The methods, devices and systems according to the invention are particularly useful for determining the lifetime of a battery on a test bench. Other applications are possible. Background Art

[0002] Batteries are used in different technical fields, such as for powering vehicles, especially electric vehicles, in consumer devices, and even as energy storage devices for storing energy obtained from renewable energy sources, for example. Battery performance may change over time. In particular, battery performance may degrade depending on the number of charge and discharge cycles. A reliable and fast prediction of battery performance is necessary in all of the above technical fields.

[0003] In order to develop new and advanced battery materials, especially cathode materials, for specific applications, battery performance is tested by performing multiple experiments. Subsequently, the experimental results are analyzed and the cathode material under test can be classified as a "good" or "bad" cathode material. The test results obtained can be used to determine whether the cathode material needs to be adjusted.

[0004] The use of known test methods for determining battery performance during the development of a battery requires a long test time. In particular, determining battery performance on a test bench typically requires multiple charge and discharge cycles over the lifetime of the battery and may take several months. For example, it usually takes six months until it can be determined whether the cathode material used can be classified as a "good" or "bad" cathode material. This takes a long time and results in a long development time. The long test time may even pose problems for the logical planning of the test bench capacity. It is usually possible to manage more than 100 channels and a reliable capacity prediction of the test bench is required.

[0005] "Data-driven prediction of battery cycle life before capacity degradation" by Kristen A. Severson et al., Nature Energy, https: / / doi.org / 10.1038 / s41560-019-0356-8 describes predicting the life of lithium-ion batteries to accelerate technology development. Discharge voltage curves based on early cycles are used to show capacity degradation, and machine learning tools are described to predict and classify battery cells by cycle life.

[0006] US2019 / 0115778A1 describes a method for exploring the multi-dimensional parameter space of a battery cell test protocol, which includes defining a parameter space for a plurality of battery cells to be tested, discretizing the parameter space, collecting a preliminary set of cells cycled to failure to sample strategies from the parameter space and including multiple repetitions of the strategies, specifying resource hyperparameters, parameter space hyperparameters, and algorithm hyperparameters, selecting a random subset of charging strategies, testing the random subset of charging strategies until the number of cycles required for an early prediction of battery life is reached, inputting the early predicted cycle data into an early prediction algorithm to obtain an early prediction, inputting the early prediction into an optimal experimental design (OED) algorithm to obtain a recommendation for running at least one next test, running the recommended test by repeating the above random subset test step, and validating the final recommended strategy.

[0007] WO2019 / 017991A1 describes a battery management system (BMS) for a vehicle, including a module for real-time estimating the state of a rechargeable battery, such as its state of charge. The module includes a learning model for predicting the battery state based on vehicle usage and related factors specific to the vehicle in addition to the sensed voltage, current, and temperature of the battery.

[0008] Problems to be Solved

[0009] Therefore, it is desirable to provide methods and devices for solving the above technical challenges. Specifically, devices and methods for determining battery performance during battery construction development in a test environment should be provided, which ensures a reduction in measurement time in a test bench for battery materials. In addition, the number of required experiments should be ensured to be reduced in order to accelerate the modification of cathode materials for specific applications in less time. Summary of the Invention

[0010] This problem is solved by a test system for determining battery performance during battery construction development in a test environment, having the features of the independent claims, a test bench configured to perform at least one battery performance test on at least one battery based on at least one test protocol, a method for determining at least one data-driven model for determining battery performance during battery construction development in a test environment, a computer-implemented method for determining battery performance during battery construction development in a test environment, and a computer program for determining battery performance during battery construction development in a test environment. Advantageous embodiments that can be implemented independently or in any arbitrary combination are listed in the dependent claims.

[0011] As used hereinafter, the terms "having", "comprising", "including" or any grammatical variants thereof are used in a non-exclusive manner. Thus, these terms can refer both to a situation where no other features exist in the entity described in this context apart from the features introduced by these terms, and to a situation where one or more other features exist. For example, the expressions "A has B", "A includes B" and "A contains B" can all refer to a situation where no other elements exist in A apart from B (i.e., the situation where A consists solely and exclusively of B), or to a situation where, apart from B, one or more other elements also exist in entity A, such as element C, elements C and D or even other elements.

[0012] Furthermore, it should be noted that the terms "at least one", "one or more" or similar expressions indicating that a feature or element can occur once or more than once are generally used only once when introducing the corresponding feature or element. In most cases hereinafter, the expressions "at least one" or "one or more" will not be repeated when referring to the corresponding feature or element, although the corresponding feature or element can occur once or more than once.

[0013] Furthermore, as used hereinafter, the terms "preferably", "more preferably", "particularly", "more particularly", "specifically", "more specifically" or similar terms are used in conjunction with optional features without restricting the possibilities of alternatives. Thus, the features introduced by these terms are optional features and are not intended to limit the scope of the claims in any way. As will be recognized by those skilled in the art, the present invention can be implemented by using alternative features. Similarly, features introduced by "in an embodiment of the present invention" or similar expressions are intended to be optional features, without any limitation to alternative embodiments of the present invention, without any limitation to the scope of the present invention, and without any limitation to the possibility of combining the features introduced in this way with other optional or non-optional features of the present invention.

[0014] In a first aspect of the present invention, a test system is proposed for determining battery performance during battery construction development in a test environment. The test system includes at least one communication interface and at least one processing device. The test system is configured to receive operation data indicating at least one test protocol via the communication interface. The test system is configured to receive battery performance input data via the communication interface. The processing device is configured to determine at least one predicted time series of at least one state variable indicating battery performance based on the battery performance input data and the operation data using at least one data-driven model. The test system is configured to provide at least a partial predicted time series of the state variable.

[0015] During battery development, different materials and material compositions are used and tested with respect to performance and / or characteristics and / or behavior, in particular the development of charge and discharge characteristics and / or behavior over time. These tests can be performed in a test environment within predefined test conditions such as temperature conditions. Specifically, the test environment can be a test bench configured to apply at least one test to a battery, in particular to a plurality of batteries. During these tests, the battery under test can undergo a plurality of charge-discharge cycles according to at least one test protocol. Battery testing in battery development is more complex compared to battery quality control where only the number of charge cycles is determined before the capacity decays to 80% of the initial value of the battery. For battery testing during battery development, it is necessary to determine the temporal characteristics and the development over time of battery characteristics. The present invention proposes a test system that allows predicting battery performance during battery development based on battery performance input data without long-term testing and only by using a trained data-driven model.

[0016] As used herein, the term "test system" is a broad term and will be given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a proprietary or custom meaning. The term can specifically refer to, but is not limited to, a device including one or more units that is configured to determine and / or predict battery performance, in particular the development of charge and discharge characteristics and / or behavior over time.

[0017] As used herein, the term "battery" is a broad term and will be given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a proprietary or custom meaning. Specifically, the term may refer to an electrochemical cell including, but not limited to, at least one anode, at least one cathode, and at least one electrolyte. Specifically, the battery is configured to convert chemical energy into electrical energy and vice versa. The battery may be an energy storage device. The battery may be a rechargeable battery. As used herein, the term "charging" refers to converting electrical energy provided from an external source into chemical energy in the battery. As used herein, the term "discharging" refers to converting the chemical energy of the battery into electrical energy. The battery may be selected from the group consisting of: lithium-ion battery (Li-Ion); nickel-cadmium battery (Ni-Cd); nickel-metal hydride battery (Ni-MH). For example, the battery may include at least one cathode material selected from the group consisting of: LiCoO2 (lithium cobalt oxide); LiNixMnyCozO2 (lithium nickel manganese cobalt oxide), and LiFePO4 (lithium iron phosphate). For example, the battery may include at least one anode material selected from the group consisting of: graphite, silicon. For example, the battery may contain at least one electrolyte selected from the group consisting of: LiPF6, LiBF4, or LiClO4 in an organic solvent such as ethylene carbonate, dimethyl carbonate, and diethyl carbonate.

[0018] As used herein, the term "battery performance during battery development" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a special or custom meaning. Specifically, the term may refer to, but is not limited to, the development of battery behavior over time. Battery performance may be characterized by the development over time of one or more parameters such as charge capacity, discharge capacity, discharge current, charge-discharge curve, average voltage, open-circuit voltage, differential capacitance, Coulombic efficiency, or internal resistance. In particular, battery performance during battery development is determined by determining a predicted time series of at least one of these parameters. As used herein, the term "time series" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a special or custom meaning. Specifically, the term may refer to, but is not limited to, a data stream arranged in chronological order.

[0019] As used herein, the "capacity" of a battery refers to the amount of electric charge transferred by the battery at a rated voltage. As used herein, the term "charge capacity" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a special or custom meaning. Specifically, the term may refer to, but is not limited to, the capacity that can be charged into the battery. As used herein, the term "discharge capacity" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a special or custom meaning. Specifically, the term may refer to, but is not limited to, the capacity that can be discharged from the battery.

[0020] As used herein, the term "discharge current" (also labeled as C-rate) is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a specific or customized meaning. Specifically, this term can refer to, but is not limited to, a measure of the rate at which a battery is discharged or charged relative to its maximum capacity.

[0021] As used herein, the term "charge-discharge curve" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the development of charge and discharge voltages based on battery capacity. The shape of the charge and / or discharge curve can be parameterized by the change in capacity over a predefined voltage interval. The shape of the charge and / or discharge curve can contain information about electrochemical and degradation mechanisms.

[0022] As used herein, the term "average voltage" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the rated voltage at the midpoint of its discharge cycle.

[0023] As used herein, the term "open-circuit voltage" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the voltage between the terminals of a battery with no load applied.

[0024] As used herein, the term "differential capacitance" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the first derivative dQ / dE of the capacitance Q with respect to the voltage E.

[0025] As used herein, the term "coulombic efficiency" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the charging efficiency through which electrons are transferred.

[0026] As used herein, the term "internal resistance" is a general term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, the resistance of a battery that increases as the battery ages. In addition, the internal resistance can give a measure of the degree of aging of the material surface.

[0027] The parameters listed above can be used to determine battery performance, particularly to predict the development of battery performance over time. The relationships between these parameters and battery performance are generally known to those skilled in the art.

[0028] As used herein, the term "determining battery performance" is a broad term and is given its ordinary and accustomed meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, one or more of predicting, estimating, and classifying a time series of future values and / or the development of at least one state variable over time. The determination of battery performance can have as an output or result at least one prediction of the battery performance for the time series. To determine the predicted time series, the values of the state variables at different time points can be iteratively determined for multiple future time points. As used herein, the term "prediction" refers to the expected value of a state variable in the future. The result of the battery performance determination can be a predicted time series of state variables, such as a histogram showing the development of the state variable over time. The determination of battery performance can include predicting battery life and / or classification. As used herein, the term "classification" can refer to the classification of a battery and / or cathode material, for example as "good" or "bad". For example, if the state variable at a future time point meets a predetermined or predefined state and / or the determined life is higher than a predetermined or predefined limit, the battery can be classified as "good". For example, if the predicted time series of the state variable does not meet a predetermined or predefined condition, the battery can be classified as "bad". For example, if the state variable at a future time point does not meet a predetermined or predefined condition and / or the determined life is lower than a predetermined or predefined limit, the battery can be classified as "bad". The test system can include at least one output interface configured to output at least one output that includes information about at least a portion of the predicted time series of the state variable.

[0029] As used herein, the term "processing device" is a broad term and is given its ordinary and accustomed meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, any logic circuit configured to perform the basic operations of a computer or system, and / or a device generally configured to perform computational or logical operations. The processing device can include at least one processor and / or processing unit. In particular, the processing device can be configured to process the basic instructions that drive the computer or system. For example, the processing device can include at least one arithmetic logic unit (ALU), at least one floating point unit (FPU) such as a math coprocessor or a digital coprocessor, and multiple registers and memories such as a cache. In particular, the processing device can be a multi-core processor. Specifically, the processing device can be or can include a central processing unit (CPU). Alternatively or additionally, the processing device can be or can include one or more application specific integrated circuits (ASICs) and / or one or more field programmable gate arrays (FPGAs), etc.

[0030] As used herein, the term "communication interface" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, an item or element that forms a boundary configured to convey information. In particular, a communication interface can be configured to transfer information from a computing device such as a computer, for example, to send or output information to another device such as another device. Additionally or alternatively, a communication interface can be configured to transfer information onto a computing device, such as onto a computer, for example, to receive information. A communication interface can specifically provide a channel for transmitting or exchanging information. In particular, a communication interface can provide a data transfer connection, such as Bluetooth, NFC, inductive coupling, etc. As an example, a communication interface can be or can include at least one port, which includes one or more of a network or Internet port, a USB port, and a disk drive. A communication interface can be at least one network interface.

[0031] A test system can include at least one database. As used herein, the term "database" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, any collection of information, such as information stored in at least one data storage device. A database can include at least one data storage device having information stored therein. In particular, a database can contain any collection of information. A database can be or can include at least one database selected from the group consisting of: at least one server, at least one server system including a plurality of servers, at least one cloud server, or a cloud computing architecture. A database can include at least one storage unit configured to store data received via a communication interface.

[0032] As used herein, the term "operational data" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, data related to information about at least one test program performed on a battery in a test bench, particularly one or more of operating conditions, test sequence, process, etc. Operational data indicating at least one test protocol includes, for example, at least one sequence of different charge cycles and / or discharge cycles. As used herein, the term "test protocol" is a broad term and is given its ordinary and relevant meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, guidelines according to which multiple different test programs are performed on a battery in a test bench. The test protocol can be predefined. The test protocol can be known. In particular, quantities such as discharge current and duration of constant voltage operation can be controlled and / or defined by the test protocol. The test protocol can be executed according to at least one standard. The test protocol can be specific to each test setup. As will be outlined in detail below, the test protocol can allow prediction of the performance of battery materials, particularly the performance of the cathode and / or anode. The test protocol can define the order of test programs and / or the sequence of test programs and / or the duration of each test program. Each test program can include at least one charge-discharge cycle, where the charge-discharge cycles of at least two test programs are different. The test protocol can include information about at least one battery performance test. The battery performance test can include at least one sequence of different charge cycles and / or discharge cycles. In the battery performance test, a discharge-charge curve can be determined for each cycle. The battery performance test can be performed by a customer. The customer can provide battery performance input data to the test system via a communication interface. The test protocol, also referred to as a test program, can include different blocks to determine different material properties. For example, a cycling test protocol can have a charge step for each cycle, followed by a discharge step, which can include an optional rest between the steps. The charge step can apply a constant current to the battery cell until the battery cell voltage reaches a predetermined threshold, and then hold the voltage until the current drops below another predetermined threshold or reaches a time threshold. The discharge step can hold a constant current until the voltage drops below a predetermined threshold. Multiple such cycles can be performed continuously to measure the aging of the battery cell.

[0033] Specifically, the data-driven model is parameterized based on operation data indicating at least one test protocol and battery performance input data. The data-driven model can use knowledge of past and future charge-discharge cycles following at least one test protocol to predict future battery performance. In particular, knowledge of past and future charge-discharge cycles can be used as input to the data-driven model. The data-driven model can consider previous or prior measurements at each time step, previous or prior predictions made by the data-driven model, and future values of quantities controlled and / or defined by at least one test protocol. The data-driven model can use knowledge of past charge-discharge cycles to predict the future, particularly future battery performance. The data-driven model can use knowledge of future charge-discharge cycles from the test protocol to predict future battery performance. The test protocol can control and / or define variables such as discharge current, charge current, rest steps, thresholds for starting a new cycle, etc. Specifically, knowledge of future values of the quantities controlled and / or defined is used to predict future battery performance. In particular, information predefined by the test protocol about how much stress the battery will undergo during a certain future cycle can be used as additional and useful information for predicting future battery performance. Thus, the present invention particularly solves the problem of allowing the assessment of the future battery state, or in other words how much electrical energy the battery will provide after N future operating cycles, where the details of the operation can be specified by a cycling program and / or a test protocol. The present invention proposes to particularly use the test protocol and thereby use the planned cycling program as input to the data-driven model. Information using knowledge of future test parameters in the protocol can allow for further improvement of performance prediction. In particular, the use of the defined, especially predefined and / or known test protocol allows for the possession and use of knowledge of past and future charge-discharge cycles. In contrast, in a battery management system of a vehicle as described in, for example, WO 2019 / 017991A1, where the battery uses data that does not follow a test protocol, only data from previous battery use, particularly unstructured historical data, can be considered. WO 2019 / 017991A1 only predicts that the battery has a certain number of remaining cycles currently, while the present invention particularly proposes that if the battery is further processed in a predefined manner following a test protocol, how many remaining cycles in the future can be predicted. The present invention can allow for the prediction of the future capabilities of the battery based on or considering the use of a test protocol.

[0034] Specifically, the data-driven model can have a time memory and / or the data-driven model can be a time-dependent model. Input data can be fed into a data-driven model that includes information about measurement data obtained at a certain point in time. The data-driven model can provide at least one output, in particular so-called "latent" variables. The latent variables can include information about the predicted future battery behavior and / or information about the predicted future battery behavior that can be derived from the latent variables. For subsequent time steps, the data-driven model can take into account additional measurement data and / or at least one latent variable obtained up to that point in time, in particular as input data. The data-driven model can be configured to determine the relevance of the individual latent variables and to assign weights to the latent variables. The data-driven model can consider the weighted latent variables for further prediction. The determination of the relevance and the assignment of the weights can be repeatedly executed. As described above, the data-driven model can further take into account knowledge of the future values of the controlled and / or defined quantities predefined by at least one test protocol, in particular as input data. Known methods such as those described in WO2019 / 017991A1 use methods referred to as "fixed window" or "fixed level". They do not use the "time series" method proposed by the present invention, which particularly proposes a data-driven model that is time-dependent. The present invention particularly proposes using a data-driven model that steps through time and retains relevant memory. In contrast, the "fixed window" or "fixed level" methods use information from the past and see how it relates to the behavior at a fixed future time.

[0035] As used herein, the term "battery performance input data" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, data that includes information about battery behavior and / or performance generated in response to a test protocol. The battery performance input data can include data generated in response to a test protocol. The battery performance input data can be or can include raw data and / or preprocessed data. The battery performance input data can be determined by performing at least one test procedure, in particular as outlined above following at least one test protocol. The test procedure can include determining at least one discharge-charge curve of the battery. The test procedure can be performed on a test bench. The battery performance input data can be transmitted to the test system in real time via a communication interface or in batches with a delay.

[0036] Battery performance input data may include charge-discharge cycle data. As used herein, the term "cycle" refers to a discharge sequence following a charge and vice versa. The charge-discharge cycle data may include at least one charge-discharge curve. The battery performance input data may include one or more of information about discharge capacity, information about charge capacity; information about the shape of the charge-discharge curve; information about average voltage; information about open-circuit voltage, information about differential capacity, information about coulombic efficiency; and information about internal resistance.

[0037] Battery performance input data may include metadata related to one or more of the cathode material and the cell setup. As used herein, the term "metadata" refers to data that includes information about the charge-discharge cycle data. The metadata can be used to select an appropriate trained model, for example, taking into account the cathode material and / or the cell setup.

[0038] The processing device may be configured to verify the battery performance input data. The verification may include determining whether the retrieved battery performance input data is complete and / or whether it includes sufficient cycles for determining battery performance. As used herein, the term "verification" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term may specifically refer to, but is not limited to, the process of reviewing, examining, or testing one or more of the battery performance input data. Specifically, the verification includes determining whether the retrieved battery performance input data is complete and / or whether it includes sufficient cycles for determining battery performance. For example, in the case where the verification reveals that the retrieved battery performance input data is incomplete and / or does not contain sufficient cycles for determining battery performance, a request may be sent to the customer via the communication interface to provide additional data.

[0039] As used herein, the term "state variable" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term may specifically refer to, but is not limited to, a quantifiable variable of at least one parameter that can characterize battery performance. The state variable may be derived from or derivable from at least one charge-discharge curve. The state variable may be at least one variable selected from the group consisting of: discharge capacity; charge capacity; shape of the charge-discharge curve; average voltage; open-circuit voltage; differential capacitance; coulombic efficiency; and internal resistance. For each cycle, the model may predict each available feature as an input to the model for the next cycle.

[0040] The processing device can be configured to select information from the battery performance input data depending on a state variable whose future evolution is to be predicted. The processing device can be configured to sort the battery performance input data depending on a state variable whose future evolution is to be predicted. For example, the state variable can be the discharge capacity. The sorting can be as follows: the charge capacity from previous cycles and / or the discharge capacity from previous cycles can be considered the most relevant as they are very close to the state variable to be predicted. The shape of the charge and / or discharge curves and the internal resistance from previous cycles can be considered less relevant than the charge capacity from previous cycles and / or the discharge capacity from previous cycles. For other state variables such as the internal resistance, the sorting may be different.

[0041] The term “predicted time series of a state variable” used herein is a general term and is given its ordinary and customary meaning to a person of ordinary skill in the art without being limited to a special or customized meaning. The term can specifically refer to, but is not limited to, an expected time series of a state variable determined using at least one data-driven model. The battery performance input data of the battery can be experimental data. By using a data-driven model, in particular, the predicted time series of a state variable can be determined, where the battery performance input data is used as an input to the data-driven model. The battery performance input data can include experimental data based on which the predicted time series of a state variable is predicted. In addition, the predicted time series is determined according to the operation data. The test system can include at least one storage device in which multiple different data-driven models can be stored. For example, the storage device can include different data-driven models for different test protocols. In particular, the processing device can be configured to select a data-driven model based on the test protocol for the battery test. Additionally or alternatively, the data storage device can include different data-driven models depending on the state variable to be predicted. The processing device can be configured to determine the predicted time series of one state variable or multiple state variables. At the same time, the time series of any combination of state variables can also be predicted.

[0042] As used herein, the term "data-driven model" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, empirical prediction models. Specifically, a data-driven model is obtained from the analysis of experimental data. A data-driven model can be a machine learning tool. A data-driven model can include at least one trained model. As used herein, the term "trained model" is a broad term and is given its usual and customary meaning to a person of ordinary skill in the art and is not limited to a specialized or customized meaning. The term can specifically refer to, but is not limited to, a model trained on at least one training data set (also referred to as training data) for predicting battery performance. For example, a data-driven model is trained on at least one training data set, where the training data set includes a time series of historical data of charge and discharge cycles of at least one known battery configuration and at least one known test protocol. The training data can be obtained according to a well-defined test protocol. The training data set can include experimental results of multiple battery tests, such as one or more parameters for characterizing battery performance. The training data set can include multiple experimental results of battery tests determined at at least two time points (such as for different cycles). For the training of the model, multiple training data sets related to different batteries to be tested can be used, for example.

[0043] A data-driven model can be a feature-based model, where features or a subset of features are used to predict battery performance. The training data can be used to select features for the trained model. Feature selection can include selecting a subset of relevant features to be used in model construction, particularly variables and predictors. Feature selection can include deriving information about electrochemical aspects at certain time points, such as charge or discharge curves, internal resistance, open-circuit voltage, differential capacitance at relevant time points. Irregularities due to material classes can be considered for feature selection. The training data can include raw data. Feature selection can include data aggregation, such as from each step to each cycle. Electrochemical knowledge and / or traditional feature selection methods can be used to select features, such as retaining features highly correlated with the response. Electrochemical knowledge can include, for example, information about voltage intervals in which phase transitions of interest are expected and thus features should be extracted from curve regions.

[0044] The data-driven model may include at least one recurrent neural network, such as at least one echo state network. As used herein, the term "echo state network" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term may specifically refer to, but is not limited to, a recurrent neural network with leaky integrators for discrete-time continuous-valued units. For more details on the data-driven model, reference may be made to the description of the method for determining at least one data-driven model, which is described in more detail below and is used to determine battery performance during the development of battery constructs in a test environment.

[0045] The processing device may be configured to use the battery performance input data as input parameters in order to determine a predicted time series of state variables using the data-driven model. In order to determine the battery performance of the battery under test, the battery performance input data may be processed and applied as input to the data-driven model. The processing may include data aggregation. The processing may include selecting information related to at least one of the parameters characterizing the battery performance, the time series of which is predicted by the data-driven model and is also represented as the target state variable. In this case, the output generated by the data-driven model may be the predicted time series of the target state variable or the matrix of state variables.

[0046] The processing device may be configured to use the test protocol as input parameters in order to determine a predicted time series of state variables using the data-driven model. Specifically, the processing device may be configured to select at least one suitable data-driven model based on the received battery performance input data, such as based on the received information about the test protocol and / or the sequence of different charge cycles and / or discharge cycles. The processing device may include a plurality of data-driven models, wherein the processing device is configured to select one data-driven model in order to determine a predicted time series of state variables according to the test protocol. For example, each of the plurality of data-driven models may be trained on data derived by using different test protocols. The models may have different fitting parameters and different hyperparameters that control the model complexity. However, for all models, the general structure may be the same. Other criteria for selecting the most suitable data-driven model are also possible. For example, depending on differences in material properties, the processing device may include a plurality of data-driven models. The processing device may be configured to analyze the battery performance input data, where the analysis includes determining at least one material property. The processing device may be configured to select at least one of the data-driven models based on the material properties. Information about the material properties may be determined from the battery performance input data such as metadata.

[0047] The processing device can be configured to perform at least one confidence test for comparing the determined state variables with experimental test results. As used herein, the term "confidence test" is a broad term and is given its ordinary and customary meaning to a person of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, the process of reviewing, examining, or testing one or more of the outputs of a data-driven model for conformance to one or more experimental test results. In a confidence test, the predicted time series of the state variables can be compared with the experimental test results. The experimental results for different battery types or materials can be stored in at least one data memory and can be used for comparison in the confidence test.

[0048] The test system is configured to provide at least a portion of the predicted time series of the state variables. As used herein, the term "at least a portion" refers to embodiments that provide the complete predicted time series of the state variables and / or embodiments that provide at least one specific time range or time point of the predicted time series of the state variables and / or embodiments that provide other information related to the predicted time series. The test system can include at least one output interface that is configured to output at least one output including information about at least a portion of the predicted time series of the state variables. The output can include one or more of at least one histogram showing the development of the state variables over time. The output can also include information about at least one prediction and / or at least one classification of the battery life. The test system can be configured to perform at least one measure depending on the predicted time series of the state variables, where the measure includes one or more of the following: issuing a recommendation; issuing a warning; issuing an indication to modify the battery cathode material. As used herein, the term "measure" refers to any action depending on the determined result of the battery performance. The test system can be configured to provide at least one information about the determined battery performance to the customer of the battery. For example, the information can be provided to the customer by the output interface. The information can also include one or more of the ranking of the battery in the test bench, battery classification, cathode material classification, or at least one recommendation. The output can include the ranking of the battery in the test bench and / or a recommendation to stop the battery measurement. The communication interface can be configured to allow monitoring of the state of the test equipment. This can allow preparation and design of the next experiment based on the determined result of the battery performance.

[0049] In another aspect of the present invention, there is provided a test bench configured to perform at least one battery performance test on at least one battery based on at least one test protocol for determining the battery performance of at least one battery.

[0050] As used herein, the term "test bench" is a broad term and is given its ordinary and customary meaning to one of ordinary skill in the art and is not limited to a special or customized meaning. The term can specifically refer to, but is not limited to, an environment for testing the characteristics of at least one battery. The test bench can be configured to test multiple batteries. The test bench can include at least one storage device for storing batteries during a test or test sequence.

[0051] The battery performance test includes at least one sequence of different charge cycles and / or discharge cycles. The battery performance test includes determining the discharge-charge curve for each cycle. The test bench according to the present invention includes at least one communication interface configured to provide operation data indicating a test protocol and battery performance input data to at least one test system. For the definition and implementation of the test bench, reference is made to the description of the test system.

[0052] In another aspect of the present invention, a method for determining at least one data-driven model for determining battery performance during battery construction development in a test environment is proposed. The method includes training a data-driven model using at least one training data set. The training data set includes historical data of charge and discharge cycles of at least one known battery construction and at least one known test protocol. For the definition and implementation, reference is made to the description of the test system above or the further detailed description below.

[0053] The method can include the following steps:

[0054] - Generate at least one random dynamic repository;

[0055] - Determine at least one input unit and at least one output unit;

[0056] - Generate connections from the input to the repository and connections from the output to the repository;

[0057] - Select at least one input weight matrix W in and at least one repository weight matrix W;

[0058] - Train the model using the training data by training the data-driven model with the training data to determine the output weights, where regression analysis is used to determine the output weights.

[0059] For example, the data-driven model can be at least one echo state network. Any neuron model can be used to generate a random dynamic reservoir. The determination of the data-driven model can include applying training data, in particular training data experimentally determined by measuring at least one parameter characterizing the battery performance of a training test battery at a first time point, as the input u(n) to the random dynamic reservoir. Thus, the input u(n) can be filled into the input unit as an input state and applied to the reservoir such that a reservoir neuron activation vector x(n) is generated. The determination of the data-driven model can include generating connections from the output to the reservoir to allow output feedback. The output unit can be filled with the so-called training data, in particular the output y(n) of training data experimentally determined by measuring at least one parameter characterizing the battery performance of a training test battery at a second time point delayed from the first time point. The output weights W out can be determined using regression analysis (in particular linear regression) of the taught output y(n) with respect to the reservoir state x(n). The output weights W out for the connections from the output to the reservoir. The training can include multiple training cycles, such as using different training data sets for different cathode materials, etc.

[0060] The echo state network can use the following system equations:

[0061]

[0062]

[0063] where u(n) is the input matrix of the echo state network, is the vector of reservoir neuron activations, and is its update at time n, f is an activation function, in particular a sigmoid function such as the logistic sigmoid or tanh function, and is applied element-wise, [;] represents the concatenation of vertical vectors or matrices, and are the input weight matrix and the recurrent weight matrix, respectively, α ∈ (0, 1] is the leakage rate, see "A Practical Guide to Applying Echo State Networks", Mantas Lukosevicius, published in Neural Networks: Tricks of the Trade 2012. DOI: 10.1007 / 978-3-642-35289-8_36 and http: / / minds.jacobs-university.de / uploads / papers / PracticalESN.pdf. The taught output y(n) can be given by the following formula:

[0064] y(n) = g(W out z(n)),

[0065] where g is the output activation function, W out is the output weight matrix, and z(n) is the extended system state, where z(n) = [x(n); u(n)] at time n is the concatenation of the reservoir and input states. As outlined above, during the training phase, the echo state network output weight W out can be determined using regression analysis on the desired output derived from the training dataset. After training, the echo state network can be used to determine the battery performance of the battery under test.

[0066] To determine the battery performance of the battery under test, the battery performance input data can be processed and applied as input to the echo state network. This processing can include data aggregation and / or selection of information related to one or more of the parameters characterizing the battery performance, which battery parameters can be determined by the trained echo state network and are also represented as target state variables. In this case, the output y(n) of the echo state network can be the target state variable or a matrix of state variables at some future time point or future period.

[0067] In another aspect of the present invention, a computer-implemented method for determining battery performance during battery construction development in a test environment is proposed. In this method, at least one test system according to the present invention is used. Therefore, for the implementation manner and definition of the method, reference is made to the above description of the test system or the further detailed description below.

[0068] As used herein, the term "computer-implemented" is a broad term and is given its ordinary and customary meaning to those of ordinary skill in the art and is not limited to a special or customized meaning. Specifically, this term can refer to, but is not limited to, a process that is fully or partially implemented by using a data processing device (such as a data processing device including at least one processor). Therefore, the term "computer" generally can refer to a device or a combination or network of devices having at least one data processing device, such as at least one processor. In addition, a computer can include one or more additional components, such as at least one of a data storage device, an electronic interface, or a human-machine interface.

[0069] The method includes the following method steps. Specifically, the method steps can be executed in a given order. However, different orders are also possible. Two or more method steps can also be executed fully or partially simultaneously. Further, one or more or even all of the method steps can be executed once or can be repeatedly executed, such as being repeated one or more times. In addition, the method can include additional method steps not listed.

[0070] The method comprises the following steps:

[0071] a) Retrieving operation data indicating at least one test protocol via at least one communication interface;

[0072] b) Retrieving battery performance input data via the communication interface;

[0073] c) Determining a predicted time series of state variables indicating battery performance based on the battery performance input data and the operation data by using a data-driven model with a processing device;

[0074] d) Providing at least a part of the predicted time series of the state variables.

[0075] Method steps a) to d) can be performed in whole or in part in a computer-implemented manner. In method steps a) and b), the operation data indicating at least one test protocol and the battery performance input data can be transmitted via the communication interface, for example, to the test system through a web interface, especially to the processing device for analysis in step c).

[0076] In another aspect of the present invention, a computer program for determining battery performance during the development of a battery configuration in a test environment is proposed. The computer program comprises instructions that, when the program is executed by a computer or a computer system, cause the computer or the computer system to perform the method for determining battery performance as described above or in more detail hereinafter. In particular, one, more than one or even all of the method steps a) to d) of the method as indicated above can be performed by using a computer or a computer network, preferably by using a computer program. Therefore, for the possible definitions of most of the terms used herein, reference can be made to the computer-implemented method for determining battery performance during the development of a battery configuration in a test environment as described above or in more detail hereinafter.

[0077] Specifically, one or both of the computer programs can be stored on a computer-readable data carrier and / or a computer-readable storage medium. As used herein, the terms "computer-readable data carrier" and "computer-readable storage medium" can specifically refer to a non-transitory data storage device, such as a hardware storage medium on which computer-executable instructions are stored. The computer-readable data carrier or storage medium can specifically be or can include storage media such as random access memory (RAM) and / or read-only memory (ROM).

[0078] Further disclosed and presented herein is a computer program product having program code components for performing, when the program is executed on a computer or computer network, a method for determining battery performance during battery construction development in a test environment according to one or more embodiments disclosed herein. Specifically, the program code components may be stored on a computer-readable data carrier and / or a computer-readable storage medium. According to one or more embodiments disclosed herein, further disclosed and presented herein is a data carrier having a data structure stored thereon, which, after being loaded into a computer or computer network, such as into the working memory or main memory of a computer or computer network, can perform a method for determining battery performance during battery construction development in a test environment according to one or more embodiments disclosed herein.

[0079] Further disclosed and presented herein is a computer program product having program code components stored on a machine-readable carrier for performing, when the program runs on a computer or computer network, a method for determining battery performance during battery construction development in a test environment according to one or more embodiments disclosed herein. As used herein, a computer program product refers to a program as a tradable product. The product can generally exist in any format, such as in paper format, or on a computer-readable data carrier. Specifically, the computer program product can be distributed over a data network.

[0080] In another aspect of the present invention, a computer-implemented method for determining battery performance during battery construction development in a test environment according to the present invention is proposed for use in optimizing synthesis and / or manufacturing parameters, such as battery materials and / or battery geometry.

[0081] In another aspect of the present invention, a method is provided for providing to a system at least a part of a predicted time series of state variables determined by a computer-implemented method for determining battery performance during battery construction development in a test environment according to the present invention in order to optimize battery materials and / or battery geometry.

[0082] The optimization may include performing at least one simulation.

[0083] The methods, systems, and programs of the present invention have many advantages compared to methods, systems, and programs known in the art. In particular, the methods, systems, and programs disclosed herein can allow reducing the measurement time on a test bench during battery material development from several months to several weeks. Additionally, if the battery is further processed as defined by at least one test protocol, the methods, systems, and programs of the present invention can not only allow predicting the state of health of the battery, but can also additionally predict the future characteristics of the battery.

[0084] Summarizing and without excluding further possible embodiments, the following embodiments can be envisaged:

[0085] Embodiment 1: A test system for determining battery performance during the development of a battery configuration in a test environment, the test system comprising at least one communication interface and at least one processing device, wherein the test system is configured to receive operation data indicating at least one test protocol via the communication interface, wherein the test system is configured to receive battery performance input data via the communication interface, wherein the processing device is configured to determine at least one predicted time series of at least one state variable indicating battery performance based on the battery performance input data and the operation data using at least one data-driven model, and wherein the test system is configured to provide at least a portion of the predicted time series of the state variable.

[0086] Embodiment 2: The test system according to the preceding embodiment, wherein the data-driven model comprises at least one recurrent neural network, such as at least one echo state network.

[0087] Embodiment 3: The test system according to any one of the preceding embodiments, wherein the state variable is derived from at least one charge-discharge curve, and wherein the state variable is at least one variable selected from the group consisting of discharge capacity, charge capacity, shape of the charge-discharge curve, average voltage, open circuit voltage, differential capacitance, coulombic efficiency, and internal resistance.

[0088] Embodiment 4: The test system according to any one of the preceding embodiments, wherein the operation data indicating at least one test protocol comprises at least one sequence of different charge cycles and / or discharge cycles.

[0089] Embodiment 5: The test system according to any one of the preceding embodiments, wherein the data-driven model is trained on at least one training data set, and wherein the training data set comprises a time series of historical data of charge and discharge cycles of at least one known battery configuration and at least one known test protocol.

[0090] Embodiment 6: The test system according to any one of the preceding embodiments, wherein the processing device is configured to use the test protocol as an input parameter for determining the predicted time series of the state variable using the data-driven model, and / or wherein the processing device comprises a plurality of data-driven models, and wherein the processing device is configured to select one of the data-driven models for determining the predicted time series of the state variable according to the test protocol.

[0091] Embodiment 7: The test system according to any one of the preceding embodiments, wherein the processing device is configured to use the battery performance input data as an input parameter for determining the predicted time series of the state variable using the data-driven model.

[0092] Example 8: The test system according to the foregoing embodiments, wherein the processing device includes a plurality of data-driven models, wherein the processing device is configured to analyze battery performance input data, wherein the analysis includes determining at least one material property, and wherein at least one of the data-driven models is selected based on the material property.

[0093] Example 9: The test system according to any one of the foregoing embodiments, wherein the battery performance input data includes data generated in response to a test protocol.

[0094] Example 10: The test system according to any one of the foregoing embodiments, wherein the test protocol is predefined.

[0095] Example 11: The test system according to any one of the foregoing embodiments, wherein the data-driven model is parameterized based on operation data indicating at least one test protocol and battery performance input data.

[0096] Example 12: The test system according to the foregoing embodiments, wherein the data-driven model uses knowledge of past and future charge-discharge cycles following at least one test protocol to predict future battery performance.

[0097] Example 13: The test system according to any one of the foregoing embodiments, wherein the data-driven model has temporal memory and / or the data-driven model is a time-dependent model.

[0098] Example 14: The test system according to any one of the foregoing embodiments, wherein the test protocol includes information about at least one battery performance test, wherein the battery performance test includes at least one sequence of different charge cycles and / or discharge cycles, and wherein a discharge-charge curve is determined for each cycle in the battery performance test.

[0099] Example 15: The test system according to the foregoing embodiments, wherein the battery performance test is performed by a customer, wherein the customer provides battery performance input data to the test system via a communication interface, and wherein the test system is configured to provide at least a portion of a predicted time series of state variables to the customer at least.

[0100] Example 16: The test system according to any one of the foregoing embodiments, wherein the test system includes at least one output interface configured to output at least one output, the output including information about at least a portion of a predicted time series of state variables.

[0101] Example 17: The test system according to the foregoing embodiments, wherein the output includes one or more of at least one histogram showing the development of a state variable over time, and wherein the output further includes at least one prediction and / or at least one classification regarding battery life.

[0102] Example 18: The test system according to any one of the preceding examples, wherein the test system is configured to perform at least one measurement depending on a predicted time series of state variables, wherein the measurement comprises one or more of the following: issuing a recommendation; issuing a warning; issuing an indication to modify the cathode material of the battery.

[0103] Example 19: The test system according to any one of the preceding examples, wherein the battery performance input data comprises discharge-charge cycle data, wherein the discharge-charge cycle data comprises at least one charge-discharge curve.

[0104] Example 20: The test system according to any one of the preceding examples, wherein the battery performance input data comprises one or more of information on discharge capacity, charge capacity information, information on the shape of the charge-discharge curve, information on average voltage, information on open circuit voltage, information on differential capacitance, information on Coulomb efficiency, and information on internal resistance.

[0105] Example 21: The test system according to any one of the preceding examples, wherein the battery performance input data comprises metadata related to one or more of the cathode material and the battery settings.

[0106] Example 22: The test system according to any one of the preceding examples, wherein the processing device is configured to verify the battery performance input data, wherein the verification comprises determining whether the retrieved battery performance input data is complete and / or whether it comprises a sufficient number of cycles for determining battery performance.

[0107] Example 23: The test system according to any one of the preceding examples, wherein the processing device is configured to perform at least one confidence test, wherein in the confidence test the determined state variables are compared with experimental test results.

[0108] Example 24: A test bench configured to perform at least one battery performance test on at least one battery based on at least one test protocol, wherein the battery performance test comprises at least one sequence of different charge cycles and / or discharge cycles, wherein the battery performance test comprises determining the charge-discharge curve for each cycle, wherein the test bench comprises at least one communication interface configured to provide operational data indicating the test protocol and battery performance input data to at least one test system according to any one of the preceding examples.

[0109] Example 25: A method for determining at least one data-driven model for determining battery performance during the development of a battery configuration in a test environment, the method comprising training a data-driven model with at least one training data set, the training data set including historical data of charge and discharge cycles of at least one known battery configuration and at least one known test protocol.

[0110] Example 26: The method according to the foregoing example, wherein the method comprises the steps of:

[0111] - Generating at least one random dynamic repository;

[0112] - Determining at least one input unit and at least one output unit;

[0113] - Generating connections from the input to the repository and connections from the output to the repository;

[0114] - Selecting at least one input weight matrix and at least one repository weight matrix;

[0115] - Determining output weights by training the data-driven model with training data, wherein regression analysis is used to determine the output weights.

[0116] Example 27: A computer-implemented method for determining battery performance during the development of a battery configuration in a test environment, wherein at least one test system according to any of the foregoing examples related to a test system is used in the method, the method comprising the steps of:

[0117] a) Retrieving operation data indicating at least one test protocol via at least one communication interface;

[0118] b) Retrieving battery performance input data via the communication interface;

[0119] c) Using a processing device to determine a predicted time series of state variables indicating battery performance based on the battery performance input data and the operation data using the data-driven model;

[0120] d) Providing at least a part of the predicted time series of the state variables.

[0121] Example 28: A computer program for determining battery performance during the development of a battery configuration in a test environment, the computer program being configured to cause a computer or a computer network to perform, fully or partially, the method for determining battery performance during the development of a battery configuration in a test environment according to the foregoing examples when executed on the computer or the computer network, wherein the computer program is configured to perform at least steps a) to d) of the method for determining battery performance during the development of a battery configuration in a test environment according to the foregoing examples.

[0122] Example 29: Use of a computer-implemented method for determining battery performance during the development of a battery configuration in a test environment according to Example 27 for optimizing battery materials and / or battery geometry.

[0123] Example 30: A method of providing at least a portion of a predicted time series of state variables determined by the method of Example 27 to a system for optimizing battery materials and / or battery geometry. Description of the Drawings

[0124] Brief Description of the Drawings

[0125] Further optional features and embodiments will be preferably disclosed in more detail in the following description of the embodiments in connection with the dependent claims. Wherein, as will be appreciated by those skilled in the art, the corresponding optional features can be implemented in isolation and in any feasible combination. The scope of the present invention is not limited by the preferred embodiments. The embodiments are schematically illustrated in the drawings. Wherein, the same reference numerals in these figures refer to the same or functionally comparable elements.

[0126] In the drawings:

[0127] Figure 1 Schematically illustrates an embodiment of a test system and a test bench according to the present invention;

[0128] Figure 2 Schematically illustrates an embodiment of a method for determining battery performance during the development of a battery configuration in a test environment according to the present invention;

[0129] Figure 3 Shows a comparison of experimental results and predictions using the method according to the present invention;

[0130] Figure 4A and 4B Shows the extraction of exemplary charge-discharge curves and battery performance input data. Detailed Description of the Invention

[0131] Figure 1A highly exemplary embodiment of a test system 110 for determining battery performance during the development of a battery configuration under test is shown. An exemplary battery 112 is shown. The battery 112 can be a rechargeable battery. The battery 112 can be selected from the group consisting of: lithium-ion battery (Li-Ion), nickel-cadmium battery (Ni-Cd), nickel-metal hydride battery (Ni-MH). For example, the battery can include at least one cathode material selected from the group consisting of: LiCoO2 (lithium cobalt oxide), LiNixMnyCozO2 (lithium nickel manganese cobalt oxide), and LiFePO4 (lithium iron phosphate). For example, the battery can include at least one anode material selected from the group consisting of: graphite, silicon. For example, the battery can contain at least one electrolyte selected from the group consisting of: LiPF6, LiBF4, or LiClO4 in organic solvents such as ethylene carbonate, dimethyl carbonate, and diethyl carbonate.

[0132] The test system 110 includes at least one communication interface 114. The test system 110 is configured to receive battery performance input data of the battery 112 via the communication interface 114. The communication interface 114 can include at least one data storage device configured to store the battery performance input data. The communication interface 114 can specifically provide components for transmitting or exchanging information. In particular, the communication interface 114 can provide a data transmission connection, such as Bluetooth, NFC, inductive coupling, etc. As an example, the communication interface 114 can be or can include at least one port, which includes one or more of a network or internet port, a USB port, and a disk drive. The communication interface 114 can be at least one network interface. The communication interface 114 can be or can include at least one database selected from the group consisting of: at least one server, at least one server system including multiple servers, at least one cloud server, or a cloud computing architecture. The communication interface 114 can include at least one storage unit configured to store the received battery performance input data.

[0133] The battery performance input data can be data including information on the behavior and / or performance of the battery generated in response to a test protocol. The battery performance input data can include data generated in response to at least one test protocol. The battery performance input data can be or can include raw data and / or preprocessed data. The battery performance input data can be determined by executing at least one test program. The test program can include determining at least one discharge-charge curve of the battery 112. The test program can be executed in a test bench 116. The battery performance input data can be transmitted to the test system 110 in real time or in batches with a delay via the communication interface 114.

[0134] Battery performance input data may include charge-discharge cycle data. The charge-discharge cycle data may include at least one charge-discharge curve. The battery performance input data may include one or more of information on discharge capacity, charge capacity, information on the shape of the charge-discharge curve, information on average voltage, information on open-circuit voltage, information on differential capacitance, information on Coulomb efficiency, and information on internal resistance. The battery performance input data may include metadata related to one or more of the cathode material and the cell settings. The metadata can be used to select an appropriate trained model, for example, taking into account the cathode material and / or the cell settings.

[0135] The test system 110 is configured to receive operation data indicating a test protocol via the communication interface 114. The operation data may include information on at least one test procedure performed on the battery 112 in the test bench 116, in particular one or more of operating conditions, test sequence, process, etc. The test bench 116 may include a battery storage device 118 configured to accommodate the battery during testing and a cycling machine 120 configured to apply a plurality of charge-discharge cycles to the battery. The cycling machine 120 may be configured to record the battery performance input data. The operation data indicating at least one test protocol includes, for example, at least one sequence of different charge cycles and / or discharge cycles. The test protocol may define the order of the test procedures and / or the sequence of the test procedures and / or the duration of each test procedure. Each test procedure may include at least one charge-discharge cycle, and the charge-discharge cycles of at least two test procedures are different. The test protocol may include information on at least one battery performance test. The battery performance test may include at least one sequence of different charge cycles and / or discharge cycles. In the battery performance test, a charge-discharge curve may be determined for each cycle. The battery performance test may be performed by a customer. The customer may provide the battery performance input data to the test system 110 via the communication interface 114.

[0136] The test system 110 includes at least one processing device 122. The processing device 122 may be configured to verify the battery performance input data. The processing device 122 is configured to use as Figure 1At least one data-driven model, as indicated by reference numeral 124 therein, determines at least one predicted time series of at least one state variable indicative of battery performance based on battery performance input data and operational data. Validation may include determining whether the retrieved battery performance input data is complete and / or whether it includes sufficient cycles for determining battery performance. Validation may include one or more of reviewing, inspecting, or testing the battery performance input data. Specifically, validation includes determining whether the retrieved battery performance input data is complete and / or whether it includes sufficient cycles for determining battery performance. For example, in the case where validation reveals that the retrieved battery performance input data is incomplete and / or does not include sufficient cycles for determining battery performance, a request may be sent to the customer to provide additional data by using the communication interface 114.

[0137] The state variable(s) may be derived from or derivable from at least one charge-discharge curve. The state variable(s) may be at least one variable selected from the group consisting of: discharge capacity, charge capacity, shape of the charge-discharge curve, average voltage, open-circuit voltage, differential capacitance, coulombic efficiency, and internal resistance. The state variable(s) may be used simultaneously.

[0138] The processing device 122 may be configured to select information from the battery performance input data depending on the state variable(s) whose future evolution is to be predicted. The processing device 122 may be configured to rank the battery performance input data depending on the state variable(s) whose future evolution is to be predicted. For example, the state variable may be the discharge capacity. The ranking may be such that the charge capacity from the previous cycle and / or the discharge capacity from the previous cycle may be considered the most relevant as they are very close to the state variable to be predicted. The shape of the charge and / or discharge curve and the internal resistance from the previous cycle may be considered less relevant compared to the charge capacity from the previous cycle and / or the discharge capacity from the previous cycle. For other state variables such as the internal resistance, the ranking may be different.

[0139] The predicted time series of the state variables can be the expected time series of the state variables determined using at least one data-driven model. The battery performance input data of the battery can be experimental data. It can be determined by using a data-driven model, in particular to predict the predicted time series of the state variables, where the battery performance input data is used as the input of the data-driven model. The battery performance input data can include experimental data based on which the predicted time series of the state variables is predicted. Additionally, the predicted time series is determined according to the operation data. The test system 110 can include at least one storage device, where multiple different data-driven models can be stored. For example, the storage device can include different data-driven models for different test protocols. In particular, the processing device 122 can be configured to select a data-driven model based on the test protocol for battery testing. Additionally or alternatively, the data storage device can include different data-driven models depending on the state variables to be predicted. The processing device 122 can be configured to determine the predicted time series of one state variable or multiple state variables. At the same time, the time series of any combination of the state variables can also be predicted.

[0140] The data-driven model can be derived from the analysis of experimental data. The data-driven model can be a machine learning tool. The data-driven model can include at least one trained model. For example, the data-driven model is trained on at least one training dataset, where the training dataset includes a time series of historical data of charge and discharge cycles of at least one known battery configuration and at least one known test protocol. The training dataset can include experimental results of multiple battery tests, such as one or more parameters for characterizing battery performance. The training dataset can include multiple experimental results of battery tests determined at at least two time points (such as for different cycles). For the training of the model, multiple training datasets related to different batteries to be tested can be used, for example.

[0141] The data-driven model can be a feature-based model, where features or subsets of features are used to predict battery performance. The training data can be used to select features for the trained model. Feature selection can include selecting a subset of relevant features, especially variables and predictors, for model construction. Feature selection can include deriving information on electrochemistry at certain time points, such as at relevant time points of charge or discharge curves, internal resistance, open-circuit voltage, differential capacitance. Irregularities due to material categories can be considered for feature selection. The training data can include raw data. Feature selection can include data aggregation from each step to each cycle. Features can be selected using electrochemistry knowledge and / or traditional feature selection methods, such as retaining features highly correlated with the response. Electrochemistry knowledge can include, for example, information on voltage intervals where phase transitions of interest are expected and thus features should be extracted from this curve region.

[0142] The data-driven model may include at least one recurrent neural network, such as at least one echo state network.

[0143] The processing device 122 may be configured to use the battery performance input data as input parameters for determining a predicted time series of state variables using a data-driven model. To determine the battery performance of the battery 112 under test, the battery performance input data may be processed and applied as input to the data-driven model. The processing may include data aggregation. The processing may include selecting information related to at least one of the parameters characterizing the battery performance whose time series is predicted by the data-driven model, also referred to as the target state variable. In this case, the output generated by the data-driven model may be a predicted time series of the target state variable or a matrix of state variables.

[0144] The processing device 122 may be configured to use the test protocol as input parameters for determining a predicted time series of state variables using a data-driven model. More specifically, the processing device may be configured to select at least one suitable data-driven model based on the received battery performance input data, such as based on the received information about the test protocol and / or the sequence of different charge cycles and / or discharge cycles. The processing device 122 may include a plurality of data-driven models, where the processing device 122 is configured to select one of the data-driven models for determining the predicted time series of state variables depending on the test protocol. For example, each of the plurality of data-driven models may be trained on data obtained by using different test protocols. The models may have different fitting parameters and different hyperparameters for controlling complexity. However, the general structure may be the same for all models. Other criteria for selecting the most suitable data-driven model are possible. For example, depending on the material properties, the processing device 122 may include a plurality of data-driven models. The processing device 122 may be configured to analyze the battery performance input data, where the analysis includes determining at least one material property. The processing device 122 may be configured to select at least one of the data-driven models based on the material properties. Information about the material properties may be determined from the battery performance input data such as metadata.

[0145] The processing device 122 may be configured to perform at least one confidence test, where the determined state variables are compared with experimental test results. In the confidence test, the predicted time series of the state variables may be compared with the experimental test results. The experimental results for different battery types or materials may be stored in at least one data memory and may be used for comparison in the confidence test. The test system 110 is configured to provide at least a portion of the predicted time series of the state variables. The test system 110 may include at least one output interface 126, which is configured to output at least one output including information about at least a portion of the predicted time series of the state variables. InFigure 1 In this case, the communication interface 114 and the output interface 126 are the same. The output may include one or more of at least one histogram showing the development of a state variable over time. The output may also include at least one prediction regarding battery life and / or information regarding at least one classification. The test system 110 may be configured to perform at least one measure depending on the predicted time series of the state variable, where the measure includes one or more of the following: issuing a recommendation, issuing a warning, issuing an indication that the cathode material of the battery needs to be modified. The test system 110 may be configured to provide at least one information regarding the determined battery performance to the customer of the battery 112. For example, the information may be provided to the customer by the output interface 126. The information may also include one or more of a ranking of the batteries in the test bench 116, a battery classification, a classification of the cathode material, or at least one recommendation. The output may include a ranking of the batteries in the test bench 116 and / or a recommendation to stop the battery measurement. The communication interface 114 may be configured to allow monitoring of the state of the test bench 116. This may allow preparation and design of the next experiment based on the determination result of the battery performance.

[0146] In Figure 2 An embodiment of a computer-implemented method for determining battery performance during battery construction development in a test environment is schematically shown.

[0147] The method includes the following steps:

[0148] -(represented by reference numeral 128) retrieving, via at least one communication interface, operation data indicating at least one test protocol and retrieving battery performance input data via the communication interface;

[0149] -(represented by reference numeral 130) using a data-driven model by the processing device 122 to determine a predicted time series of state variables indicating battery performance based on the battery performance input data and the operation data.

[0150] The method may also include a decision-making step 132, where depending on the output of the test system 110, the customer may decide whether the battery 112 and the cathode material used are considered "good" or "bad". In the case where the cathode material is considered "bad", the customer may continue to replace or modify the cathode material and restart method step a). Thus, method steps a) and b) may be performed once or may be repeatedly performed, such as repeated one or more times, particularly for different batteries 112 and / or different cathode materials.

[0151] Figure 3Shows a comparison of experimental results and predictions using the method according to the present invention. In particular, for the training data set 134 and the prediction 136, the battery capacity c is depicted as a function of time t in days. For the predictions shown here, the early data 138 is used as input. Good agreement between the predictions and the measurements can be observed. In Figure 3 it, the deviation between the model and the experiment is about 2% over the next 200 cycles, which is approximately 3 weeks.

[0152] Figure 4A Shows exemplary charge-discharge curves, in particular the voltage as a function of capacity determined in the test bench 116. The charge-discharge curves can be used to determine several parameters indicating the battery performance such as the discharge capacity, indicated by arrow a. In addition, the voltage drop indicated by arrow b, which is used to calculate the internal resistance, is shown. Further, in FIG. 4, the average voltage denoted by c, the dQ / dV curve denoted by d, the discharge open-circuit voltage denoted by e, and the charge open-circuit voltage denoted by f are shown. Figure 4A Shows the charge capacity denoted by g, where the thin dashed line represents the constant charge capacity and the thick dashed line represents the constant voltage charge capacity.

[0153] Figure 4B Shows the extraction of battery performance input data from the experimental data determined from 2130 charge and discharge tests. Specifically, the development of voltage and current over time is described. The thin solid line shows the current changing in a stepwise manner, as the current remains constant for most of each charge and discharge step. The double arrows centered on the voltage curve show some features that can be used as model inputs. They correspond to the capacity change over a predetermined voltage interval during charging, given by the Figure 4B horizontal line in. The current drop corresponds to the constant voltage and the remaining steps, where the battery is allowed to relax to the equilibrium state.

[0154] List of reference signs:

[0155] 110 Test system

[0156] 112 Battery

[0157] 114 Communication interface

[0158] 116 Test bench

[0159] 118 Battery storage device

[0160] 120 Cycling machine

[0161] 122 Processing device

[0162] 124 Data-driven model

[0163] 126 Output interface

[0164] 128 Retrieval data

[0165] 130 Determining prediction

[0166] 132 Decision making

[0167] 134 Training data set

[0168] 136 Prediction

[0169] 138 Early data

[0170] References

[0171] Kristen A. Severson et al., “Data-driven prediction of battery cycle life before capacity degradation”, Nature Energy, https: / / doi.org / 10.1038 / s41560-019-0356-8

[0172] US2019 / 0115778A1

[0173] “A Practical Guide to Applying Echo State Networks”, Mantas Lukosevicius, published in Neural Networks: Tricks of the Trade 2012. DOI: 10.1007 / 978-3-642-35289-8_36 and http: / / minds.jacobs-university.de / uploads / papers / PracticalESN.pdf

[0174] WO2019 / 017991A1

Claims

1. A test system (110) for determining battery performance during the development of a battery configuration in a test environment, the test system (110) comprising at least one communication interface (114) and at least one processing device (122), wherein the test system (110) is configured to receive operation data indicating at least one test protocol via the communication interface (114), wherein the test system (110) is configured to receive battery performance input data via the communication interface (114), wherein the processing device (122) is configured to determine at least one predicted time series of at least one state variable indicating battery performance based on the battery performance input data and the operation data using at least one data-driven model, and wherein the test system (110) is configured to provide at least a portion of the predicted time series of the state variable.

2. The test system (110) according to claim 1, wherein the data-driven model comprises at least one recurrent neural network, such as at least one echo state network.

3. The test system (110) according to any one of claims 1 to 2, wherein the state variable is derivable from at least one charge-discharge curve, and wherein the state variable is at least one variable selected from the group consisting of: discharge capacity, charge capacity, shape of the charge-discharge curve, average voltage, open circuit voltage, differential capacitance, coulombic efficiency, and internal resistance.

4. The test system (110) according to any one of claims 1 to 2, wherein the operation data indicating at least one test protocol comprises at least one sequence of different charge cycles and / or discharge cycles.

5. The test system (110) according to any one of claims 1 to 2, wherein the data-driven model is trained on at least one training data set, and wherein the training data set comprises a time series of historical data of charge and discharge cycles of at least one known battery configuration and at least one known test protocol.

6. The test system (110) according to any one of claims 1 to 2, wherein the processing device (122) is configured to use the test protocol as an input parameter for determining the predicted time series of the state variable using the data-driven model, and / or wherein the processing device (122) comprises a plurality of data-driven models, and wherein the processing device (122) is configured to select one of the data-driven models for determining the predicted time series of the state variable depending on the test protocol.

7. The test system (110) according to any one of claims 1 to 2, wherein the processing device (122) is configured to use the battery performance input data as an input parameter for determining the predicted time series of the state variable using the data-driven model.

8. The test system (110) according to claim 7, wherein the processing device (122) includes a plurality of data-driven models, wherein the processing device (122) is configured to analyze the battery performance input data, wherein the analysis includes determining at least one material property, and wherein at least one of the data-driven models is selected based on the material property.

9. The test system (110) according to any one of claims 1 to 2, wherein the battery performance input data includes data generated in response to the test protocol.

10. The test system (110) according to any one of claims 1 to 2, wherein the test protocol is predefined.

11. The test system (110) according to any one of claims 1 to 2, wherein the data-driven model is parameterized based on the operation data indicating the at least one test protocol and the battery performance input data.

12. The test system (110) according to claim 11, wherein the data-driven model uses the knowledge of past and future charge-discharge cycles following the at least one test protocol to predict future battery performance.

13. The test system (110) according to any one of claims 1 to 2, wherein the data-driven model has temporal memory and / or the data-driven model is a time-dependent model.

14. The test system (110) according to any one of claims 1 to 2, wherein the test protocol includes information about at least one battery performance test, wherein the battery performance test includes at least one sequence of different charge cycles and / or discharge cycles, and wherein a discharge-charge curve is determined for each cycle in the battery performance test.

15. The test system (110) according to any one of claims 1 to 2, wherein the battery performance input data includes discharge-charge cycle data, and wherein the discharge-charge cycle data includes at least one charge-discharge curve.

16. The test system (110) according to any one of claims 1 to 2, wherein the battery performance input data includes one or more of information about discharge capacity, information about charge capacity, information about the shape of the charge-discharge curve, information about average voltage, information about open-circuit voltage, information about differential capacitance, information about Coulomb efficiency, and information about internal resistance.

17. The test system (110) according to any one of claims 1 to 2, wherein the battery performance input data includes metadata related to one or more of the cathode material and the battery cell setup.

18. A test bench (116) configured to perform at least one battery performance test on at least one battery (112) based on at least one test protocol, wherein the battery performance test includes at least one sequence of different charge cycles and / or discharge cycles, and wherein the battery performance test includes determining a discharge-charge curve for each cycle, and wherein the test bench (116) includes at least one communication interface configured to provide operation data indicating the test protocol and battery performance input data to at least one test system according to any one of claims 1 to 17.

19. A method for determining at least one data-driven model for determining battery performance during battery construction development in a test environment, the data-driven model being configured to be used by a processing device of a test system according to any one of claims 1 to 17 to determine at least one predicted time series of state variables indicating battery performance based on battery performance input data and operation data, wherein the method includes training the data-driven model using at least one training data set, and wherein the training data set includes historical data of charge and discharge cycles of at least one known battery construction and at least one known test protocol.

20. The method according to claim 19, wherein the method includes the following steps: - Generating at least one random dynamic repository; - Determining at least one input unit and at least one output unit; - Generating connections from the input to the repository and connections from the output to the repository; - Selecting at least one input weight matrix and at least one repository weight matrix; - Determining output weights by training the data-driven model using the training data, wherein regression analysis is used to determine the output weights.

21. A computer-implemented method for determining battery performance during battery construction development in a test environment, wherein at least one test system (110) according to any one of claims 1 to 17 is used in the method, and the method includes the following steps: a) Retrieving operation data indicating at least one test protocol via at least one communication interface (114); b) Retrieving battery performance input data via the communication interface (114); c) Determining a predicted time series of state variables indicating battery performance based on the battery performance input data and the operation data using a data-driven model by using a processing device (122); d) Providing at least a part of the predicted time series of the state variables.

22. A computer program for determining battery performance during battery construction development in a test environment, the computer program being configured to cause the computer or computer network to execute the method for determining battery performance during battery construction development in a test environment according to the foregoing claims when executed on the computer or computer network, and wherein the computer program is configured to execute at least steps a) to d) of the method for determining battery performance during battery construction development in a test environment according to claim 21.

23. Use of a computer-implemented method for determining battery performance during the development of a battery configuration in a test environment for optimizing battery materials and / or battery geometry according to claim 21.

24. A method for providing at least a part of a predicted time series of state variables determined by the method according to claim 21 to a system for optimizing battery materials and / or battery geometry.

Images