Cell disconnect detection

Abstract

There is described apparatus for monitoring a string of battery cells for disconnection events each of which comprises electrical disconnection of one or more of the cells from the string, the cells of the string being connected electrically in parallel with each other such that the string has an electrical capacity, the apparatus comprising: a string measurement input arranged to receive measurements of one or more physical parameters of the string of cells; a degradation-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a concurrent first estimate of the electrical capacity; a disconnect-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a concurrent second estimate of the electrical capacity; and a disconnect detector arranged to detect disconnection events using a comparison of the first and second estimates of electrical capacity. A corresponding monitoring method is also described.

Description

[0001] Cell disconnect detection

[0002] The invention relates to monitoring strings of battery cells, for example such monitoring implemented within a battery management system of a vehicle where the battery is used to provide motive power for the vehicle.

[0003] Introduction

[0004] Battery modules, for example used for motive power in vehicles as well as numerous other applications, often contain multiple strings each of several cells which are coupled electrically in parallel, so as to meet capacity and maximum current requirements which would otherwise be unachievable from a single cell. Multiple strings are typically connected electrically in series to achieve the desired module voltage and maximum power output. Such battery modules are usually designed such that electrical connections between cells of a string allow for homogenous current sharing across all the cells in parallel. The cells of a string may be connected in parallel in various ways, but frequently this is through welded connections of the cells to electrical bus bars or similar.

[0005] Various circumstances can lead to particular cells within a parallel string becoming electrically disconnected. This results in the total load current being shared across a lower number of cells in the string. Each cell therefore sees a higher current than expected, causing it to charge and discharge more rapidly. This results in reduced electrical capacity of the string, and of any module of which it forms a part, as well as faster cell degradation.

[0006] Such cell disconnections within a parallel string are challenging to detect without adding separate current sensors for each cell. In some approaches, a controlled charge and discharge cycle can be used in an attempt to accurately measure changes in total electrical capacity of a string or module, but while such testing can be used in a controlled environment such as at manufacturing and related testing, it is generally impractical to carry out such controlled cycles during normal operation of a vehicle or other application scenario.

[0007] The invention relates to this and other issues of the related prior art.

[0008] Summary of the invention

[0009] The invention provides apparatus and methods for detecting cell disconnects within a parallel string of cells, which remains reliable and uncompromised by cell capacity degradation over time, and which moreover can be readily implemented in a real-time embedded battery management system environment for example on board a vehicle or elsewhere. Detection of cell disconnects operates based on a comparison of the outputs of two model based estimators of electrical capacity of the string, a first estimator tuned to detect long term degradation of electrical capacity of the string (sometimes referred to as a state of health capacity), and a second estimator tuned for increased sensitivity to rapid or instantaneous changes in electrical capacity of the string. Operation may then be based on the following two principles:

[0010] - since a cell disconnect is an immediate event, the second estimator will respond more rapidly than the first. Observing the difference between the capacity estimates of the first and second estimators, for example as expressed in a fraction of a nominal 100% capacity or total change from that nominal capacity, can be used to flag a potential disconnect if this becomes large enough. When a disconnect is not present the capacity estimates from both estimators should be changing at roughly the same rate.

[0011] - the number of disconnects occurring at a particular time, or in a particular interval, or in total since initialisation, can be determined in various ways. However, since the capacity estimates of both estimators is expected to drop over time as cell degradation takes place, determining a number of cells which have disconnected may be based on the capacity of the second estimator in comparison to a baseline capacity which can be held constant, but reset periodically for example when the first estimator indicates a sufficient fall in capacity due to degradation.

[0012] To carry out a concurrent or real time estimate of electrical capacity of the string of cells, the first and second estimators typically requires at least measurements of the voltage across, and current through the string, and preferably also a temperature of the string. In the described apparatus and methods the first and second estimators are preferably also stateful: any form of recursive state estimation which only requires measurements of physical parameters of the string at the current time step (t = k), and the state at the previous time step (t = k-1) could be used, although particular examples of the invention use Kalman filters such as extended Kalman filters for these purposes.

[0013] The two estimators then each provide a separate estimate of electrical capacity of the string, which may be represented as a loss in capacity relative to the nominal 100% capacity, which suitable logic implemented in software then uses to flag cell disconnects from the string. For example, the logic may compute a difference between the electrical capacities estimated by the two estimators and use this difference to determine if and when a cell disconnect has taken place. The logic then also uses at least one of the estimates of electrical capacity, typically at least that from the second capacity estimator, as well as any other relevant input from the estimators to estimate the likely number of cells which have disconnected.

[0014] In particular, the invention provides apparatus for monitoring a string of battery cells for disconnection events each of which comprises electrical disconnection of one or more of the cells from the string, the cells of the string being connected electrically in parallel with each other such that the string has an electrical capacity. The apparatus may comprise a string measurement input arranged to receive measurements of one or more physical parameters of the string of cells. However such measurements are received, the apparatus then comprises a degradation-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a first estimate of the electrical capacity; a disconnect-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a second estimate of the electrical capacity; and a disconnect detector arranged to detect disconnection events using a comparison of the first and second estimates of electrical capacity.

[0015] The first and second estimates of electrical capacity of the string may be described as concurrent, or real-time estimates of electrical capacity, in the sense that the apparatus operates in real-time to receive measurements of the string and to calculate and use the two estimates of electrical capacity. Of course, the capacity estimators and disconnect detector need not run continuously, but may be executed periodically depending on need, computational resources available, and other factors, which may lead to capacity estimates and detection of disconnection events being implemented at intervals of fractions of a second up to minutes or hours.

[0016] The one or more physical measurements may comprise a voltage across the string and a current through the string, and typically also a temperature of the string. These measurements are then used to calculate updated values of the capacity estimates and to detect any disconnection event indicated by these updated capacity estimates.

[0017] In order to detect disconnection events, the disconnect-sensitive capacity estimator may be tuned for sensitivity to instantaneous changes in the electrical capacity and the degradation-sensitive capacity estimator may be tuned for sensitivity to long-term degradation in the electrical capacity. For example, the degradation-sensitive capacity estimator and the disconnect-sensitive capacity estimator and may be tuned or arranged such that, in response to a change in the electrical capacity, such as a disconnection event or step change in the actual capacity of the string, the second estimate of electrical capacity has a higher rate of change than the first estimate of electrical capacity. The capacity estimators may be tuned to optimise the differences in such rates of change for event detection and disconnect counting, but typically the second estimate of electrical capacity may have a rate of change which is at least five times higher, or at least ten times higher than a rate of change of the first estimate of electrical capacity in response to the same disconnection event or step change in actual electrical capacity.

[0018] The disconnect detector may more particularly be arranged to detect a disconnection event from differences between the first and second estimates of electrical capacity, for example when a difference between the first estimate of electrical capacity and the second estimate of electrical capacity exceeds an event detection threshold. Since the capacity estimates are likely to be noisy, the disconnect detector may be arranged to detect and flag a disconnection event only when the first and second estimates are indicative of a disconnection event for a particular interval such as a period of time or number of time steps or iterations of the estimator, for example when the difference between the two capacity estimates exceeds the event detection threshold for at least a predetermined de-bounce interval.

[0019] The disconnect detector may also be arranged to determine a number of cells which have disconnected in one or more disconnection events using at least the second estimate of electrical capacity. The determined number of disconnected cells may correspond to the number of cells disconnected in a short, particular detected disconnection event, or over a longer period for example during an interval following such an event.

[0020] More particularly, the apparatus may be arranged to maintain a baseline capacity estimate which is periodically reset, and the disconnect detector is then arranged to determine a number of cells which have disconnected in one or more disconnection events from a difference between the baseline capacity estimate and the second estimate of electrical capacity. The baseline capacity estimate may be set as constant to provide a stable reference point against which the number of disconnects can be determined, but can then be reset periodically to correspond to the concurrent first estimate of electrical capacity, thereby ensuring that the reference point reflects changes in string capacity due to gradual cell degradation. Just prior to a reset of the baseline capacity, the difference between the baseline capacity and the second estimate of electrical capacity is then indicative of loss of capacity due to disconnects, from which the number of disconnects since the last reset can then be calculated.

[0021] Each of the degradation sensitive capacity estimator and the disconnect-sensitive capacity estimator may comprise one or more state or stateful models of the string of cells, each state model representing a different respective model estimate of electrical capacity of the string, each state model being arranged to advance to a new state having the same model estimate of electrical capacity as before responsive to received measurements of the one or more physical parameters of the string of cells. In particular, each state model may be implemented as a Kalman filter, or more particularly as an Extended Kalman filter.

[0022] In more detail, each of the degradation-sensitive capacity estimator and disconnectsensitive capacity estimator may comprise an interacting multiple model structure comprising a plurality of such state models arranged to track a current state of the string of cells responsive to receiving the measurements of the one or more physical parameters. Within each of the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator, each state model may then represent the string of cells as having a different electrical capacity. For each capacity estimator, the estimate of electrical capacity of the string of cells may then correspond to that state model which best fits the current physical parameter measurements.

[0023] For the degradation-sensitive capacity estimator, the state models may represent a range of electrical capacities close to the current best estimate of electrical capacity. For the disconnect-sensitive capacity estimator, each state model may represent the string of cells with a different number of cells having disconnected, and the estimator may then maintain weights indicating the probability of each state model being the best fit to the concurrent physical parameter measurements. These weights, for example in the form of a probability mass function, may then be used by the disconnect detector in addition to using at least the second estimate of electrical capacity, to determine the number of cells which have disconnected in one or more disconnection events.

[0024] The above apparatus may be implemented in hardware, software, or a combination of the two, but typically will be implemented using software arranged to execute on a suitable computer system. Such a computer system will typically comprise one or more microprocessors, memory for storing the software and related data, and suitable input / output mechanisms which may include one or more displays, network connections, electronics for receiving the measured physical parameters of the string and other data relating the battery module, and electronics for sending control signals relating to management of the battery module. Typically the apparatus and related software may be implemented as part of a batter management system on board a vehicle or a variety of other implementation environments.

[0025] To this end, the invention also provides the above apparatus but further comprising a battery, the battery comprising the string of battery cells, and sensors arranged to make the one or more physical measurements and to pass the one or more physical measurements to the string measurement input.

[0026] The invention also provides a vehicle comprising the above apparatus wherein the battery is arranged to provide motive power for the vehicle.

[0027] The invention also comprises methods corresponding to the above apparatus, for example a method of monitoring a string of battery cells for disconnection events each of which comprises electrical disconnection of one or more of the cells from the string, the cells of the string being connected electrically in parallel with each other such that the string has an electrical capacity, the method comprising: receiving measurements of one or more physical parameters of the string of cells; calculating from the measurements a first estimate of the electrical capacity using a degradation-sensitive capacity estimator; calculating from the measurements a second estimate of the electrical capacity using a disconnect-sensitive capacity estimator; and detecting a disconnection event using a comparison of the first and second estimates of electrical capacity.

[0028] In response to a change in the electrical capacity such as a disconnection event or step change, the second estimate of electrical capacity has a higher rate of change, or a rate of change which is at least five or at least ten times higher, than a rate of change of the first estimate of electrical capacity.

[0029] The method may proceed to detect a disconnection event when a difference between the first estimate of electrical capacity and the second estimate of electrical capacity exceeds an event detection threshold, and optionally when the difference exceeds the event detection threshold for at least a predetermined de-bounce interval.

[0030] The method may further comprise determining a number of cells which have disconnected in the disconnection event, optionally using at least the second estimate of electrical capacity.

[0031] Each of the degradation sensitive capacity estimator and the disconnect-sensitive capacity estimator may comprise one or more state of stateful models of the string of cells, each state model defining a respective model estimate of electrical capacity of the string, each state model being arranged to advance to a new state responsive to received measurements of the one or more physical parameters of the string of cells. For the disconnect-sensitive capacity estimator, each state model may represent the string of cells with a different number of cells having disconnected.

[0032] The invention also provides one or more computer programs arranged to put into effect the above apparatus, or to carry out the above method steps, when executed on a suitable computer system, and one or more computer readable media carrying such computer program code.

[0033] Brief description of the drawings

[0034] Embodiments of the invention will now be described, with reference to the drawings, of which:

[0035] Figure 1 depicts an embodiment of the invention in which a monitor is implemented to monitor a string of cells in a battery module;

[0036] Figure 2 shows behaviour of first (solid) and second (dotted) estimates of electrical capacity of the string of cells as determined using the monitor of figure 1 , along with a baseline capacity estimate (dashed) and detection of a disconnection event;

[0037] Figure 3 schematically shows how logic of the disconnect detector of figure 1 may be implemented;

[0038] Figure 4 is a flow chart showing execution of tasks by the monitor of figure 1 ;

[0039] Figure 5 is a flow chart showing how the disconnect flagging logic may detect a disconnection event using the estimates of capacity provided by the two estimators of figure 1 ; and

[0040] Figure 6 illustrates how each estimator of figure 1 may be implemented using an interactive multiple model structure.

[0041] Detailed description of embodiments

[0042] Referring now to figure 1 there is shown a battery module 10, and a monitor 100 for monitoring the battery module 10. The battery module 10 may for example be installed in a road or other vehicle, and the monitor 100 may form part of a battery management system installed in such a vehicle for the purposes of managing and monitoring the battery module. A complete battery may typically comprise many such battery modules 10. It is of course possible for the monitor 100 to be provided remotely from the battery module, and in particular not in a vehicle which is carrying, or other installation comprising the battery module, and instead to be connected to the vehicle or other installation over a network connection such as a cellular data connection. However, the monitor 100 and associated battery monitoring techniques described herein have particular utility in being suitable for operation with reduced computational power within a battery management system onboard the same vehicle or local to the same installation as the battery module.

[0043] The battery module 10 comprises one or more strings 20 of battery cells 30. Typically such a battery module 10 will comprise multiple such strings arranged electrically in series in order to provide a higher voltage output at the battery module terminals 40 than would be available from a single string, and each string will comprise a plurality of battery cells arranged electrically in parallel so as to increase the electrical capacity of each string, i.e. the capacity for the string to deliver electrical power to a load, typically measured in kWh or sometimes expressed in Ah. In a typical road vehicle battery for example each string might comprise around 5 to 20 separate cells in parallel, and have an electrical capacity of from about 50 to 500 Wh. Suitable cell types could be 21700 cylindrical cells (diameter 21 mm and length 70 mm) such as the Molicel P45B cell, but larger or smaller cells, and cells of different shapes and form such as pouch or prismatic cells could be used. The battery module 10 might contain around 10 to 20 such strings in series, providing a total voltage between the battery module terminals of some 20 to 100V. To form a complete battery, multiple such battery modules may typically be connected in parallel or series, for example about 5 to 20 such modules for a typical automobile.

[0044] The individual cells 30 in a string 20 may be electrically connected in parallel in various ways, for example using bus bars 50 as illustrated in figure 1 , each bus bar being coupled to the terminals of one polarity of all of the cells in the string. These electrical connections may be made by welding or in other ways. Various circumstances such as design flaws, assembly or welding errors, rough handling of the battery module, thermal or kinetic shocks and so forth can lead to individual cells becoming electrically disconnected from the string 20. Of course, a failure in the electrical connection to just one of the terminals of a particular cell is sufficient for this to be the case. Sometimes just one cell may become disconnected from the string at any one time, but sometimes multiple cells may become disconnected substantially at the same time, for example due to a single kinetic shock. Generally, we will refer to the electrical disconnection from the string of just one cell, or the electrical disconnect of multiple cells at substantially the same time, or over a suitably limited period which may depend on how the monitor is implemented, as a disconnection event.

[0045] A disconnection event results in the total load current through the battery module 10 being shared by a reduced number of cells in the string suffering the event, so that the remaining connected cells of the string then forced to charge and discharge more rapidly than they were before the event. Not only is electrical capacity of the string and indeed the whole module reduced, but the cells in the string suffering the event degrade more quickly, and in extreme cases may operate close to or above safe current limits.

[0046] In the prior art, such disconnection events within individual strings can be difficult to detect, since most prior art battery modules only contain sensors to measure the total current through the module. Attempting to provide separate sensors to detect the current through each individual cell would lead to a much more complex battery management system.

[0047] The apparatus of figure 1 therefore includes a monitor 100 configured to monitor each of one or more of the strings 20 of battery cells. In figure 1 it is indicated how the monitor may be connected to monitor just one of such strings, but the functionality of the monitor may of course be reproduced for each string to be monitored, or operate using a multiplexing mode for example using time division multiplexing, so as to effectively function separately for each of multiple strings 20. The monitor may similarly function separately for strings 20 in each of multiple battery modules. In practice, the monitor 100 will typically be implemented largely or wholly in software on a suitable computer system, and typically as part of a larger piece of software forming part of a battery management system 110.

[0048] The monitor 100 receives one or more physical measurements / Wfrom the string 20, at a string measurement input 120, and uses these physical measurements to detect disconnection events. The physical measurements / I will typically comprise at least electrical current / passing through the string and voltage V across or at the string, and typically at least one temperature measurement T of the string. The temperature measurement could be at a single or at multiple points within or proximal to a particular string of cells, the temperature of an associated heatsink or similar structural element associated with the string and so forth, or a single temperature measurement could be used for all strings within a battery module. The current / may conveniently be measured at a single point for multiple strings 20 connected in series, rather than for each string 20. In figure 1 these physical measurements / Ware made by one or more sensors 60 disposed at appropriate positions relative to the string. The physical measurements may be communicated to the monitor 100 in various ways but typically as electrical data signals, and are received by the monitor 100 at the string measurement input 120 which may comprise various combinations of electronics and computer hardware and software, but in any case makes these measurements available to other parts of the monitor 100 as discussed below.

[0049] The monitor 100 comprises two elements 130, 140 each of which is used to provide a separate estimate the concurrent electrical capacity of the string 20 using the physical measurements M. These elements may be implemented using stateful models representing the string of cells, which can be advanced in state at discrete intervals on the basis of updates in the available physical measurements. For example such stateful models may be implemented using Kalman filters, or more particularly interactive multiple model arrangements each using multiple such Kalman filters as discussed in more detail below.

[0050] How frequently the two elements 130, 140 are used to calculate updated estimates of electrical capacity may depend for example on the computation resources of the monitor 100, as well as on factors such as whether the electrical capacity is expected to have changed rapidly for example due to rapid battery charging and discharging, and whether the particular application demands faster estimates. Typically the estimates may be updated using concurrently available measurements at intervals of a few seconds to a few hours.

[0051] More notably, the two elements are arranged to respond to actual changes in electrical capacity of the string at different response rates to each other so that they can be used together to provide better detection of disconnection events as discussed below.

[0052] To this end, the two elements mentioned above are provided as a degradationsensitive capacity estimator 130 and a disconnect-sensitive capacity estimator 140. The disconnect-sensitive capacity estimator is tuned for sensitivity to instantaneous or rapid changes in the electrical capacity such as disconnection events, which typically occur over time intervals of a few seconds at most, for example when an electrical connection weld fails. The degradation-sensitive capacity estimator is tuned for sensitivity to longer-term degradation in the electrical capacity of the string, for example due to chemical degradation and similar effects, which typically take place over multiple charging and discharging cycles, so for example over periods of hours to days or longer.

[0053] Using the measurements M received from the measurement input, the degradationsensitive capacity estimator 130 calculates and outputs a first estimate of change of electrical capacity of the string EC1, and the disconnect-sensitive capacity estimator calculates and outputs a second estimate of change of electrical capacity of the string EC2. Although sensitivity of the two estimators to degradation and disconnections respectively in a string 20 can be tuned in various ways, they may be arranged such that, in response to a change in the electrical capacity, such as a disconnection event, the second estimate of electrical capacity has a higher rate of change, for example at least five times or at least ten times higher, than a rate of change of the first estimate of electrical capacity. This could be measured in various ways, one being to measure the rate of change of the respective estimate of electrical capacity over a short time period immediately following a step change in the actual electrical capacity or disconnection event, or by determining the length of time taken for respective estimates to fall from a correct estimate of the actual capacity immediately prior to a step change or disconnection event by one half of the capacity change, or some other suitable measure. In other arrangements the capacity-sensitive estimator is configured to model more closely a disconnect - such as taking into account how cells behave differently when they have more current applied to them (e.g. with a disconnect) rather than a pure capacity loss owing to expected degradation.

[0054] The first and second estimates of change of electrical capacity are passed to a disconnect detector 150 of the monitor, which is arranged to use these detect disconnection events, in which of one or more cells from the string become disconnected so can no longer carry part of the current through the string. In particular, the disconnect detector 150 uses a comparison between the first and second estimates to detect such disconnection events. This comparison could take various forms, for example requiring one or more conditions using the estimates to be satisfied. In one form, these conditions may be met when a difference between the first estimate of electrical capacity and the second estimate of electrical capacity exceeds a predefined event detection threshold. However, because the first and second estimates of electrical capacity may be noisy, another condition may be that this exceeding of a threshold is sustained for a predefined de-bounce interval, for example over a particular period of time or a particular number of recalculations of the first and second estimates, thereby implementing a “de-bounce” condition. Another such condition could be that, once such a debounce condition has been met, the disconnection event is thereafter maintained regardless of any contrary indications in the first and second estimates.

[0055] Properties of a detected disconnection event may be output by the disconnect detector 150 as event properties P, which may for example comprise a number of cells of the string determined to have been disconnected during one or more events, and optionally other data such as time stamps, measures of certainty of aspects of the data, and so forth. These event properties may then be used by the wider battery management system 110, for example by a battery controller 160, to provide better management of charge and discharge particular strings, battery modules, or the battery as a whole, and / or to provide maintenance warnings to a user of a vehicle or other system, or other party responsible for maintenance of the vehicle or other system, within which the battery module 10 is installed.

[0056] Figure 2 illustrates graphically how the disconnect detector 150 may be arranged to determine a number of cells which have disconnected in one or more disconnection events, in particular using at least the second estimate of electrical capacity. The ordinate of the graph represents electrical capacity of a string of cells relative to a nominal or initial capacity of 100%, and the abscissa represents time, for example over a period of a few hours. The solid curve 210 then represents the first estimate of electrical capacity EC1 output by the degradation-sensitive capacity estimator 130, and the dotted curve 220 represents the second estimate of electrical capacity EC2 output by the disconnectsensitive capacity estimator 140. It can be seen that during normal operation the first estimate of capacity slowly falls, as gradual degradation of the cells of the string takes place. Estimating the electrical capacity from the concurrently available measured parameters at any one time can be challenging to do accurately, and so both the first and second estimates may be fairly noisy. During normal operation, however, absent any disconnection event or other rapid change in actual string capacity, it is expected that the second estimate of electrical capacity will roughly track the first.

[0057] Then, at the time of a disconnection event 230 it can be seen that the second estimate EC2 falls much more rapidly than the first estimate EC1, so that the difference between the two quickly exceeds the event detection threshold 240, and is maintained for longer than the predetermined de-bounce interval 250. A disconnect-event is therefore be recorded by the disconnect detector 150 and subsequently output as part of event properties Pas shown in figure 1 .

[0058] The disconnect detector 150 is arranged to determine a number of cells which have disconnected in one or more disconnection events. The number of cells disconnected will typically equate to loss of a corresponding fraction of the original, nominal 100% capacity, for example with disconnection of 1 cell of a 50 cell string giving rise to about a 2% loss in capacity as shown in figure 2. The number of cells disconnected in a particular disconnection event may therefore be determined using at least the second estimate of electrical capacity 220, and optionally other available data as discussed in more detail below. For example, the number of cells disconnected in a particular disconnection event may be determined by a comparison of the magnitude of change of the second estimate of the electrical capacity alone within a suitable time frame, or a comparison of the change of the second estimate relative to the first estimate of electrical capacity within a suitable time frame.

[0059] However, since the first estimate of capacity 210 may be fairly noisy, and typically falls gradually over time due to cell degradation, and noting that the number of cells disconnected in a disconnection event 230 may not be determined by the disconnect detector 150 immediately after a particular disconnection event, the inventors have found that it can be advantageous not to use a difference between the first and second estimates at around the time of the event, or to use an absolute change in the second estimate at around the time of the event, to determine a loss of capacity during the event and therefore a number of cells which have disconnected during the event. In the multi-model IMM approach described below in respect of figure 6, since each model of the disconnectsensitive IMM corresponds to the battery string having a different number of cell disconnects, a likeliest or best fit model of the IMM can be used to determine the number of disconnects.

[0060] Instead, the disconnect detector as illustrated in figures 1 and 2 makes use of a further baseline capacity estimate which is maintained by the monitor 100, and shown as B in figure 1 and depicted as curve 260 in figure 2. When the battery monitor 100 and related string are initialised to have a nominal capacity of 100%, this baseline capacity estimate 260 is also set to 100%. It is then maintained as a constant or substantially constant value and is only changed periodically at reset points 270, at which time it is reset to a new value corresponding to a concurrent first estimate of electrical capacity EC1, for example the particular current value, or an average of recent values in the interest of reducing the effects of noise. Reset points 270 may occur at fixed intervals, but may advantageously instead occur in response to the first estimate of electrical capacity EC1 falling by a certain amount, for example each time EC1 falls by a particular fraction of the nominal 100% capacity such as by 1%, or by a fraction of the nominal capacity representing the capacity of a particular, optionally fractional, number of cells in the string, such as 0.6 or 1 .0. The interval of time from reset point to the next is then referred to herein as a reset interval.

[0061] Following a detection of disconnection event 230, the disconnect detector 150 is then arranged to determine a number of cells which have disconnected since the last reset point 270, using a difference between the baseline capacity estimate 260 and the second estimate of electrical capacity 220. As indicated above, this difference can be converted to a number of disconnected cells by determining the fraction of nominal capacity it represents, and the number of cells that fraction represents. The fraction will rarely correspond to an exact number of cells, but the disconnect detector can be provided with various calculation rules to come to a best estimate of the number of cells disconnected, for example taking into account the total of previously calculated numbers of disconnected cells and the current fraction of nominal 100% capacity represented by the current second estimate of electrical capacity, or in other ways.

[0062] The number of cells disconnected during disconnection event 230, and any further disconnection events which then additionally takes place before the next reset point 270, can be output by the disconnect detector 150 as part of parameter output P depicted in figure 1 . When the next reset point 270 is reached this leads to a lowering of the baseline capacity 260, the number of disconnected cells since the last reset point 270 is no longer intrinsically recorded in a difference between the baseline capacity and the second estimate of capacity, so at least by this point a separate record and / or output of the number of disconnected cells since the previous reset point 270 is required. Therefore, in practice, a record of the total number of cells disconnect so far since initialisation may be retained by the disconnect detector on a continual basis, and indeed may be output as event properties P, without waiting for each reset point 270 to occur.

[0063] Logic which may be used within the disconnect estimator 150 to implement this detection of disconnection events 230 and count of disconnected cells is illustrated in figure 3. As already shown in figure 1 , the disconnect detector 150 receives as input the first and second estimates of electrical capacity EC1 and EC2 from the degradation-sensitive capacity estimator and the disconnect-sensitive capacity estimator, and maintains a value for the current baseline capacity B. The disconnect detector 150 then comprises disconnect flagging logic 310, the operation of which is illustrated in more detail in figure 5, and maintains a disconnect flag F which is true if a disconnection event has been detected within the current reset interval between resets of the baseline capacity B, or otherwise is false. As already indicated above, the disconnect flagging logic detects a disconnection event using a comparison of EC1 and EC2. Periodic reset of the baseline capacity estimate B is implemented by reset logic 320, with each reset being triggered for example by a lowering of EC1 by a particular fraction of the nominal 100% capacity or in some other way. When a periodic reset occurs, the disconnect flag is also reset to false.

[0064] The disconnect detector 150 also comprises disconnect counting logic 330 which, once the disconnect flag 315 has been set to true in a particular reset interval, determines a number of cells which have disconnected within that reset interval using at least EC2, and as illustrated in figure 3, also using the baseline capacity estimate B. The disconnect counting logic 330 then maintains at least a first disconnect count C which represents the number of cells determined to have disconnected during the current reset interval, and preferably also a second disconnect count C’ which represents the number of cells determined to have disconnected since initialisation of the monitor 100 when the nominal capacity of the string was set to 100%. The first disconnect count C is reset to zero at each reset point, whereas the second disconnect count C’ is maintained through each reset point.

[0065] The disconnect detector 150 outputs the second disconnect count, and optionally also the first disconnect count and / or the disconnect flag F, depending on what is required for use by other parts of the battery management system such as battery controller 160. These parameters may be output continuously as they are calculated and updated, or may be output periodically for example at every reset point. Figure 4 illustrates in flow chart form how the monitor 100 may operate to detect and flag disconnection events, calculate and maintain a record of the number of cells disconnected so far, and periodically reset the baseline capacity B. At step 400 the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator are initialised, such that the starting values for outputs EC1 and EC2 are both at the nominal 100% capacity. The monitor 100 then repeats execution of a loop in which inputs of measured parameters of the string of cells are received, capacity estimates are made, event detection and disconnection counts are carried out, and a baseline capacity reset may be implemented.

[0066] Within the loop at a first step 410 the physical parameters / Ware read from the string measurement input 120 shown in figure 1 . In some embodiments, the repeat of the loop may be triggered by availability of new, updated values of parameters M, but in other embodiments, the repeat may be fixed at certain suitable time intervals, or carried out in accordance with computational resources available to the monitor within the wide battery management system, or in other ways. At a second step 420 in the loop the degradationsensitive capacity estimator 130 and disconnect-sensitive capacity estimator 140 are then executed in order to determine updated values of estimates EC1 and EC2.

[0067] At a third step 430 in the loop the disconnect detector 150 executes the disconnect flagging logic which functions to maintain the disconnect flag F between consecutive resets of the baseline capacity estimate which indicates true if a disconnection event has been detected within that interval as discussed above in connection with figure 3. At a fourth step 440 in the loop the disconnect detector 150 executes the disconnect count logic which functions to maintain the first and second disconnect counts C, C’ as discussed above. At step 450 at least the second disconnect count C’, and optionally also the first disconnect count and / or the disconnect flag Fare output by the disconnect detector. Finally, at step 460 of the loop the baseline capacity is reset if the conditions to do so are met, also as discussed above.

[0068] Figure 5 is another flow chart, this time illustrating in more detail how the disconnect flagging logic 310 may be implemented within a particular reset interval. In this flow chart, the disconnect flag is indicated more simply as “flag”. At a first execution of this disconnect flagging logic after a reset point, steps 510 set the disconnect flag Fto false, and also set a count to zero and a previous value of the flag to false also. The count is used to implement de-bounce interval as already discussed above. A second value for the flag, indicated as “previous flag” is also set to false. For the first and all subsequent executions of the logic within a reset interval, values for EC1 and EC2 are then received, and compared for example by subtraction to yield a difference value A, at steps 520. If this difference A exceeds an event detection threshold T1 then count is incremented, otherwise it is remains at or is reset to zero. In this way, the value of count continues to increase while the difference value A continues to indicate a likely disconnection event.

[0069] However, is disconnection event is only flagged as true if A exceeds the threshold T1 a sufficient number of times consecutively. This is implemented using steps 540 which set flag to true only when count exceeds a de-bounce threshold T2. Once the de-bounce threshold has been exceeded, steps 550 using the values of flag and previous_flag

[0070] Steps 540 ensure that once the value of flag has been set to true, it remains as true until the next reset point. This ensures that, even if the first estimate of electrical capacity EC1 continues to fall due to degradation, and the difference between EC1 and EC2 falls beneath the event detection threshold T1, the flag remains true indicating that a disconnection event has occurred within the reset period.

[0071] The degradation-sensitive and disconnect-sensitive capacity estimators 130, 140 may be implemented in various ways in order to provide the required first and second estimates of electrical capacity of the string. However, they may in particular each be implemented using one or more state models, with each state model defining a respective concurrent estimate of electrical capacity of the string, each state model being arranged to advance to a new state responsive to received measurements of the one or more physical parameters of the string of cells. By implementing the capacity estimators in this way, the computational work required to advance the capacity estimators when new measurements of the physical parameters are received can be reduced to an acceptable level for implementation on an embedded battery management system.

[0072] In particular, either or both of the capacity estimators 130, 140 may comprise one or more Kalman filters. The inventors have found that extended Kalman filters may in particular be advantageous for these purposes.

[0073] Either or both of the capacity estimators may each be implemented using an interacting multiple model (IMM) structure 600 such as that illustrated in figure 6. As shown in figure 6, each IMM structure 600 comprises a plurality of state models 610, for example each implemented as an extended Kalman filter or other type of Kalman filter. Each state model 610 is arranged to track a current state of the string of cells responsive to receiving the measurements of the one or more physical parameters of the string. Each state model 610 represents the string of cells as having a different electrical capacity, so that a selection function 630 of the IMM structure 600 which determines which state model 610 currently best fits the concurrent physical parameter measurements can then output a value for the estimate of electrical capacity EC1 or EC2 which corresponds to the electrical capacity which that model represents. Typically therefore, at anyone time, there will be a best fit IMM model which best fits the current measurements and which therefore may be used to define the estimated electrical capacity currently output by the capacity estimator running the IMM.

[0074] However, as well as knowing the current best fit model, the selection function 630 may also maintain weights, for example, as a probability mass function PMF, of all of the state models currently being maintained by the IMM, which indicate a probability of each model being the current best fit to the concurrent physical parameter measurements. The weights or PMF of models used in the disconnect-sensitive capacity estimator 130 can then be used by disconnect counting logic 330 as an additional piece of information, in addition to the second estimate of electrical capacity, when determining a number of cells which have disconnected in one or more disconnection events.

[0075] The state models 610 of each IMM may be controlled and defined by IMM logic 620. This IMM logic operates to maintain state models which are most relevant to the current estimates of electrical capacity, and to add new state models and retire old state models to ensure that the state models of the IMM continue to represent the full range of electrical capacities within which the current electrical capacity is expected to fall, even in the event of multiple cell disconnects. In figure 6 the state models 610 are shown as defining a range of electrical capacities X-2%, X-1%, X, X+1% and X+2%, where X is the capacity of the current model, so that . If the model with capacity X-1% is currently the state model which best fits the concurrent physical parameter measurements, then the selection function 630 may output the current electrical capacity as X-1%. If the state of the string of cells changes such that the model with capacity X-2% is more consistently the state model which best fits the concurrent physical parameter measurements, then the selection function 630 may change the current model to the X-2% model, and output the current electrical capacity as X-2%. If this situation persists then the IMM logic 620 may then retire the X+2% and X+1% models, and add new state models with electrical capacities X-1% and X-2%. Which model is found by the selection function 630 to be the current model may be implemented in conjunction with the IMM logic using a state transition probability matrix, discussed in more detail below. However, in other arrangements the current electrical capacity may instead be determined from a weighted sum of the individual discrete capacities represented by the models of the IMM.

[0076] The number of state models which are maintained by the IMM at any one time, and the spacing between those state models 610 in terms of electrical capacity, may be chosen to provide sufficient resolution and scope for rapid changes in electrical capacity, while also keeping the requirement for computational resources to an acceptable level. Although five separate state models are shown in figure 6, each of the degradation-sensitive capacity estimator and the disconnect-sensitive capacity estimator may have more or less state models operating within the IMM at any one time.

[0077] For the disconnect-sensitive capacity estimator, each state model may represent the string of cells with a different number of cells having disconnected, in other words the spacing between each adjacent state models of the IMM may substantially correspond to the loss of capacity corresponding to disconnection of a single cell from the string.

[0078] Further details of how interacting multiple models may be used to implement estimation of state of batteries, or in the present case of strings of cells, may be found in the PhD thesis of Adam Smiley, University of Colorado Springs, 2019, titled “An improved approach to state-of-age estimation for lithium-ion battery cells using interacting multiple model Kalman filters.” Implementations of interacting multiple models may more generally be found discussed in E. Mazer et al., IEE Transactions on Aerospace and Electronic Systems, vol. 34, no. 1 , 1998. Kalman filters used in implementing these IMM models in the described embodiments can make use of various underlying physical and chemical models of the battery cells, for example see those described in A. Fotouhi, Renewable and Sustainable Energy Reviews 56 (2016) 1008-1021 , and Manh-Kien Tran et al., Batteries 2021 , 7, 51 . In practical terms, the extended Kalman filters mentioned above can be readily implemented using the EKF tracking filter objects available in MATLAB (RTM) and in other commercially available toolkits.

[0079] In a particular demonstrations of the invention conducted by the inventions, the two capacity estimators were implemented using two corresponding sets of interacting multiple models. Both contained several extended Kalman filters (EKFs) tracking state of charge and overpotential of a string of cells, with each interacting multiple model using multiple state models with different electrical capacities.

[0080] The degradation-sensitive capacity estimator was implemented as a variable structure interacting multiple model using five state models at any one time, with a range of modelled electrical capacities of the string distributed across the models with a resolution of ±1% about a current model with a capacity of X, for example this could at some point in time lead to the following state models being current:

[0081] Model 1 : X+2%

[0082] Model 2: X+1%

[0083] Model 3: X+0% Model 4: X-1%

[0084] Model 5: X-2%

[0085] The variable structure was prevented from running state models above X+4%, focusing instead on the tendency for electrical capacity to decrease over time.

[0086] The weights and electrical capacities of the IMM models were initialised around a current model having the estimated electrical capacity at the time of initialisation. The variable structure of the IMM automatically retires models no longer needed and adds new models with electrical capacities that become needed as the tracked capacity moves outside of the current capacity range represented by the models. For example, after running for some time, and with the cells in the parallel string losing capacity over time, the models will tend to represent the following distribution around the currently mode with estimated electrical capacity X, since electrical capacity generally expected to fall over time:

[0087] Model 1 : X + 0%

[0088] Model 2: X - 1%

[0089] Model 3: X - 2%

[0090] Model 4: X - 3%

[0091] Model 5: X - 4%

[0092] This scheme allows for tracking of electrical capacity relative to an initialised nominal 100% capacity of between +104% down to 0% without using more than 5 state models per parallel string.

[0093] In these particular demonstrations, the disconnect-sensitive capacity estimator used a fixed structure IMM with 6 state models, each representing a string with from 0 to 5 of the string cells disconnected relative to a nominal current electrical capacity. This range of 0-5 disconnects was chosen empirically as the most common disconnect occurrence range. Severity of more than 6 disconnects was deemed comparable to 5. Implementation can of course vary based on application or observed number of disconnects, or a variable structure IMM could be used.

[0094] In these particular demonstrations, aspects of the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator that were set up differently, allowing each estimator to be sensitive to its designated behaviour (typical usage capacity fade vs disconnection events) included the rate at which interaction between state models within an IMM happens, and details of a state transition probability matrix for transition of the current model between the existing multiple models. Regarding the rate at which interaction between state models within an IMM happens, model interaction within a capacity estimator can be allowed to happen only under certain conditions, in order for prediction error to accumulate, allowing the correction to have a meaningful impact on interaction as well as the IMM states and outputs. If the disconnect-sensitive capacity estimator IMM is allowed to correct more often than the degradation-sensitive capacity estimator VSIMM, the estimate of electrical capacity output by the degradation-sensitive capacity estimator also varies faster. This ensures that instantaneous or rapid electrical capacity losses lead to the disconnect-sensitive capacity estimator output to drop faster than that of the degradation-sensitive capacity estimator.

[0095] Regarding the state transition probability matrix (represented as p( i , j ) in Mazer et al., 1998) this determines both the transitions for the current model between existing state models as well as the magnitude of state correction that must happen for its model likelihood to change. The transition can be tuned in such a way that reflects either degradation or disconnect behaviour. For example:

[0096] For the degradation-sensitive capacity estimator :

[0097] - Probability of the current model staying the same is higher, as electrical capacity degradation in this scenario happens slowly. This also results in a lower sensitivity and slower response to sudden changes in electrical capacity.

[0098] - Transition of the current model is only allowed to neighbouring models. This means that with a difference in electrical capacity between adjacent models of about 1% of the nominal, original capacity of the string, if the current model is currently X-1%, the current model can only jump to X or X-2%. This is because capacity degradation happens gradually, and it is unlikely that due to typical use it would jump by a discrete and high percentage.

[0099] For the disconnect-sensitive capacity estimator:

[0100] - Probability of the current model staying the same lower, as disconnect capacity loss is essentially instantaneous and discrete. This also allows for higher sensitivity to sudden losses in capacity.

[0101] - Change of the current model to be any of the other existing models is allowed, since it is possible for multiple cells to become disconnected at one given time, causing an instant loss of capacity rather than gradual.

[0102] There are some conditions on IMM operation, however, which are preferably the same both estimators to ensure output stability and reliability. Any condition that enables / disables the IMM from running completely (for example in certain parts of the state of charge curve for the modelled string in which either string or model behaviour is unreliable, inconsistent or unstable) should be applied to both IMMs equally. This prevents the capacity estimates one estimator from having large variations relevant to the other without any real large underlying capacity changes, which can lead to false disconnect detections. An example of this would be preventing IMMs from run within a certain range of values of state of charge, in which the model error is known to be large.

[0103] Similarly any conditions, unrelated to the difference in behaviour between cell disconnects and typical degradation capacity loss, which prevent an IMM from interacting (only prediction is allowed), should also be equalised between the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator. This prevents any behaviour unrelated to disconnect detection from impacting the relative IMM interaction rate (mentioned above). For example, these conditions can be:

[0104] - allowing a decreased number of capacity estimator executions within a particular time period due to hardware computational limitations;

[0105] - only allowing capacity estimator execution when estimated cell overvoltage is too large to prevent dynamic and less predictable voltage behaviour from affecting capacity estimation and model interaction accuracy; and / or

[0106] - only allowing capacity estimation to run under conditions which are known to be effectively represented by the cell model used.

[0107] The last of these conditions essentially avoids running capacity estimation when under conditions that known to present dynamics not well-represented by the cell model since the IMM capacity and disconnect tend to run more open loop than regular kalman filters.

[0108] As mentioned above, the described monitor 100 for detecting disconnection events will typically be implemented a part of a wide battery management system, for example on board a vehicle or within some other installation scenario. The functionality described above, from receiving measurements of physical parameters of a string of cells through to determining disconnection events, detecting numbers of cells which have disconnected, and providing such information to the wider battery management system, may be implemented in hardware, software, or a combination of the two, but typically will be implemented using software arranged to execute on a suitable computer system, and typically on a computer system used by a battery management system. Such a computer system will typically comprise one or more microprocessors, memory for storing the software and related data, and suitable input / output mechanisms which may include one or more displays, network connections, electronics for receiving the measured physical parameters of the string and other data relating the battery module, and electronics for sending control signals relating to management of the battery module. Such computer software may be provided on one or more computer readable media, be transmitted as a signal over a network connection, and be stored within the battery management system within suitable non-volatile memory. Although particular embodiments of the invention have been described, the skilled person will appreciate that various alternatives and variations may be implemented without departing from the scope of the claims.

Claims

CLAIMS:1 . Apparatus for monitoring a string of battery cells for disconnection events each of which comprises electrical disconnection of one or more of the cells from the string, the cells of the string being connected electrically in parallel with each other such that the string has an electrical capacity, the apparatus comprising: a string measurement input arranged to receive measurements of one or more physical parameters of the string of cells; a degradation-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a concurrent first estimate of the electrical capacity; a disconnect-sensitive capacity estimator arranged to receive the measurements and to calculate from the measurements a concurrent second estimate of the electrical capacity; and a disconnect detector arranged to detect disconnection events using a comparison of the first and second estimates of electrical capacity.

2. The apparatus of claim 1 wherein the one or more physical measurements comprise a voltage across the string and a current through the string, and optionally also a temperature of the string.

3. The apparatus of claim 1 or 2 wherein the disconnect-sensitive capacity estimator is tuned for sensitivity to instantaneous changes in the electrical capacity and the degradation-sensitive capacity estimator is tuned for sensitivity to long-term degradation in the electrical capacity4. The apparatus of any preceding claim wherein the degradation-sensitive capacity estimator and the disconnect-sensitive capacity estimator and are arranged such that, in response to a change in the electrical capacity, such as a disconnection event, the second estimate of electrical capacity has a higher rate of change, or a rate of change which is at least five times higher, than a rate of change of the first estimate of electrical capacity.

5. The apparatus of any preceding claim wherein the disconnect detector is arranged to detect a disconnection event when a difference between the first estimate of electrical capacity and the second estimate of electrical capacity exceeds an event detectionthreshold, and optionally when the difference exceeds the event detection threshold for at least a predetermined de-bounce interval.

6. The apparatus of any preceding claim wherein the disconnect detector is arranged to determine a number of cells which have disconnected in one or more disconnection events using at least the second estimate of electrical capacity.

7. The apparatus of claim 6 wherein the apparatus is arranged to maintain a baseline capacity estimate which is periodically reset to correspond to the concurrent first estimate of electrical capacity, and the disconnect detector is arranged to determine a number of cells which have disconnected in one or more disconnection events since the previous reset of the baseline capacity estimate using a difference between the baseline capacity estimate and the second estimate of electrical capacity prior to the next reset of the baseline capacity estimate.

8. The apparatus of any preceding claim wherein each of the degradation sensitive capacity estimator and the disconnect-sensitive capacity estimator comprises one or more state models of the string of cells, each state model representing a different respective model estimate of electrical capacity of the string, each state model being arranged to advance to a new state responsive to received measurements of the one or more physical parameters of the string of cells.

9. The apparatus of claim 8 where in each state model is implemented as a Kalman filter, or more particularly as an Extended Kalman filter.

10. The apparatus of claim 8 or 9 wherein each of the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator comprises an interacting multiple model structure comprising a plurality of state models arranged to track a current state of the string of cells responsive to receiving the measurements of the one or more physical parameters,11 . The apparatus of claim 10 wherein for each of the degradation-sensitive capacity estimator and disconnect-sensitive capacity estimator, each state model represents the string of cells with a different electrical capacity, and wherein for each capacity estimator,the estimate of electrical capacity of the string of cells corresponds to the state model which best fits the concurrent physical parameter measurements.

12. The apparatus of claim 11 wherein, for the disconnect-sensitive capacity estimator, each state model represents the string of cells with a different number of cells having disconnected.

13. The apparatus of claim 11 or 12 when dependent on claim 6, wherein the disconnect-sensitive capacity estimator maintains weights indicating the probability of each state model being the best fit to the concurrent physical parameter measurements, and the disconnect detector is arranged to also use the weights to determine the number of cells which have disconnected in one or more disconnection events.

14. The apparatus of any preceding claim further comprising a battery comprising the string of battery cells, and sensors arranged to make the one or more physical measurements and to pass the one or more physical measurements to the string measurement input.

15. A vehicle comprising the apparatus of claim 14 wherein the battery is arranged to provide motive power for the vehicle.

16. A method of monitoring a string of battery cells for disconnection events each of which comprises electrical disconnection of one or more of the cells from the string, the cells of the string being connected electrically in parallel with each other such that the string has an electrical capacity, the method comprising: receiving measurements of one or more physical parameters of the string of cells; calculating from the measurements a concurrent first estimate of the electrical capacity using a degradation-sensitive capacity estimator; calculating from the measurements a concurrent second estimate of the electrical capacity using a disconnect-sensitive capacity estimator; and detecting a disconnection event using a comparison of the first and second estimates of electrical capacity.

17. The method of claim 16 wherein, in response to a change in the electrical capacity such as a disconnection event, the second estimate of electrical capacity has a higher rateof change, or a rate of change which is at least five times higher, than a rate of change of the first estimate of electrical capacity.

18. The method of claim 16 or 17 wherein a disconnection event is detected when a difference between the first estimate of electrical capacity and the second estimate of electrical capacity exceeds an event detection threshold, and optionally when the difference exceeds the event detection threshold for at least a predetermined de-bounce interval.

19. The method of any of claims 16 to 18 further comprising determining a number of cells which have disconnected in the disconnection event, optionally using at least the second estimate of electrical capacity.

20. The method of any of claims 16 to 19 wherein each of the degradation sensitive capacity estimator and the disconnect-sensitive capacity estimator comprises one or more state models of the string of cells, each state model defining a respective model estimate of electrical capacity of the string, each state model being arranged to advance to a new state responsive to received measurements of the one or more physical parameters of the string of cells.21 . The method of claim 20 wherein, for the disconnect-sensitive capacity estimator, each state model represents the string of cells with a different number of cells having disconnected.

22. One or more computer readable media carrying computer program code arranged to carry out the method of any of claims 16 to 21 when executed on a suitable computer system.

Images