EP4158176A1

EP4158176A1 - Method for the model-based open-loop and closed-loop control of an internal combustion engine

Info

Publication number: EP4158176A1
Application number: EP21729461.0A
Authority: EP
Inventors: Jens Niemeyer; Roman GEISELHART; Knut GRAICHEN; Daniel Bergmann
Original assignee: Rolls Royce Solutions GmbH
Current assignee: Rolls Royce Solutions GmbH
Priority date: 2020-05-27
Filing date: 2021-05-25
Publication date: 2023-04-05
Also published as: CN115720605A; US20230093283A1; DE102020003174A1; WO2021239752A1; DE102020003174B4; US11788484B2

Abstract

What is proposed is a method for the model-based open-loop and closed-loop control of an internal combustion engine, in which a pre-optimized quality measure (J(VO)) is calculated on the basis of the operating situation (BS) by an optimizer in a first step, wherein, in the calculation of the pre-optimized quality measure (J(VO)), discrete manipulated variables having discrete settings are interpreted as continuous manipulated variables (SG(k)) having a continuous settings range, in which these continuous manipulated variables (SG(k)) are quantized and set as new discrete manipulated variables (SG(new)) having discrete settings in a second step, in which a post-optimized quality measure (J(NA)) is calculated on the basis of the new discrete manipulated variables (SG(new)) and the operating situation (BS) of the internal combustion engine (1) by the optimizer in a third step, and the post-optimized quality measure (J(NA)) is set as critical for the operating point of the internal combustion engine (1) by the optimizer (21).

Description

Method for model-based control and regulation of an internal combustion engine

The invention relates to a method for model-based control and regulation of an internal combustion engine, in which an optimizer calculates a quality measure and sets it as decisive for the operating point of the internal combustion engine.

The behavior of an internal combustion engine is largely determined by an engine control unit as a function of a desired performance. For this purpose, corresponding characteristic curves and maps are usually applied in the software of the engine control unit. The manipulated variables of the internal combustion engine, for example the start of injection and a required rail pressure, are calculated from the desired output. These characteristic curves / maps are equipped with data at the manufacturer of the internal combustion engine on a test bench. However, the large number of these characteristic curves / maps and the correlation of the characteristic curves / maps with one another cause a high level of coordination effort.

In practice, attempts are therefore made to reduce the coordination effort by using mathematical models. For example, DE 10 2006004516 B3 describes a Bayesian network with probability tables for determining an injection quantity and US 2011/0172897 A1 describes a method for adapting the start of injection and the injection quantity using combustion models using neural networks. Since trained data is mapped here, it must first be learned during a test bench run.

A method for model-based control and regulation of an internal combustion engine is known from DE 10 2017005783 A1, in which the setpoint values for the injection system actuators are calculated using a combustion model and the setpoint values for the gas path actuators are calculated using a gas path model. Both the combustion model and the gas path model are based on Gaussian process models. An optimizer in turn determines a quality measure from the setpoint values and predicts how the quality measure will change if the setpoint values change within a prediction horizon would develop. Once the best possible quality measure has been calculated, the optimizer sets the injection system target values and the gas path target values as decisive for the operating point of the internal combustion engine.

Tests on the test bench have shown that the integration of manipulated variables with discrete switching states in the previously described model-based method is not yet satisfactory. Manipulated variables with discrete switching states are, for example, the activation of the second exhaust gas turbocharger during register charging, a cylinder bank deactivation, the activation of a pre- or post-injection and the opening or closing of various flaps. So-called branch and bound methods for optimal solution finding in the case of discrete manipulated variables are very computationally expensive, since in the worst case all combinatorial possibilities of the discrete manipulated variables have to be examined. Their use in an internal combustion engine quickly leads to very complex structures which cannot be displayed on an engine control unit.

The invention is based on the object of improving the previously described model-based method with regard to the integration of manipulated variables.

This object is achieved by the features of claim 1. The refinements are presented in the subclaims.

The procedure is carried out in three steps. In the first step, the optimizer calculates a pre-optimized quality measure as a function of the operating situation, the discrete manipulated variables with discrete setting values being interpreted as continuous manipulated variables with a continuous setting range. The pre-optimized quality measure is a computational variable, that is, it is not applied to the internal combustion engine. In the second step, these continuous manipulated variables are then quantized and set as new discrete manipulated variables with discrete setting values. The quantization takes place on the basis of switching thresholds and hysteresis. Finally, in the third step, the optimizer calculates a post-optimized quality measure as a function of the new discrete manipulated variables and the operating situation of the internal combustion engine and sets it as decisive for the operating point of the internal combustion engine.

When calculating the post-optimized quality measure, however, the new discrete manipulated variable are assumed to be fixed. To this extent, they no longer represent a degree of freedom for optimization within the predicted horizon. The rest Continuous manipulated variables are re-optimized in such a way that the solution with regard to the fixed new manipulated variables is the best possible.

The operating situation of the internal combustion engine is understood to mean both the external framework conditions, in particular the emission limit values or the desired output, and the current operating point. Both the pre-optimized quality measure and the post-optimized quality measure are determined by using the combustion model to calculate the injection system setpoints for controlling the injection system actuators, for example the setpoint rail pressure, and using a gas path model to calculate the gas path setpoints for actuating the gas path actuators are calculated and then these target values are changed by the optimizer with the aim of finding a minimum.

The invention allows optimization tasks to be solved with input variables that are partially continuous in value and partially discrete value with limited computing capacity for the optimization method used. Instead of a parallel calculation of the manipulated variables, as is necessary for the implementation of branch-and-bound methods, the invention uses a serial method. Only then can the quality measure and the resulting values for the manipulated variables be fully calculated on an engine control unit.

A preferred embodiment is shown in the figures. Show it:

1 shows a system diagram,

2 shows a model-based system diagram,

3 is a block diagram,

4 shows a program flow chart,

Fig. 5 shows a subroutine,

6 shows a subroutine,

7 shows a subroutine,

Fig. 8 timing diagrams and

Fig. 9 timing diagrams.

FIG. 1 shows a system diagram of an electronically controlled internal combustion engine 1 with a common rail system. The common rail system comprises the following mechanical components: a low-pressure pump 3 for delivering fuel from a fuel tank 2, a variable suction throttle 4 for Influencing the fuel volume flow flowing through, a high-pressure pump 5 for delivering the fuel with an increase in pressure, a rail 6 for storing the fuel and injectors 7 for injecting the fuel into the combustion chambers of the internal combustion engine 1. Optionally, the common rail system can also be designed with individual stores, in which case, for example, an individual memory 8 is integrated in the injector 7 as an additional buffer volume. The further functionality of the common rail system is assumed to be known. The gas path shown includes both the air supply and the exhaust gas discharge. In the air supply are arranged: the compressor of an exhaust gas turbocharger 11, a charge air cooler 12, a throttle valve 13, a junction 14 for merging the charge air with the recirculated exhaust gas and a variably controllable inlet valve 15. In the exhaust gas duct are arranged: a variably controllable exhaust valve 16, an EGR actuator 17, the turbine of the exhaust gas turbocharger 11 and a turbine bypass valve 18.

The mode of operation of the internal combustion engine 1 is determined by an electronic control unit 10 (ECU). The electronic control unit 10 contains the usual components of a microcomputer system, for example a microprocessor, I / O modules, buffers and memory modules (EEPROM, RAM). The operating data relevant to the operation of the internal combustion engine 1 are applied as models in the memory modules. The electronic control unit 10 uses this to calculate the output variables from the input variables. The following input variables are shown as examples in FIG. 1: A target torque M (SOLL), which is specified by an operator, the actual rail pressure pCR, which is measured by means of a rail pressure sensor 9, the engine speed nIST, the charge air pressure pLL, the charge air temperature TLL, the humidity phi of the charge air, the exhaust gas temperature TAbgas, the air-fuel ratio lambda, the NOx actual value, optionally the pressure pES of the individual accumulator 8 and an input variable EIN. The additional sensor signals (not shown) are combined under the input variable IN, for example the coolant temperatures. In Figure 1 are shown as output variables of the electronic control unit 10: a signal PWM to control the suction throttle 4, a signal ve to control the injectors 7 (start / end of injection), a control signal DK to control the throttle valve 13, a control signal VVT to control the Inlet or outlet valve, a control signal AGR for controlling the EGR actuator 17, a control signal TBP for controlling the turbine bypass valve 18 and an output variable AUS. The output variable AUS is representative of the further actuating signals for controlling and regulating the internal combustion engine 1, for example for an actuating signal for activating a second exhaust gas turbocharger a register charge. In the illustration of FIG. 1, for example, the throttle valve 13, the EGR actuator 17, the turbine bypass valve 18 or the suction throttle 4 can be controlled with a continuous control signal and can therefore be set in a continuous range of values. A discrete manipulated variable, on the other hand, would be the control signal for activating a second exhaust gas turbocharger, since this control signal can only assume individual discrete values, i.e. intermediate values do not exist.

FIG. 2 shows a model-based system diagram. In this illustration, a combustion model 19, a gas path model 20 and an optimizer 21 are listed within the electronic control device 10. Both the combustion model 19 and the gas path model 20 map the system behavior of the internal combustion engine as mathematical equations, for example in the form of Gaussian process models. The combustion model 19 statically depicts the processes during the combustion. In contrast to this, the gas path model 20 depicts the dynamic behavior of the air routing and the exhaust gas routing. The combustion model 19 contains individual models, for example for the formation of NOx and soot, for the exhaust gas temperature, for the exhaust gas mass flow and for the peak pressure. These individual models, in turn, depend on the boundary conditions in the cylinder and the parameters of the injection. The combustion model 19 is determined for a reference internal combustion engine in a test bench run, the so-called DoE test bench run (DoE: Design of Experiments). During the DoE test run, operating parameters and manipulated variables are systematically varied with the aim of mapping the overall behavior of the internal combustion engine as a function of engine variables and environmental conditions. The optimizer 21 evaluates the combustion model 19 with regard to the target torque M (SOLL), the emission limit values, the environmental boundary conditions, for example the humidity phi of the charge air, and the operating situation of the internal combustion engine. The operating situation is defined by the engine speed nIST, the charge air temperature TLL, the charge air pressure pLL, etc. The function of the optimizer 21 now consists in evaluating the injection system setpoints for controlling the injection system actuators and the gas path setpoints for activating the gas path actuators . The optimizer 21 selects that solution in which a quality measure is minimized. The measure of quality is calculated as the integral of the quadratic target / actual deviations within the prediction horizon. For example in the form:

(1) J = f [w1 (NOx (TARGET) -NOx (ACTUAL)] ² + [w2 (M (TARGET) -M (ACTUAL))] ² + [w3 (....)] + Here, w1, w2 and w3 mean a corresponding weighting factor. As is known, the nitrogen oxide emissions result from the humidity phi of the charge air, the charge air temperature TLL, the start of injection SB and the rail pressure pCR.

The optimizer 21 determines the best possible quality measure by finding a minimum by calculating a first quality measure at a first point in time, varying the injection system setpoints and the gas path setpoints, and using these to forecast a second quality measure within the prediction horizon. On the basis of the deviation of the two quality measures from one another, the optimizer 21 then defines a minimum quality measure and sets this as decisive for the internal combustion engine. For the example shown in the figure, this is the set rail pressure pCR (SL) for the injection system. The set rail pressure pCR (SL) is the reference variable for the subordinate rail pressure control loop 22. The manipulated variable of the rail pressure control loop 22 corresponds to the PWM signal to act on the suction throttle. The optimizer 21 indirectly determines the target gas path values for the gas path. In the example shown, these are a lambda setpoint LAM (SL) and an EGR setpoint AGR (SL) for specifying the two subordinate control loops 23 and 24. The measured variables MESS are to be understood as meaning both directly measured physical variables and auxiliary variables calculated from them. In the example shown, the actual lambda value LAM (IST) and the actual EGR value AGR (IST) are read in. The manipulated variables of the internal combustion engine are combined with the reference symbol SG. This includes both the continuous manipulated variables with a continuous setting range and the discrete manipulated variables with discrete setting values. Continuous manipulated variables can be adjusted steplessly between a minimum and maximum value, for example the start of injection and the end of injection with which the injector (Fig. 1: 7) is directly acted upon. Discrete manipulated variables with discrete setting values can only be set in stages in fixed values, for example cylinder deactivation.

FIG. 3 shows a block diagram with the operating situation BS of the internal combustion engine as the input variable and the quality measure as the output variable, here referred to as the post-optimized quality measure J (NA). A pre-optimization 25, a quantization 26 and a post-optimization 27 are shown within the block diagram. In a first step, the pre-optimization 25 calculates a pre-optimized quality measure J (VO) in which the discrete manipulated variables with discrete setting values are interpreted as continuous manipulated variables with a continuous setting range. An example of a discrete manipulated variable is the pre-injection, which can only be activated or deactivated. By using pre-injection, the peak pressure of the combustion can be reduced significantly. In addition, when the pre-injection is activated, all other combustion parameters such as NOx emissions or the number of particles also change. The internal combustion engine is measured once with activated preinjection and once with deactivated preinjection. This results in two separate combustion models. When calculating the pre-optimized quality measure J (VO), the optimizer interpolates intermediate values, which means that the preinjection status activated or deactivated is artificially converted into a continuous input variable through the interpolation between these two combustion models. This variable is then used continuously in the pre-optimization 25. In FIG. 3, these continuous manipulated variables are designated as SG (k). The pre-optimized quality measure J (VO) is a purely internal calculation variable which has no access to the actuators of the internal combustion engine. In other words: the pre-optimized quality measure J (VO) is access-free and is not connected to the internal combustion engine.

In a second step, 26 new discrete manipulated variables SG (new) are calculated from the continuous manipulated variables SG (k) via the quantization. For the preinjection, a fixed assignment to preinjection is activated again in the quantization, or preinjection is deactivated. The quantization 26 offers the advantage that, for example, the variable valve control is set to three discrete values, namely minimum, mean value and maximum, for example 450 °, 495 ° and 540 ° crankshaft angle. This considerably reduces the computational effort in the subsequent determination of the post-optimized quality measure. During the quantization 26, the calculated values are also stabilized via optional hysteresis bands. In a third step, the new discrete manipulated variables SG (new) and the operating situation are merged and a post-optimized quality measure J (NA) is calculated by the optimizer. When calculating the post-optimized quality measure J (NA), the new discrete manipulated variables SG (new) are not changed. In this respect, these are not a degree of freedom when calculating the post-optimized quality measure J (NA). In the post-optimization, the actually continuous manipulated variables are adapted to the course specified from the quantization, for example the pre-injection. In other words: In the case of post-optimization, the manipulated variables that are actually described by continuous manipulated variables are varied. The post-optimized quality measure J (NA) corresponds to the minimum quality measure J (min), which the optimizer uses as is set significantly for the operating point of the internal combustion engine (1), so the internal combustion engine is switched on.

The method is shown in a program flow chart in FIG. After the initialization at S1, it is checked at S2 whether the start process has ended. If this is still running, query result S2: no, it branches back to point A. If the starting process has ended, the operating situation of the internal combustion engine is recorded at S3. The operating situation is defined by the engine speed nIST, the charge air temperature TLL, the charge air pressure pLL, etc. At S4, the optimizer subroutine is called and the initial values, for example the start of injection, are generated at S5. In steps S6 to S8, the subroutines pre-optimization, quantization and post-optimization are called up one after the other. These subroutines are described in connection with FIGS. The post-optimized quality measure calculated in the post-optimization subroutine is set as the minimized quality measure J (min), which determines the operating point of the internal combustion engine. Subsequently, at S10 it is checked whether an engine stop has been initiated. If this is not the case, query result S10: no, the system branches back to point B. Otherwise the program schedule is ended.

In FIG. 5, the pre-optimization subroutine is shown as a program flow chart. At S1, a first quality measure J1 (VO) of the pre-optimization is calculated using equation (1). An essential feature here is that when calculating the first quality measure J1 (VO), in addition to the continuous manipulated variables with a continuous manipulating range, the discrete manipulated variables with discrete values are interpreted as continuous manipulated variables via interpolation. At S2 a running variable i is set to zero. The initial values are then changed at S3 and calculated as new setpoint values for the manipulated variables. At S4 the running variable i is increased by one. On the basis of the new setpoint values, a second quality measure J2 (VO) of the pre-optimization within the prediction horizon, for example for the next 8 seconds, is then forecast at S5. At S6, the second quality measure J2 (VO) is subtracted from the first quality measure J1 (VO) and compared with a limit value GW. The further progress of the quality measure is checked by calculating the difference between the two quality measures. Alternatively, the comparison of the running variable i with a limit value iGW is used to check how often an optimization has already been carried out.

In this respect, the two limit value considerations are a termination criterion for further optimization. If further optimization is possible, query result S6: no, the system branches back to point A. Otherwise, the optimizer will use the The second quality measure J2 (VO) is output as a pre-optimized quality measure J (VO) together with the manipulated variables calculated in the process, and the program returns to the main program in FIG. The pre-optimized quality measure J (VO) is a pure arithmetic variable, that is, the calculated injection system target values, the calculated gas path target values and the calculated manipulated variables are not applied to the internal combustion engine by the optimizer.

The quantization subroutine is shown in FIG. The pre-optimized quality measure J (VO) with the associated manipulated variables is read in at S1. Then those manipulated variables are discretized with original discrete setting values. At S2, this takes place on the basis of corresponding threshold values with a hysteresis band. Oscillating calculation values are avoided via the hysteresis band. Instead of a hysteresis band, other logics can also be used which prevent rapid switching, for example a time control. The new discrete manipulated variables SG (new) are then output at S3 and a return is made to the main program in FIG.

In FIG. 7, the post-optimization subroutine is shown as a program flow chart. A post-optimized quality measure is determined from the operating situation of the internal combustion engine and the new discrete manipulated variables SG (new) via the post-optimization subroutine. When calculating the post-optimized quality measure, the new discrete manipulated variables are not updated. At S1, a first quality measure J1 (NA) of the post-optimization is calculated using equation (1). At S2 a running variable i is set to zero. The initial values are then changed at S3 and calculated as new setpoint values for the manipulated variables. At S4 the running variable i is increased by one. On the basis of the new setpoint values, a second quality measure J2 (NA) of the post-optimization within the prediction horizon, for example for the next 8 seconds, is then forecast at S5. At S6, the second quality measure J2 (VO) is subtracted from the first quality measure J1 (VO) and compared with a limit value GW. The further progress of the quality measure is checked by calculating the difference between the two quality measures. Alternatively, the comparison of the running variable i with a limit value iGW is used to check how often an optimization has already been carried out. In this respect, the two limit value considerations are a termination criterion for further optimization. If further optimization is possible, query result S6: no, the system branches back to point A. Otherwise, at S7 the optimizer outputs the second quality measure J2 (VO) as the minimum quality measure J (min) and returns to the main program in FIG. The two FIGS. 8 and 9 show, in a comparison, the course of selected variables over time in seconds. The variables shown are: the variable valve control VVT in degrees crankshaft angle, the start of injection SB in degrees before top dead center (TDC), the combustion pressure pZYL in the cylinder and the engine speed nMOT. For the combustion pressure pZYL, the maximum permissible combustion pressure pMAX is also shown as a dashed line. On the left-hand half of the drawing, these sizes are shown when the previous optimization is applied, while these sizes are shown on the right-hand half of the drawing when the invention is used. The representation in FIG. 8 and FIG. 9 is based on a step-like increasing setpoint torque as the input variable. First, the variables according to FIG. 8 are described. In a first step, the optimizer uses the pre-optimization to calculate a pre-optimized quality measure based on the operating situation. In this calculation, the discrete manipulated variables with discrete setting values are interpreted as continuous manipulated variables with a continuous setting range. For the variable valve control VVT there is a continuous curve with any intermediate values over the entire time range. For the VVT actuator for controlling the variable valve with three defined actuator positions, however, such a course cannot be represented. A calculated start of injection SB and the corresponding cylinder pressure pZYL correspond to the pre-optimized quality measure. The maximum value pMAX is maintained for the cylinder pressure pZYL. The manipulated variable results in an increasing engine speed nMOT in the period under consideration. FIG. 9 is described below. The VVT curve shown corresponds to the curve after quantization. It becomes clear here that, in contrast to the illustration in FIG. 8, the VVT curve shows only three discrete values, namely 450 °, 495 ° and 540 ° crankshaft angles. The advantage is that the VVT actuator can be controlled with only three values, which significantly reduces the computational effort. The post-optimized quality measure is calculated from the VVT curve based on the operating situation of the internal combustion engine. The course of the start of injection SB and the cylinder pressure pZYL, which in this case also remains below the maximum value pMAX, correspond to this. Reference number

1 internal combustion engine

2 fuel tank

3 low pressure pump

4 suction throttle

5 high pressure pump

6 rail

7 injector

8 individual memories

9 Rail pressure sensor

10 Electronic control unit

11 exhaust gas turbocharger

12 intercoolers

13 throttle valve

14 Junction point

15 inlet valve, variably controllable

16 exhaust valve, variably controllable

17 EGR actuator (EGR: exhaust gas recirculation)

18 turbine bypass valve

19 Combustion model

20 gas path model

21 optimizers

22 Rail pressure control circuit

23 Lambda control loop

24 EGR control circuit

25 Pre-optimization

26 quantization

27 Fine tuning

Claims

1. A method for model-based control and regulation of an internal combustion engine (1), in which an optimizer (21) calculates a pre-optimized quality measure (J (VO)) as a function of the operating situation (BS) in a first step pre-optimized quality measure (J (VO)), discrete manipulated variables with discrete setting values are interpreted as continuous manipulated variables (SG (k)) with a continuous setting range, in which in a second step these continuous manipulated variables (SG (k)) are quantized and as new discrete manipulated variables (SG (new)) can be set with discrete setting values, in which the optimizer (21) in a third step a post-optimized quality measure (J (NA)) depending on the new discrete manipulated variables (SG (new)) and the operating situation (BS) of the internal combustion engine (1) is calculated and the optimizer (21) sets the post-optimized quality measure (J (NA)) as decisive for the operating point of the internal combustion engine (1).

2. The method according to claim 1, characterized in that the pre-optimized quality measure (J (VO)) is determined in that the injection system setpoints for controlling the injection system actuators are calculated using a combustion model (19) by using a gas path model (20) the gas path setpoints for controlling the gas path actuators are calculated and in that the continuous manipulated variables (SG (k)) are calculated via interpolation from the discrete setting values of the discrete manipulated variables.

3. The method according to claim 2, characterized in that the pre-optimized quality measure (J (VO)) is set without access for the actuators of the internal combustion engine.

4. The method according to claim 3, characterized in that in the second step the continuous manipulated variables (SG (k)) are quantized via switching thresholds together with hysteresis.

5. The method according to claim 1, characterized in that the post-optimized quality measure (J (NA)) is determined in the third step by calculating the injection system setpoint values for controlling the injection system actuators via the combustion model (19), via the gas path model ( 20) the gas path setpoints for controlling the gas path actuators are calculated and calculated by the optimizer (21) using the change in the injection system setpoints and the gas path setpoints with constant new discrete manipulated variables (SG (new)) by finding a minimum within a prediction horizon will.