KR20190141092A - Method and device for determining a crankshaft torque in a combustion engine - Google Patents

Abstract

The present invention relates to a method for determining crankshaft torque for the operation of a combustion engine (10), wherein the crankshaft torque is determined through a data-based statistical function model according to the operating point of the combustion engine (10).

Description

METHOD AND DEVICE FOR DETERMINING Crankshaft Torque in Combustion Engines

The present invention relates to a combustion engine and in particular to a control device of such a combustion engine. The invention also relates to a method for determining the crankshaft torque of a combustion engine, in particular a method for operating such a combustion engine.

With regard to the operation of the combustion engine, knowledge of the crankshaft torque is in fact important for performing different functions. In particular, the crankshaft torque is used for the monitoring and control of the propulsion torque.

In conventional engine systems the crankshaft torque is not measured directly, but is modeled in particular through characteristic curves based on the measured or modeled state variables of the engine system. This modeling is carried out on the basis of torque balance, which takes into account the individual driving torque and load torque acting on the crankshaft.

The drive torque directly related to the combustion engine corresponds to a high pressure torque, and in general the high pressure torque is modeled via a high pressure torque model provided, for example in the form of a characteristic curve. In addition, the load torque inside the engine is often explained through a low pressure torque model provided in the form of characteristic curves because of the loss of charge and load exchange, the friction loss is explained through a friction model, and the friction model is dependent on the physical relevance Correspondingly, it can be mapped through modeling or a suitable characteristic curve model. Application of such characteristic curve models is costly, because the characteristic curve models must be valid over a wide operating area of the combustion engine, or must consider the foregoing. This requires measuring the engine system over the maximum operating area, which makes the method described above expensive.

The prior art known general control devices, which are suitable for calculating the state variables of the engine system according to the computational model. Such control devices generally include a microprocessor, which is equipped with appropriate model computing software. When such a control device implements a characteristic curve model, the control device can access the corresponding characteristic curve memory device to read out the required values.

The prior art also known a control device, which has a built-in control module comprising a main computing device and a separate model calculator for calculating the data-based functional model. For example, DE 10 2010 028 266 A1 discloses a control module comprising an additional logic circuit as a model calculator, which calculates an exponential function and an addition operation and a multiplication operation based only on hardware. It is formed for. This can support, in particular, Bayesian-Regression methods in hardware devices that are needed to calculate Gaussian-process-models.

In total, the model calculator is provided to execute a mathematical process for calculating a data-based function model based on parameters and sample points or training data. In particular, the function of the model calculator for efficiently calculating the exponential and sum functions can be performed entirely in hardware, so that the Gaussian-process-model can be calculated at a higher computational speed than that performed in the main control unit of software control. .

It is an object of the present invention to provide a method and a system according to the dependent claims for determining the crankshaft torque of a combustion engine for operating a combustion engine according to claim 1.

Another embodiment is described in the dependent claims.

According to a first aspect, a method for determining crankshaft torque for combustion engine operation is provided, wherein the crankshaft torque is determined via a data based statistical function model according to the operating point of the combustion engine.

The idea of the method is to use a statistical function model, which in effect takes into account a number of state variables for determining the crankshaft torque, where the state variables represent the individual states of the engine system. The functional model can be used to implement statistical function models and control devices, in particular engine control devices of combustion engines.

In addition, the data-based statistical function model may correspond to a Gaussian process model. A variant of the data-driven function model is not a parametric model and can be generated from training data, i.e., a certain number of training data points, without special designation. An example of a data-based functional model is a so-called Gaussian-process-model, which is based on Gaussian-process-regression. Gaussian-process-regression is a variety of statistical methods for modeling complex physical systems based on data, typically based on large amounts of training data.

Further, the crankshaft torque may be determined as the sum of the drive torque and the plurality of load torques, and at least one drive torque as the high pressure torque is determined by the data based function model.

The method described above is for using a data-based functional model for modeling the crankshaft torque. This crankshaft torque is generally determined through the balance of drive torque and load torque. The drive torque corresponds to the high pressure torque determined by the high pressure torque model. The load torque is the low pressure torque determined by the low pressure torque model, that is, the low pressure torque corresponding to the load torque generated due to the exchange of charge in the combustion engine, the friction torque determined by the friction model and representing the friction loss in the combustion engine, and Another assembly load torque, wherein the assembly load torque occurs through the assembly coupled with the crankshaft, such as load torque due to the drive of a high pressure pump, oil pump, coolant pump, generator and / or air conditioner, for example. In particular, the assembly load torque is influenced by another factor and there is a suitable physical model to account for the corresponding assembly load torque.

The above-described physical models, in particular the physical models for the modeling of high and low pressure torques, are partially inaccurate and generally are not provided for practically all possible operating areas of the combustion engine. Evaluations performed by extrapolation in the operating area according to the corresponding characteristic curve model are inaccurate in determining the crankshaft torque. By using a statistical function model for the modeling of high pressure torques and / or low pressure torques, the effort required by the characteristic curve models used so far can be significantly reduced. In addition, due to the nature of the statistical function model to provide reliable model values in the previously unmeasured operating range, the effort for modeling the crankshaft torque can be significantly reduced, further increasing the crankshaft torque. Accuracy can be improved when calculating. Overall, the use of statistical function models makes it possible to better map the complex behavior of combustion engines than would be possible with simple characteristic curve models.

In particular, a statistical function model for calculating the high pressure torque may be calculated based on the input variable, which may correspond to the following input variable or is selected from the following input variable: crankshaft rotational speed , Injection volume of main injection, injection angle of main injection, rail pressure, air mass, EGR-mass, gas temperature of suction mass flow, pilot injection quantity, post injection quantity, post-injection crankshaft angle, oil temperature and swirl valve position .

According to an embodiment, one of said load torques corresponds to a low pressure torque, said low pressure torque representing a gas exchange loss of a combustion engine and is determined via a data based function model.

In particular, a statistical function model for calculating the low pressure torque may be calculated based on the input variable, which may correspond to the following input variable or is selected from the following input variable: crankshaft revolutions , Exhaust back pressure, suction line pressure, partial flap position, EGR-ratio, suction temperature and oil temperature.

As the load torque, one or a plurality of load torques of the following load torques may be considered: friction torque and assembly load torque.

According to another aspect of the invention, a control module for an engine system comprising a combustion engine is provided, which control module is configured to carry out the method.

Embodiments are described in detail below with reference to the accompanying drawings:

1 schematically illustrates an embedded control module including a hardware based model computing device,
2 schematically shows an engine system including a combustion engine operated via an embedded control module.

Fig. 1 schematically shows the hardware architecture for the embedded control module 1, for example in the form of a microcontroller, where the main computational unit 2 and the model calculator are only for hardware-based calculation of the data-based functional model. (3) is provided in a form embedded in the microcontroller. The main computing device 2 and the model calculator 3 are communicatively connected to one another via an internal communication link 4, for example a system bus.

Basically, the model calculator 3 is built in hardware and is not formed to execute software code like the main computing unit 2. Optionally, the model calculator 3 for computing a data-based functional model is limited, but a highly specialized instruction set can be used according to the solution. The model calculator 3 is not provided with a processor. This makes it possible to realize resource-optimization for this model calculator 3, or to enable a space-optimized configuration in an integrated manner.

The model calculator 3 has an arithmetic kernel 31, which performs the calculation of a given algorithm only in hardware.

The model calculator 3 may also include a local SRAM 33 for storing configuration data. Similarly, the model calculator 3 may include a local DMA-unit 34 (DMA = Direct Memory Access). This DMA-unit 34 allows access to the built-in resources of the control module 1, in particular the internal memory device 5.

The control module 1 may comprise an internal memory device 5 and another DMA-unit 6 (DMA = Direct Memory Access). The internal memory device 5 and another DMA-unit 6 are connected in communication with each other in a suitable manner, for example via an internal communication link 4. This internal memory device 5 flashes for common SRAM-memory devices and configuration data (parameter and sample point data) (for the main computing unit 2, the model calculator 3 and optionally another unit). It may include a memory.

The use of function models that are not based on parameters and data is based on Bayesian-Regression. The basis of such Bayesian regression analysis is described, for example, in C. E. Rasmussen et al., “Gaussian Processes for Machine learning”, MIT Press 2006. The Bayes-Regression method is a data based method based on a model. In order to generate the model, measurement points of the training data and corresponding output data of the output variable to be modeled are required. The generation of the model is carried out via sample point data, which may be completely or partially corresponding to, or generated from, the training data. In addition, abstract hyperparameters are determined, which parameterize the space of the model function and focus on effectively influencing the model predictions that individual measurement points of the training data are made later.

The abstract hyper parameter is determined through an optimization method. The possibility for such an optimization method is in marginal likelihood. This marginal likelihood represents the determined output value of the training data and the effectiveness of the measurement points of the training data in a given model parameter. By obtaining a suitable hyperparameter in the model training, the surrounding likelihood is maximized, where the aforementioned hyperparameters advance the model function determined through the hyperparameter and the training data and map the training data as accurately as possible. To simplify the operation, the logarithm of the surrounding likelihood is maximized because the log does not change the continuity of the effective area.

The formula for creating a function model that is not based on parameters and data includes the following functions:

In the formula a function value z is obtained. Where D is the dimension of the input data space / training data space / sample data space, and z is a test that includes u _{1 ... D} , x _i, or (χ _i ) _d corresponding to the sample point of the sample point data. The model value (output value) of point u (the input variable vector of dimension D), N is the number of sample points in the sample point data, l _d , σ _f and the parameter-vector (Q _z ) are the It is a hyper parameter.

In addition, input normalization and output normalization can be performed, since the calculation of the Gaussian-process-model is generally performed in the normalized space.

In order to start the calculation, in particular, the computing device 2 passes the configuration data corresponding to the function model to be calculated to the model calculator 3 and starts the calculation performed through the configuration data. Another DMA-unit 6 can be designated. The configuration data includes hyperparameter and sample point data of a Gaussian-process-model, preferably the sample point data is stored in an address area of the internal memory device 5 specified in the model calculator 3 via an address pointer. Is displayed. In particular, an SRAM-memory device 33 may be used for the model calculator 3, which can be arranged for the above, in particular in the model calculator 3 or inside the model calculator 3. The internal memory device 5 and the SRAM memory device 33 may also be used in combination.

The calculation in the model calculator 3 is carried out in the hardware architecture of the model calculator 3, which is executed via a pseudo-code, which then corresponds to the calculation rules. It can be seen from the above pseudo code that the calculation is performed in the inner and outer loops and the subtotal values are accumulated. To start model calculation, the typical value for the counter start variable (Nstart) is zero.

/ * Calculate in outer loop * /

001: for (i = Nstart; i <N; i ++) {

002: j = i * D;

/ * Calculate in inner loop * /

003: t = 0.0;

004: for (d = 0; d <D; d ++) {

005: m = u [d]-x [j + d];

006: m = m * m;

007: t + = m / l [d];

008:}

/ * Calculate Exponential Function * /

009: e = exp (-0.5 * t);

/* Sum*/

010: z + = Q _z [i] * e;

011:}

012: return z;

The model data required to calculate the data-driven function model includes parameter vectors and sample point data, and the sample point data of the storage area belonging to the data-based function model is stored in the memory device. Corresponding to the aforementioned pseudo code, the parameter vector of the data-driven function model calculates the length scale-vector (l) for the parameter-vector (Qz) and all normalization processes (d) of the input variable corresponding to the input variable vector. Include. Also, the number of sample point data points (N) is given the start value (Nstart) of the outer loop when resuming the inner loop (generally = 0) calculation.

2 schematically shows an engine system comprising a control module 1 and a combustion engine 10 controlled by such a control module. The combustion engine 10 is formed in the form of a conventional four-stroke cycle operation of the combustion engine, and has a plurality of cylinders 11, the cylinders having a reciprocating piston coupled with the crankshaft 12. do. The crankshaft 12 supplies the drive torque during operation of the combustion engine 10.

In order to control the combustion engine 10, in fact, knowledge of the crankshaft torque is required. This crankshaft torque is defined as the torque at the crankshaft 12 that is used to drive the vehicle or is used at the transmission input. This crankshaft torque results from the torque balance of the drive torque and the load torque affecting the crankshaft 12.

The aforementioned drive torque corresponds to the high pressure torque provided by the combustion engine 10, the so-called internal torque, which internal torque arises from the combustion process and the compression process, in particular the piston operation, in the combustion engine.

In addition, a load torque is generated through a gas exchange process in the combustion engine 10, and the load torque is expressed as a low pressure torque, and is generated through a gas flow passing through the combustion engine 10.

In addition, when the movable component moves in the combustion engine 10, frictional losses occur, and frictional losses as friction torque likewise reduce the crankshaft torque provided to the crankshaft. Friction torque can be determined through a friction model, which maps engine temperature, engine speed, and engine load to friction torque.

In addition, one or more of the assembly 14, such as one of the following assemblies 14, namely: a high pressure pump, an oil pump, a coolant pump, a generator and an air conditioner, may be connected to the crankshaft 12. Can be combined, and this assembly can be driven according to the operating point of the combustion engine 10, or optionally activated or deactivated. An assembly load torque corresponding to each of these assemblies 14 is provided, which is the torque considered in balancing to determine the crankshaft torque in accordance with the assembly load torque model.

The high pressure torque model may be calculated through a statistical function model, ie based on input variables, which may correspond to the following input variables or are selected from the following input variables: crankshaft rotational speed, Injection volume of main injection, injection angle of main injection, rail pressure, air mass, EGR-mass, gas temperature of suction mass flow, pilot-injection amount, post-injection amount, post-injection crankshaft angle, oil temperature and swirl valve position.

In addition, a low pressure torque model for determining working medium exchange loss can be calculated via a statistical function model, taking into account the following input variables, or the input variables are selected from the following input variables: crankshaft revolutions, Exhaust gas back pressure, suction line pressure, partial flap position, EGR-ratio, suction temperature and oil temperature.

Using a Gaussian-process-model for calculating high and low pressure torques, one can simply create a model by measuring the combustion engine 10 in different operating regions.

In connection with the above-described high pressure torque and / or low pressure torque determination, data-based function model setting may be performed based on training data determined on a test bench. To this end, the combustion engine 10 is operated, for example, for different operating regions, and corresponding state variables are determined. This is used for training the corresponding Gaussian-process-model according to known methods.

Claims (11)

A method of determining crankshaft torque for operation of a combustion engine (10), wherein the crankshaft torque is determined via a data based statistical function model according to the operating point of the combustion engine (10). The method of claim 1,
The data-based statistical function model corresponds to a Gaussian process model. The method according to claim 1 or 2,
The crankshaft torque is determined as the sum of the drive torque and the plurality of load torques, and at least one drive torque as the high pressure torque is determined via a data based function model. The method of claim 3,
A statistical function model for calculating the high pressure torque is calculated based on the input variable, the input variable being:
Crankshaft rotation speed, injection rate of main injection, injection angle of main injection, rail pressure, air mass, EGR mass, gas temperature of suction mass flow, pilot injection quantity, post injection quantity, post injection injection crankshaft angle, oil temperature And swirl valve position
Corresponding to or selected from. The method according to claim 3 or 4,
One of the load torques is a low pressure torque, the low pressure torque representing a gas exchange loss of the combustion engine (10), characterized in that it is determined through a data based function model. The method of claim 5,
A statistical function model for calculating low pressure torque is calculated based on an input variable, the input variable being:
Crankshaft rotation speed, exhaust gas back pressure, suction line pressure, partial flap position, EGR-ratio, suction temperature and oil temperature
Corresponding to or selected from. The method according to any one of claims 1 to 6,
As additional load torque, at least one of friction torque and assembly load torque is considered. A control module (1) for an engine system comprising a combustion engine (10), said control module being configured for carrying out the method according to any one of the preceding claims. A computer program set up to carry out the method of claim 1. A storage medium readable by an apparatus in which the computer program of claim 9 is stored. An electronic control device comprising the electronic storage medium of claim 10.

Images