Cost-effective and accurate determination of the degradation state of rechargeable batteries

Abstract

This invention relates to a method for approximating and / or predicting the true degradation state of a rechargeable battery, comprising the steps of: • providing a time series of values ​​of the degradation state obtained by measurement techniques at past time points, discretely over a pre-given time step; • providing a trained Hidden Markov Model (HMM) that, based on the true degradation state, indicates: - with what probability which value of the degradation state is observed in the case of measurement techniques and - with what probability and for how long the true degradation state is maintained, and / or with what probability the true degradation state transitions to which worse degradation state in the next time step; • determining the most probable trend of the true degradation state in the past from the observed time series and the HMM, the trend being consistent with the observed time series; • analyzing the approximation and / or prediction sought from the most probable trend.

Description

Technical Field

[0001] The present invention relates to the monitoring and prediction of the degradation state of rechargeable batteries, particularly for vehicles that are at least partially electrically powered. Background Technology

[0002] During charging, a rechargeable battery converts electrical energy supplied from an external source into chemical energy. Conversely, during discharging, the chemical energy is converted back into electrical energy, which powers an external load. In particular, in the case of traction batteries in at least partially electrically driven vehicles, such as pure battery electric vehicles or hybrid vehicles, a significant portion of the stored energy is typically extracted and subsequently recharged.

[0003] Current electrochemical batteries undergo a degradation process. With increased operating time and, in particular, with an increased number of completed charge and discharge cycles, usable capacity decreases. This aging process is largely dependent on usage characteristics and operating conditions. Thus, it is known, for example, that batteries in the hands of some users provide excellent service to smartphones for many years, while in the hands of others, the battery proves to be the first worn-out component to fail.

[0004] A method for determining the kinetic parameters of a battery model is known from EP2852848B1. Based on this model, the degradation state and SOH (state of health) can be determined. Summary of the Invention

[0005] Within the scope of this invention, an improved method is provided for approximating and / or predicting the true degradation state of a rechargeable battery.

[0006] This method provides a time series of the degraded state, with values ​​obtained by measurement techniques at past time points, discretely distributed over a pre-given time step.

[0007] Here, "obtained by measurement techniques" should be understood as various detections with subsequent analysis of the measured values, which provides a measure of the degradation state. The degradation state can be approximated in many ways. Therefore, sensors that directly provide values ​​of the degradation state are not necessary.

[0008] For example, the observed value of the degradation state can be determined from the amount of charge released into the battery as it is charged from a first terminal voltage to a second, higher terminal voltage. A typical single cell of a traction battery for a vehicle can be charged, for example, from a terminal voltage of 3V to its charging termination voltage of 4.2V. The amount of charge released into the battery can then be determined by means of a "coulomb count," for example, by integrating the current flowing into the battery. The more charge the battery can absorb, the greater its remaining usable capacity and the better its degradation state.

[0009] The observed value of the degradation state is determined based on the duration during which the battery's terminal voltage drops from a first value to a second, lower value under a predetermined load. The longer the terminal voltage remains above the second, lower value, the greater the capacity and the better the degradation state, from which the battery supplies power to the load.

[0010] For example, during the discharge of the battery, time series of the battery's terminal voltage, discharge current, and state of charge can also be detected. Then, based on a physical model of the degradation state, the observed values ​​of the degradation state can be derived from this time series. Additionally, the temperature within the battery can be detected, particularly during discharge, and this temperature can be incorporated into the physical model. Typically, the aforementioned parameters are particularly useful in at least partially electrically powered vehicles.

[0011] The method exemplified here, like all other methods for approximating the state of degradation, is limited in its accuracy in obtaining the approximation by the accuracy required to detect the measured values ​​used. However, measuring instruments installed in vehicles are typically designed, more precisely, for short-term monitoring of the operating status of the electrical system at every moment, relative to long-term monitoring of the state of degradation. For the latter purpose, measurement data is sometimes relatively uncertain and elusive. Applying high-quality measuring instruments is expensive and does not guarantee an arbitrary increase in accuracy because there are numerous sources of interference in the confined space of a vehicle's electrical system for measurement.

[0012] Therefore, the scope of the method specifies that accuracy is improved by means of a trained Hidden Markov Model (HMM). Based on observations, an HMM can make probabilistic conclusions about "hidden" (implicit) variables that cannot be directly measured, the probabilities of which depend on the state of the "hidden" variable. In this application, the "hidden" variable from which the conclusions should be derived is the true degradation state of the battery, and the value of the degradation state obtained by measurement techniques reflects the observations.

[0013] Here, the observation is, in principle, time-discrete within a pre-defined time step. A raster of this time step can be superimposed, in which the values ​​of the degradation state obtained by measurement techniques are also discretized. However, this is not mandatory. The degradation state itself is also discretized, for example, as a percentage ranging from 100% for a new battery to 70% for a battery that should be replaced.

[0014] A trained Hidden Markov Model (HMM) indicates, based on the actual degradation state, the probability of observing which value of the degradation state, when obtained by measurement techniques. This probability can, for example, constitute a so-called "observation matrix." The more strongly the model of the physics used for measurement techniques simplifies, and the more uncertain the measurements used, the more dispersed the measurement-obtained values ​​of the degradation state are around the actual degradation state. Therefore, the observation matrix can particularly reflect predictions about the model of the physics used for measurement techniques and about the uncertainties of the measurements used.

[0015] Furthermore, the trained HMM indicates, based on each true degradation state, the probability with which it will transition to a worse degradation state in the next time step. Alternatively, or in combination with this, the trained HMM can indicate, for example, in the form of a distribution, the probability and duration with which the true degradation state is maintained. This probability can, for example, constitute a "transition matrix." Therefore, the dynamics of degradation can be reflected in the transition matrix. Moreover, the model mentioned therein can demonstrate, for example, that the degradation is predictable, and similarly, as with boundary conditions, irreversible and monotonically progressive. That is, with good handling, the current degradation state of the battery can be maintained to the maximum extent, but the battery is no longer recovered from the degradation it has already suffered.

[0016] The most probable trend of the true degradation state in the past is obtained from the observed time series of values ​​of the degradation state obtained by measurement techniques and the Hidden Markov Model (HMM), and the trend is consistent with the observed time series. An approximation is found by analyzing this most probable trend. Alternatively, or also in combination with this, the inference of the most probable trend mentioned can also provide a prediction of future degradation states. This task is somewhat similar to error correction during the reading of a data carrier, where one or more bits can be "flipped". Therefore, the most probable trend of the true degradation state can also be obtained, in particular, for example, using the Viterbi algorithm for error correction.

[0017] Prior to using the HMM, the time series of observed values ​​in a degraded state can optionally be smoothed, filtered, or otherwise preprocessed in any way.

[0018] It is known that by using the HMM, it is possible to suppress not only the effects of inaccuracies in the physical models used and uncertainties in the measurements. More specifically, the method also provides information on whether the battery is still operating well and for an extended period under its current operating conditions, or whether the battery is being used by "battery abusers," which contribute to its degradation. Therefore, it is particularly effective in responding to situations where, for example, predictions of the actual degradation state obtained for future time points meet pre-defined criteria, up to a period when that future time point is assessed as the remaining usable lifespan of the battery.

[0019] In a vehicle's traction battery, there are many possibilities that could delay, prevent, or even promote degradation. Therefore, it is not good for the battery, for example, when it is frequently deeply discharged or when it is subjected to high current at high temperatures.

[0020] Therefore, even under similar driving power conditions, the degradation state of the battery can be distributed, for example, very unevenly within a fleet of electric vehicles. Future predictions suggest that traction battery replacement should be carried out precisely at the point in time when it is actually necessary. Because the traction battery is arguably the most expensive component in an electric vehicle, premature replacement would be costly. Conversely, delayed replacement could result in the vehicle having to be abandoned and towed away at considerable expense. Currently, there is no provision for on-site replacement of the traction battery via roadside assistance.

[0021] As explained above, the uncertainty of the degradation state of the battery obtained by measurement techniques depends particularly on the uncertainty of the measurements used and the physical model. The physical model determines the propagation of the individual uncertainty of the measurements to the overall uncertainty of the obtained degradation state. For application in the methods described above, this has been taken into account in the training of the HMM. Therefore, for this application, the present invention also provides a method for training an HMM.

[0022] In this method, a large time series of actual degradation trends of batteries during their use is provided. This time series can be recorded, for example, on an electrically driven vehicle and transmitted in real time to the cloud, or read at regular intervals at the factory. The actual degradation state can be determined, in particular, by measurement techniques, such as in a laboratory under optimal conditions. For this purpose, sensors that directly provide values ​​of the degradation state are not necessary; for example, measurements of the battery's capacity loss are sufficient. For example, the operating battery can be measured at defined intervals in a test vehicle or within a simulated range of use.

[0023] The cost of training the HMM is a single, one-time investment. Subsequently, the HMM can be benefited from multiple times in mass production.

[0024] Therefore, a physical model is provided within the scope of the method, which maps measurements of at least the terminal voltage, discharge current, and state of charge of the battery to values ​​of the degradation state.

[0025] Taking into account the uncertainties inherent in the measurement techniques used in the physical model, the determination is made based on the actual degradation state: with what probability is a particular value of the degradation state observed, given that the measurement values ​​are obtained using measurement techniques and the physical model is employed. This allows the aforementioned observation matrix to be constructed.

[0026] Considering the trend of the true degenerate state, determine: with what probability and for what time the true degenerate state remains, and / or with what probability the true degenerate state transitions to a worse degenerate state in the next time step.

[0027] This probability depends on the current degradation state, but also on the current load of the battery with the already mentioned and other stress factors that are detrimental to the battery's lifespan. Therefore, in addition to the battery's actual degradation state, this stress factor can also be considered as another "hidden" variable in the HMM. The current state of the HMM is then characterized not only by the actual degradation state, but also by the stress factor. In this way, transition probabilities other than zero can be indicated, particularly between states characterized by different actual degradation states but the same stress factor. That is, information on how quickly degradation progresses further while maintaining the stress factor can be collected and aggregated during the training of the aforementioned transition matrix for many different stress factors. Then, the aforementioned determination of the most probable trajectory of the actual degradation state provides not only the actual degradation state itself, but also the respective stress factors. It is then possible to define, or even precisely determine, what exactly caused the observed degradation in a feasible manner, thus avoiding, for example, operational errors in the future.

[0028] As explained above, the physical model allows the degradation state to additionally depend on at least one temperature within the battery.

[0029] The method can be implemented, in particular, entirely or partially, in software. Therefore, the invention also relates to a computer program with machine-readable instructions that, when executed on one or more computers, drive the one or more computers to implement one of the described methods. Downloading a product refers to a digital product that is transferable via a data network, i.e., downloadable by a user of the data network, and that the product can be offered, for example, in an online store for timely download.

[0030] In addition, computers can be equipped with computer programs, machine-readable data carriers, or downloadable products. Attached Figure Description

[0031] Furthermore, measures to improve the present invention are shown in more detail below, together with the description of preferred embodiments of the present invention, and with reference to the accompanying drawings. Wherein:

[0032] Figure 1 An embodiment of a method 100 for approximating and / or predicting the degradation 2a of battery 1 is shown;

[0033] Figure 2 An exemplary most probable trend 2* of the actual degradation 2 is shown, which is based on observations 2b of the given measurement technique of said degradation;

[0034] Figure 3 An embodiment of method 200 for training HMM 4 applied in method 100 is shown. Detailed Implementation

[0035] Figure 1 This is an illustrative flowchart of an embodiment of method 100. In step 105, at least a partially electrically driven traction battery of the vehicle 1 is selected.

[0036] In step 110, a time series 3 discretized over a pre-given time step is provided for the value 2b of the degradation state obtained by measurement techniques at past time points. In step 120, an Hidden Markov Model (HMM) 4 is provided, which correlates the true degradation state 2 with, optionally, indirectly measurable stress factors affecting that degradation state, and probabilities, either when the determined value 2b of the degradation is observed or when the true degradation state 2 is changed in a determined manner. Based on this, in step 130, the most probable past trend 2* of the true degradation state 2 is obtained, which is consistent with the observed time series 3. Here, the Viterbi algorithm can be used, particularly according to block 131.

[0037] In step 140, the approximation and / or prediction 2a sought by the most likely trend 2* analysis is used. When the prediction 2a for a future point in time meets a pre-given criterion 150 (true value 1), the time period up to that future point in time is evaluated in step 160 as the remaining usable lifespan 6 of the battery 1.

[0038] Within box 110, some exemplary possibilities are illustrated: how the time series 3 of the degraded value 2b obtained by the measurement technique can be obtained.

[0039] According to block 111, the observed value 2b of the degradation state can be obtained based on the amount of charge, which is released into the battery 1 when the battery 1 is charged from a first terminal voltage to a second, higher terminal voltage.

[0040] According to block 112, the observed value 2b of the degradation state can be obtained based on the duration during which the terminal voltage of the battery 1 decreases from a first value to a second, lower value under a predetermined load.

[0041] According to block 113, during the discharge of the battery 1, the time sequence of the terminal voltage 5a, the discharge current 5b, and the state of charge 5c of the battery 1 can be detected. According to block 113a, the temperature 5d of the battery 5 can also be detected.

[0042] According to block 114, the observed value 2b of the degraded state can be obtained from the time series of measured parameters 5a-5c based on the physical model 5 of the degraded state. According to block 114a, the temperature 5d can also be considered here.

[0043] Figure 2 An example is shown on how the most likely trend 2* of the actual degradation 2 can be obtained based on a time series 3 of the value 2b observed by measurement techniques of the degradation of the battery 1.

[0044] Curve a, shown exemplarily, reflects the time trend of the degradation, which corresponds only to the slow aging of battery 1. This time trend only coincides with a smaller value 2b of the time series 3, and is therefore not very reliable.

[0045] The exemplary curve b reflects the time trend of the degradation, which corresponds to the very rapid aging of the battery 1. This time trend also only coincides with a smaller value 2b of the time series 3, and is therefore similarly unreliable.

[0046] The most probable trend 2* derived from the actual degradation 2 based on HMM 4 best matches the currently observed value 2b. This trend 2* is not only from the start of operation of the battery 1 at time t0 to the current time t. A Each past time provides an approximation 2a for the true degradation 2. More precisely, this approximation 2a is for times beyond the current time t. A The future time t F Seamlessly transition to forecasting.

[0047] Figure 3 An embodiment of the method 200 for training HMM 4 is shown. Similar to... Figure 1 In step 105, in step 205, the vehicle's traction battery is selected as battery 1.

[0048] In step 210, a time series 2c of the actual degradation state 2 is provided for a large number of such batteries. In step 220, a physical model 5 is provided, which maps the measurements of at least the terminal voltage 5a, the discharge current 5b, and the state of charge 5c of the battery 1 to the value 2b of the degradation state.

[0049] In step 230, taking into account the measurement technique uncertainty Δ inherent in the measurement of the measured values ​​5a-5c, the probability of observing which value 2b of the degradation state is determined based on the actual degradation state 2, given that the measured values ​​are obtained using measurement techniques and a physical model is used.

[0050] In step 240, considering the trend of the true degenerate state 2, we determine: for which time t, with what probability, and for how long, the true degenerate state 2 remains, and / or with what probability, in the next time step, the true degenerate state transitions to which worse degenerate state 2'.

[0051] The information obtained in steps 230 and 240 characterizes the HMM 4.

Claims

1. A method (100) for approximating and / or predicting (2a) the true degradation state (2) of a rechargeable battery (1), comprising the following steps: • Provides (110) a time series (3) of the values ​​(2b) of the degradation state obtained by measurement techniques at past time points, which are discrete in a pre-given time step; ·in, During the discharge of the battery (1), the time series of the terminal voltage (5a), discharge current (5b) and charging state (5c) of the battery (1) are detected (113), and wherein the observed value (2b) of the degradation state is obtained (114) from the time series according to the physical model (5) of the degradation state. • In addition, during discharge, a time series of at least one temperature (5d) in the battery (1) is detected (113a), and wherein the temperature (114a) is considered in the physical model (5); • Provides (120) trained Hidden Markov Models, HMM(4), which indicate based on the true degenerate state(2): —In the case of obtaining the value of the degenerate state by measurement techniques, with what probability are those values ​​observed (2b) and —With what probability and for how long the true degenerate state (2) is maintained, and / or with what probability the true degenerate state transitions to which worse degenerate state (2') in the next time step; • The most probable trend (2*) of the actual degradation state (2) in the past is obtained from the observed time series (3) and HMM (4), which is consistent with the observed time series (3); • Approximate and / or prediction (2a) sought by the analysis (140) of the most likely trend (2*).

2. The method (100) according to claim 1, wherein, The observed value (2b) of the degradation state is obtained based on the amount of charge, which is released into the battery (1) when the battery (1) is charged from a first terminal voltage to a second, higher terminal voltage.

3. The method (100) according to claim 1, wherein, The observed value (2b) of the degradation state is obtained based on the duration during which the terminal voltage of the battery (1) decreases from a first value to a second, lower value under a pre-given load.

4. The method (100) according to any one of claims 1 to 3, wherein, The Viterbi algorithm is used to find the most likely trend (2*) of the true degenerate state (2) described in (131).

5. The method (100) according to any one of claims 1 to 3, wherein, In response to the following situation: the prediction (2a) of the actual degradation state (2) obtained for a future time point meets a pre-given criterion (150) until the time period of the future time point is evaluated (160) as the remaining usable lifespan (6) of the battery (1).

6. A method (200) for training a Hidden Markov Model, HMM (4), comprising the following steps for application in the method (100): • Provide (210) a time series (2c) of the trends of the actual degradation state (2) of a large number of batteries (1) during their use; • Provides a physical model (5) that maps measurements of at least the terminal voltage (5a), discharge current (5b), and state of charge (5c) of the battery (1) to values ​​of the degradation state (2b), wherein, The physical model (5) additionally makes the degradation state (2b) dependent on at least one temperature (5d) in the battery (1); • Taking into account the uncertainty (Δ) of the measurement technique involved in the measurement of the measured values ​​(5a-5c), (230) is obtained based on the actual degradation state (2): with what probability is which value of the degradation state (2b) observed when the measured values ​​are obtained by measurement technique and a physical model is used; • Taking into account the trend of the true degenerate state (2), determine (240): with what probability and for how long the true degenerate state (2) is maintained, and / or with what probability the true degenerate state transitions to which worse degenerate state (2') in the next time step.

7. The method (100, 200) according to claim 1 or claim 6, wherein, Select (105, 205) at least partially ground-electrically driven vehicles as batteries (1).

8. A machine-readable data carrier having a computer program containing machine-readable instructions that, when executed on one or more computers, drive the one or more computers to perform the method (100, 200) according to any one of claims 1 to 7.

9. A computer equipped with a machine-readable data carrier according to claim 8.

Images