Method and device for evaluating a battery condition in the event of an internal or external disturbance to the device operation

Abstract

Computer-implemented method for predictively detecting a fault in a device battery (41) of a technical device (4) after the occurrence of an operating fault in the technical device (4), comprising the following steps: - Providing (S1) temporal operating variable profiles that characterize the operation of the technical device (4) and the device battery (41); - Determination (S2) of an operational disturbance by evaluating the temporal operating variables; - Upon detection of an operational fault, determining at least one fault event feature (S4) indicating a state of the technical device (4) and at least one battery state variable (S5) indicating a state of the device battery (41), and evaluating (S7) the at least one fault event feature and the at least one battery state variable using a data-based fault event model in order to obtain an output vector with output variables whose values ​​are each assigned to a fault type of a fault; - Signaling (S8) an instruction to a user of the technical device (4) depending on the output vector.

Description

Technical area

[0001] The invention relates to the use of device batteries for operating technical devices, in particular the evaluation of the battery status after the occurrence of a malfunction in the operation of the technical device, such as after an accident or any other situation deviating from the regular operation of the technical device. Technical background

[0002] While the behavior of a device battery is largely predictable during normal operation of a technical device, such as an electric vehicle, the behavior of a device battery is generally well predictable. However, in the case of unforeseeable events, such as an external malfunction, such as an accident, or an internal malfunction of a technical device, which may occur, for example, when a diagnostic error is detected during a workshop visit, the impact on the device battery due to the malfunction cannot be easily predicted. For example, in the case of a diagnostic error that diagnoses a device fault not directly related to the battery, it may be difficult to assess the extent to which the device battery might be affected and thus pose an operational risk.

[0003] It is therefore desirable to carry out a quantified assessment of the battery condition when a malfunction occurs and, if necessary, to propose concrete recommendations for action for the continued operation of the device battery.

[0004] The document DE 10 2020 212 277 A1 discloses a computer-implemented method for providing a remaining useful life based on a diagnosis of components of an electric drive system in a vehicle.

[0005] Document CN 1 12 396 156 A discloses predictive maintenance of vehicle batteries, whereby the condition of the battery is analyzed and maintenance requirements are predicted using sensor data and artificial neural networks. Disclosure of the invention

[0006] This object is achieved by the method for evaluating a battery state of a device battery for operating a technical device when an operating fault of claim 1 occurs by a corresponding device according to the independent claim.

[0007] Further embodiments are specified in the dependent claims.

[0008] According to a first aspect, a method is provided for predictively detecting a fault in a device battery of a technical device after an operating fault has occurred in the technical device, comprising the following steps: - Providing temporal operating variables that characterise the operation of the technical device and the device battery; - Determination of an operational disturbance by evaluating the temporal course of operational variables; - Upon detection of an operational fault, determining at least one fault event characteristic that indicates a state of the technical device and at least one battery state variable that indicates a state of the device battery, and evaluating the at least one fault event characteristic and the at least one battery state variable using a data-based fault event model in order to obtain an output vector with output variables whose values ​​are each assigned to a fault type of a fault; - Signaling an instruction to a user of the technical device depending on the output vector.

[0009] The above method is based on a data-based fault event model, which can be designed as a classification model. The fault event model is provided or trained in such a way that battery state variables and malfunction characteristics that can characterize a malfunction can be assigned to a probability of a fault or a probability of a specific fault type of the device battery. If several detectable fault types are present, the fault can be determined, for example, using a softmax function followed by a threshold comparison and can then be used to derive recommended actions, such as an indication that the battery is unaffected by the malfunction or an indication that a visit to the workshop should be scheduled to conduct a more intensive inspection of the device battery.

[0010] The evaluation of the fault event model is based on fault event characteristics and battery state variables that characterize a situation during the malfunction and / or immediately after the malfunction occurs. The fault event characteristics can, for example, represent variables that can describe the nature of the malfunction. Fault event characteristics in the case of an external malfunction, such as a vehicle accident as an example of a device, can be the maximum smoothed longitudinal acceleration at impact, the maximum smoothed lateral acceleration at impact, diagnostic information from the airbag control unit, a vehicle impact location, or a maximum measured impulse. Furthermore, in the case of internal malfunctions, the fault event characteristics can include diagnostic information.

[0011] Battery state variables can include operating variables of the device battery at one or more specific points in time during and after the fault event, such as temperatures, voltages, currents, states of charge, balancing states and / or their respective curves at cell, module or overall battery level and / or aggregated variables thereof. Furthermore, the battery state variables can be determined by evaluating operating variable curves, such as the battery voltage(s) at cell, module or pack level, the battery current, the battery temperature and the state of charge, during a period after the occurrence of the operational fault, in particular as internal battery states determined using an electrochemical battery model, such asSEI layer thickness, change in cyclable lithium due to anode / cathode side reactions, rapid uptake of electrolyte solvents, slow uptake of electrolyte solvents, lithium deposition, loss of anode active material and loss of cathode active material, and the like.

[0012] Furthermore, battery condition variables can include operating characteristics that indicate stress factors for battery operation. The operating characteristics can, for example, include characteristics related to the evaluation period and / or accumulated characteristics and / or statistical variables determined over the entire service life to date. In particular, the operating characteristics can, for example, include: histogram characteristics, such as temperature versus state of charge, charging current versus temperature and discharging current versus temperature, in particular multi-dimensional histogram data relating to the battery temperature distribution versus state of charge, the charging current distribution versus temperature and / or the discharging current distribution versus temperature, the current throughput in ampere-hours, the accumulated total charge (Ah), an average capacity increase during a charging process (in particular for charging processes where the charge increase exceeds a threshold proportion [e.g.20% ΔSOC] of the total battery capacity), the charging capacity and an extreme value (e.g. maximum) of the differential capacity during a measured charging process with a sufficiently large change in the state of charge (smoothed curve of dQ / dU: change in charge divided by change in battery voltage) or the accumulated mileage.

[0013] Furthermore, the battery state variables can include the anomaly status of battery cells and / or modules from multiple battery cells and / or the entire battery. The anomaly status can be determined by a known rule-based or data-based anomaly detection model and indicates whether the operating behavior of a battery cell and / or a battery module and / or the entire battery corresponds to or deviates from the conventional operating behavior.

[0014] Furthermore, an aging state can also be specified as a battery state variable, which indicates a measure of the degradation of the device battery.

[0015] If the device battery is still in discharge or charge mode, the voltage residual can be determined as the difference between the modeled voltage from a fractional battery model or electrochemical battery model and a measured voltage value. The voltage residual can be determined as a battery state variable, particularly immediately after the occurrence of the operational fault.

[0016] The battery condition variables should preferably be determined within a specified period of time after the occurrence of the malfunction, which should in particular not be more than three days after the occurrence of the malfunction. The malfunction event characteristics and the battery condition variables of the device battery can then be evaluated using a malfunction event model, which can be developed based on data.

[0017] The disturbance event model can be designed as a classification model, preferably as a multi-label logistic regression or as a neural network.

[0018] If necessary, redundant information can be extracted in advance from the fault event characteristics and the battery state variables using principal component analysis to reduce the dimensionality of the input variable vector resulting from the fault event characteristics and the operating state variables. Preferably, the fault event model is designed as a classification model so that different fault types can be classified, for example, the fault type of thermal runaway and the fault type of sudden capacity loss.

[0019] Depending on the evaluation results of the fault event model, recommended actions can be signaled. For example, depending on the determined probability of a certain fault type occurring, the malfunction model can signal whether continued use of the device battery is possible or whether a visit to the workshop or maintenance is recommended. Particularly in an area where the fault event model cannot reliably determine the condition of the device battery, such as a classification result between 20 and 80%, a manual retest in a workshop can be suggested.

[0020] The method has the advantage that damage to the device battery can be detected or predicted immediately after the fault occurs, before this damage can be detected later, e.g. during subsequent diagnostics or maintenance.

[0021] The malfunction model is trained using labels that indicate an assessment of the consequential damage to the device battery after the occurrence of an external or internal malfunction for a specified observation period, for example, between three and twelve months, preferably after six months. In particular, based on an evaluation of the battery condition variables, the device battery can be examined for anomalies using rules or data. Thus, a specific device battery can be assigned a label that indicates either a proper condition of the device battery or a specific type of fault that was not yet detectable at the time the malfunction occurred.

[0022] The training of the disturbance event model is carried out in a conventional manner, in particular using backpropagation methods and the like.

[0023] Due to the large number of devices to be monitored in this way, the model training and evaluation can take place in a remote central unit, whereby a corresponding model evaluation, i.e. a specific type of error or the resulting recommended action, which is communicated back to the device in question in order to inform the user of the device in question of possible measures, is carried out. Brief description of the drawings

[0024] Embodiments are explained in more detail below with reference to the attached drawings. They show: Fig. 1 a schematic representation of a system for providing driver- and vehicle-specific operating variables for determining operating fault characteristics of a vehicle and battery state variables of a vehicle battery in a central unit; Fig. 2 is a flowchart illustrating a method for diagnosing the condition of a vehicle battery after a malfunction has occurred; and Fig. 3 is a flowchart illustrating a method for training a fault detection model; and Fig. 4 a diagram illustrating the classification between batteries assessed as faulty and those assessed as faulty. Description of embodiments

[0025] The method according to the invention is described below using vehicle batteries as device batteries in a large number of motor vehicles as similar devices. A data-based fault event model for the respective vehicle can be trained and implemented in a central unit connected to the motor vehicles. The fault event model can, as described below, be continuously updated or retrained in a vehicle-external central unit based on operating variables of the vehicle batteries from the vehicle fleet and executed in the vehicle or in the central unit. The fault event model is operated in the central unit and used to determine internal battery states.

[0026] The above example represents a multitude of stationary or mobile devices with off-grid energy supply, such as vehicles (electric vehicles, pedelecs, etc.), systems, machine tools, household appliances, IOT devices and the like, which are connected to a device-external central unit (cloud) via a corresponding communication connection (e.g. LAN, Internet).

[0027] Fig. Figure 1 shows a system 1 for collecting fleet data in a central unit 2 for the creation, operation and evaluation of a fault event model. The fault event model is designed as a hybrid

[0028] Differential equation system is designed and is used to determine internal battery states of a vehicle battery in a motor vehicle. Fig. 1 shows a vehicle fleet 3 with several motor vehicles 4.

[0029] One of the motor vehicles 4 is in Fig. 1. The motor vehicles 4 each have a vehicle battery 41, an electric drive motor 42, and a control unit 43. The control unit 43 is connected to a communications module 44, which is suitable for transmitting data between the respective motor vehicle 4 and a central unit 2 (a so-called cloud).

[0030] The motor vehicles 4 send the operating variables F to the central unit 2, which at least indicate variables that influence or depend on the internal battery states of the vehicle battery 41. In the case of a vehicle battery 41, the operating variables F can include time series of a battery current, a battery voltage, a battery temperature, and a state of charge (SOC), both at the pack, module, and / or cell level. The operating variables can further include variables of vehicle operation, such as speeds, accelerations in the longitudinal and transverse directions, diagnostic information from the airbag control unit, a vehicle impact location, and the like. The operating variables F can be recorded in a fast time frame of 1 Hz to 100 Hz and regularly transmitted to the central unit 2 in uncompressed and / or compressed form.

[0031] In particular, the time series of the operating variables can be transmitted to the central unit 2 in blocks at intervals of several hours to several days using compression algorithms in order to minimize the data traffic to the central unit 2.

[0032] For the time series of operating variables, compression algorithms can be used to minimize data traffic to central unit 2. Furthermore, event-based transmission can be used, so that the data transfer is triggered and occurs when, for example, a stable or known Wi-Fi network connection has been identified.

[0033] The central unit 2 has a data processing unit 21 in which the method described below can be carried out, and a database 22 for storing data points, model parameters, states and the like.

[0034] A data-based fault event model implemented as a classification model can be implemented in the central unit 2 or in the control unit 43. The fault event model can be used to provide an indication of a type of potential impairment or failure of the vehicle battery 41 caused by the fault event, based on an input vector that may include battery state variables and operational fault characteristics.

[0035] The battery state variables and operational fault characteristics indicate a state of the vehicle 4 at the time the fault occurs or is detected or at a time within a specified period after the fault occurs or is detected.

[0036] Fig. 2 illustrates, using a flowchart, a method for assessing the condition of a vehicle battery after an operational fault occurs, for example an accident.

[0037] In step S1, operating parameter profiles are continuously transmitted from the vehicles in the fleet to the central unit, advantageously using a digital twin to monitor the battery. The operating parameter profiles include both the operating parameters of the vehicle batteries and the vehicle's operating parameters, in particular, longitudinal and transverse speeds and accelerations, as well as diagnostic data from one or more vehicle systems.

[0038] In step S2, a check is carried out to determine whether a malfunction has occurred. This step advantageously takes place in the digital twin in the central unit, where computing power and historical data are available for evaluation. An external malfunction can be detected, for example, if the vehicle has been in an accident. The accident can be detected based on an evaluation of the lateral and longitudinal acceleration and / or the diagnostic data, for example a diagnostic signal from the airbag control unit. Furthermore, an internal malfunction can be signaled, for example by the battery management system of the respective vehicle battery, for example if an unfavorable state of the vehicle battery 41 is permitted due to a control error, such as a voltage within the vehicle battery that is too high or too low, or battery currents or temperatures that are too high or too low.Furthermore, a malfunction in the battery management system can be signaled if the operation of the vehicle battery lies outside the limit values ​​set by the manufacturer for one or more of the operating parameters.

[0039] The detected malfunction can be signaled by the relevant vehicle 4 to the central unit 2 in step S3. If no malfunction is detected (alternative: No), the process continues with step S1.

[0040] Alternatively, the check for whether an operational fault has occurred can also be carried out in the central unit 2 after the time series of the operational variables have been transmitted.

[0041] In step S4, disturbance event characteristics are first extracted from the operating variables.

[0042] Fault event features may be features that indicate an accident involving the vehicle 4, such as an accident detection based on an evaluation of the lateral and longitudinal acceleration and / or diagnostic data, for example, a diagnostic signal from the airbag control unit. These fault event features may include a maximum longitudinal acceleration upon impact, a maximum lateral acceleration upon impact, diagnostic information from an airbag control unit, an impact location in the vehicle (e.g., front / side), a maximum measured impulse, and / or the like.

[0043] Furthermore, fault event characteristics can be variables of the battery management system, such as balancing states, model estimates and diagnostic values ​​at the pack, module and cell level.

[0044] Fault event characteristics in the event of an internal fault (and possibly an external fault) can include OBD-relevant diagnostic information and can be signaled, for example, by the battery management system of the vehicle battery 41, for example, if a control error allows an unfavorable condition of the vehicle battery 41, such as an excessively high or too low voltage within the vehicle battery, or excessively high or too low battery currents or temperatures. Furthermore, a malfunction in the battery management system can be signaled if the operation of the vehicle battery lies outside the limit values ​​set by the manufacturer for one or more of the operating variables.

[0045] Furthermore, in step S5, battery state variables can be derived from the courses of the operating variables relating to the vehicle battery 41, such as the time series of a battery current, a battery voltage, a battery temperature and a state of charge, in particular internal battery states which result from a parameterization of an electrochemical battery model.

[0046] To evaluate the temporal battery-related operating variables, especially in physical or electrochemical battery models that may be based on differential equations, for example, an internal battery state can be determined using a time integration method. An electrochemical battery model based on a differential equation system with a plurality of nonlinear differential equations can be used to evaluate the operating variables. The operating variable data enable the modeling of a current internal battery state using a time integration method. Such electrochemical battery models are known, for example, from the publications US 2016 / 023,566, US 2016 / 023,567, and US 2020 / 150,185.

[0047] Furthermore, operating characteristics can be determined from the temporal battery-related operating characteristic curves as additional battery condition variables that indicate stress factors in battery operation. The operating characteristics can include, for example, characteristics related to the evaluation period and / or accumulated characteristics and / or statistical variables determined over the entire service life to date.In particular, the operating characteristics may include, for example: histogram characteristics, such as temperature versus state of charge, charging current versus temperature and discharging current versus temperature, in particular multi-dimensional histogram data relating to the battery temperature distribution versus state of charge, the charging current distribution versus temperature and / or the discharging current distribution versus temperature, the current throughput in ampere-hours, the accumulated total charge (Ah), an average capacity increase during a charging process (in particular for charging processes where the charge increase is above a threshold proportion [e.g. 20% ΔSOC] of the total battery capacity), the charging capacity as well as an extreme value (e.g. maximum) of the differential capacity during a measured charging process with a sufficiently large swing in the state of charge (smoothed curve of dQ / dU: charge change divided by change in battery voltage) or the accumulated mileage.

[0048] In a subsequent step S6, the fault event characteristics and the battery state variables are combined into an input vector and evaluated using a data-based fault event model. The fault event model can be designed as a classification model, in particular as a multi-label logistic regression model or as a neural network.

[0049] Fig. Figure 4 shows, using two variables M1 and M2 from the fault event characteristics and battery state variables, a boundary between batteries assessed as fault-free and faulty. Circles indicate batteries assessed as fault-free, while diamonds and squares represent batteries assessed as faulty.

[0050] The fault event characteristics and the battery condition variables are preferably determined and evaluated immediately after the occurrence of the operational fault or within a specified period of 1-3 days.

[0051] In step S7, the classification model can output one or more output variables, each associated with a fault type, depending on the input vector. The classification result is determined by a threshold comparison of the values ​​of the output variables for each fault type, which indicate a probability of the fault associated with the output variable occurring in the vehicle battery 41 due to the malfunction. The different fault types are distinguished in the figure by diamonds and squares.

[0052] For example, the fault event may output fault types of thermal runaway of the vehicle battery 41, accelerated aging of the vehicle battery 41.

[0053] Depending on the previously estimated probability for the respective fault type, an action instruction can be determined from a lookup table in step S8, which is then signaled to the driver. For this purpose, the determined fault type or the associated action instruction can be transmitted from the central unit 2 to the vehicle 4 if the method has been executed in the central unit 2, and signaled there to the user or driver. Action instructions can, for example, indicate a likely fault-free battery or indicate that the battery may be damaged and an imminent visit to a workshop is recommended.

[0054] In particular, the result of the fault event model can be provided, for example, via API to a workshop (in the event of an internal fault) or to the vehicle (after an accident), where it can be used directly to make informed and data-driven decisions. Rule-based recommendations are then derived, e.g., taking into account the workshop capacity, the fault type, and the fault probability. For example, recommendations such as "Continuing the journey is possible" or that a workshop visit is planned for a specific time and location.

[0055] Furthermore, in an area where the model is uncertain, e.g., with an error probability between 20 and 80%, it is possible to conduct a manual review in the workshop before an automated decision is made. These additionally determined labels are then written directly to the database and used to further train the failure event model.

[0056] Alternatively, the method can be executed entirely or partially in the vehicle. For example, the data-based fault event model can be implemented in vehicle 4 and automatically evaluated there after a malfunction is detected. The fault event model can then be updated regularly by transmitting model parameters from the central unit 2.

[0057] To train the disturbance event model, a procedure can be used, such as the flow chart of the Fig. 3. For this purpose, training data sets are first determined in step S11. As previously described, after the detection of a malfunction, the training data sets provide for the recording of the malfunction characteristics and the battery state variables at the time of the malfunction event or shortly thereafter.

[0058] In order to identify which type of fault may develop after the occurrence of the malfunction, an assessment of the battery condition of the vehicle battery is performed after a predefined waiting period of between three and twelve months, preferably after six months. This involves assessing the consequential damage, which can be performed, for example, using known anomaly detection models and the like. If no significant fault in the vehicle battery 41 is detected after the waiting period has expired, and if no anomaly in the vehicle battery 41 is detected during a battery condition check, appropriate labels can be assigned to the fault types to indicate that no fault has occurred. Alternatively, if a fault or anomaly occurs, a label corresponding to the fault type can be assigned to determine a corresponding training data set.A labeled expansion of the feature database is planned in order to transfer systematic field observations from test fleets and vehicles that have been in fleet operation for a longer period of time to knowledge of the events in younger fleet participants and to be able to carry out better anomaly assessments based on historical data.

[0059] The input vectors resulting from the operating event characteristics and the battery state variables and the associated labels for the different fault types then form a training data set.

[0060] If it is determined in step S12 that a sufficient number of training data sets is available, the data-based disturbance event model can be trained in step S13. The disturbance event model can be designed, as previously described, as a multi-label logistic regression model or as a neural network and trained in a conventional manner using the training data sets.

[0061] Subsequently, in step S14, if the evaluation is to be carried out in the vehicles, the model parameters of the fault event model can be transmitted to the vehicles 4.

[0062] It can be provided that training data sets are continuously transmitted from the vehicles 4 to the central unit 2 after a malfunction has occurred and the waiting period has elapsed. The malfunction event model can then be retrained regularly, e.g., every two months, in order to better predict the detection of a potential vehicle battery fault after a malfunction has occurred. The malfunction event model is operated in a live mode, or in real time or near real time, based on the recorded characteristics (malfunction event characteristics and battery state variables), so that a comprehensive assessment based on at least one explainable characteristic is possible immediately after a potential malfunction event, such as an accident.

Claims

[1] Computer-implemented method for predictive detection of a fault in a device battery (41) of a technical device (4) after the occurrence of an operating fault in the technical device (4), comprising the following steps: - Providing (S1) temporal operating variable profiles that characterize the operation of the technical device (4) and the device battery (41); - Determination (S2) of an operational disturbance by evaluating the temporal operating variables; - Upon detection of an operational fault, determining at least one fault event feature (S4) indicating a state of the technical device (4) and at least one battery state variable (S5) indicating a state of the device battery (41), and evaluating (S7) the at least one fault event feature and the at least one battery state variable using a data-based fault event model in order to obtain an output vector with output variables whose values ​​are each assigned to a fault type of a fault; - Signaling (S8) an instruction to a user of the technical device (4) depending on the output vector. [2] The method according to claim 1, wherein the data-based fault event model is or is trained as a classification model to provide, depending on the at least one fault event feature and the at least one battery state variable, an output vector whose element values ​​indicate a probability of the later occurrence of a fault associated with the corresponding element. [3] Method according to claim 2, wherein the action instruction is assigned and signaled depending on a threshold value comparison of the element values ​​of the respective elements of the output vector. [4] Method according to one of claims 1 to 3, wherein an operational malfunction is detected in the event of an accident or by evaluating diagnostic data. [5] Method according to one of the preceding claims, wherein the instructions for action comprise an indication that the device battery (41) is unaffected by the malfunction, or an indication that a visit to the workshop should be planned in order to carry out a check of the device battery (41). [6] Method according to one of the preceding claims, wherein the signaling (S8) of the action instruction to a user of the technical device (4) takes place in real time or near real time depending on the output vector, wherein data streaming is used to obtain a comprehensive assessment based on at least one explainable feature immediately after a possible fault event. [7] Method according to one of the preceding claims, wherein the instruction to a user suggesting a workshop visit signals a time of the workshop visit depending on the output vector and a workshop capacity utilization. [8] Method according to one of the preceding claims, wherein a malfunction of a vehicle is detected by threshold value comparisons of a maximum of the longitudinal acceleration during an impact, a maximum of the lateral acceleration during an impact, by diagnostic information from an airbag control unit, an impact location on the vehicle and / or a maximum measured impulse. [9] Method according to one of the preceding claims, wherein battery state variables comprise operating variables of the device battery (41) at one or more specific points in time during and after the fault event, such as temperatures, voltages, currents, charge states, balancing states and / or their respective profiles at cell, module or overall battery level and / or aggregated variables thereof, and / or an aging state of the device battery and / or internal battery states, such asSEI layer thickness, change in cyclable lithium due to anode / cathode side reactions, rapid uptake of electrolyte solvents, slow uptake of electrolyte solvents, lithium deposition, loss of active anode material and / or loss of active cathode material and / or operational characteristics indicative of stress factors on the operation of the battery, in particular histogram characteristics such as temperature versus state of charge, charge current versus temperature and discharge current versus temperature, in particular multi-dimensional histogram data relating to the battery temperature distribution versus state of charge, charge current distribution versus temperature and / or discharge current distribution versus temperature, the current throughput in ampere-hours, the total accumulated charge (Ah), an average capacity increase in a charge event (in particular for charges where the charge increase is above a threshold fraction [e.g.20% ΔSOC] of the total battery capacity), the charging capacity and / or an extreme value (e.g. maximum) of the differential capacity during a measured charging process and / or a voltage residue as the difference between a modeled battery voltage from a fractional battery model or electrochemical battery model and a measured voltage value. [10] Method according to one of the preceding claims, wherein the determination of the at least one fault event feature and the at least one battery state variable is preferably to be determined in a period of time after the occurrence of the operational fault, which in particular is not more than 1 to 3 days after the time of occurrence of the operational fault. [11] Method according to one of the preceding claims, wherein the disturbance event model is designed as a multi-label logistic regression or as a neural network. [12] Method according to one of the preceding claims, wherein the method is carried out in a device-external central unit (2) in which the temporal operating variable profiles are previously transmitted, wherein an indication of a type of error and / or an indication of the action instruction is transmitted to the device (4) in question. [13] Device, in particular a data processing device, for predictively detecting a fault in a device battery (41) of a technical device (4) after the occurrence of an operational fault in the technical device (4), wherein the device is designed to: - Providing temporal operating variable profiles that characterize the operation of the technical device (4) and the device battery (41); - Determination of an operational disturbance by evaluating the temporal course of operational variables; - Upon detection of an operational fault, determining at least one fault event feature indicating a state of the technical device (4) and at least one battery state variable indicating a state of the device battery (41), and evaluating the at least one fault event feature and the at least one battery state variable using a data-based fault event model in order to obtain an output vector with output variables whose values ​​are each assigned to a fault type of a fault; - Signaling an instruction to a user of the technical device (4) depending on the output vector. [14] Computer program product comprising instructions which, when the program is executed by at least one data processing device, cause the device to carry out the steps of the method according to one of claims 1 to 12. [15] Machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause it to carry out the steps of the method according to one of claims 1 to 12.

Images