AT526104A1 - Method for identifying a law for forward control of a control parameter of a drive train from measurement data - Google Patents

Abstract

The present invention relates to a computer-implemented method for identifying a law for the forward control of a control parameter of a drive train of a vehicle, comprising the steps: - determining (20a) process data, comprising an input signal (u) and an output signal (y), on a drive train test bench or a Simulation of a powertrain test bench or an engine test bench or a simulation of an engine test bench; - Modeling (20b) the input signal (u) and the output signal (y) as a weighted linear combination of time-dependent basis functions (φk); - Determining (20c) at least one time derivative of the modeled input signal (u) and at least one time derivative of the modeled output signal (y) as a derivative of the time-dependent basis functions (φk); - Determining (20d) first independent weighted linear combinations between the modeled input signal (u) and the time derivatives of the modeled input signal (u'), whereby first model parameters are used for weighting, and second independent weighted linear combinations between the modeled output signal (m) and the time derivatives of the modeled output signal (m), with second model parameters being used for weighting; and - determining (20e) the first model parameters (θu) from the first independent weighted linear combinations and the second model parameters (θy) from the second independent weighted linear combinations using the least squares method.

Description

Procedure for identifying a law for forward control of a

Control parameters of a drive train from measurement data

The present invention relates to a computer-implemented method for identifying a law for forward control of a control parameter of a powertrain of a vehicle, as well as a method for forward control of a powertrain of a vehicle, a control unit, a powertrain of a vehicle, a computer program and a computer-readable medium based on the method.

Recent social, political and technological developments have a significant impact on the further development of vehicle powertrain control. On the one hand, the requirement to reduce energy consumption and emissions in real driving and not just for a few predefined driving cycles implies the need to continuously optimize vehicle operation in real driving. On the other hand, a variety of new vehicle functions have emerged in recent years, such as autonomous driving, ADAS, predictive functions or fleet management. New additional factors and a complicated combination of goals influence the

optimization decisions significantly.

This optimization problem can no longer be solved by conventional rule-based controls with offline calibrated rules, in particular for the following reasons.

The calibration rules approach is generally based on offline optimizations that are carried out for a few driving cycles. Optimization in real driving operations, which regularly differ greatly from this, is not possible.

The advantage of transparency and traceability of the control that exists with a few calibration rules in simple closed control loops with, for example, static feedforward control without cross-decoupling largely disappears in a multi-input-multi-output system when the number of

2/26 1/18

necessary calibration rules and their mutual influence increases sharply. As a result, the effort and costs for calibration increase exponentially with the number of optimization parameters.

In contrast, a control system that has only feedforward control responds to its control signal in a predefined manner, without responding to the way the controlled variable responds. This is in contrast to a system that also has feedback that adjusts the input to take into account how it affects the controlled variable and how the controlled variable itself may vary unpredictably; the controlled variable becomes

considered as part of the external environment of the system.

In a system with feedforward control, the adjustment of the controlled variable is not error dependent. Instead, it is based on knowledge of the process in the form of a mathematical model of the process and knowledge of the

Process disruptions or their measurement.

Powertrain feedforward control models can be used wherever systems behave approximately linearly near an operating point, the number of input data and output data are equal, and a desired trajectory of the output data is available or

can be calculated.

The advantages of forward control over other controls include higher control accuracy, higher reliability and lower power consumption, which means the control unit can be manufactured at lower costs and can operate at high speed. Furthermore, hysteresis effects are reduced in forward controls.

Since disturbances in a system to be controlled cannot be ruled out and because the model for forward control does not perfectly represent the dynamics of the system, a control unit with a dynamic feedback loop is advantageous for practical application. The dynamic feedback loop can be a known controller, such as a PL controller, a PID controller, a static feedback controller, or a rule-based controller.

3/26 2/18

The application of inventory controls is in various places

Drivetrain of a vehicle is particularly helpful.

Possible applications in an internal combustion engine include controlling the air intake or the mixing ratio of fresh air and exhaust gas recirculation (EGR) of a diesel engine. This includes the control of the exhaust gas recirculation and the turbocharger with variable turbine geometry (VTG turbocharger), control of the output, e.g. exhaust pressure/NOx or the pressure at the intake manifold (P2) and the fresh air mass flow.

Another possible application is gasoline air path control and lambda precontrol through injection mass. This can be used for the control inputs of the throttle valve, exhaust gas recirculation and/or turbo position, as well as for the intake manifold pressure, boost pressure and/or EGR mass flow outputs.

If the forward control is used for the temperature management of a diesel engine, for example, the control of the engine output temperature and/or the catalytic converter can be predictively controlled depending on the requirements of the selective catalytic reduction.

Another control option is to control the temperature of a diesel particulate filter. Hydrocarbon dosing can be done at the entrance

and a flow temperature of the particle filter can be regulated at the output.

Another application option is fuel cell air path control, for example for a compressor and a throttle valve of a polymer electrolyte fuel cell (PEMFC compressor). In this application, the control inputs are throttle and compressor speed, the control outputs are pressure and air mass flow.

Previously known methods for identifying a law for forward control are based on analyzing a system and using the data obtained to invert a model system, from which the control parameters are in turn derived. The disadvantage of this method is that inversion is not possible with all systems and often requires a lot of effort

connected is.

4/26 3/18

Classic methods for determining a feedforward control law are based on identifying a control-based model and inverting it. This method requires a lot of computing effort and does not lead to the desired success for all models, as not all models

let it be inverted.

The identification of control-based models fundamentally does not enable automatic calibration of a model for a system that functions as a black box, i.e. a system in which only its input data and output data are known.

One object underlying the invention is to overcome these disadvantages and thus to identify a law for forward control

To simplify control parameters of a drive train of a vehicle.

According to a first aspect of the invention, this object is achieved by a new computer-implemented method for identifying a law for forward control of a control parameter of a drive train

vehicle, comprising the steps:

- Determining process data, comprising an input signal and an output signal, on a powertrain test bench or a simulation of a powertrain test bench or an engine test bench or a simulation of an engine test bench;

- Modeling the input signal and the output signal as a weighted linear combination of time-dependent basis functions;

- Determining at least one time derivative of the modeled input signal and at least one time derivative of the modeled output signal as a derivative of the time-dependent basis functions;

- Determining first independent weighted linear combinations between the modeled input signal and the time derivatives of the modeled input signal, where first model parameters are used for weighting and second independent weighted linear combinations

between the modeled output signal and the time derivatives of the

5/26 4/18

modeled output signal, with second model parameters being used for weighting; and

- Determining the first model parameters from the first independent weighted linear combinations and the second model parameters from the second independent weighted linear combinations using the method of

smallest squares.

A feedforward control is an element or path within a control system that passes a control signal from a source to a controlled variable. A feedforward control responds to the control signal in a predefined manner without responding to the way the controlled variable responds. This is in contrast to a system that also has feedback that adjusts the input to take into account how it affects the controlled variable and how the controlled variable itself may vary unpredictably; the controlled variable is considered part of the external environment of the system.

In automotive technology, the drive train of a vehicle refers to all the components that generate the power for the drive in the vehicle and transmit it to the ground. A powertrain control parameter may be a control parameter of the engine or another part of the powertrain.

The first and second model parameters serve as control parameters of the drive train.

Preferably, the method can provide that the output signal comprises an actuator signal of the drive train and/or a sensor signal of the drive train.

In a further advantageous special embodiment of the invention it can be provided that the time derivatives of the modeled input signals

include first, second and third order temporal derivatives.

More preferably, the time derivatives of the modeled output signals can include first, second and third order time derivatives.

6/26 5/18

According to a further particularly advantageous embodiment of the invention, the time derivatives of the modeled output signals are limited to first, second and third order time derivatives.

According to a further advantageous embodiment of the invention, the include

Basis functions Polynomial functions and/or Gaussian radial basis functions.

According to a second aspect, the invention provides a computer-implemented method for configuring a feedforward controller with model parameters that vary with an operating point of a powertrain of a vehicle, comprising the steps of:

- Determining model parameters associated with a first operating point using a method according to one of the preceding claims;

- Determining model parameters associated with a second operating point by means of a method according to one of the preceding claims, wherein the second operating point differs from the first operating point; and

- Configuring the forward-controlled controller with the model parameters associated with the first operating point and the model parameters associated with the second operating point.

According to a third aspect, the invention provides a computer-implemented method for forward control of a powertrain of a vehicle, comprising the steps of:

- Determining first and second model parameters using a method according to one of claims 1 to 7; and - controlling the powertrain with the first and second model parameters.

According to a fourth aspect, the invention provides a control unit comprising a feedforward controller for carrying out the method according to any one of claims 1 to 7.

The control unit can preferably further comprise a feedback controller, the output of which is connected to an output of the feedforward control controller

is connected together.

7126 6/18

According to a fifth aspect, the invention provides a powertrain

Vehicle with a control unit according to the fourth aspect of the invention.

According to a sixth aspect, the invention provides a computer program comprising instructions which, when the program is executed by a computer, cause the computer to perform the method according to any one of claims 1 to 7

to carry out.

According to a seventh aspect, the invention provides a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 7

to carry out.

With the method according to the second aspect of the invention, it can preferably be provided that, in addition to the model parameters associated with the first operating point and the model parameters associated with the second operating point, model parameters associated with one or more further operating points according to one of the methods according to one of claims 1 to 5 can be determined. In this case, the forward-controlled controller can also be configured with these model parameters associated with one or more additional operating points

become.

Further advantages, features and details of the invention emerge from the following description, in which exemplary embodiments of the invention are described in detail with reference to the drawings. It shows schematically:

1 shows an overview of the differences for identifying a law for forward control in a classic and inventive manner;

2 shows a computer-implemented method for identifying a law for forward control of a control parameter of a powertrain of a vehicle, according to a particular embodiment of the invention;

Fig. 3 shows an operating range of a control parameter of an internal combustion engine

with multiple operating points, and

8/26 7/18

Fig. 4 is a schematic representation of a drive train, according to one

special embodiment of the invention.

1 shows an overview of the differences for identifying a law for forward control of a control parameter of a drive train of a vehicle in a classic and inventive manner. A technical system 10 reacts to a suggestion according to its configuration. The excitation can be described as a time-dependent function and is referred to as the input signal u. The response of the system can also be described as a time-dependent function and is referred to as the output signal y.

In a first step 12, the input signal u and the output signal y are determined.

The path shown on the left in Figure 1 shows the classic method for determining a feedforward control law. In a step 14 following the first step 12, the input signal u and the output signal y are used to carry out a backward-controlled model by estimating y given u. The model is then inverted in step 16 in order to derive a model for forward control from the backward-controlled model, in which u is estimated by y. The problem with this approach is that an inversion generally requires a matrix inversion, which can lead to numerical problems and therefore

is not possible in every case.

If the derivation of the model for forward control is successful, the model for forward control is obtained in step 18, with which a forward control input signal uf is derived from an output reference signal yr

can be.

According to the invention, after the first step 12, the model for forward control is instead derived directly from the known input signal u and the known output signal y in step 20. A potentially problematic matrix inversion is not part of the procedure in step 20. Instead, only a least squares fit is performed for each

Input signal and each output signal made, as well as to the

9/26 8/18

Identify model parameters. No numerical problems can arise here. Further details on this process are provided in connection with Fig. 2

discussed.

Figure 2 shows a computer-implemented method for identifying a law for forward control of a control parameter of a powertrain of a vehicle, according to a particular embodiment of the invention. In principle, it is possible to carry out the process automatically.

In a first step 20a, process data is determined. The process data includes an input signal u and an output signal y. The determination can be carried out on a powertrain test bench or a simulation of a powertrain test bench. Both the input signal u and the output signal y are determined as time series. The input signal u is a trajectory recorded at X sampling times. The output signal y is a trajectory recorded at Y sampling times. The sampling times X

and Y can be the same or different.

In a second step 20b, the input signal u and the output signal y are used as a weighted linear combination U; and Y; from time-dependent basis functions

x modeled. The modeling is carried out according to formulas (1) and (2). A wN . Ü; = Xk=1 YupkOr, VIE [1L,m] ©

Yı = Zhe1Vyık@rK, Vie [1,m] @)

The basis functions px are chosen so that they form an approximation to the infinite dimensional space in which the input and the output lie. By choosing N basis functions, u is approximated as the sum of these basis functions. The basis functions are chosen as universal function approximators. Classically, Gaussian basis functions or polynomials come into question. In step 20b, the output u is approximated by a sum of these basis functions. The weighting coefficients of the basis functions are determined automatically by minimizing the sum of the

squared difference between the input signal u and the modeled one

10/26 8

Input signal {;. The approximation of the output signal y with the modeled output signal 9; takes place accordingly. Each of the modeled signals and Y; has its own coefficients Y,, x and Yy,,x, which correspond to the respective weighting of the

Basic functions @, which correspond to the corresponding measured signal u am

best approximate.

The coefficients Yı,,,x and Yy,x are identified to represent the input signal u and

represent the output signal y as the sum of the basis functions “x. Every sentence

of coefficients Yı,,,x and Yy,,g is a vector of N components representing N

Weight basic functions x.

However, the coefficients Y,,,x and Yy,x are not model parameters, but

just an intermediate result for identifying the model parameters.

In a third step 20c, at least one time derivative of the

modeled input signal Ur and at least one time derivative of the modeled output signal Dir as a derivative of the time-dependent basis functions DO. The determination is carried out according to the following formulas (3) and (4).

as AT _d U;

d%$; do. Al Do N k Yıi — dt“ — Xk-ı Yyık dt ” Vi € [1, m] (4)

do. — Xkeı Yuzk PIE VLE L1, m] (3)

In a fourth step 20d, the first independent weighted ones are weighted

Linear combinations between the modeled input signal U; and the

time derivatives of the modeled input signal Qi and second

11/26 10/18

independent weighted linear combinations between the modeled output signal Yi and the time derivatives of the modeled output signal Dir are created, with first and second model parameters 0,,, 0y each being used for weighting. The relationships are shown in equations (5) and (6). [a Q u QO]0.=D Yo F0O10, © in 8 4 a 9 O4] = 0m 6

The first and second model parameters are matrices Oyik and Oyık with the respective dimensions (p,m) and (8,m).

The quantity ‘p’ is a vector with m elements that determines how often each of the m

Input signals are differentiated. The size '8' also has m elements and lays

pP and ‘8’ are generally unknown. If set to low values,

the number of derivatives of each of the output signals. The values of

the system dynamics can be poorly estimated. The advantage of this method is that p and 5 can be chosen large without numerical difficulties. Additionally, one can use least squares properties to look at the singular values and see how many of them are insignificant to get an estimate of the number of input/output derivatives needed. Another possible method is to create multiple models with different p and 5 and select the model with the lowest values of p and 5 with acceptable accuracy and/or performance. The minimum value 0 (zero) can be selected for the size p. For size 5, the minimum value 1 can be selected. The number of maximum differentiations

of a size is also referred to as the order of the model.

In a fifth step 20e, the first model parameters are 0,, from the first independent weighted linear combinations and the second model parameters

0y from the second independent weighted linear combinations using the

Least squares method determined.

12/26 11/18

By determining the model parameters, a trajectory of the input signal u is determined so that it follows a predetermined trajectory of the output signal y. This describes a law for the forward control of a control parameter of a drive train of a vehicle. With the model order, the accuracy of the model as well as the computational effort to determine it increases

and, if necessary, also for its application.

3 shows a schematic representation of an operating range 30 of a control parameter of a drive train with several operating points 22a-22k. The operating points 22a-22k differ from each other in the speed Nice and/or torque Tice of an internal combustion engine.

For each operating point 22a-22k, the model parameters O,, and 0, are determined using a method according to one of claims 1 to 5. Each operating point 22a-

22k has its own model parameters Ol, and 65. The model parameters OL, and 6)

different operating points 22a-22k have the same function in that each parameter is intended to multiply an input signal u or an output signal y or a derived input signal u 'or a derived output signal y'. In this sense, the model parameters

Oi, and 63 of the individual operating points are combined with one another to create a

to form a parameter-varying model as the weighting sum of the local model parameters u and 0}.

The type of weight sum can be set manually, for example as

a Euclidean norm. Alternatively, the weighting sum can be from measurement data in

determined in another form. Mathematically, this connection can be expressed in such a way that the n,

local operating parameters Oi associated with operating points, and 6)

combined operating parameters 0..(X) and 0, (X), according to the following

Equations (7), (8), (9) and (10) can be determined.

13/26 12/18

Lu u uP)]0.,(X) — [Yref Yref' U Yref®] 0, (X) (7) X= Nice» Tice} (8) 0.,.(X) = Xo:0u (9 ) 0,(X) = X6:0) (10)

Figure 4 shows a schematic representation of a drive train 31 according to a special embodiment of the invention.

The main task of controlling a drive train 31 is to find operating points for the components of the drive train 31 that meet the desired optimization goals and at the same time meet certain specifications with regard to safety, diagnostics and protection of the drive train components. Possible optimization goals include a certain fuel consumption, overall system efficiency, emissions, or the durability of various components. The components of this drive train 31 include an internal combustion engine 32, a transmission 40, a

Coupling 42 and a control unit 44.

The control unit 44 includes a feedforward controller 46 and a feedback controller 48 with a feedback loop. The feedback controller 48 can be any classic controller and as such can be configured with, for example, PI control, PID control, state feedback or other rule-based control. In a practical implementation, it may be advantageous to combine the two controllers 46, 48 in order to be able to react to system disturbances or to be able to react to an imperfect representation of the system dynamics by the feedforward controller 46 with the feedback controller 48.

The feedforward controller 46 receives an output reference signal yret and outputs a feedforward input signal uf. An output signal y is transmitted from the internal combustion engine 32 to the feedback controller 48. Furthermore, the feedback controller 48 also receives the output reference signal yrer. The feedback controller 48 determines a feedback signal from both output signals y, yrer.

14/26 13/18

Input signal upı, which in this example is determined with a proportional-integral control (PL control). From the forward control input signal ur and the feedback input signal upı, an input signal u is calculated and/or selected in an addition device 50, which is then transmitted to the input of the internal combustion engine 32 for control.

Such control can be carried out for different devices of the internal combustion engine 32 or the drive train 31.

The above explanations of the embodiments describe the

present invention exclusively within the scope of examples.

15/26 14/18

Claims (12)

Expectations 1. Computer-implemented method for identifying a law for forward control of a control parameter of a drive train (31) of a vehicle, comprising the steps: - Determining process data, comprising an input signal (u) and an output signal (y), on a drive train test bench or a simulation of a drive train test bench or an engine test bench or a simulation of an engine test bench; - Modeling the input signal (u) and the output signal (y) as a weighted linear combination of time-dependent basis functions (dk); - Determining at least one time derivative of the modeled input signal (u) and at least one time derivative of the modeled output signal (y) as a derivative of the time-dependent basis functions (dk); - Determining first independent weighted linear combinations between the modeled input signal (u) and the time derivatives of the modeled input signal (u'), where first (Su) model parameters are used for weighting and second independent weighted linear combinations between the modeled output signal (m) and the time Derivatives of the modeled output signal (m'), with second model parameters (8y) being used for weighting; and - Determining the first model parameters (u) from the first independent weighted linear combinations and the second model parameters (6y) from the second independent weighted linear combinations using the least squares method. 2. The method according to claim 1, wherein the output signal (y) is an actuator signal of the drive train (31) and / or a sensor signal of the drive train (31). includes. 16/26 15/18 3. Method according to one of the preceding claims, wherein the time derivatives of the modeled input signals are time derivatives of first, second and third order. 4. Method according to one of the preceding claims, wherein the time derivatives of the modeled output signals are time derivatives of first, second and third order. 5. The method according to any one of the preceding claims, wherein the basis functions (@k) are polynomial functions and/or Gaussian radials Include basic functions. 6. Computer-implemented method for configuring a feedforward controller (46) with model parameters (8,, 9,) that vary with an operating point (22a-22k) of a drive train (31) of a vehicle, comprising the steps: - Determining model parameters (9u1, 0.1) associated with a first operating point by means of a method according to one of the preceding claims; - Determining model parameters (9u2, 0.2) associated with a second operating point by means of a method according to one of the preceding claims, wherein the second operating point differs from the first operating point; and - Configuring the forward-controlled controller (46) with the model parameters (0.1, 9.1) associated with the first operating point and the Model parameters associated with the second operating point (9.2, 9.2). 7. Computer-implemented method for forward control of a drive train (31) of a vehicle, comprising the steps: - determining first and second model parameters (8,, 9,) with a Method according to one of claims 1 to 5; and 17/26 16/18 - Controlling the drive train (31) with the first and second Model parameters (8,, 0,). 8. Control unit (44), comprising a forward-controlled controller (46), for carrying out a method according to one of claims 1 to 7. 9. Control unit (44) according to claim 8, further comprising a feedback controller (48), wherein an output of the forward-controlled controller (46) is connected to an output of the feedback controller (48) in electrical and / or there is a signaling connection. 10. Drive train (31), comprising a control unit (44) according to one of claims 8 or 9 for controlling the drive train (31). 11.Computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 7. 12. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 7. 18/26 17/18