Method for determining an actuation correction, method for teaching a computer-implemented algorithm, method for operating a power provider, control device, and power arrangement

Abstract

The invention relates to a method for determining an actuation correction (25) of at least one fuel valve (9) of a power provider (3), in particular of an internal combustion engine, wherein - during operation of the power provider (3), a sensor signal (19) representing an actuation of the at least one fuel valve (9) is detected in a specified time interval by means of a sensor (11, 17), - a setpoint fuel quantity (21) to be dispensed by the at least one fuel valve (9) is specified in the specified time interval, and - by means of a computer-implemented algorithm (23) based on a neural network using the sensor signal (19) and the target fuel quantity (21) an actuation correction (25) for the at least one fuel valve (9) is determined.

Description

[0001] 2021P00007DE

[0002] 1

[0003] Rolls-Royce Solutions GmbH January 17, 2025

[0004] DESCRIPTION

[0005] Method for determining a control correction, method for teaching a computer-implemented algorithm, method for operating a power supply device, control device and power arrangement

[0006] The invention relates to a method for determining a control correction, a method for teaching a computer-implemented algorithm for determining a control correction, a method for operating a power provision device, a control device and a power arrangement with such a control device.

[0007] It is known that in a power supply device, a variation in the actual fuel quantity of a fuel valve, in particular a variation over time of the fuel valve, and / or a drift in the injection behavior, for example due to aging of the fuel valve, has a direct influence on emission values, running characteristics and operating ranges of the power supply device. The greater the variation, the worse, in particular higher, the average emission values.

[0008] Furthermore, the greater the variation, the less efficiently and, in particular, less smoothly the power supply device operates. Additionally, the operating ranges in which the power supply device can be operated safely and with low risk become smaller as the variation increases.

[0009] Furthermore, it is known that the timing of the fuel injection via the fuel valve directly influences the properties of the power delivery device. The greater the time lag from the optimal injection point, the worse the properties, especially the combustion properties.

[0010] At the same time, the fuel valve of the power supply device is subjected to high mechanical stress during operation. This causes high mechanical wear of the 2021P00007DE

[0011] 2

[0012] Fuel valve wear increases both the variation in the actual fuel quantity and the timing of injection. Therefore, it may be necessary to monitor the mechanical wear of the fuel valve and, if necessary, take corrective action or adhere to comparatively short replacement intervals.

[0013] A deceleration calibration method is known for monitoring the fuel valve. In this method, the fuel valves are individually calibrated during deceleration. Normally, the fuel valves do not introduce any fuel during deceleration. Then, the activation duration of the selected fuel valve is increased until combustion is detected in a combustion chamber associated with that valve. The corresponding activation duration is stored as the base point of a control characteristic curve for the selected fuel valve.

[0014] The disadvantage is that not every power supply device can be operated in thrust mode.

[0015] For monitoring purposes, each fuel valve can also be equipped with a separate sensor for detecting electrostatic discharges. A disadvantage of this approach is that these sensors require complex calibration and are expensive. Furthermore, such methods are technically demanding, as the fuel valves must be partially deactivated and calibrated separately during the process. This makes the methods expensive and not feasible while the power supply system is in operation.

[0016] The invention is based on the objective of creating a method for determining a control correction, a method for teaching a computer-implemented algorithm for determining a control correction, a method for operating a power supply device, a control device and a power arrangement with such a control device, wherein the aforementioned disadvantages do not occur, at least in part.

[0017] The problem is solved by providing the present technical teaching, in particular the teaching of the independent claims and the embodiments disclosed in the dependent claims and the description. 2021P00007DE

[0018] 3

[0019] The problem is solved, in particular, by creating a method for determining a control correction of at least one fuel valve of a power supply device, especially an internal combustion engine. During operation of the power supply device, a sensor signal representing the control of the at least one fuel valve is acquired at a predetermined time interval. Furthermore, a target fuel quantity to be delivered by the at least one fuel valve is specified within the same time interval.

[0020] Subsequently, a computer-implemented algorithm incorporating a neural network determines a control correction for the at least one fuel valve based on the sensor signal and the target fuel quantity. Advantageously, the method detects whether the at least one fuel valve is delivering an actual fuel quantity that is greater or less than the target fuel quantity. Furthermore, the control correction reduces the variation of the at least one fuel valve, resulting in lower emissions, improved running characteristics of the power generation device, and wider operating ranges in which the power generation device can be operated safely and with minimal risk. Additionally, no overrun operation of the power generation device and / or partial deactivation of the fuel valves is required to monitor the at least one fuel valve.Furthermore, the procedure can advantageously be carried out during regular operation of the power supply device – in real time.

[0021] In one embodiment, the target fuel quantity is a quantity of fuel to be introduced. In another embodiment, the target fuel quantity is a duration of energization for the at least one fuel valve, or is represented by the duration of energization.

[0022] In one embodiment, a control duration correction is used as the control correction. The control duration correction is a shortening or lengthening of the control duration, whereby a control start and / or control end point is changed based on the control duration correction. Preferably, the control duration correction ranges from -2 ms to 2 ms, where a negative control duration correction means a shortening of the control duration and a positive control duration correction means a lengthening of the control duration. 2021P00007DE

[0023] 4

[0024] In particular, the control signal is applied by energizing at least one fuel valve. Therefore, an energizing duration correction is used as the control correction, specifically as the control duration correction.

[0025] In a further embodiment, a fuel quantity correction is used as the control correction. The fuel quantity correction is either an increase or a decrease in the target fuel quantity. Preferably, a negative fuel quantity correction indicates a reduction in the control signal, and a positive fuel quantity correction indicates an increase in the control signal. In this case, the control signal is specifically a fuel quantity.

[0026] In the context of this technical teaching, a power supply device is understood to be, in particular, a device configured to supply power, especially electrical and / or mechanical power. Preferably, a power supply device is understood to be a device that supplies power—for example, electrical and / or mechanical power—using electrical, mechanical, chemical, or electrochemical energy, or another form of energy. The power supply device may be an internal combustion engine or an internal combustion engine-generator combination device, i.e., a genset. However, the power supply device may also be a larger, more complex system, for example, consisting of a plurality of the aforementioned devices.

[0027] In particular, the power supply device has two cylinder banks, each with an identical number of combustion chambers. The procedure is carried out separately for each cylinder bank, with at least one sensor being used for each cylinder bank of the power supply device.

[0028] In one embodiment, the power supply device is an internal combustion engine with at least one combustion chamber, wherein the at least one fuel valve is assigned to the at least one combustion chamber in order to supply fuel to the at least one combustion chamber.

[0029] In the context of the present technical teaching, a fuel valve is in particular a device for introducing a fuel into a combustion chamber of the 2021P00007DE

[0030] 5

[0031] The term "power supply device" is understood to mean the fuel valve. Preferably, the fuel valve is configured as a fuel injector. Alternatively or additionally, the fuel valve is configured to introduce, in particular inject, or introduce gaseous or liquid fuels, in particular natural gas, hydrogen, gasoline, or diesel, into the combustion chamber of the power supply device.

[0032] In particular, the sensor signal is acquired with time resolution or as a function of a crankshaft angle of the power supply device. Preferably, the crankshaft angle of a crankshaft of the power supply device is acquired for this purpose, in particular by means of a crankshaft speed angle sensor.

[0033] In particular, the sensor signal is filtered, preferably by means of a low-pass filter.

[0034] In a preferred embodiment, the sensor signal represents the injection of a specific quantity of fuel dispensed by the at least one fuel valve. This utilizes the fact that each injection generates an event, in particular a deflection, in the sensor signal. The injection causes a drop in pressure, at least locally in the area of ​​a fluid connection of the fuel valve, in a fuel reservoir, hereinafter also referred to as the rail or common rail. More generally, a local disturbance occurs, which propagates as a sound wave to the sensor. The arrival time of the disturbance at the sensor and the signal amplitude then allow conclusions to be drawn about the fuel valve and the injected quantity of fuel.

[0035] Furthermore, if the power delivery device has multiple combustion chambers, each with at least one fuel valve, an insertion sequence is specified for the individual fuel valves. This allows each insertion to be assigned a crankshaft angle range. Additionally, this also allows each combustion chamber and each fuel valve to be assigned a crankshaft angle range.

[0036] In particular, the predetermined time interval is determined based on the crankshaft angle of the power supply device. Preferably, the predetermined time interval is selected as a working interval of the power supply device, wherein the working interval has a length of 360 kW or 720 kW. 2021P00007DE

[0037] 6

[0038] In the context of this technical teaching, a working interval of the power supply device is understood to be a time period in which exactly one working cycle is carried out in each combustion chamber, without regard to the temporal sequence of the strokes in the individual combustion chambers. Thus, the working interval corresponds to the number of crankshaft revolutions equivalent to the length of one working cycle in a combustion chamber of the power supply device. In a four-stroke internal combustion engine configuration, the working interval has a length of 720° crankshaft rotation. Alternatively, in a two-stroke internal combustion engine configuration, the working interval has a length of 360° crankshaft rotation.

[0039] In a further embodiment, during the operation of the power supply device, a first sensor signal representing the control of at least one fuel valve is acquired by a first sensor within the specified time interval. Furthermore, during the operation of the power supply device, a second sensor signal representing the control of at least one fuel valve is acquired by a second sensor within the specified time interval. The first and second sensor signals are then combined to obtain the sensor signal. Alternatively, the control correction of the at least one fuel valve is determined using the computer-implemented algorithm based on the first sensor signal, the second sensor signal, and the target fuel quantity.Advantageously, the use of two sensors allows potential overlaps of combustion events in multiple combustion chambers to be detected and compensated for within a single sensor signal. In particular, overlaps of combustion events in multiple combustion chambers within a single sensor signal can occur in power delivery devices with a large number of combustion chambers per cylinder bank, especially more than five, more than seven, or more than nine combustion chambers per cylinder bank.

[0040] In particular, the target fuel quantity is specified and / or calculated by a control device of the power supply device.

[0041] In particular, the neural network is an artificial neural network. 2021P00007DE

[0042] 7

[0043] In particular, a feedforward network is used as the neural network. In the context of this technical teaching, a feedforward network is understood to be a forward-directed network without feedback. The network comprises an input layer, a plurality of intermediate layers, and an output layer. The input layer is connected to the first intermediate layer of the plurality of intermediate layers. Furthermore, each intermediate layer following the first intermediate layer of the plurality of intermediate layers is connected to the preceding intermediate layer. Finally, the last intermediate layer of the plurality of intermediate layers is connected to the output layer.

[0044] In particular, the input layer of one embodiment of the neural network has at least one input, in particular the sensor signal. Optionally, the input layer of the neural network also has a rotational speed of the power supply device as a further input. Furthermore, the neural network has at least four intermediate layers, in particular of different sizes. In addition, the output layer of the neural network has two outputs, in particular the control correction and an identifier for the fuel valve to be corrected.

[0045] Preferably, an elliotsig function is used as a transfer function of the neural network, in particular the feedforward network.

[0046] In particular, the control correction is estimated using the computer-implemented algorithm.

[0047] In particular, the rotational speed is measured using a speed sensor. Alternatively or additionally, the rotational speed is calculated.

[0048] According to a further development of the invention, the control correction is stored. In particular, the control correction is used to control the at least one fuel valve. Additionally, the control correction is not fed back to the neural network. Advantageously, a detected malfunction and the associated control correction are thus stored and made available for future control of the at least 2021P00007DE

[0049] 8

[0050] A fuel valve is used, at least until a new control correction is issued by the neural network and then preferably added to the stored control correction. The at least one fuel valve is then controlled with the sum thus calculated, and this sum is stored as a new control correction.

[0051] According to a further development of the invention, it is provided that the control correction for a main injection is determined by means of the at least one fuel valve.

[0052] Alternatively or additionally, the control correction for pre-injection is determined by means of the at least one fuel valve. This advantageously improves, in particular, the cold-start capability of the power delivery device. The combustion of the fuel quantity introduced by the pre-injection raises the gas temperature in a combustion chamber associated with the at least one fuel valve. Due to the higher temperature, the subsequent main injection has a shorter ignition delay. This advantageously reduces the pressure rise gradient and thus protects a connecting rod bearing of the power delivery device. Furthermore, the power delivery device generates less noise emissions as a result, since diesel knock is eliminated or at least reduced.

[0053] In one embodiment, the power supply device is operated during the execution of the process without the subsequent introduction of fuel.

[0054] According to a further development of the invention, it is provided that a fuel pressure sensor is used as the sensor, so that a fuel pressure profile is recorded as the sensor signal.

[0055] In particular, an injection event generates a pressure pulse in a fuel line of the power supply device, especially a common rail, such that the fuel pressure advantageously represents the control of the at least one fuel valve. In particular, the power supply device has the fuel pressure sensor independently of the method according to the invention, so that advantageously no additional sensors need to be installed to carry out the method, and thus the costs of the method are low. 2021P00007DE

[0056] 9

[0057] In particular, a rail pressure sensor is used as the fuel pressure sensor, so that a rail pressure profile is recorded as the sensor signal. Preferably, at most two, and more preferably exactly two, rail pressure sensors are used for a power delivery device with a plurality of combustion chambers, wherein the at least two rail pressure sensors are preferably arranged at opposite ends of the common rail along a long side of the rail. In particular, exactly two rail pressure sensors are used for each cylinder bank of the power delivery device.

[0058] In one embodiment of the neural network, the input layer has, as its at least one input, the fuel pressure profile, specifically the rail pressure profile. In another embodiment, the input layer has, as its at least one input, a first fuel pressure profile from the first sensor. Alternatively or additionally, the input layer has, as its at least one input, a second fuel pressure profile from the second sensor. Alternatively or additionally, the input layer has, as its at least one input, a fuel pressure profile that combines the first and second fuel pressure profiles.

[0059] According to a further development of the invention, it is provided that the specified time interval is at most a working interval of the power supply device.

[0060] Advantageously, this ensures that in the specified time interval, only one injection event and one combustion event are recorded for each fuel valve and the respective associated combustion chambers, in particular for each cylinder bank of the power supply device.

[0061] In a preferred embodiment, the specified time interval is exactly one operating interval of the power supply device. Advantageously, this ensures that within the specified time interval, exactly one injection event and exactly one combustion event are recorded for each fuel valve and its respective associated combustion chamber, particularly for each cylinder bank of the power supply device. 2021P00007DE

[0062] 10

[0063] Alternatively, a plurality of operating intervals of the power supply device is selected as the predetermined time interval. For each operating interval of the plurality of operating intervals, a partial sensor signal is extracted, particularly taking the crankshaft angle into account. Subsequently, the plurality of partial sensor signals, particularly taking the crankshaft angle into account, are averaged, resulting in an averaged sensor signal. Preferably, the target fuel quantity to be delivered by the at least one fuel valve is further specified for each operating interval of the plurality of operating intervals, and the plurality of target fuel quantities are averaged, resulting in an averaged target fuel quantity. Finally, the control correction of the at least one fuel valve is determined using the computer-implemented algorithm based on the averaged sensor signal and the averaged target fuel quantity.Optionally, before the averaged sensor signal is passed to the computer-implemented algorithm, the majority of partial sensor signals and / or the averaged sensor signal is filtered, preferably using a low-pass filter.

[0064] According to a further development of the invention, a plurality of discrete, time-dependent sensor measurements are acquired as the sensor signal. Advantageously, this makes it possible to analyze the control of the at least one fuel valve with temporal resolution.

[0065] In particular, at least 720 sensor measurements and at most 14,400 sensor measurements, in particular at least 900 sensor measurements, in particular at least 1,200 sensor measurements, in particular at least 1,440 sensor measurements, in particular at least 1,800 sensor measurements, in particular at least 3,600 sensor measurements, in particular at most 12,000 sensor measurements, in particular at most 9,000 sensor measurements, and especially preferably 7,200 sensor measurements, are recorded as a sensor signal per working interval.

[0066] According to a further development of the invention, it is provided that a control correction and / or a control compensation is determined for at least two fuel valves of the power supply device. The sensor signal represents a control action of each of the at least two fuel valves. Furthermore, the target fuel quantity is specified for each fuel valve. Using the computer-implemented algorithm, the majority of control corrections and / or the majority of control compensation are determined.

[0067] 11

[0068] The control of at least two fuel valves is balanced such that the actual torque of the power delivery device remains constant. Advantageously, this allows the control of at least one fuel valve to be adjusted to the actual torque and / or the control to be carried out in such a way that the actual torque is maintained. Furthermore, this advantageously balances the fuel quantities of the majority of fuel valves so that the actual torque applied to the crankshaft remains constant.

[0069] In particular, a target torque is specified, and a torque control module determines the target fuel quantities for all fuel valves of the power delivery device as a function of this target torque. If the method according to the invention is not carried out and the actual fuel quantity at at least one fuel valve does not correspond to the target fuel quantity, the actual torque will also not correspond to the target torque. Therefore, the torque control module will change the target fuel quantities for all fuel valves simultaneously and uniformly until the actual torque is aligned with the required target torque. A disadvantage of this is that the combustion chambers are subjected to uneven loading due to the differing actual fuel quantities.

[0070] In particular, the computer-implemented algorithm includes a torque compensation module in addition to the neural network. The neural network determines the control correction, which is specifically non-zero, for all fuel valves where the actual fuel quantity does not equal the target fuel quantity. Furthermore, based on the control corrections for all other fuel valves, the torque compensation module determines the control compensation such that the actual torque of the power delivery device remains constant. Subsequently, each fuel valve is controlled by a combination, specifically a sum, of the control correction and the control compensation.

[0071] In one embodiment, the control compensation is stored. In particular, the control compensation is used to control at least one fuel valve. Additionally, the control compensation is not fed back to the neural network.

[0072] Advantageously, this allows a compensation value determined by a detected malfunction to be stored and used for future control of at least one fuel valve. 2021P00007DE

[0073] 12

[0074] In particular, the actual torque is measured using a torque sensor. Alternatively or additionally, the actual torque is calculated.

[0075] In an exemplary implementation of the method in a power supply device with n fuel valves controlled by identical target fuel quantities, the control correction of the first fuel valve is determined such that the target fuel quantity of the first fuel valve is changed by x%. Thus, the control correction of the first fuel valve is x% of the target fuel quantity, and the control corrections of the n-1 other fuel valves are 0. Simultaneously, the target fuel quantities of the n-1 other, otherwise uncorrected fuel valves are each changed by x / (nl)% relative to the uncorrected target fuel quantity of the first fuel valve. Thus, the control adjustments of the n-1 other fuel valves are each x / (nl)% of the target fuel quantity, and the control adjustment of the first fuel valve is 0.In this process, if the target fuel quantity for controlling the first fuel valve is increased, the target fuel quantities for controlling the n-1 other fuel valves are reduced, and if the target fuel quantity for controlling the first fuel valve is reduced, the target fuel quantities for controlling the n-1 other fuel valves are increased.

[0076] According to a further development of the invention, the method is carried out cyclically. Advantageously, this makes it possible to react quickly to changes, in particular wear and / or signs of aging, in the at least one fuel valve and to adapt the control quickly and easily to the changed operating principle of the at least one fuel valve.

[0077] In a preferred embodiment, the procedure is carried out in each operating interval of the power supply device.

[0078] The task is also solved by a method for teaching a computer-implemented algorithm for determining a control correction, in particular a control duration correction, of at least one fuel valve of a 2021P00007DE

[0079] 13

[0080] A power supply device is created. This device is operated in a plurality of predefined time intervals, with a predetermined actual fuel quantity being set for the at least one fuel valve during each interval. A sensor captures a sensor signal at each of these intervals, representing the activation of the at least one fuel valve. Furthermore, a sensor signal is assigned to each predefined actual fuel quantity set for the at least one fuel valve. The computer-implemented algorithm is then trained based on the plurality of sensor signals and the plurality of predefined actual fuel quantities.In connection with the procedure for learning the computer-implemented algorithm for determining the control correction, the advantages that have already been explained in connection with the procedure for determining the control correction become particularly apparent.

[0081] In one embodiment, the actual fuel quantity is the amount of fuel supplied. In another embodiment, the actual fuel quantity is the duration of the current flow to the at least one fuel valve, or is represented by the duration of the current flow.

[0082] In one embodiment, for at least one time interval, preferably all time intervals, or the plurality of time intervals, at most one operating interval of the power supply device is selected. Particularly preferably, for at least one time interval, preferably all time intervals, or the plurality of time intervals, exactly one operating interval of the power supply device is selected.

[0083] In an alternative embodiment, for at least one time interval, preferably all time intervals, a plurality of operating intervals of the power supply device is selected, wherein preferably one load point is not changed during the plurality of operating intervals and thus the actual fuel quantity is set almost identically. For each operating interval of the plurality of operating intervals, a partial sensor signal is extracted, particularly taking the crankshaft angle into account. Subsequently, the plurality of partial sensor signals, particularly taking the crankshaft angle into account, are averaged, resulting in an averaged sensor signal. Furthermore, optionally, for each operating interval of the 2021P00007DE

[0084] 14

[0085] For multiple operating intervals, the actual fuel quantity is specified, and the multiple actual fuel quantities are averaged to obtain an averaged actual fuel quantity. The computer-implemented algorithm is then trained using the at least one averaged sensor signal, preferably the multiple averaged sensor signals, and the at least one, and preferably the multiple, averaged actual fuel quantity. Optionally, before the averaged sensor signal is used to train the computer-implemented algorithm, the multiple partial sensor signals and / or the averaged sensor signal are filtered, preferably using a low-pass filter.

[0086] In one embodiment, for the at least one time interval, preferably all time intervals, the plurality of time intervals, at least 50 to at most 500, preferably 250, working intervals are selected.

[0087] Preferably, the computer-implemented algorithm is trained on the power supply device for at least one load point, in particular for a plurality of load points.

[0088] In particular, a nominal actuation value of the at least one fuel valve is used as the energization duration for each load point to be taught. Additionally, the nominal actuation value is varied for each load point to be taught. It is particularly preferred that the nominal actuation value is varied in 5% increments, with the nominal actuation value being changed by a maximum of 20%.

[0089] In one embodiment, the nominal control, 80% of the nominal control, 85% of the nominal control, 90% of the nominal control, 95% of the nominal control, 105% of the nominal control, 110% of the nominal control, 115% of the nominal control, and 120% of the nominal control are used for each load point to be measured.

[0090] Preferably, a fuel pressure sensor is used, in which case a fuel pressure curve is detected as the sensor signal. Particularly preferably, a rail pressure sensor is used, in which case a rail pressure curve is detected as the sensor signal. 2021P00007DE

[0091] 15

[0092] Preferably, the weighting and bias values ​​of the computer-implemented algorithm are updated using Levenberg-Marquardt optimization. In particular, the training function `trainlm` is used to train the computer-implemented algorithm.

[0093] In one implementation, the quality of the learning process is measured using the mean value of a squared error.

[0094] In particular, the learning process is carried out using test bench data, whereby the predetermined actual fuel quantity is set on a test bench and the sensor signals are also recorded on the test bench. Alternatively or additionally, the learning process is carried out during operation of the power supply device. Preferably, a power supply device is used for learning in which all fuel valves function correctly. This means, in particular, that each control signal is assigned a unique and correct actual fuel quantity.

[0095] The problem is also solved by providing a method for operating a power supply device with at least one fuel valve. In this method, a control correction, in particular a control duration correction, of the at least one fuel valve is determined using a method according to the invention or a method according to one or more of the embodiments described above. Furthermore, the power supply device is operated depending on the determined control correction of the at least one fuel valve. In particular, the control of the at least one fuel valve is modified based on the control correction. In connection with the method for training the computer-implemented algorithm for determining the control correction, the advantages that have already been explained in connection with the method for determining the control correction become particularly apparent.

[0096] The problem is also solved by creating a control device configured to carry out a method according to the invention or a method according to one or more of the embodiments described above. The control device is preferably a computing device, particularly preferably a 2021P00007DE

[0097] 16

[0098] The device is designed as a computer or as a control unit, preferably as a control unit of an internal combustion engine. In connection with the control device, the advantages are realized in particular those previously described in connection with the method for determining the control correction and the method for operating the power supply device.

[0099] The control device is preferably configured to operate the power supply device. In one embodiment, the control device is configured to operate an internal combustion engine, an internal combustion engine-generator combination device (i.e., a genset), or a fuel cell.

[0100] In one embodiment, the control device is configured to operate the power supply device, in particular at least one fuel valve of the power supply device, with a control correction, i.e. preferably to control it.

[0101] In one embodiment, the control device is configured to detect, determine, or receive a target fuel quantity, a sensor signal, and / or a control command.

[0102] The problem is also solved by creating a power arrangement with a power supply device and a control device according to the invention or a control device according to one or more of the embodiments described above, wherein the power supply device has at least one fuel valve. In connection with the power supply device, those advantages are realized in particular that have already been described previously in connection with the method and the control device.

[0103] The control device is preferably operatively connected to the power supply device in order to control the power supply device. Furthermore, the control device is operatively connected to the at least one fuel valve. 2021P00007DE

[0104] 17

[0105] In particular, the power arrangement also includes at least one sensor. This at least one sensor is configured to detect a sensor signal representing the actuation of the at least one fuel valve and to transmit it to the control device, which is configured to detect the sensor signal.

[0106] Preferably, the at least one fuel valve is designed to be controlled by means of control correction.

[0107] In one embodiment, the power supply device is designed as an internal combustion engine or as an internal combustion engine-generator combination device, i.e., a genset.

[0108] The invention will be explained in more detail below with reference to the drawing. The drawing shows:

[0109] Figure 1 shows a schematic representation of an exemplary embodiment of a power arrangement,

[0110] Figure 2 is a schematic representation of an embodiment of a method for determining a control duration correction in the form of a flowchart, Figure 3 is a schematic representation of an embodiment for teaching a computer-implemented algorithm in the form of a flowchart, and Figure 4 is a schematic representation of an insertion sequence using two fuel valves and the generation of sensor signals.

[0111] Fig. 1 shows a schematic representation of an embodiment of a power arrangement 1. The power arrangement 1 comprises a power supply device 3, in particular an internal combustion engine, and a control device 5.

[0112] Furthermore, the power supply device 3 has at least one combustion chamber 7, in particular four combustion chambers 7, at least one fuel valve 9, in particular four fuel valves 9, and at least one sensor 11, in particular two sensors 11. Each combustion chamber 7 is assigned at least one, in particular exactly one, fuel valve 9. The at least one fuel valve 9 is configured to introduce a quantity of fuel into the combustion chamber 7 assigned to the fuel valve 9.

[0113] 18

[0114] Preferably, the majority of fuel valves 9 are fluidically connected to a common fuel supply line 13, in particular a common rail 15.

[0115] The at least one sensor 11 is configured to detect a sensor signal 19, representing an actuation of the at least one fuel valve 9, and to transmit it to the control device 5. Preferably, a fuel pressure sensor is used as sensor 11, so that a fuel pressure profile is detected as sensor signal 19.

[0116] Preferably, a rail pressure sensor 17 is used as sensor 11, such that a rail pressure loss is detected as sensor signal 19. The two rail pressure sensors 17 are preferably arranged at opposite ends of the common rail 15.

[0117] The control device 5 is configured to carry out a method for determining a control correction 25 of the at least one fuel valve 9, in particular a method for operating the power supply device 3. An embodiment of the method is described in more detail in Figure 2. Alternatively or additionally, the control device 5 is configured to carry out a method for learning a computer-implemented algorithm 23 for determining the control correction 25 of the at least one fuel valve 9. An embodiment of the method is described in more detail in Figure 3.

[0118] The control device 5 is connected to the at least one sensor 11, in particular the two rail pressure sensors 17, in a manner not explicitly shown, in order to detect the sensor signal 19 detected by the sensor 11, in particular the rail pressure profile.

[0119] The control device 5 is preferably configured to operate the power supply device 3 with a target fuel quantity 21 for the at least one fuel valve 9. Alternatively or additionally, the control device 5 is operatively connected to the fuel valve 9 in a manner not explicitly shown and configured to control it in order to set a target fuel quantity 21, a start of injection, a duration of injection, and / or an end of injection. 2021P00007DE

[0120] 19

[0121] The control device 5 is specifically configured to specify and / or calculate the target fuel quantity 21. Alternatively or additionally, the control device 5 is configured to detect and / or calculate the rotational speed of the power supply device 3. Alternatively or additionally, the control device 5 is configured to detect and / or calculate the crankshaft angle of a crankshaft of the power supply device 3. Alternatively or additionally, the control device 5 is configured to detect and / or calculate the actual torque 29 of the power supply device 3.

[0122] Fig. 2 shows a schematic representation of an embodiment of the method for determining the control correction 25 of the fuel valve 9 in the form of a flowchart.

[0123] Identical and functionally equivalent elements are provided with the same reference symbols in all figures, so reference is made to the preceding description.

[0124] In step S1, during the operation of the power supply device 3, a sensor signal 19 is detected at a specified time interval by means of at least one sensor 11, which represents a control of at least one fuel valve 9.

[0125] In particular, the sensor signal 19 is detected as a function of a crankshaft angle of the power supply device 3. A working interval of the power supply device 3 has a length of 360°KW or 720°KW.

[0126] Preferably, during operation of the power supply device 3, a sensor signal 19 is detected by each of the two sensors 11 at the specified time interval, representing the control of the at least one fuel valve 9. The two sensor signals 19 are then combined, resulting in the sensor signal 19.

[0127] Preferably, the specified time interval is selected to be at most one working interval, and particularly preferably exactly one working interval, of the power supply device 3. 2021P00007DE

[0128] 20

[0129] Preferably, a plurality of discrete, time-dependent sensor measurements are acquired as the sensor signal 19. In particular, at least 720 sensor measurements and at most 14,400 sensor measurements, in particular at least 900 sensor measurements, in particular at least 1,200 sensor measurements, in particular at least 1,440 sensor measurements, in particular at least 1,800 sensor measurements, in particular at least 3,600 sensor measurements, in particular at most 12,000 sensor measurements, in particular at most 9,000 sensor measurements, and most preferably 7,200 sensor measurements, are acquired as the sensor signal 19 per working interval.

[0130] In step S2, a target fuel quantity 21 to be discharged by the at least one fuel valve 9 is specified within the given time interval.

[0131] In step S3, a control correction 25 of the at least one fuel valve 9 is determined using a computer-implemented algorithm 23, which has a neural network, based on the sensor signal 19 and the target fuel quantity 21.

[0132] In step S4, the power supply device 3 is operated by means of the control correction 25.

[0133] In an optional step S5, the sensor signal 19 is filtered, in particular by means of a low-pass filter, resulting in a filtered sensor signal 27. In particular, the control correction 25 is then determined in step S3 based on the filtered sensor signal 27.

[0134] In an optional step S6, the actual torque 29 of the power supply device 3 is specified. A control compensation 31 of an additional fuel valve 9 is determined based on the control correction 25 of the at least one fuel valve 9 such that the actual torque 29 of the power supply device 3 is constant. In particular, the power supply device 3 is then additionally operated by means of the control compensation 31.

[0135] Preferably, the method is carried out cyclically. Particularly preferably, the method is carried out in each operating interval of the power supply device 3. 2021P00007DE

[0136] 21

[0137] In particular, steps S1 and S2 can be performed in any order, and it is not necessary to perform all optional steps.

[0138] Fig. 3 shows a schematic representation of an embodiment for teaching the computer-implemented algorithm 23 in the form of a flowchart.

[0139] In step TI, a predetermined actual fuel quantity 33 of the at least one fuel valve 9 is set. In particular, a load point of the power supply device 3 is set.

[0140] In step T2, the power supply device 3 is operated with the set actual fuel quantity 33 for a predetermined time interval. In one embodiment, the predetermined time interval is selected to be at most one operating interval, and in particular exactly one operating interval, of the power supply device 3. In an alternative embodiment, the predetermined time interval is selected to be a plurality of operating intervals, particularly preferably 250 operating intervals, of the power supply device 3, wherein the load point is preferably not changed during the plurality of operating intervals, and thus the actual fuel quantity 33 is set almost identically.

[0141] In step S1, analogous to Figure 2, the sensor signal 19, which represents an actuation of the at least one fuel valve 9, is detected by means of the at least one sensor 11 in the specified time interval.

[0142] The optional step S5 is identical to Figure 2.

[0143] In the configuration where a plurality of work intervals are chosen as the specified time interval, steps T5 and T6 are also carried out.

[0144] In step T5, a partial sensor signal 35 is extracted for each of the majority of working intervals, taking into account the crankshaft angle in particular. 2021P00007DE

[0145] 22

[0146] In step T6, the majority of partial sensor signals 35 are then averaged, particularly taking into account the crankshaft angle, resulting in an averaged sensor signal 37.

[0147] In step T3, the predetermined actual fuel quantity 33 is assigned the sensor signal 19, in particular the averaged sensor signal 37. Alternatively or additionally, a filtered sensor signal 27 and / or an averaged sensor signal 37 is assigned to the predetermined actual fuel quantity 33.

[0148] Steps TI, T2, S1, and T3 are performed multiple times so that an associated sensor signal 19 is obtained for each of a plurality of predetermined actual fuel quantities 33. The power supply device 3 is operated in a plurality of predetermined time intervals, with a predetermined target fuel quantity 33 of the at least one fuel valve 9 being set for the operation of each predetermined time interval. In particular, the steps are performed for at least one load point, and especially for a plurality of load points, of the power supply device, with a plurality of actual fuel quantities 33 being used for each load point. In particular, a nominal control signal of the at least one fuel valve 9 is used for each load point to be taught, with the actual fuel quantity 33 being determined based on the nominal control signal.Additionally, the nominal control signal is preferably varied for each load point to be taught. Particularly preferably, the nominal control signal is varied in 5% increments, with the nominal control signal being changed by a maximum of 20%.

[0149] In step T4, the computer-implemented algorithm 23 is trained using the plurality of preset, predetermined actual fuel quantities 33 and the plurality of sensor signals 19, in particular the plurality of filtered sensor signals 27 and / or the plurality of averaged sensor signals 37. Preferably, the weighting and bias values ​​of the computer-implemented algorithm 23 are updated by means of a Levenberg-Marquardt optimization. In particular, the training function trainlm is used to train the computer-implemented algorithm 23. In one embodiment, the quality of the training is measured using a mean squared error.

[0150] 23

[0151] Fig. 4 shows a schematic representation of an injection sequence using two fuel valves 9 and the generation of the sensor signals 19.

[0152] The common fuel supply line 13, in particular the common rail 15, with two fuel valves 9, namely a first fuel valve 9.1 and a second fuel valve 9.2, is shown. Additionally, two sensors 11, in particular two rail pressure sensors 17, are preferably arranged at opposite ends of the fuel supply line 13.

[0153] Vertically below the fuel supply line 13, a temporal sequence of injection events and the generation of the sensor signals 19 is shown. The vertical direction represents a time axis, and the horizontal direction a spatial axis. The horizontal dashed lines mark the beginning, specifically marked 0°KW, and the end of each operating interval. Within each operating interval, each fuel valve 9 introduces fuel into the combustion chamber 7 exactly once. These injection events are marked with circles.

[0154] Furthermore, each injection event generates a pressure fluctuation 39, represented as a curve. This pressure fluctuation 39 propagates in the fuel supply line 13, particularly at the speed of sound, as shown by slanted dashed lines emanating from the circles. When these pressure fluctuations 39 encounter the sensors 11, the sensor signal 19 is detected. The point where the pressure fluctuations 39 impact the respective sensor 11 is marked with a square.

[0155] Depending on the spatial distance between the fuel valves 9 and the time interval between two injection events, it is possible for two pressure fluctuations 39 to superimpose, and thus only one superimposed pressure fluctuation 39 arrives at one of the sensors 11 instead of two separate pressure fluctuations 39. Advantageously, in such a case, the pressure fluctuations 39 assigned to the different fuel valves 9.1, 9.2 can be distinguished by comparing the sensor signals of the two sensors 11.

Claims

2021P00007DE 24 REQUIREMENTS 1. Method for determining a control correction (25) of at least one fuel valve (9) of a power supply device (3), in particular an internal combustion engine, wherein - during the operation of the power supply device (3) a sensor signal (19) is detected at a specified time interval by means of a sensor (11, 17) which represents an actuation of the at least one fuel valve (9), wherein - a target fuel quantity (21) to be discharged by the at least one fuel valve (9) is specified within the specified time interval, and wherein - by means of a computer-implemented algorithm (23) incorporating a neural network, a control correction (25) of the at least one fuel valve (9) is determined based on the sensor signal (19) and the target fuel quantity (21).

2. Method according to claim 1, wherein a fuel pressure sensor is used as the sensor (11, 17) such that a fuel pressure profile is detected as the sensor signal (19).

3. Method according to one of the preceding claims, wherein the specified time interval is selected to be at most one working interval, in particular exactly one working interval, of the power provision device (3).

4. Method according to one of the preceding claims, wherein a plurality of discrete, time-dependent sensor measurements, in particular 7200 sensor measurements per working interval, are recorded as the sensor signal (19).

5. Method according to any one of the preceding claims, wherein - for at least two fuel valves (9) of the power supply device (3) a control correction (25) and / or a control compensation (31) is determined, wherein - the sensor signal (19) represents a control of at least two fuel valves (9), wherein - for each fuel valve (9) the target fuel quantity (21) is specified, where 2021P00007DE 25 - the majority of control corrections (25) are determined using the computer-implemented algorithm (23), and wherein - based on the majority of control corrections (25) the majority of control compensations (31) of the at least two fuel valves (9) is determined such that an actual torque (29) of the power supply device (3) is constant.

6. Method according to one of the preceding claims, wherein the method is carried out cyclically, preferably in each working interval of the power provision device (3).

7. Method for teaching a computer-implemented algorithm (23) for determining a control correction (25) of at least one fuel valve (9) of a power supply device (3), wherein - the power supply device (3) is operated in a plurality of predetermined time intervals, wherein - for the operation of each predetermined time interval, a predetermined actual fuel quantity (33) of the at least one fuel valve (9) is set, wherein - by means of a sensor (11, 17) a sensor signal (19) is detected in the plurality of predetermined time intervals, which represents an actuation of the at least one fuel valve (9), wherein - a sensor signal (19) is assigned to each set predetermined actual fuel quantity (33), wherein - the computer-implemented algorithm is trained on the basis of the majority of sensor signals (19) and the majority of set predetermined actual fuel quantities (33).

8. Method for operating a power supply device (3) with at least one fuel valve (9), wherein - by means of a method according to one of claims 1 to 6, a control correction (25) of the at least one fuel valve (9) is determined, wherein 2021P00007DE 26 - the power supply device (3) is operated depending on the specific control correction (25) of the at least one fuel valve (9), in particular the control of the at least one fuel valve (9) is changed.

9. Control device (5) for carrying out a method according to any one of claims 1 to 8.

10. Power arrangement (1) with a power supply device (3) with at least one fuel valve (9) and a control device (5) according to claim 9.

Images