Determining peaks in incremental capacity curves of a lithium-ion cell, in particular for determining a degradation mode of this cell

Abstract

The invention relates to a method (606) for determining peaks in incremental capacity curves of a battery cell, the method being characterised in that it comprises: - inputting a series of actions (A1…AN), each characterised by predefined parameter values (p, T, c, d, lc, r1, r2, SoCmin, SoCmax, DoD, N, F), into a learning model (208) that has been previously trained by supervised learning to provide, in response, for each action (A1…AN), at least one main peak (P11, P21…P1N, P2N) in the incremental-capacity curve associated with the relevant action.

Description

Description TITLE: DETERMINATION OF PEAKS IN INCREMENTAL CAPACITY CURVES OF A LITHIUM-ION CELL, IN PARTICULAR TO DETERMINE A DEGRADATION MODE OF THIS CELL Technical field of the invention

[0001] The present invention relates to a method for determining peaks of incremental capacity curves of a lithium-ion cell, a method for determining a degradation mode of a lithium-ion cell of a motor vehicle battery, a method for training a learning model, a corresponding computer program, a monitoring system for a lithium-ion cell of a motor vehicle battery and a motor vehicle comprising such a monitoring system. Technological background

[0002] A lithium-ion cell can age, meaning it degrades over time. Such aging is a complex process resulting from the interaction of multiple causes. Several modes of degradation can occur depending on the cell's usage, particularly the specifics of the different charge and discharge cycles the cell has undergone in the past.

[0003] Incremental capacity analysis is a known method for detecting three modes of degradation: loss of conductivity, loss of lithium inventory, and loss of active material. More specifically, this method uses the evolution of peaks in the incremental capacity curve to perform this detection.

[0004] However, this method requires determining incremental capacity curves over successive time intervals and detecting peaks in the resulting curves. These operations are computationally expensive.

[0005] It may therefore be desirable to provide a method for determining peaks that allows us to overcome at least some of the aforementioned problems and constraints. Summary of the invention

[0006] A method for determining the peaks of incremental capacity curves of a battery cell is therefore proposed, characterized in that it comprises: an input of a sequence of actions each characterized by predefined parameter values, to a learning model previously trained by supervised training to provide in response, for each action, at least one principal peak of the incremental capacity curve associated with the action in question.

[0007] The invention may further include one or more of the following optional features, in any technically feasible combination.

[0008] Optionally, the model is a random forest, preferably with multiple exits.

[0009] Optionally, the predefined parameters also include one or more of the following parameters: a charge / discharge protocol; a temperature; a charge current; a discharge current; a voltage limit; a rest period before a discharge; a rest period before a charge; and a compression force applied to a pack containing the cell.

[0010] Optionally, the actions form a series of charge / discharge cycles, and the predefined parameters of each action include one or more of the following parameters: the number of cycles; a charge state at the beginning of the action cycle; a charge state at the end of the action cycle; and a discharge depth equal to the difference between the charge states at the beginning and end of the action cycle.

[0011] Optionally, the cell is also a lithium-ion cell.

[0012] A method for determining the degradation mode of a cell in a motor vehicle battery is also proposed, characterized in that it comprises: a measurement of the voltage and current of the cell as a function of time in vehicle use; a determination of a sequence of actions from the measured voltage and current; a obtaining of peaks by a process according to the invention; and a determination of one among several predefined modes of cell degradation from the determined peaks.

[0013] Optionally, the determination of the sequence of actions includes determining successive charge / discharge cycles and breaking down each cycle into actions.

[0014] Also proposed is a computer program downloadable from a communication network and / or recorded on a computer-readable medium, characterized in that it includes instructions for executing the steps of a process according to the invention, when said program is executed on a computer.

[0015] Also proposed is a monitoring system for a cell in a motor vehicle battery, designed to implement a method according to the invention.

[0016] Also proposed is a motor vehicle comprising: a battery comprising several lithium-ion cells; and a monitoring system for at least one of the cells, according to the invention.

[0017] A method for training a learning model is also proposed, characterized in that it comprises: for each of several identical battery cells: • an application of a sequence of actions, each action being characterized by predefined parameter values, • during the application of the sequence of actions, a measurement of cell voltage and capacitance as a function of time, and • for each action, a calculation of an incremental capacity curve and a determination of at least one principal peak of this incremental capacity curve; a formation of a training corpus associating, for each sequence of actions, each action of this sequence of actions with the peak(s) determined for this action; and a supervised training of the model from the training corpus. Brief description of the figures

[0018] The invention will be better understood with the aid of the following description, given solely by way of example and made with reference to the accompanying drawings in which: Figure 1 is a schematic side view of a motor vehicle according to the invention; Figure 2 is a functional diagram of a monitoring system provided in the motor vehicle of Figure 1; Figure 3 is an example of an incremental capacity curve; Figure 4 is a curve illustrating the charge / discharge cycles of a monitored cell; Figure 5 is a functional diagram illustrating the input and output of a previously trained learning model; Figure 6 is a functional diagram of the model of Figure 5, in the case where this model is a random forest; Figure 7 is a block diagram of a method for determining a degradation mode of the cell; Figure 8 is an example of two timing diagrams.Figure 9 shows an example of peaks in incremental capacitance curves, Figure 10 illustrates linear regressions on the peaks of Figure 9, Figure 11 illustrates linear regressions on another example of peaks in incremental capacitance curves, Figure 12 is a block diagram illustrating a possible use of the method in Figure 7 for determining degradation mode(s), Figure 13 is a block diagram illustrating a method for training the model, and Figure 14 is a functional diagram illustrating a possible implementation of a data processing device for the system in Figure 2. Detailed description of the invention

[0019] With reference to Figure 1, an example of a motor vehicle 100 according to the invention will now be described.

[0020] The motor vehicle 100 includes drive wheels 102 and an electric motor 104 for driving the drive wheels 102.

[0021] The motor vehicle 100 further includes a battery 106 designed to provide a DC voltage U and an inverter 108 designed to convert the DC voltage U into phase voltages for the electric motor 104. The DC voltage U is preferably a high voltage, i.e. a voltage greater than 100 V, for example 400 V or 800 V.

[0022] The battery 104 comprises several cells 110, for example lithium-ion, connected in series and / or parallel to collectively provide the DC voltage U. For example, the cells 110 are distributed into several 110A-C packs, the cells of each 110A-C pack being stacked and compressed against each other with a certain compression force. This compression is intended to maintain contact between the cells of the 110A-C pack.

[0023] The motor vehicle 100 further includes a monitoring system 114 for at least one 112 of the cells 110 of the battery 104, preferably all of them.

[0024] The motor vehicle 100 further includes a battery 116 into which the monitoring system 114 is designed to write, and a human / machine interface 118, for example a dashboard, which the monitoring system 114 is designed to use.

[0025] With reference to Figure 2, the 114 monitoring system will now be described in more detail, in relation to the monitoring of cell 112, the monitoring of the other monitored cells being similar.

[0026] The monitored cell 112 is designed to present a voltage V and to exchange, that is to say to supply when discharging and to receive when charging, a current I. The monitoring system 114 thus includes a voltage sensor 202 designed to measure the voltage V of the monitored cell 112, a current sensor 204 designed to measure the current I exchanged by the monitored cell 112, and a temperature sensor 205 designed to measure an ambient temperature of the cell 102.

[0027] The 114 surveillance system also includes a 206 data processing device, the operation of which will be described later.

[0028] The monitoring system 114 also includes, for example in the device 206, a learning model 208, previously trained to provide, as will be described later in more detail, peaks of incremental capacity curves.

[0029] With reference to Figure 3, the concept of the peak of the incremental capacity curve will now be described in more detail.

[0030] The monitored cell 112 evolves over time and thus exhibits a capacitance Q that varies with time, and therefore with the voltage V, which also varies with time. It is thus possible to calculate the derivative dQ / dV of this capacitance Q as a function of the voltage V, over any time interval called an "action." This derivative dQ / dV generally exhibits two main peaks Pi and P2, defined in particular by their respective positions. The evolution of these positions during successive actions can provide information on the degradation mode of cell 112, as will be explained later.

[0031] More specifically, with reference to Figure 4, the monitored cell 112 undergoes successive partial or complete charge and discharge cycles during operation, labeled C1-C5 in Figure 4. Each cycle C1-C5 consists of a charge (partial or complete) and a discharge (partial or complete) of the monitored cell 112. Each cycle C1-C5 can be broken down, as will be explained later, into several successive actions A (for clarity, only some are indicated by an arrow in Figure 4). Each action A is characterized by predefined parameter values ​​that define the evolution of the monitored cell 112 during action A, that is, over the time interval corresponding to that action A.

[0032] These predefined parameters defining each action include, for example, one or more of the following parameters: a charge / discharge protocol (denoted p): this parameter indicates the one among several predefined protocols that most closely matches the evolution of the voltage V and / or the current I during the action. The predefined protocols include, for example, one or more of the following: constant current I, freely varying voltage V; constant voltage V, freely varying current I; voltage V and current I following each a predefined evolution; the current I following a predefined (non-constant) evolution, the voltage V varying freely; the voltage V following a predefined (non-constant) evolution, the current I varying freely; a predefined evolution is, for example, a linear or polynomial evolution of degree two or higher; the ambient temperature (denoted T) during the action; a charging current (denoted c): when the protocol is for the current I to be constant, this parameter indicates the value of the constant (positive) current I; a discharging current (denoted d): when the protocol is for the current I to be constant, this parameter indicates the value of the constant (negative) current I; a voltage limit (denoted le): value of the voltage V at the end of the action; a rest period, i.e., with the current I zero, before a discharging action (denoted r1); a rest period, i.e., with the current I zero, before a charging action (denoted r2);a state of charge of the monitored cell 112 at the beginning of the action cycle (noted SoCmin); a state of charge of the monitored cell 112 at the end of the action cycle (noted SoCmax); a depth of discharge (noted DoD): it is equal to the difference between the states of charge at the beginning and end of the action cycle, DoD = SoCmax - SoCmin; the number of charge / discharge cycles (noted K); and the compression force (noted F).

[0033] Thus, the parameters p, T, c, d, le, r1 and r2 can have different values ​​from one action to another), the parameters SoCmin, SoCmax and DoD have the same values ​​for the actions of the same cycle, and the compression force has the same value for all actions of all cycles.

[0034] Thus, the 208 model is designed, after training, to receive the values ​​of the predefined parameters of each action of a series of actions on the monitored cell 112, forming for example K charge / discharge cycles, and to provide, in response, for each action, at least one peak of the incremental capacity curve corresponding to that action.

[0035] With reference to Figure 5, the 208 model is more precisely designed to receive as input a list of groups of values ​​of predefined parameters and to provide as output a list of principal peaks, i.e. their coordinates in the dQ / dV - V plane. For example, the first two principal peaks P1, P2 are provided for each action (group of parameter values).

[0036] Referring to Figure 6, Model 208 is, for example, a random forest. Model 208 then comprises J decision trees Ti...Tj designed to provide independent predictions. The decision trees Ti...Tj are trained on a training corpus that associates, for several training actions, each of these training actions with one or more peaks. Preferably, each decision tree Ti...Tj is trained on only a subset of the data in the training corpus, that is, only a subset of the parameters and / or only a subset of the training actions. The output prediction is then obtained from the individual predictions of the decision trees Ti...Tj. The output prediction is, for example, an average of the individual predictions or obtained by a majority vote of the individual predictions.

[0037] Preferably, a random forest with multiple outputs is used. In this case, each individual prediction is a prediction of all the peaks being sought, as illustrated in Figure 6.

[0038] With reference to Figure 7, an example of a method 700 for searching for mode(s) of degradation of the monitored cell 112 will now be described.

[0039] During a step 702, the device 206 receives the voltage V, the current I and the temperature T, measured by the sensors 202, 204, 205 over time, during the use of the motor vehicle 100. The device 206 thus obtains in particular the voltage V and the current I measured as a function of time.

[0040] An example of voltage V and current I curves as a function of time t, over a charge / discharge cycle Ck (complete in this example) is illustrated in Figure 8.

[0041] For example, device 206 allows the time interval to continue until K consecutive charge / discharge cycles are achieved, K being a predefined number.

[0042] Alternatively, device 206 determines, within the time interval which is then of predefined duration, the number K of consecutive charge / discharge cycles present in that time interval.

[0043] Returning to Figure 7, during a step 704, the device 206 determines a sequence of actions AI ...AN corresponding, over the time interval, to the measured voltage V and current I.

[0044] To achieve this, the device 206, for example, breaks down each charge / discharge cycle into actions, each characterized by predefined parameter values. The actions (time intervals) are determined so that the parameter values ​​remain approximately constant during each action. For example, an optimization algorithm is used to determine the number of actions and their durations, ensuring that the parameter values ​​remain approximately constant during each action.

[0045] For example, in Figure 8, actions An...An+6 are represented, the five actions A n ...HAS n+4 forming the charge / discharge cycle Ck, with their parameters p, T, c, d, le, r1, r2. When a parameter is not indicated, it is irrelevant to the action being considered. For example, action An is defined as implementing a constant current charging protocol p (indicated by the value p1), so the discharge current value parameter d is meaningless. In this case, the parameter is set to a value, for example zero, signifying its non-use.

[0046] The parameters SoCmin, SoCmax, and DoD have the same values ​​for all A actions. n ...HAS n +4 of the Ck cycle, and are thus denoted SoCmink, SoCmaxk, DoDk.

[0047] Returning to Figure 7, during a step 706, the device 206 provides the list of actions AI ...AN to the previously trained model 208, so that the latter provides in response, for each action AI ...AN, at least one peak of the incremental capacity curve corresponding to that action AI ...AN.

[0048] An example of peak generation is illustrated in Figure 9, where the 208 model is pre-trained to provide the two peaks of each incremental capacity curve. These peaks are labeled P1. n , P2 n or together (P1, P2) n , with n the index of the corresponding AI ...AN stock.

[0049] Returning to Figure 7, during a step 708, the device 206 determines at least one of several predefined modes of cell 112 degradation from the P1 peaks n , P2 n determined.

[0050] For example, the predefined degradation modes are: a loss of conductivity (from the English "conductivity loss"), which is reflected in the fact that the P2n peaks are located approximately on a horizontal line; a loss of cyclable lithium (from the English "loss of lithium inventory"), which is reflected in the fact that the P2n peaks are located approximately on a vertical line; and a loss of active material (from the English "loss of active material"), which is reflected in the fact that the P1n peaks are located approximately on an oblique line.

[0051] Thus, device 206 performs a linear regression of the P1n peaks on a line D1 and a linear regression of the P2n peaks on a line D2. Then, device 206 compares the slope of each line D1, D2 with predefined values: if the slope of line D1 is not zero (below or above zero), degradation by loss of active material is detected; if the slope of line D2 is zero, degradation by loss of conductivity is detected – degradation by loss of conductivity can also be detected when the slope of line D2 is not zero but within a predefined range around zero; and if the slope of line D2 is infinite, degradation by loss of cyclable lithium is detected – degradation by loss of conductivity can also be detected when the slope of line D2 is not infinite but simply greater than a predefined threshold.

[0052] This determination of the degradation mode(s) is illustrated in Figure 10 in the case of the peaks in Figure 9. As can be seen, degradation by loss of active material and degradation by loss of cyclable lithium are detected.

[0053] Another example of peaks obtained and determination of degradation mode(s) are illustrated in Figure 11. In this case, degradation by loss of active material and degradation by loss of conductivity are detected.

[0054] With reference to Figure 12, the determination of degradation mode(s) can be used in a battery management process 1200 106, as will now be described.

[0055] During a step 1202, one or more degradation modes are sought by implementing a process according to the invention, such as process 700.

[0056] During step 1204, if a degradation mode is detected, the monitoring system 114 triggers an alert for the driver on the human-machine interface 118 of the motor vehicle 100. Generally, the alert may also include more detailed information about the nature of the problem and the recommended actions to be taken. For example, the alert may include a visual indicator that could specify the detected degradation mode(s). The visual indicator may include, for example, an icon or illuminated symbol on the motor vehicle's instrument panel, an alert message on a motor vehicle display, or a notification sent to the driver's portable device, such as a mobile phone, for example, to a mobile application installed on the device.This latest alert method allows the battery 106 to be monitored remotely and alerts to be received even remotely from the motor vehicle 100.

[0057] Alternatively or in addition, during a step 1206, the monitoring system 114 records the detected degradation mode(s) in memory 116.

[0058] Thus, during a step 1208, a repairer of the motor vehicle 100 can consult the memory 116 and undertake the necessary repairs on the battery 106 according to the mode(s) of degradation recorded.

[0059] Alternatively or in addition, during a step 1210, parameters of the motor vehicle 100 may be modified according to the mode(s) of degradation detected, for example in order to extend the life of the battery 106.

[0060] Alternatively or in addition, during a step 1212, a customized cycling / storage strategy based on the detected degradation mode(s) may be recommended.

[0061] With reference to Figure 13, an example of a 1300 method for training the 208 model will now be described.

[0062] The 1200 process first includes steps 1302 to 1312 carried out for each of several identical lithium-ion cells, i.e. of the same model.

[0063] During step 1302, predefined actions are applied to the current cell to perform several charge / discharge cycles. For example, The same sequence of actions performing a charge / discharge cycle is repeated several times.

[0064] Each action is characterized by values ​​of predefined parameters, for example those detailed above.

[0065] During step 1304, during the application of actions 902, the voltage V and current I of the current cell are measured over time.

[0066] During a step 1306, a capacitance Q of the current cell as a function of time is obtained, for example calculated from the voltage V and the current I (the capacitance Q is then indirectly measured).

[0067] In an optional step 1308, the voltage V and capacitance Q are smoothed. For this, the Savitzky-Golay algorithm can be used for each curve. As is well known, this algorithm involves fitting successive portions of the curve with a polynomial. Preferably, a polynomial of degree three is used.

[0068] During step 1310, for each action, an incremental capacity curve is calculated.

[0069] During a step 1312, for each action, two main peaks of the associated incremental capacity curve are determined.

[0070] During step 1314, a training corpus is formed. The training corpus associates, for each cycling plan Sk, each action Ank of that cycling plan Sk with the peaks determined for that action Ank.

[0071] During step 1316, model 208 is trained in a supervised manner from the training corpus.

[0072] During step 1318, the most important parameters are selected and the process returns to step 1316.

[0073] With reference to Figure 14, the data processing device 206 is, for example, a computer system comprising a data processing unit 1402 (such as a microprocessor) and a main memory 1404 (such as RAM, or Random Access Memory) accessible by the processing unit 1402. The computer system further comprises, for example, a network interface and / or a computer-readable medium, such as a local medium (such as a local hard drive 1406) or a remote medium (such as a A remote hard drive accessible via the network interface through a communication network) or a removable storage medium (such as a USB flash drive, a CD, a Compact Disc, or a DVD, a Digital Versatile Disc) readable by means of a suitable drive on the computer system (such as a USB port or a CD and / or DVD drive). A computer program 1408 containing instructions for the processing unit 1402 is stored on the storage medium 1406 and / or downloadable via the network interface. This computer program 1408 is, for example, intended to be loaded into main memory 1404, so that the processing unit 1402 executes its instructions, in order to implement process 700 of Figure 7 and / or process 1200 of Figure 12. The model 208 is thus in the form of digital data, for example recorded on the medium 1406.

[0074] Alternatively, all or part of the steps of processes 700, 1200 could be implemented in the form of hardware modules, i.e. in the form of an electronic circuit, for example micro-wired, not involving a computer program.

[0075] Furthermore, process 1300 can be implemented by a computer program designed to be executed by a computer system similar to that in Figure 13.

[0076] In conclusion, it is clear that the invention makes it possible to determine peaks of incremental capacity curves.

[0077] It should also be noted that the invention is not limited to the embodiments described above. Indeed, it will be apparent to those skilled in the art that various modifications can be made to the embodiments described above, in light of the information just disclosed to them.

[0078] In the detailed presentation of the invention given above, the terms used shall not be interpreted as limiting the invention to the embodiments set forth in this description, but shall be interpreted as including all equivalents which can be foreseen by a person skilled in the art by applying their general knowledge to the implementation of the teaching which has just been disclosed to them.

Claims

Demands [1] Method (606) for determining peaks of incremental capacity curves of a battery cell, characterized in that it comprises: an input, for each of a series of successive time intervals (AI ...AN), of values ​​of predefined parameters (p, T, c, d, le, r1, r2, SoCmin, SoCmax, DoD, N, F) characterizing an evolution of the battery cell over the time interval considered, in a learning model (208) previously trained by supervised training to provide in response, for each time interval (AI ...AN), at least one main peak (P11, P2I ... P1 N, P2N) of the incremental capacity curve associated with the time interval considered, the predefined parameters comprising at least one charge / discharge protocol (p) of the battery cell over the time interval considered. [2] Method (606) according to claim 1, wherein the model (208) is a random forest, preferably with multiple exits. [3] Method (606) according to claim 1 or 2, wherein the predefined parameters include one or more of the following parameters: a temperature (T); a charging current (c); a discharging current (d); a voltage limit (denoted le); a rest period (r1) before a discharge; a rest period (r2) before a charge (denoted r2); and a compression force (denoted F) applied to a pack comprising the cell (112). [4] Method (606) according to any one of claims 1 to 3, wherein the successive time intervals form a series of charge / discharge cycles, and wherein the predefined parameters of each action include one or more of the following parameters: the number of cycles (N); a state of charge at the beginning of the cycle of which the time interval considered is a part (SoCmin); a state of charge at the end of the cycle which includes the time interval considered (SoCmax); and a depth of discharge (DoD) equal to the difference between the states of charge at the beginning and end of the cycle which includes the time interval considered. [5] Method (606) according to any one of claims 1 to 4, wherein the cell is a lithium-ion cell. [6] Method (700) for determining a degradation mode of a cell (112) of a battery (106) of a motor vehicle (100), characterized in that it comprises: a determination (704) of a series of successive time intervals from a measurement (702) of a voltage (V) and a current (I) of the cell as a function of time in use of the vehicle (100); a obtaining (706) of peaks by a method according to any one of claims 1 to 5; and a determination (708) of one among several predefined modes of cell degradation from the determined peaks. [7] Method (700) according to claim 6, wherein the determination (704) of the sequence of time intervals comprises a determination of successive charging / discharging cycles and a division of each cycle into time intervals. [8] Computer program (1408) downloadable from a communication network and / or stored on a computer-readable medium, characterized in that it includes instructions for carrying out the steps of a process (606, 700) according to any one of claims 1 to 7, when said program is executed on a computer. [9] System (114) for monitoring a cell (112) of a battery (106) of a motor vehicle (100), designed to implement a method (700) according to claim 6 or 7. [10] Motor vehicle (100) comprising: a battery (106) comprising several lithium-ion cells (110); and a monitoring system (114) for at least one (112) of the cells (110), according to claim 9. [11] Method (1300) for training a learning model (208), characterized in that it comprises: for each of several identical battery cells: • an application (1302) of a sequence of actions over time intervals respectively, each action being characterized by predefined parameter values, including a cell charge / discharge protocol (p) during the action considered, • during the application of the sequence of actions, a measurement (1304, 1306) of a voltage (V) and a capacitance (Q) of the cell as a function of time, and • for each time interval, a calculation (1310) of an incremental capacity curve and a determination (1312) of at least one principal peak (P1, P2) of this incremental capacity curve; a training (1314) of a training corpus associating, for each sequence of actions, the values ​​of the predefined parameters of each action of this sequence of actions with the peak(s) determined for this action; and a supervised training (1316) of the model (208) from the training corpus.

Images