DE102006004516B3 - Bayes network for controlling and regulating internal combustion engine, has measuring variables that are assigned probabilities, and correction value that is calculated for correcting control variable of controller using correction table - Google Patents

Abstract

The network has measuring variables (MG1, MG2) that are assigned probabilities by a table. The probabilities are assigned factors (F1, F2) using operational instructions. The factor (F1) is weighed against the other factor (F2) using weight-operational instructions. A correction value (KORR) is calculated for correcting a control variable of a controller (13) using a correction table. A weight factor is calculated depending on a third factor (F3) that is determined based on an operating time of an injector.

Description

The The invention relates to a Bayesian network for controlling and regulating a Internal combustion engine according to claim 1 and a Bayesian network for control and control of an internal combustion engine with at least one exhaust gas turbocharger according to claim 6.

The state of an internal combustion engine is determined via an electronic control unit as a function of input variables, for example a desired performance. Characteristics and maps are used to convert the input variables to the corresponding control variables. For example, in a common rail system, an operating point (speed, engine torque) is assigned a desired rail pressure via a characteristic map, which is then converted by a rail pressure regulator into a corresponding manipulated variable, in particular volumetric flow. To adapt to changing conditions a map-based feedforward control, map switching and correction variables are also provided. Such a system is for example from DE 101 57 641 A1 known.

To increase the performance of an internal combustion engine, exhaust gas turbochargers are provided. These can be designed as turbochargers with variable turbine geometry or as a switchable exhaust gas turbocharger, for example in a register charging. In the text below, an exhaust gas turbocharger whose geometry is changeable is referred to as a VTG exhaust gas turbocharger. In these systems too, the electronic control unit determines a manipulated variable from the input variables via characteristic curves and characteristic diagrams. When a register charging the electronic control unit determines from the speed of the permanently operated exhaust gas turbocharger as a manipulated variable an activation signal for a switching flap. About this then the switchable exhaust gas turbocharger is activated. Such a system is for example from the DE 103 08 075 A1 known.

In In practice, the characteristics and maps of the manufacturer of Internal combustion engine in the initial configuration with a standard parameter set stocked. These data are then used in bench tests customized. For this would have to each individual operating point of each map is measured and be adjusted. This is extremely expensive so often the originals Values unchecked become. In addition, a review of the functionality of the engine control designed as very difficult.

Of the The invention is therefore based on the object to show measures with which the coordination effort for an internal combustion engine with at least one exhaust gas turbocharger reduced can be.

The The object is achieved by the features of claim 1 and the features of claim 6 solved.

In both cases The invention provides a Bayesian network for controlling and regulating the Internal combustion engine before. The special feature of a Bayes network consists among other things, that each measure a probability table over all conditions is deposited.

From the DE 103 33 181 A1 is a Bayesian network-based expert system known in which the technical relationships and dependencies are mapped by an expert. The probability tables stored for the nodes (parameters / variables) are adapted from data to the technical system in a learning process. However, the Bayesian network is geared to the diagnosis of, for example, a vehicle.

Out of WO 2005/033809 A1 is also known a Bayesian network, in which at the beginning all nodes (parameters) are interconnected are what an initial assumed dependence between means all nodes. This made of knots and edges (connections) existing Bayes network represents a statistical model of the data thus described, ie the Probability tables. By analyzing the measurement data, connections, which are statistically irrelevant, are separated, causing the Bayesian network learns from data. The Bayesian network is used to analyze influencing variables a burning process in a gas turbine. The reference therefore provides no immediate indication of how to solve the problem.

In the first embodiment of the invention, a geodetic height-dependent correction of an injection quantity is provided. The first measured variable of the Bayes network corresponds to an ambient air pressure. The second measurand corresponds to an ambient temperature. For the purposes of the invention, measured variable means both directly measured quantities and derived quantities. The Bayesian network now provides that each measure is assigned a probability via a probability table, which is assigned a factor via an action instruction. Thereafter, the factors are weighted against each other, and the result is corrected via a correction table to correct the injection assigned quantity.

In the second embodiment the invention is the calculation of a control signal for an exhaust gas turbocharger executed. When a register is charged by the control signal, the switchable Exhaust gas turbocharger activated or deactivated. For a VTG turbocharger the angular position of the guide grid is interposed by the actuating signal open and closed changed. For one VTG exhaust gas turbocharger provides the Bayes net that an air-fuel ratio as well Engine speed over one probability table each and one instruction each the corresponding factors are assigned. After that, the weighted against each other. A supercharger speed is over a Probability table a corresponding probability from which then with the result of the weighting Manipulated variable for loading the exhaust gas turbocharger is determined. For a switchable turbocharger is complementary a time level with probability table setting the Manipulated variable provided.

The Advantages of the invention over consist of characteristic curve and map-controlled engine management systems in a much lower coordination effort, since the technical relationships graphically and graphically. In addition, the Probability tables are learned from data, causing the Effort is reduced again. The system is diagnosable and predictable, blurred Information can also be processed.

In In the drawings, as a first embodiment of a Bayesian network for calculating a correction value and as a second embodiment a Bayes network for determining a manipulated variable for an exhaust gas turbocharger shown. Show it:

1 a system diagram;

2 a block diagram, first embodiment;

3 a block diagram, second embodiment.

The 1 shows a system diagram of an internal combustion engine 2 with common rail system. The common rail system comprises the following components: a low-pressure pump 5 for pumping fuel from a fuel tank 4 , a variable suction throttle 6 for influencing the flow through the fuel volume, a high-pressure pump 7 to promote the fuel under pressure increase, a rail 8th as well as single memory 9 for storing the fuel and injectors 10 for injecting the fuel into the combustion chambers of the internal combustion engine 2 , To increase the performance of the internal combustion engine 2 is at least one exhaust gas turbocharger 3 intended. The exhaust gas turbocharger 3 can be designed as a VTG exhaust gas turbocharger or as a switchable exhaust gas turbocharger in a register charging.

The operation of the internal combustion engine 1 is controlled by an electronic control unit (ADEC) 12 certainly. The electronic control unit 12 includes the usual components of a microcomputer system, such as a microprocessor, I / O devices, buffers and memory devices (EEPROM, RAM). In the memory modules are those for the operation of the internal combustion engine 1 applied relevant operating data. This is calculated by the electronic control unit 12 from the input variables the output variables. In 1 The following input variables are shown by way of example: a rail pressure pCR, which is determined by means of a rail pressure sensor 11 is measured, an engine speed nMOT, a supercharger speed nATL the exhaust gas turbocharger 3 and an input ON. The input variable ON, for example, is the charge air pressure of the exhaust gas turbocharger 3 , an air-fuel ratio, an ambient air pressure, an ambient temperature and the temperatures of the coolants / lubricants as well as the fuel subsumes.

In 1 are the output variables of the electronic control unit 12 a signal PWM for controlling the suction throttle 6 , a signal ve for controlling the injectors 10 , an actuating signal SG for controlling the exhaust gas turbocharger 3 and an output OUT shown. The output variable OFF is representative of the other control signals for controlling and regulating the internal combustion engine 2 , The signal ve is representative of a power-determining signal, for example for an injection quantity. About the control signal SG, the guide grid of the VTG exhaust gas turbocharger is either opened or closed or activated or deactivated in a switchable exhaust gas turbocharger this.

In 2 is a Bayesian network 1 for controlling and regulating an internal combustion engine 2 shown in a first embodiment. The input variables of the Bayes network 1 are a first measured variable MG1 and second measured variable MG2. The illustrated Bayes network 1 is used for height correction. Therefore, the first measured variable MG1 corresponds to the ambient air pressure p0 and the second measured variable MG2 corresponds to the ambient temperature T0. The output of the Bayes network in this embodiment corresponds to a correction value KORR. The correction value KORR is used to correct a control signal RSG, for example injection quantity. The control signal RSG is via a regulator 13 determined from a control deviation dR. The output signal ve corresponds to the corrected controller manipulated variable RSG.

the Ambient air pressure p0 is over a first probability table WTB1 a first probability Assigned to W1. In the first probability table WTB1 are three pressure ranges scaled. In the example shown, the exact one Ambient air pressure p0 known, here: 850 ± 29 mbar. The probability, that the ambient air pressure p0 is in the range of 800 to 900 mbar, is therefore 100. The probability W1 is about a first action HAW1 a first factor F1 zugeodnet. The first instruction HAW1 contains three assignments. These stand for a higher one Injection quantity (plus), maintaining the injection quantity (holding) and Reduce the injection quantity (minus). For the example, this means that the first factor F1 is a higher one Injection amount due to the low pressure request.

Of the Ambient temperature T0 is over a second probability table WTB2 a second probability Assigned to W2. The second probability table WTB2 contains three temperature ranges. For the given example is the ambient temperature is 35 ± 1.4 ° C. The probability, that this is in the range of 30 to 80 ° C, therefore, is 100%. The second probability W2 is a second action HAW2 assigned a second factor F2. The second instruction HAW2 contains The same assignments as the first instruction HAW1. For the example this means that the injection quantity due to the high ambient temperature is reduced.

The first factor F1 and the second factor F2 are the input variables for a weighting instruction GWHA. In 2 the weighting action instruction GWHA is supplied as a further input variable, a third factor F3. This is determined from an operating time BZT of the injector. The operating time BZT is assigned a third probability W3 via a third probability table WTB3. In the example shown, the accumulated operating time of the injector is 250 ± 140 operating hours. The probability that this operating time is in the range of zero to five hundred operating hours is therefore 100. The third factor W3 is then assigned the third factor F3 via a third action instruction HAW3. The third instruction HAW3 is representative of the wear on the injector. This is analogous to the two other instructions HAW1 and HAW2. In the example shown, the third factor F3 corresponds to the instruction that the injection quantity is maintained.

About the Weighting Instructions GWHA become the three factors F1 to F3 weighted against each other. This contains the same information as the previously described instructions. When presented Example is the highest proportion 70%. This corresponds to a weighting factor FGW, which is the input quantity for a correction table KTB represents. The correction table KTB determines which injection quantity in all likelihood the most meaningful. For this purpose, the Maximum value evaluated, here 50.9. As correction value KORR results range of values from 0.99 to 1.01. At a multiplication point the controller manipulated variable RSG calculated by the controller 13 is multiplied by the value KORR. The output corresponds then the corrected injection quantity ve.

In 3 is a Bayesian network 1 for controlling and regulating an internal combustion engine 2 shown in a second embodiment. The input variables of the Bayes network 1 are four measured variables MG1 to MG4. The illustrated Bayes network is used to activate or deactivate a switchable exhaust gas turbocharger. The first measured variable MG1 corresponds to the air-fuel ratio LAM, the second measured variable MG2 corresponds to the engine rotational speed nMOT, the third measured variable MG3 corresponds to the supercharger rotational speed nATL of the permanently operated exhaust-gas turbocharger and the fourth measured variable MG4 corresponds to a time step dT.

the Air fuel ratio LAM will have one first probability table WTB1 a first probability Assigned to W1. In the example shown, the air-fuel ratio is LAM z. Eg one. The probability that the air-fuel ratio LAM is in the range of 1 to 2, is therefore 100%. The probability W1 will over a first action instruction HAW1 assigned a first factor F1. The first instruction HAW1 contains the assignment rule Shut down and hold. Shut down stands for the activation of the switchable exhaust gas turbocharger. Holding stands for retention the current state.

The engine speed nMot is assigned a second probability W2 via a second probability table WTB2. The second probability table WTB2 contains several engine speed ranges. For the example given, the engine speed is for example 1500 revolutions / minute. The probability that this is in the range of 1000 to 1700 revolutions / minute is therefore 100. The second probability W2 is assigned a second factor F2 via a second action instruction HAW2. The second action instruction HAW2 contains the same assignment rules as the first action instruction HAW1. The first factor F1 and second factor F2 are weighed against each other via a third course of action HAW3, z. B. on arithmetic mean value formation. The output corresponds to a third factor F3.

Of the Time step dT is over a fourth probability table WTB4 assigned a fourth probability W4. The time step dT stands for the time since the last deactivation of the switchable Exhaust gas turbocharger has passed. In the example shown was this only recently disabled, for example, 3 seconds ago. The fourth probability W4 is therefore 100%. The fourth Probability W4 and the third factor F3 is then over a fifth Instruction HAW5 assigned a fourth factor F4. On reason the small period since the last deactivation of the switchable Exhaust gas turbocharger contains the fourth factor the instruction that maintained the current state shall be.

Of the Supercharger speed nATL is over a third probability table WTB3 a third probability Assigned to W3. This then becomes together with the fourth factor F4 over a fourth instruction HAW4 associated with a control signal SG. In the example shown, the previous state is retained, d. H. The switchable exhaust gas turbocharger remains deactivated.

Becomes instead of a switchable turbocharger, a VTG turbocharger is used, it is not necessary the signal path for the time step dT and the fifth instruction HAW5. In this case, the control signal SG becomes the third factor F3 and the fourth probability W4 formed.

• – Die technischen Zusammenhänge können grafisch und damit anschaulich abgebildet werden;
• – Die Strukturen der Motorsteuerung können ähnlich einem neuronalen Netz aus den Daten gelernt werden;
• – Die den Variablen (Messgrößen) hinterlegten Wahrscheinlichkeits-Tabellen können aus Daten gelernt werden;
• – Das System ist diagnosefähig und kann daher mit einem Motordiagnose-System einfach gekoppelt werden;
• – Die zukünftige Entwicklung einer Messgröße kann abgeschätzt werden, d. h. das System ist prognosefähig;
• – Unscharfe Informationen können ebenso verarbeitet werden;
• – Eine Interaktion mit Kennfeldern und Sensordaten ist möglich;
• – Die Daten werden reduziert und verdichtet.

From the description, the following advantages result for the invention:

• - The technical relationships can be depicted graphically and thus clearly;
• The structures of the motor control can be learned from the data similar to a neural network;
• - The probability tables stored for the variables (measured quantities) can be learned from data;
• - The system is diagnosable and therefore can be easily coupled with a motor diagnostic system;
• - The future development of a measurand can be estimated, ie the system is predictable;
• - Blurred information can also be processed;
• - An interaction with maps and sensor data is possible;
• - The data is reduced and condensed.

1

Bayesian network

2

Internal combustion engine

3

turbocharger

4

Fuel tank

5

Low pressure pump

6

interphase

7

High pressure pump

8th

Rail

9

Single memory

10

injector

11

Rail pressure sensor

12

electronic control unit (ADEC)

13

regulator

Claims (10)

Bayesian network ( 1 ) for controlling and regulating an internal combustion engine ( 2 ), in which a first probability (W1) is assigned to a first measured variable (MG1) via a first probability table (WTB1), the first probability (W1) is assigned a first factor (F1) via a first operating instruction (HAW1), in which a second probability (W2) is assigned to a second measured variable (MG2) via a second probability table (WTB2), the second probability (W2) is assigned a second factor (F2) via a second operational instruction (HAW2), in which the first factor (F1) is weighted against the second factor (F2) via a weighting instruction (GWHA) and a weighting factor (FGW) is assigned and in which a correction value (KORR) for correcting a controller input variable (RSG), in particular an injection quantity, is calculated from the weighting factor (FGW) via a correction table (KTB). Bayesian network ( 1 ) according to claim 1, characterized in that the weighting factor (FGW) is additionally calculated as a function of a third factor (F3) which depends on the operating time (BZT) of an injector (BZT). 10 ). Bayesian network ( 1 ) according to claim 2, characterized in that the third factor (F3) is calculated by assigning a third probability (W3) to the operating time (BZT) via a third probability table (WTB3) and the third probability (W3) via one third action instruction (HAW3) the third factor (F3) is assigned. Bayesian network according to claim 3, characterized that the first (HAW1), the second (HAW2) and the third action (HAW3) three regulations each in the sense of an enlargement, Preserve or decrease the controller setpoint (RSG). Bayesian network ( 1 ) according to claim 3 or 4, characterized in that the first measured variable (MG1) corresponds to an ambient air pressure (p0) and the second measured variable (MW2) corresponds to an ambient temperature (T0). Bayesian network ( 1 ) for controlling and regulating an internal combustion engine ( 2 ) with at least one exhaust gas turbocharger ( 3 ), in which a first probability (W1) is assigned to a first measured variable (MG1) via a first probability table (WTB1), the first probability (W1) is assigned a first factor (F1) via a first operating instruction (HAW1), in which a second probability (W2) is assigned to a second measured variable (MG2) via a second probability table (WTB2), the second probability (W2) is assigned a second factor (F2) via a second operational instruction (HAW2), in which the first (F1) and second factor (F2) are assigned a third factor (F3) via a third instruction (HAW3), in which a third probability (W3) is assigned to a third measured variable (MG3) via a third probability table (WTB3) , and in which a manipulated variable (SG) for acting on the exhaust gas turbocharger ( 3 ) is calculated from the third factor (F3) and the third probability (W3) via a fourth action instruction (HAW4). Bayesian network ( 1 ) according to claim 6, characterized in that the manipulated variable (SG) acts on an exhaust gas turbocharger with variable turbine geometry in the sense of opening or closing. Bayesian network ( 1 ) according to claim 6, characterized in that a fourth probability (W4) is assigned to a fourth measured variable (MG4) via a fourth probability table (WTB4), and the manipulated variable (SG) for acting upon the exhaust-gas turbocharger ( 3 ) is additionally calculated as a function of the fourth probability (W4). Bayesian network ( 1 ) according to claim 8, characterized in that the manipulated variable (SG) a switchable exhaust gas turbocharger ( 3 ) is activated or deactivated. Bayesian network ( 1 ) according to claim 8, characterized in that the first measured variable (MG1) an air-fuel ratio (LAM), the second measured variable (MG2) of an engine speed (nMOT), the third measured variable (MG3) of a supercharger speed (nATL) and the fourth measured variable (MG4) correspond to a time step (dT).