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

Abstract

Method for model-based control of an internal combustion engine (1), - in which, during normal operation, the injection system setpoints for controlling the injection system actuators are calculated via an adaptable combustion model (3) depending on the specified values ​​for the operation of the internal combustion engine (1), - in which an optimizer (6) calculates a quality measure at least as a function of the injection system setpoints, - the optimizer (6) minimizes the quality measure by changing at least the injection system setpoints within a prediction horizon and - in which the injection system setpoints for adjusting the operating point of the internal combustion engine (1) are specified by the optimizer (6) on the basis of the minimized quality measure (J(MIN)), characterized in that - in stationary operation, the system cyclically switches from normal operation to exploration mode, - in exploration operations, an exploration quality measure (J(EXP)) is calculated as a function of the combustion model (3) and its variance (VAR), - the control variables of the internal combustion engine resulting from the exploration quality measure (J(EXP)) are specified for setting the operating point of the internal combustion engine, and - the combustion model (3) is adjusted based on the operating parameters of the internal combustion engine and the system switches back to normal operation.

Description

[0001] The invention relates to a method for model-based control and regulation of an internal combustion engine according to the preamble of claim 1.

[0002] The behavior of an internal combustion engine is largely determined by an engine control unit (ECU) based on a desired power output. For this purpose, corresponding characteristic curves and maps are applied to the ECU software. These are used to calculate the engine's control parameters from the desired power output, particularly a target torque, such as the injection timing and the required rail pressure. These characteristic curves / maps are populated with data by the engine manufacturer during a test bench run. However, the large number of these characteristic curves / maps and their interactions with each other result in a significant tuning effort.

[0003] In practice, attempts are therefore made to reduce the coordination effort by using mathematical models. For example, DE 10 2006 004 516 B3 describes a Bayesian network with probability tables for determining an injection quantity, and US 2011 / 0 172 897 A1 describes a method for adapting the injection timing and injection quantity via combustion models using neural networks. Since only trained data is used, this data must first be learned during a test bench run.

[0004] From DE 103 11 269 A1, a method for controlling a piezoelectric element is known. In this method, a control variable is used. To determine the control variable, a quality factor for the transient response of a state variable of the piezoelectric element to be controlled, which depends on the control variable, is minimized.

[0005] From DE 10 2005 018 272 A1, a method and a device for operating an internal combustion engine are known. The method is intended to enable the precise determination of the cross-sectional area through which a component located in a gas channel flows. The internal combustion engine has an adjustable component through which a gas flows, and the position of this component influences the flow of the gas. Specific values ​​of the component are determined using models.

[0006] German patent DE 10 2018 001 727 A1 discloses a method for model-based control of an internal combustion engine, in which the injection system setpoints for controlling the injection system are calculated using an adaptable combustion model. The combustion model comprises a first Gaussian process model for representing a basic grid and a second Gaussian process model for representing adaptation data points. The data values ​​for the first and second Gaussian process models are determined during a Design of Experiments (DoE) test bench run of the complete engine and during a single-cylinder test bench run. The adaptation procedure is implemented such that a current adaptation data point is transferred to the second Gaussian process model if the adaptation data point lies within the current confidence interval. The confidence interval is calculated from the variance.If the adaptation data point lies outside the confidence interval, previously stored adaptation data points are iteratively removed from the second Gaussian process model until the current adaptation data point lies within the modified confidence interval. Test bench trials have shown that, in less frequently used operating ranges, the adaptation can cause an excessive change to the second Gaussian process model and thus to the combustion model.

[0007] The invention is therefore based on the objective of further developing the previously described method for adapting the combustion model with regard to improved quality and, in addition, simplifying the data acquisition.

[0008] This problem is solved by the features of claim 1 and claim 7. The embodiments are described in the dependent claims.

[0009] In the inventive method according to claim 1, during steady-state operation, the system cyclically switches from normal operation to exploration mode, whereby in exploration mode an exploration quality measure is calculated as a function of the combustion model and its variance. Furthermore, the exploration quality measure is used as the determining factor for setting the operating point of the internal combustion engine, and the combustion model is adapted via the second Gaussian process model based on the operating parameters of the internal combustion engine. Afterwards, the system switches back to normal operation.

[0010] The central idea of ​​the invention is to systematically utilize the knowledge of variance in exploration operations. By additionally considering the variance, those operating points are identified where a new measurement could lead to the greatest possible improvement in future operating points after the adaptation of the second Gaussian process model.

[0011] The exploration goodness-of-fit measure is calculated by minimizing a membership function. This function is determined by subtracting an "expected improvement" function from the expected value of the combustion model. Additionally, the procedure assesses the variance by excluding operating ranges with high variance through a limit test. Since areas of the combustion model with very high uncertainty are not considered, the adaptation is effective in the typical operating range of the internal combustion engine and not in extreme, irrelevant boundary regions. The "expected improvement" function is calculated by comparing the expected value of the combustion model and its variance with a reference value, such as a minimum fuel consumption. The reference value corresponds to a measured data value or was previously determined during normal operation using the minimized goodness-of-fit measure.

[0012] One option provides that, before activation, the target values ​​calculated using the exploration quality measure are checked against inequality constraints and either blocked or enabled depending on whether the value of the target value leads to a violation of the inequality constraints. Inequality constraints include, for example, the maximum combustion pressure. Considering these constraints determines the degree of confidence in the calculated operating limits.

[0013] In the inventive method according to claim 7, the model of the overall behavior of the internal combustion engine during a test bench run is determined by acquiring the data in an exploratory operation according to the procedure described above, based on an expected improvement, a membership function, and a variance check. Optionally, compliance with equation and inequality conditions can also be taken into account.

[0014] The figures show a preferred embodiment. They show: Fig. 1 a model-based system diagram, Fig. 2 a block diagram, Fig. 3 a diagram of the combustion model, Fig. 4 a diagram of the EI function, Fig. 5 a diagram evaluation of the variance, Fig. 6 a diagram of the membership function, Fig. 7 a diagram of the inequality condition, Fig. 8 a diagram evaluation of the variance and Fig. 9 a program schedule.

[0015] The Fig. Figure 1 shows a model-based system diagram for the control and regulation of an internal combustion engine 1 via an electronic control unit 2. The input variables of the electronic control unit are: a first library Biblio1, a second library Biblio2, measured variables MESS, and the collective reference symbol EIN, which represents the other setpoint variables, for example, a target torque or a target speed specified by an operator. The first library, Biblio1, identifies the operation of the internal combustion engine according to the IMO's MARPOL (Marine Pollution) emission class or according to the EU IV / Tier 4 final emission class. The second library, Biblio2, identifies the internal combustion engine type and a maximum mechanical component load, for example, the maximum combustion pressure or the maximum speed of the exhaust gas turbocharger. The input variable MESS identifies both directly measured physical quantities and auxiliary variables calculated from them.The output variables of the electronic control unit are: the setpoints for the subordinate control loops, the injection start (SB), and the injection end (SE). The subordinate control loops shown are a rail pressure control loop (7), a lambda control loop (8), and an EGR control loop (9). Within the electronic control unit (2) are a combustion model (3), an adaptation (4), a gas path model (5), and an optimizer (6). Both the combustion model (3) and the gas path model (5) represent the system behavior of the internal combustion engine as mathematical equations. The combustion model (3) statically represents the combustion processes. In contrast, the gas path model (5) represents the dynamic behavior of the air intake and exhaust gas flow. The combustion model (3) includes individual models, for example, for NOx and soot formation, exhaust gas temperature, exhaust gas mass flow, and peak pressure.These individual models, in turn, depend on the boundary conditions in the cylinder and the injection parameters. Combustion model 3 is determined using a reference internal combustion engine in a DoE (Design of Experiments) test bench run. During the DoE test bench run, operating parameters and control variables are systematically varied with the aim of modeling the overall behavior of the internal combustion engine as a function of engine parameters and environmental boundary conditions. Combustion model 3 is supplemented by adaptation 4. The aim of adaptation is to reduce the production variation of an internal combustion engine.

[0016] After activation of the internal combustion engine 1, the optimizer 6 first reads the emission class from the first library, Biblio1, and the maximum mechanical component loads from the second library, Biblio2. The optimizer 6 then evaluates the combustion model 3 with regard to the target value, for example, the target torque, the emission limits, the environmental boundary conditions, for example, the humidity of the charge air, the operating situation of the internal combustion engine, and the adaptation data points. The operating situation is defined in particular by the engine speed, the charge air temperature, and the charge air pressure. The function of the optimizer 6 is then to evaluate the injection system target values ​​for controlling the injection system actuators and the gas path target values ​​for controlling the gas path actuators. In doing so, the optimizer 6 selects the solution that minimizes a specific efficiency measure.The quality measure is calculated as the integral of the squared target-actual deviations within the prediction horizon. For example, in the form: J=∫[w1(NOx(SOLL)−NOx(IST)]2+[w2(M(SOLL)−M(IST)]2+[w3(….)]+…

[0017] Here, w1, w2, and w3 are weighting factors, and M(SOLL) corresponds to the specified target moment. As is known, nitrogen oxide emissions result from the charge air humidity, charge air temperature, injection timing (SB), and rail pressure (pCR). Adaptation 4 intervenes in the actual values, for example, the NOx value or the exhaust gas temperature.

[0018] The performance measure is minimized by the optimizer 6 calculating a first performance measure at an initial time point, varying the injection system setpoints and the gas path setpoints, and using these variations to predict a second performance measure for the system behavior within the prediction horizon. From the deviation of the two performance measures, the optimizer 6 then defines a minimum performance measure and sets this as the decisive factor for the internal combustion engine. For the example shown in the figure, these are the target rail pressure pCR(SL), the injection start SB, and the injection end SE for the injection system. The target rail pressure pCR(SL) is the reference variable for the subordinate rail pressure control loop 7. The manipulated variable of the rail pressure control loop 7 corresponds to the PWM signal for actuating the intake throttle. With the injection start SB and injection end SE, the injector is directly actuated for fuel injection.For the gas path, optimizer 6 indirectly determines the gas path setpoints. In the example shown, these are a lambda setpoint LAM(SL) and an EGR setpoint AGR(SL) for the subordinate lambda control loop 8 and the subordinate EGR control loop 9. The manipulated variables of the two control loops 8 and 9 correspond to the signal TBP for controlling the turbine bypass, the signal AGR for controlling the EGR actuator, and the signal DK for controlling the throttle valve. The feedback measured variables MESS are read by the electronic control unit 2. The measured variables MESS include both directly measured physical quantities and auxiliary variables calculated from them. In the example shown, the actual lambda value and the actual EGR value are read.

[0019] The Fig. Figure 2 shows a block diagram illustrating the interaction of the two Gaussian process models for adapting the combustion model. Gaussian process models are familiar to those skilled in the art, for example from DE 10 2014 225 039 A1 or DE 10 2013 220 432 A1. In general, a Gaussian process is defined by a mean value function and a covariance function. The mean value function is often assumed to be zero or a linear / polynomial curve is introduced. The covariance function describes the relationship between arbitrary points. A first function block 10 contains the DoE (Design of Experiments) data for the complete engine. This data is determined for a reference internal combustion engine during a test bench run by recording all variations of the input variables across the entire operating range of the engine in steady-state driving conditions. This data characterizes the behavior of the internal combustion engine in steady-state driving conditions with high accuracy.A second functional block 11 contains data acquired on a single-cylinder test bench. This single-cylinder test bench allows for the setting of operating ranges, such as high geodetic altitude or extreme temperatures, that cannot be tested during a DoE (Design of Experiments) test bench run. These limited measurement data serve as the basis for parameterizing a physical model, which roughly accurately represents the global combustion behavior in the form of trend information, reference numeral 12. The physical model roughly depicts the behavior of the internal combustion engine under extreme boundary conditions. Extrapolation is used to complete the physical model, so that a normal operating range is roughly accurately described. In the . Fig. 2 is the extrapolation-capable model labeled with reference symbol 13. From this, the first Gaussian process model 14 (GP1) is generated to represent a basic grid.

[0020] The merging of the two sets of data points forms the second Gaussian process model 15. This means that the operating ranges of the internal combustion engine, which are described by the DoE data, are also defined by these values, and the operating ranges for which no DoE data is available are represented by data from the physical model. Since the second Gaussian process model is adapted during operation, it also serves to represent the adaptation data points. In general, the following applies to the combustion model 3 overall: E(x)=GP1+GP2

[0021] Here, GP1 corresponds to the first Gaussian process model for representing the basic grid, GP2 to the second Gaussian process model for representing the adaptation data points, and E(x) to the combustion model. The combustion model is the input variable for the optimizer, for example, an actual NOx value or an actual exhaust gas temperature value. The double arrow in the figure represents two information paths. The first information path denotes the provision of the basic grid data from the first Gaussian process model 14 to the combustion model. The second information path denotes the back-adaptation of the first Gaussian process model 14 via the second Gaussian process model 15.

[0022] The block diagram is supplemented by the optimizer 6, an exploration unit 16, and a switch S. Both the optimizer 6 and the exploration unit 16 have access to the combustion model 3 with the first and second Gaussian process models. In normal operation, the switch S is in position 1. In position 1, the input parameters for the internal combustion engine 1 are specified by the optimizer 6 via the minimized efficiency factor J(MIN). The switch S changes to position 2 when steady-state operation is established and a time stage has elapsed. In position 2, the exploration unit 16 determines the input parameters for the internal combustion engine 1 via the exploration efficiency factor J(EXP). The input parameters are those specified in the Fig. The two quantities shown are used to define an operating point for the internal combustion engine 1, for example, the injection start SB or the target rail pressure pCR(SL). The measured parameters of the internal combustion engine 1 ( Fig. 2: MESS) are fed back to the second Gaussian process model 15 via a feedback path and form the basis for the adaptation of the second Gaussian process model. In the Fig. Section 2, designated with reference numeral 10A, represents an alternative. In this alternative, the DoE data are determined on the test bench analogously to the procedure for calculating the exploration quality measure, including the inequality constraints. This alternative offers the advantage of a shortened test bench trial.

[0023] Further explanation regarding the determination of the exploration quality measure J(EXP) is provided based on the Fig. 3, Fig. 4, Fig. 5 to Fig. 6. The Fig. Figure 3 shows a component E1(x) of the combustion model as a function of a manipulated variable x. For clarity, in the following description, the manipulated variable x corresponds to the injection timing SB, and the component E1(x) of the combustion model 3 represents fuel consumption. The goal is to achieve minimal fuel consumption while adhering to emission targets and other constraints. Within the diagram, the expected value 17 is represented by a solid line, and the variance VAR, a measure of uncertainty (e.g., the confidence interval), is shown as a hatched area, indicating a 95% probability that the actual system behavior lies within this uncertainty. Points A, B, and C correspond to measured data values, i.e., real data values. The expected value 17, in turn, was calculated within the combustion model. Under normal operating conditions, the optimizer determines the operating point of the internal combustion engine using the minimized performance measure J(min).To set the minimum fuel consumption, the optimizer determines the expected value during normal operation at which this requirement is met.

[0024] In exploratory mode, unlike normal operation, the variance is also considered. The first step involves finding the minimum consumption. When evaluating the component E1(x) of the combustion model and its variance VAR, one can see in the Fig. Three further points where minimum fuel consumption could occur are identified, for example, at an abscissa value of x=0.55 or at the outer edges, here: data values ​​(0 / -1) or (1 / -1). The idea of ​​the exploration is now to check whether lower fuel consumption is actually possible at these points. Ultimately, one drives to points deviating from the previous minimum to test whether lower fuel consumption is indeed possible there. In the Fig. Figure 3 shows an example of test point D. In a second step of the exploration, a function EI ("Expected Improvement") is calculated. Fig. Figure 4 shows this function EI(x) over the quantity x. The function El(x) is calculated by substituting the range (0, 1) of the quantity x into the Fig. The route passes through 3 points, and for each point, its expected value and variance relative to point B are evaluated. Point B is a measured data value that serves as a reference value. From the Fig. 4. For test point D, an expected improvement of approximately -0.13 is obtained with respect to the reference value, i.e., point B. For the data values ​​A, B, and C from the Fig. 3 results in the Fig. 4 an El value of zero with respect to the optimum at point B.

[0025] In a third step, the variance of component E1(x) of the combustion model is evaluated. This corresponds to the representation of the Fig. Figure 5, where the variance VAR(x) is plotted against the quantity x. Regions with very high variance are excluded. The goal is to exclude regions where the internal combustion engine is not operating and to remain within the range of the usual solution.

[0026] In the Fig. Figure 5 shows an example of the maximum permissible variance value, MAX. The hatched areas indicate where the variance, VAR(x), exceeds this maximum value.

[0027] In a fourth step, an affiliation function (AF) is determined. This is in the Fig. Figure 6 shows the membership function. The membership function is determined from the difference between the expected value 17 of the combustion model of the Fig. 3 and the function EI(x) of the Fig. 4. The behavior of the membership function AF(x) yields a minimum consumption, point H, for an injection start at x=0.55. In other words, the membership function AF(x) is expected to show the greatest possible improvement in consumption. The selected operating point H is then set as the target value for the internal combustion engine. In the example shown, the selected operating point H, or the resulting manipulated variables, corresponds to the exploration quality measure J(EXP). More generally, the exploration quality measure J(EXP) can also be defined by other criteria. The subsequent procedure then corresponds to the model adaptation method known from DE 10 2018 001 727 A1; that is, the new point is incorporated into the second Gaussian process model based on the measured variables MESS and set to normal mode ( Fig. 2: S=1) switched back.

[0028] In the Fig. Figure 7 presents an optional supplement to the exploration operation. This supplement improves safety by incorporating equation and inequality conditions. An equation condition corresponds to a fixed value, for example, NOx = 10 g / kWh. An inequality condition corresponds to a range, for example, NOx < 10 g / kWh or the measured combustion pressure must be less than the maximum combustion pressure. An inequality condition h(x) is shown as a function of the quantity x, here: the start of injection, and a variance VAR with a 95% confidence interval is represented as a hatched area. Three data points E, F, and G are plotted. The inequality condition can be evaluated at the data points included in the model; for the data points between these, the combustion model is interpolated with the corresponding uncertainty (variance). Additionally, the inequality condition h(x) must be less than zero.In other words, the combustion pressure calculated in the combustion model must be less than the maximum combustion pressure stored in the Biblio2 library. Therefore, the area above the ordinate value of zero, corresponding to data point F, is not permissible. In a second step, the variance is then evaluated, see [reference]. Fig. 8, and a probability function P(x) is calculated from the variance and the expected value. The probability function describes the probability that the restriction is violated. In the Fig. Figure 8 shows a maximum value MAX. Larger values ​​of the probability function P(x) than the maximum value MAX are excluded. The hatched areas therefore correspond to the impermissible areas. For example, if the value of the membership function AF ( Fig. 6) determined point H, i.e. the point of minimum consumption, within the permissible variance range of the Fig. 8, the exploration quality measure J(EXP) is derived from this and applied to the internal combustion engine. If, however, point H lies in one of the three impermissible areas of the Fig. 8, so a new point is sought that is within the permissible range of the Fig. 8 is located.

[0029] In the Fig. Figure 9 illustrates the procedure in a program flowchart. After the program starts, a query at S1 checks whether the conditions for changing the operating mode are met. The conditions are met when the internal combustion engine is in a steady state and a time interval has elapsed. The exploration mode is set cyclically based on this time interval. A steady state exists, for example, at a constant engine speed or constant torque. If the condition at S1 is not met (query result: no), normal operation remains set at S2. In normal operation, the optimizer calculates the minimized efficiency and sets the resulting target values ​​as the relevant ones for the internal combustion engine. At S3, a check is performed to see if an engine stop has been initiated. If so (query result: yes), the program flowchart terminates. Otherwise, the program returns to point A.If the condition at S1 is met (query result: yes), then exploration mode is activated at S4. Subsequently, the EI (Expected Improvement) function is calculated at S5. The EI function is calculated using the probability that the expected value ( Fig. 3: 17) of the combustion model and its variance below the previous optimum, i.e. the reference value ( Fig. 6: Point B). At S6, the variance is evaluated by comparing it to a maximum permissible value. Areas with very high variance are excluded. The goal is to eliminate areas where the internal combustion engine is not operated and to remain within the range of the usual solution. At S7, the membership function AF is calculated from the difference between the expected value of the combustion model and the function EI (expected improvement). The operating point that is likely to meet the requirement, e.g., minimum fuel consumption, is then ultimately determined using the membership function AF. At S8, it is checked whether the inequality conditions are set. If the inequality conditions are not set, the program flow continues at S11. Otherwise, the program section with steps S9 and S10 is executed.Steps S9 and S10 correspond to a safety check, for example, whether the minimum consumption calculated for exploration operations or the exploration quality measure can be achieved within permissible values ​​of the control variables, in particular a maximum combustion pressure. Accordingly, in S9, an inequality function h(x) and its variance are calculated. Additionally, it is checked which ranges of the inequality function h(x) exceed a predefined value. In S10, the variance VAR is then evaluated by calculating a probability function P(x). The probability function P(x) is calculated from the expected value and the variance of the inequality function h(x). The goal is to exclude larger values ​​of the probability function P(x) than a maximum value. In the example shown, it is assumed that the data value calculated via the membership function ( Fig.6: Point H) is permissible. The program flow then continues with S11 and sets the exploration quality measure as decisive for the operating point of the internal combustion engine. Decisive means that the control variables resulting from the exploration quality measure, such as the target rail pressure or the start of injection, etc., of the internal combustion engine are specified. At S12, the operating parameters of the internal combustion engine are recorded, transferred to the second Gaussian process model GP2 at S13, and the second Gaussian process model GP2 is adapted. Subsequently, normal operation is restored at S14, and the program branches back to point A.

Claims

[1] Method for model-based control of an internal combustion engine (1), - in which, during normal operation, the injection system setpoints for controlling the injection system actuators are calculated via an adaptable combustion model (3) depending on the specified values ​​for the operation of the internal combustion engine (1), - in which an optimizer (6) calculates a quality measure at least as a function of the injection system setpoints, - the optimizer (6) minimizes the quality measure by changing at least the injection system setpoints within a prediction horizon and - in which the injection system setpoints for adjusting the operating point of the internal combustion engine (1) are specified by the optimizer (6) based on the minimized quality measure (J(MIN)), characterized by , that - in stationary operation, the system cyclically switches from normal operation to exploration mode, - in exploration operations, an exploration quality measure (J(EXP)) is calculated as a function of the combustion model (3) and its variance (VAR), - the control variables of the internal combustion engine resulting from the exploration quality measure (J(EXP)) are specified for setting the operating point of the internal combustion engine, and - the combustion model (3) is adjusted based on the operating parameters of the internal combustion engine and the system switches back to normal operation. [2] Method according to claim 1, characterized by , that the exploration quality measure (J(EXP)) is determined by finding the minimum of a membership function (AF), wherein the membership function (AF) is determined by subtracting an Expected Improvement (EI) function from the expected value (17) of the combustion model (3). [3] Method according to claim 2, characterized by, that the variance (VAR) of the combustion model (3) is assessed with respect to a limit value and operating ranges of high variance (VAR>MAX) are not taken into account when calculating the membership function (AF). [4] Method according to claims 2 and 3, characterized by , that the Expected Improvement function is calculated by comparing the expected value (17) of the combustion model and its variance with a reference value, where the reference value was previously recorded as a measured data value or is determined in normal operation based on the minimized goodness-of-fit measure. [5] Method according to claim 1, characterized by , that the exploration quality measure (J(EXP)) is checked using inequality conditions and the exploration quality measure (J(EXP)) is either set as authoritative for setting the operating point of the internal combustion engine or a new exploration quality measure (J(EXP)) is calculated. [6] Method according to claim 5, characterized bythat the inequality conditions are calculated from the specified values ​​for the operation of the internal combustion engine. [7] Method for determining the overall behavior of an internal combustion engine, in which data points for the combustion model (3) according to one of claims 1 to 6 are determined in a test bench run in an exploration operation.

Images