EP1076764A1

EP1076764A1 - Method for automatically generating smoothed characteristic diagrams for an electronic engine control of an internal combustion piston engine

Info

Publication number: EP1076764A1
Application number: EP00907623A
Authority: EP
Inventors: Günter Schmitz; Werner Hansen
Original assignee: FEV Motorentechnik GmbH and Co KG
Current assignee: FEV Europe GmbH
Priority date: 1999-03-08
Filing date: 2000-02-25
Publication date: 2001-02-21
Anticipated expiration: 2020-02-25
Also published as: DE19910035A1; US6425374B1; EP1076764B1; ATE257905T1; DE50005008D1; WO2000053911A1

Abstract

The invention relates to a method for automatically generating smoothed characteristic diagrams for electronic engine controls of internal combustion piston engines. The invention is characterized in that the adjustment variable combination of the individual successive operating points are input by means of a motor control to a reference internal combustion piston engine by entering specified values of the boundary conditions for the operation of an internal combustion piston engine. According to the invention, the reference internal combustion piston engine is operated in this operating point, and the actual values and/or boundary conditions occurring during the same are acquired and compared with the specified values of the boundary conditions in an optimization system assigned to the engine control and, in the instance of variations, the adjustment variable combinations are optimally altered in a progressive manner by the optimization system. A quality function for the respective alteration of the adjustment variable combination is predetermined in the optimization system, and the quality function is corrected while taking into account already established values of the adjustment variable combination of at least one adjacent operating point.

Description

Description Process for the automatic creation of smoothed maps for an electronic engine control of a piston internal combustion engine

description

The invention relates to a method for the automatic creation of smoothed maps for an electronic engine control of a piston internal combustion engine.

In modern industrial society, mobility plays a major role in the transportation of goods and for trips to work. A large part of these movements take place on the road, and the piston internal combustion engine plays the dominant role as the drive source.

The emissions of piston internal combustion engines have recently become the focus of public discussion. This is reflected in the legislation in the form of ever lower emission limit values. Furthermore, the prices for the required fuels are increasing. Both lead to the fact that lower-emission and lower-consumption piston internal combustion engines are required.

In order to achieve this goal, piston internal combustion engines have to be developed and constructed according to the latest knowledge. Not only does a modern mechanical design play a role here, but the electronics are becoming increasingly important due to the enormously increasing possibilities and flexibility.

Where mechanical centrifugal force adjusters used to adapt the ignition timing to the requirements, an electronic control unit is now in use. This can take influencing factors into account much more precisely and can be adapted more easily to different purposes. In these control units, the dependencies between input variables, for example speed, and the output variables, that is to say the actuating variables, such as, for example, ignition angle, injection quantity, etc., are stored in characteristic maps which contain corresponding characteristic map points for each operating state of a piston internal combustion engine, which contain the current values for the Specify manipulated variables.

When developing a piston internal combustion engine, the necessary maps must be filled with values. So far, the maps have been created by particularly experienced developers based on test bench measurements, heuristic methods and sometimes also intuitively on the basis of measurements on a reference machine. This took considerable development time and usually did not produce optimal results.

The effort for coordinating the characteristic maps depends heavily on the number of parameters to be calibrated. The number of degrees of freedom in control units is increasing, for example due to the introduction of exhaust gas recirculation (EGR), camshaft adjustment, and a variable intake system, to name just a few. The then required solution of a more than three-dimensional optimization task with many parameters can hardly be overlooked by people.

For this reason, systems for automatic map optimization and corresponding software were developed. These create maps based on test bench measurements and mathematically based algorithms. Fewer road tests with vehicles are therefore necessary and the piston internal combustion engine can be optimized even if the entire vehicle is not yet available. On the one hand, this shortens the development time and thus the "time-to-market" and consequently costs are saved. On the other hand, the results generated are reproducible and do not depend on a human optimizer who works with intuition. The optimization system is also easier to adapt and adapt to other requirements. Due to the relatively short time required, the automatic optimization can be carried out several times with different configurations. This opens up the possibility of running through different scenarios that would not be feasible in a practical experiment with reasonable effort.

With the methods used up to now, it is possible to create "mother" maps for a given design of a piston internal combustion engine, according to which the corresponding map data carriers can be created for later series production and also for the engine control to be manufactured in series. However, the disadvantage of the method used hitherto is that a value is generated for each interpolation point or for each operating point of a map while the automatic optimization is being carried out, but without taking into account the relationships between adjacent interpolation points. This results in jumps in the calibration data of neighboring support points, which endanger the transferability of the optimization result and the driveability in practical vehicle use. Strong jumps in calibration data from neighboring operating points must therefore be avoided.

Jumps occur in two phases of optimization: On the one hand, there is the problem that coordination results within such a map area optimized according to the same criteria show such jumps in adjustment values. On the other hand, there is a further problem of abrupt transitions when merging map areas optimized according to different criteria.

The invention is based on the object of finding a method which already prevents excessive jumps in the calibration data during the optimization run and nevertheless permits a good optimization result and enables the creation of a smoothed characteristic map.

This object is achieved with the method steps specified in claim 1. Inventive modifications of the method are given in claims 2 to 4. The invention is explained in more detail below with the aid of schematic drawings. Show it:

1 is a block diagram for a test bench with map optimization,

2 shows the workflow of the test stand according to FIG. 1 as a block diagram,

3 shows an unsmoothed map, created by the method according to the prior art,

4 shows a smoothed map, created according to the method according to the invention,

5 shows the representation of an adjustment variable jump for a variable manipulated variable,

6 the representation of the adjustment variable jump according to FIG. 5 in a coordination system for two variables,

7 shows a flow chart for a map optimization by means of a predetermined quality function,

8 shows a detailed flow chart to explain the optimization of the target and limit values,

9 the mode of operation of a superposition of a quality function with an incentive function,

10 shows a detailed flow chart for a map optimization when a quality function is superimposed with an incentive function,

11 shows a detailed flow chart for a map optimization by means of quality function and adjustment variable difference detection, Fig. 12 is a detailed flow chart for a map optimization to limit the roughness in each operating level.

Fig. 1 shows a test rig with automatically-working map optimization system 1, to input information I _and output information I _from as well as a map output K _from, electric motor control device 2, reference piston-type internal combustion engine 3 for a series of the required measuring devices and 4. The input information of the system part predefined by the user (limit values, targets and map points to be optimized) and partially requested by the system during the optimization of the engine test bench (measured values). For this purpose, the system specifies calibration variables that are automatically set on the piston internal combustion engine 3, and then evaluates the measured values for determining optimal calibration variables. Finally, the system generates maps as a result, which are transferred to the engine control unit 2 of the piston internal combustion engine 3 and for which the optimization was carried out. The engine control unit 2 also takes into account all the values that are relevant for the use of the piston internal combustion engine 3 in a given vehicle.

FIG. 2 shows the workflow of the test stand from FIG. 1 with exemplary input information and examples for calibration variables, for each of which a characteristic diagram has to be created and which measured values can be recorded here. The individual components of the test bench are identified here with the reference symbol from FIG. 1. It is indicated both for the engine control unit 2 of the test bench and for the measuring device 4 that further control elements and measuring devices can be provided. During the automatic map optimization, the map optimization system determines an adjustment variable value for every map point (map point = a combination of the input variables), for example load and speed, for example the ignition timing. However, no relationships between adjacent map points are taken into account. As FIG. 3 shows, with this type of map generation there are jumps in the adjustment variable values to neighboring map points, which endanger the transferability of the optimization result into the engine control unit and the drivability in practical vehicle use. Large jumps in the size of neighboring map points must therefore be avoided. A "smoothed" map must be generated, as shown in FIG. 4 for comparison. A smooth map is characterized by small jumps in the adjustment size.

The change in the adjustment quantity, which is used to evaluate the smoothness of a characteristic diagram, is explained with reference to FIG. 5. For reasons of clarity, an example is shown in which only one adjustment parameter, here the ignition timing ZZP, is considered. The ignition timing in this example depends only on a variable input variable, here the speed n, while the value for the torque is kept constant. Shown are a speed n _a , called "current speed" and two neighbors "nl" and "n2". The current speed has the ignition timing ZZP _a and the two neighbors have the ignition timing ZZP1 and ZZP2.

At the current engine speed n _a , an "ideally smooth ignition timing" is determined, which leads to a smooth map. To determine this "ideal ignition timing" at the current speed, an interpolation between the ignition timing of the neighbors is carried out, shown in FIG. 5 by a dashed line between ZZPl and ZZP2. The difference between this straight line and the ignition timing ZZP _a at the current speed is defined as an adjustment variable jump. The smaller the adjustment variable jump (here the ignition timing jump), the smoother the map at the current point (here at the current speed) in relation to its neighbors.

For calibration variables that usually change linearly, the ideal calibration variable value is determined by linear interpolation. Generally speaking, other interpolations can also be used. In the normal case, the map points depend on (at least) two input variables, for example the ignition point on speed n and load M. In this case, there are more than two adjacent map points between which the ideal calibration variable value must be interpolated, as shown in FIG. 6. 5 is drawn into the coordinate system of FIG. 6. In order to obtain a smooth map, it is not sufficient to carry out the interpolation shown in FIG. 5, but the values of the other adjacent map points, for example N7 and N3, must also be taken into account.

The same procedure is followed for other adjustment variables, e.g. Injection quantity, start of injection, exhaust gas recirculation rate, etc. In these cases, an interpolation between the neighboring map points is carried out for each adjustment variable in order to determine the ideal adjustment variable value.

A so-called quality function is used to determine the most favorable adjustment variable combination. The optimization goal is to fall below the specified limit values (e.g. for exhaust gas emissions). The quality function is made up of all the variables G _: to G _n to be optimized (e.g. consumption, emissions, ...) and the associated limit values GVI ₁ to GW _n . The weight of the individual quantities in the quality function is determined by factors λ _x to λ _n . Thus the quality function is:

Quality = λ ₁ (G _l - GW _X ) + λ ₂ (G ₂ - GW ₂ ) + λ ₃ (G ₃ - GW ₃ ) + ... + λ ₃ (G _n - GW _n )

As an example, a quality function for an optimization of the fuel consumption b _e with a simultaneous requirement for compliance with a nitrogen oxide limit value (N0 _X ) is given.

If N0 _X denotes the currently measured NO _x value and NO ^ _^ the limit value to be observed and b _e the currently measured fuel consumption, the quality function for this application is:

Quality _example = λ _x (NO _x - NO _max ) + λ ₂ * b _e A minimum of the quality function is determined during the optimization. The sequence of such an optimization in the map optimization system 4 is explained and illustrated in FIG. 7 in the form of a flow chart. In the example mentioned, the ZZP is varied until the minimum of the quality function is found. If the limit for NO _{x is} still exceeded at this minimum, the quality function can be trimmed to a greater sensitivity to the nitrogen oxide value by varying the Lagrangian factors λ _j and λ ₂ and a minimum can be sought again.

The variables to be optimized are a function of the calibration variables and the map point:

G _n = f (calibration variables, input variables)

For the example given, this means:

NO _x = f ZZP, n, M) and b _e = f ₂ (ZZP, n, M)

The minimum of the quality function for the entire map is determined by determining the minimum of the quality function in each map point by varying the adjustment variables, as shown in FIG. 8. In the selected exemplary embodiment, it applies to a map point that n and M are kept constant and the minimum of the ZZP is determined. The minimum is determined in each map point. The adjustment variable values belonging to these minima are the optimal adjustment variable values with regard to the optimization goals in the respective map point. The result of this procedure is an unsmoothed map in accordance with FIG. 3, which still has considerable jumps in the adjustment variable.

To avoid jumps in the adjustment variable, the quality function must now be influenced during the optimization process. This avoids the occurrence of map jumps in the course of the optimization. For this purpose, the smoothness of the map to be created is taken into account as an additional boundary condition in the optimization. In a first embodiment of the method according to the invention, this is done by "rewarding" an adjustment variable combination which leads to a smooth characteristic diagram in the calculation process, so that it is preferred in the optimization over other adjustment variable combinations, the same or even better results with respect to the rest Deliver boundary conditions, but lead to larger jumps in the adjustment size.

The quality function is influenced by a so-called incentive function for rewarding favorable adjustment variable combinations with regard to smoothness, which can be formulated as follows:

VG1 to VGx denote the adjustment variables, Optl to Optx the optima of the corresponding adjustment variables in the neighboring operating levels, a to d are factors that determine the influence of the respective adjustment variable in the incentive function.

As an example, the incentive function for the ignition timing ZZP is shown as an adjustment variable, where Ml is the optimum of the ignition timing from the neighboring operating stages. The optimum is the "ideal variable value", i. H. the interpolated value from the optima of the neighboring operating stages:

Incentive _example = | a (ZZP-Ml) |

This incentive function is superimposed on the quality function. The result is a new quality function that has a different minimum and thus leads to a different combination of adjustment variables:

Kindness _incentive = kindness + incentive

For the example, the overlaid function is:

Quality _{incentive Bgp} = 1 λ ₂ (NO _x - NO _xMax ) + λ ₂ * b _e | + | a (ZZP - Ml) The mode of operation of such an incentive function is shown in FIG. 9. The ignition timing (adjusting variable) should be optimized taking into account the minimum consumption (target variable). In this case, the quality function is the course of the consumption over the ignition point. Smooth transitions to neighboring map points are to be created.

At a map point "a", the ignition point x was determined to be optimal with regard to consumption (FIG. 9). In the neighboring map point 2, an optimization of the ignition timing should now be carried out taking into account the smoothness. In this characteristic point "b", the ignition point y would be determined as optimal with regard to the consumption, because the minimum M2 is smaller than the minimum M1 (FIG. 9).

However, the adjustment variable combination in the minimum Ml leads to a greater smoothness than the adjustment variable combination in the minimum M2, since for the adjacent map point 1 the optimal adjustment variable combination lies with the minimum Ml and not with the minimum M2.

In order to influence the smoothness, the incentive function incentive _{example is} added to the quality function quality _example , which has its minimum at the ignition point x of the map point "a", the function value of which becomes less favorable the further the ignition point deviates from the ignition point x (Fig. 9). The addition results in the new quality function, quality _{incentive example,} for the map point "b" (FIG. 9). When optimizing the map point "b", the ignition point is found in the minimum Ml, which is closer to the ignition point x of the neighboring map point than the ignition point y. This leads to a more favorable adjustment size combination in terms of smoothness.

This procedure is now applied iteratively to all map points. During the optimization runs, each map point alternately becomes both a neighbor, which has an influence on the point to be optimized, and a point to be optimized, which is influenced by its neighbors. In the general case with several neighbors, an incentive function is used, which accordingly has several minima depending on the optimal adjustment variables of the neighbors. The example uses a linear incentive function. Depending on the course of the quality function and other variables involved, depending on requirements, non-linear incentive functions can also be used to achieve the described influence on the quality function.

The iterative sequence of a map optimization with influence by an incentive function is explained and shown in FIG. 10.

In another embodiment of the method, a measure of the smoothness in this point is determined from the change in the size of a map point.

For this purpose, the difference between the ideal value and the value found during optimization is formed in the current map point. This difference is called the adjustment variable difference. The adjustment variable difference, like other boundary conditions, e.g. B. the emission values included in the optimization.

The adjustment variable difference is included in the optimization as an additional constraint instead of the incentive function. For this purpose, it is treated like a measured value of the piston internal combustion engine. With each measurement on the piston internal combustion engine, it is calculated from the calibration variables of the neighboring and the current operating stage. The adjustment variable difference, like the exhaust gas emissions, is included in the quality function. So one of the values G _j to G _n can contain the smoothness information:

Quality = - GW + λ ₂ (G ₂ - GW ₂ ) + λ ₃ (G ₃ - GW ₃ ) + ... + λ ₃ (G _n - GW _n )

For an optimization of fuel consumption and nitrogen oxide development with specification of a maximum roughness R_, _ax (roughness = opposite of smoothness) one obtains:

Quality _vdvBsp = λ _x * b _e + λ ₂ (NO _x - NO _xMax ) + λ ₃ (R - R _max ) if R is the currently determined value for the roughness.

The procedure for an adjustment variable is described below. If there are several calibration variables, the procedure is used for each calibration variable.

One considers an operating level in the map, called the current operating level BS, and its neighbors. During the optimization of this operating level, the calibration variable values of the neighbors are constant, since only the calibration variable value of the current operating level is varied. The optimal calibration variable value for the current operating level is calculated from the calibration variable values of the neighbors.

A minimum of the quality function is sought in the current operating stage. The adjustment variable of the current point is varied in order to find the minimum, as can be seen from the flow chart according to FIG. 11. This results in a different adjustment variable difference for each adjustment variable value in accordance with the differently smooth adjustment variable curve to the neighbors.

At the end of an optimization cycle (optimization of all map points), a global value R is calculated for the roughness. "Global" means: for the entire map. To do this, all the differences in the calibration variables are added up. This roughness value is compared with the global limit value for the roughness R _^ . A small limit value corresponds to a small roughness corresponding to a good smoothness of the map.

If this limit value is exceeded, the factor λ (λ ₃ in the above example) of the roughness in the quality function is modified, preferably increased, such that the roughness has a stronger influence on the quality function. In the next optimization run, the optimal calibration variables for the changed quality function are determined. Since this quality function is more dependent on the roughness, more favorable values for the adjustment variables with regard to smoothness are achieved. The roughness is limited for the entire map by specifying a global limit value. It does not matter what proportion the individual operating levels have in the overall result, but only that the value falls below the limit. This process is repeated until all optimization goals are achieved.

If the roughness is limited by a global limit value, local, existing adjustment variable differences can be compensated for by smooth parts of the map in the total value of the roughness. "Local" means: In a map point. However, local roughness is undesirable.

In order to keep these local adjustment variable differences small, the roughness of the map is limited in each individual operating stage by the introduction and specification of a local limit value R (n, M). The result of this is that adjustment variable combinations that exceed this limit value are immediately discarded when this map point is optimized, as indicated in FIG. 12.

The maps of piston internal combustion engines are divided into several areas in which different boundary conditions and optimization goals apply. An area is specified by the legally prescribed driving cycle (to limit emissions) and is called the driving cycle area. Other areas are the full load curve, on which maximum power is required, and the rest of the map, in which minimal consumption is usually desired, called the consumption minimum area.

In order to be able to make a statement about the roughness in the entire map, the roughness values of the different areas must be summarized accordingly. The following procedure is used:

After the optimization is complete, a value for the roughness is available for each area. The optimization system calculates this value for each area using the dwell times from the results of the individual operating levels corresponding to

Consumption and emissions.

Dwell times are only specified for the driving cycle area by the driving cycle. The number of operating levels and the dwell times in the individual operating levels (for the driving cycle area) are determined by converting the driving cycle into stationary operating levels. There are no corresponding requirements for the full load curve and the consumption minimum area.

In order to be able to carry out an optimization on the full load curve and in the consumption minimum area, dwell times are also required there. In principle, any length of stay can be assumed. However, since the dwell times are also used to extrapolate the roughness, the following procedure is used to determine the dwell times for the full load curve and the minimum consumption area:

The average length of stay in an operating level for the driving cycle area can be calculated from the length of stay and the number of operating levels in the driving cycle area:

Average length of stay = seconds in the driving cycle area / number of operating levels in

Driving cycle area

This average length of stay is also used for the operating levels on the full load curve and in the consumption minimum area. This makes it possible to calculate the roughness for the entire map: the results of all operating levels are extrapolated (on average) with the same dwell time. The share of an area in the overall result is therefore the ratio of the number of operating levels in the area to the total number of operating levels in the map.

A smoothed map can be generated with the method shown, as can be seen from the comparison between FIGS. 3 and 4. This smoothed map not only enables emission limit values to be met, as the map according to Fig. 3, but by the smooth transitions between the operating levels, transferability to the engine control unit and driveability are ensured. The smoothed characteristic maps created in this way during the operation of a reference piston internal combustion engine then serve as “mother” characteristic maps for the production of engine control units for piston internal combustion engines of this type.

Claims

Expectations

1. A method for the automatic creation of smoothed maps for electronic engine controls on piston internal combustion engines, characterized in that the calibration variable combination of the individual successive operating points are entered on a reference piston internal combustion engine by means of an engine control by specifying target values of the boundary conditions for the operation of a piston internal combustion engine, the reference -Piston internal combustion engine driven at this operating point and the actual values and / or boundary conditions that occur are recorded and compared with the target values of the boundary conditions in an optimization system assigned to the engine control and, in the event of deviations by the optimization system, the adjustment variable combinations are changed step by step in an optimizing manner, a quality function for the respective change in the combination of adjustment variables is specified in the optimization system, and that the quality function is in each case under consideration correcting already set values of the adjustment variable combination of at least one neighboring operating point is corrected.

2. The method according to claim 1, characterized in that the quality of the respective combination variable is defined by the function

Quality = λ ₁ (G ₁ - GW + λ ₂ (G ₂ - GW ₂ ) + λ ₃ (G ₃ - GW ₃ ) + ... + λ _n (G _n - GW _n )

3. The method according to claim 1 and 2, characterized in that an incentive function taking into account fixed map values of at least one adjacent operating point of the quality function

is superimposed.

4. The method according to claim 1 and 2, characterized in that the respective adjustment variable is given a maximum roughness for the map to be created and is taken into account in the quality function.