JP4835483B2 - Prediction device, control device, and prediction program - Google Patents

Description

  The present invention relates to performance prediction for devices such as engines.

  Conventionally, various apparatuses are actually operated under various conditions and their performances are examined. However, if there are many operating conditions such as an engine, it is difficult to confirm the performance under all conditions.

  Therefore, a model for simulating the apparatus is created, and the performance of the apparatus under various conditions is predicted by simulation using the model.

  And the operation control of the apparatus suitable for various conditions is performed using the result of such a simulation.

  For example, in Patent Document 1, in order to achieve early activation of the catalyst while maintaining good drivability, the required torque generated by engine combustion is calculated based on the accelerator opening degree operated by the driver. At the same time, MBT (optimum ignition timing) estimated torque and retard limit estimated torque are calculated.

  Further, in Patent Document 2, an improvement is made so that an internal combustion engine model can be converted into data as accurately as possible with as little measurement cost as possible, and an optimum characteristic amount can be obtained for the control of the internal combustion engine. In order to do so, at least one local model of a relatively low order, for example, is converted into data for each predetermined operation of the internal combustion engine.

JP-A-2005-171979 JP 2000-248991 A

  However, in the above conventional example, there is no torque and torque fluctuation estimation model that satisfies the required accuracy of adaptation, and therefore, sufficient man-hour reduction has been realized for adaptation to achieve both drivability and emissions. Absent. Further, the required control performance (for example, compatibility between drivability and emission) is not realized.

The present invention provides a prediction apparatus for predicting an output torque or output torque fluctuation of the engine, in addition to the operating conditions of the engine, the minimum is a parameter that takes into account the physical mechanisms of generation phenomena in the operation of the engine of the engine flame Regression formulas with parameters including propagation energy as regression factors are stored. Regression factors during actual operation of the engine and output torque or output torque fluctuation data are input. The coefficient of the regression factor is estimated, and the output torque of the engine or the output torque fluctuation is predicted using the regression equation specified by the estimated coefficient.

  Moreover, it is a control apparatus which memorize | stored the regression equation which specified the coefficient obtained in the said prediction apparatus, Comprising: It is suitable for a vehicle to control the engine of the vehicle.

Further, the present invention provides a prediction program for predicting output torque or output torque fluctuation of the engine, the computer, in addition to the operating conditions of the engine, met the parameters taking into account the physical mechanisms of generation phenomena in the operation of the engine The regression equation with the parameters including the minimum flame propagation energy of the engine as a regression factor is stored, the regression factor during actual operation of the engine and the output torque or output torque fluctuation data are input, and the regression equation is regressed. to estimate the coefficients of the regression factors in the formula, characterized in that to predict the output torque or the output torque variation of the engine using the identified regression equation by the coefficient obtained by estimation.

  As described above, according to the present invention, more accurate prediction or estimation can be performed by adding information that considers not only the actual regression factors such as operating conditions but also the physical mechanism of the phenomenon to the regression factors. Can do. Further, by improving the accuracy, it becomes possible to satisfy the required accuracy required for a significant reduction in the number of test points and test man-hours.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  In the present embodiment, in order to appropriately control the engine, a torque fluctuation of the engine and a model for estimating the torque are obtained. All models in the present embodiment are regression equations.

  To determine the regression equation, first consider which regression factor to use. Table 1 shows the regression factors examined. A test is performed to identify the engine torque fluctuation and the regression equation for estimating the torque based on these regression factors. The regression equation is identified by using Partial Last Squares (PLS).

*回帰では、各回帰要因の変動幅で規格化（スケーリング）を行う。

* In regression, normalization (scaling) is performed with the fluctuation range of each regression factor.

  Here, in Table 1, as a result of examination by PLS, there are many regression factors that are not used in the final regression equation. There are three types of torque fluctuations and torques, respectively, but the torque fluctuation is not used as a regression factor in the regression of torque fluctuations, even if the types are different. The same applies to the torque.

[Physical Base Regression Formula]
Here, in the present embodiment, not only operating conditions and numerical values of test data are used as regression factors, but important physical quantities are found by considering the physical meaning or origin of torque fluctuation and torque itself. Is added to the regression factors. That is, in Table 1, the minimum flame propagation energy is a regression factor. This flame propagation energy is an important physical quantity for engine torque fluctuation and torque, and the regression equation of this embodiment using this as a regression factor is called a physics-augmented regression.

[Regression data]
The engine test data used for identifying the regression equation is data of 754 operating conditions (27 items excluding the minimum flame propagation energy in Table 1). Table 2 outlines the test conditions. Thus, each regression factor, torque fluctuation, and torque were measured under various operating conditions.

[Torque fluctuation estimation model]
If the cause of engine torque fluctuation is combustion fluctuation, the factor that determines the stability of the flame seems to be important. However, the mechanism of torque fluctuation, that is, cycle fluctuation is not clear, and a physical model that can predict it is not known.

  Therefore, based on the collected engine test data, a regression equation of torque fluctuation was created using the operating conditions and various data at that time as regression factors. However, in the present embodiment, in order to consider a physical mechanism, the minimum flame propagation energy is added to the regression factor to improve the prediction accuracy of the regression equation.

[Minimum flame propagation energy]
The torque fluctuation of a gasoline engine is considered to be one of the manifestations of cycle fluctuation, but when it is extreme, a misfire cycle appears. There are two possible causes of misfire. One is a spark ignition failure, and the other is a case where the flame propagation is too slow and the exhaust valve opens before the flame covers the entire combustion chamber.

  First, consider the failure of spark ignition. If the spark ignition energy is insufficient, a flame that can be propagated by spark ignition is not formed. The minimum energy required for ignition of the air-fuel mixture is called minimum ignition energy, which is generally expressed by the following equation (1) ("Combination Theory, Second Edition", FA Williams, Benjamin / Cummings). , Menlo Park, 1985.).

Where, α: minimum ignition energy, Q _spark : spark energy, Q _loss : energy loss to the ignition electrode, δ: laminar flame thickness, ρ _u : unburned mixture density, C _pu : unburned mixture The constant-pressure specific heat of the gas, T _f : the adiabatic flame temperature of the mixture, and T _u : the temperature of the unburned mixture.

The term δ ^{2 in} equation (1) is a case where it is assumed that the shape of the flame kernel formed by spark ignition is cylindrical, and the unit of α is [energy / length]. If the shape of the flame kernel is flat, δ should be used instead of δ ² , and the unit of α in this case is [square of energy / length]. If the shape of the flame kernel is a sphere, δ ³ should be used, and the unit of α in that case is [energy].

  On the other hand, the situation where the exhaust valve opens before the flame propagation ends corresponds to a case where the ignition timing is greatly retarded. Test data confirms that such a cycle also exists. However, there is no cycle in which spark ignition has failed. That is, torque fluctuations exist even under operating conditions without these misfiring cycles. Therefore, it is necessary to search for a regression factor that can describe combustion fluctuations or flame stability under such operating conditions.

  Therefore, as described below, the minimum ignition energy of Equation 1 was reinterpreted, and a regression factor describing the combustion fluctuation or flame stability was found.

[Interpretation of minimum ignition energy]
The minimum ignition energy is the minimum energy required to give energy to the unburned mixture by some means and raise its temperature to the adiabatic flame temperature. Therefore, the value is not limited to spark ignition but also holds in flame propagation after spark ignition. That is, as shown in FIG. 1, in the laminar flame, energy is transferred from the burned side to the unburned side by heat conduction, and ignition (flame propagation) of the unburned portion occurs. Therefore, the minimum energy required for flame propagation is equal to the minimum ignition energy.

  The flame in engine combustion is a turbulent flame, but in many cases it is said that the structure can be described as a saddle-like laminar flame, so the energy required for flame propagation in the engine cylinder (minimum flame propagation energy) ) Is considered to be equal to the minimum ignition energy.

  Here, the mathematical relationship between the minimum flame propagation energy and the stability or combustion fluctuation of the flame is unknown. However, if, for some reason, the amount of energy transfer from the burned side to the unburned side fluctuates around the value of the minimum flame propagation energy, the minimum flame propagation energy is large in terms of flame stability or combustion fluctuation. It is considered to have an influence. Therefore, the minimum flame propagation energy (= minimum ignition energy: MFPE) is selected as a candidate for a regression factor that describes combustion fluctuation or flame stability.

  In the present embodiment, the value of the minimum flame propagation energy is obtained based on the assumption that the fuel is normal heptane based on the temperature, pressure, and composition of the unburned mixture at the ignition timing. However, in practice, it is only necessary to select and obtain conditions as appropriate according to the target device.

[Regression of torque fluctuation]
The torque fluctuation was considered to be caused by combustion fluctuation or flame instability, and regression using minimum flame propagation energy (MFPE) was performed.

  FIG. 2 shows the regression accuracy of torque fluctuation (torque fluctuation shown) in the compression + expansion process. It has been confirmed that the regression accuracy of torque fluctuation for one cycle (torque fluctuation shown in the figure) is almost the same as the torque fluctuation in the compression + expansion process.

  The test data obtained under the above-mentioned 754 operating conditions was processed by PLS to remove the regression factors having little influence on the torque fluctuation, and the regression equation (2) was obtained. In this regression equation, the operating conditions shown in Table 2 and the exhaust temperature are excluded from the regression factors. In regression, normalization (scaling) is performed based on the fluctuation range of each regression factor.

Torque fluctuation (Nm) = C ₀ + C ₁ * KL (%) + C ₂ * Indicated torque (Nm) + C ₃ * Ga (g / s) + C ₄ * Surge tank pressure (abs.kPa) + C ₅ * MFPE (erg / cm ³ ) + C ₆ * cooling water temperature (° C.) + C ₇ * ignition timing (degBTDC) (2)

  Here, KL: load, Ga: intake air flow rate, MFPE: minimum flame propagation energy. * Represents multiplication.

  FIG. 2A shows an estimated value and actual value (observed value) of torque fluctuation obtained from the regression equation of Expression (2). On the other hand, FIG. 2B shows a regression result when the minimum flame propagation energy (MFPE) is removed from Equation 2.

  Comparing FIG. 2A and FIG. 2B, it can be seen that the regression accuracy is remarkably improved by including MFPE in the regression factors. That is, it can be seen that MFPE is one of the important factors that determine the value of torque fluctuation.

  Also, looking at the regression factors included in equation (2), factors other than the indicated torque, MFPE, cooling water temperature, and ignition timing are the load KL, the intake flow rate Ga, and the surge tank pressure, all of which are related to intake air. It is. That is, it can be seen that the influence of the intake process on the torque fluctuation is great. 2 is a standard deviation of error (regression value−observed value), and is about 0.294 Nm when MFPE is included, and about 0.395 Nm when MFPE is not included. Consider MFPE. Thus, it is understood that the regression accuracy becomes high.

[Return of torque]
Since torque itself is not at least directly related to combustion fluctuations, adding MFPE to the regression factor does not seem meaningful at first glance. However, since MFPE is also the amount of heat generated by the flame (see FIG. 1), it should be greatly involved in determining the torque. Thus, MFPE was added to the regression factor to determine the torque regression equation of equation (3). This determination was made using PLS.

  FIG. 3 shows the regression accuracy of torque (illustrated torque) in the compression + expansion process. It has been confirmed that the regression accuracy of the torque for one cycle (illustration torque) is almost the same as the torque of the compression + expansion process.

Torque (Nm) = C ₀ + C ₁ * Engine speed (rpm) + C ₂ * Load (%) + C ₃ * Air / fuel ratio + C ₄ * Ignition timing (degBTDC) + C ₅ * Cooling water temperature (° C) + C ₆ * MFP (Erg / cm ³ ) (3)

  FIG. 3A shows an estimated torque value and an actual value (observed value) obtained from the regression equation. On the other hand, FIG. 3B shows a regression result when the minimum flame propagation energy (MFP) is removed from the equation (3). Comparing FIG. 3A and FIG. 3B, it can be seen that the regression accuracy is remarkably improved by including MFPE as a regression factor. That is, it can be seen that MFPE is one of the important factors that determine the torque value. Note that σ in FIG. 3 is a standard deviation of an error (regression value−observed value), which is about 5.149 Nm when MFPE is included, and about 8.459 Nm when MFPE is not included. Thus, it is understood that the regression accuracy becomes high.

[Unit of minimum flame propagation energy]
As described above, the minimum ignition energy is expressed by using δ or δ ³ instead of the term δ ^{2 in} the equation (1) depending on the shape of the flame kernel in the spark ignition, and the unit is different. However, it was initially unknown what unit should be used as minimum flame propagation energy (MFP) to describe flame stability in flame propagation rather than spark ignition. Therefore, the regression accuracy was examined by adding MFPE when δ or δ ^{3 is} used or when nothing is used to the term δ ² of Equation 1 as a regression factor.

As a result, it was found that MFPE (erg / cm ³ ) in the case where the term δ ² or δ or δ ³ of Equation (1) is not used at all improves the torque fluctuation and the torque regression accuracy most.

  That is, the use of MFPE represented by Equation (4) as a regression factor improves torque fluctuation and torque regression accuracy.

In addition, instead of MFPE (erg / cm ³ ) in Equation (4), the minimum flame propagation energy MFPE (erg / g) per unit mass of the unburned mixture and the density ρ _u of the unburned mixture are simultaneously regression factors. It was also confirmed that an improvement in regression accuracy equivalent to MFPE (erg / cm ³ ) can be obtained. That is, for the minimum flame propagation energy MFPE, it is also preferable to use the equation (5) and add the density ρ _u of the unburned mixture to the regression factor.

  As described above, it was found that a regression equation with high regression accuracy can be obtained by using the minimum flame propagation energy MFPE as a regression factor of a regression equation for estimating engine torque fluctuation and torque.

  FIG. 4 shows a flowchart of processing in the present invention. First, a regression factor that determines the target to be estimated is determined (S11). Then, a regression equation using the determined regression factor is input to the computer. Here, the regression factors include factors that take into account the target physical mechanism.

  Next, tests are performed under various conditions using an actual machine, and data is collected (S12). Then, the obtained data is input to the computer. The computer executes the stored application program, determines the coefficient of each regression factor of the regression equation by PLS, and obtains the regression equation (S13). Then, the obtained regression equation is mounted on an actual machine and used for target control.

  FIG. 5 is an example applied to engine control. A memory 12 that stores a regression equation is connected to the engine ECU 10. The engine ECU 10 controls fuel supply to the engine 14, ignition timing, and the like. For example, the engine ECU 10 determines the target output torque of the engine 14 from the depressed state of the accelerator, and determines the fuel supply and ignition timing when controlling the engine 14, but is stored in the memory 12 at this time. The torque fluctuation and torque are estimated using the regression equation. Then, by controlling the engine 14 using this estimation result, it is possible to control the operation of the engine 14 so as to suppress the torque fluctuation and output the target torque while maintaining the emission at an appropriate level. .

  As described above, by mounting the ECU on the model vehicle having the above regression equation, it is possible to estimate engine torque fluctuation and torque in the vehicle ECU. Therefore, it is possible to easily perform engine control that achieves a given target setting (for example, compatibility between emission and drivability) while estimating engine torque fluctuation and torque. In other words, it is preferable to use an estimated value of torque fluctuation and torque for control of various actuators such as engine torque control, engine speed control, air-fuel ratio control, ignition timing control, intake / exhaust valve control, and the like.

  In addition, instead of being implemented in the ECU, the computer uses the regression equation obtained as described above to estimate the target motion under various conditions, and uses this estimation result to determine the target control method. It is also suitable.

  In this way, by adding information that takes into account not only the obvious regression factors such as operating conditions but also the physical mechanism of the phenomenon to the regression factors, more accurate prediction or estimation can be performed. Further, by improving the accuracy, it becomes possible to satisfy the required accuracy required for a significant reduction in the number of test points and test man-hours.

Furthermore, for the following phenomenon, it is preferable to add information that takes into account the physical mechanism of the phenomenon to the regression factor.
(1) As a regression factor of the regression equation for estimating the hydrocarbon concentration in the engine exhaust, the thickness of the flame extinguishing layer in the engine combustion chamber, the gas mixture density, or the product thereof is added. Or, instead of the thickness of the quenching layer, one of the factors that determine the thickness of the quenching layer is the exhaust temperature, laminar combustion speed, calorific value, mixture density, mixture pressure, combustion chamber wall temperature, etc. . The envisaged physical mechanism is that extinguishing the fuel in the air-fuel mixture is thought to be burned and released into the engine exhaust.
(2) The exhaust gas temperature is added as a regression factor of the regression equation for estimating the NOx concentration in the engine exhaust gas. The assumed physical mechanism is a so-called extended Zeldovich mechanism in which the NOx concentration increases as the temperature of the burned gas increases.

  In addition, the identification of the coefficient in the regression equation in the present embodiment is performed by installing a program for performing a calculation process including a PLS calculation in a general-purpose computer and executing the program. The test data is input by storing the output from the measuring device in a memory and reading it from the memory, but a keyboard or the like may be used. Data that has already been input may be input via the communication port. The processing result is displayed on a display and output from a printer or the like. The program is generally provided by a storage medium such as a CD-ROM or DVD, but may be provided by downloading from a predetermined site.

It is a figure showing the mechanism of laminar flame propagation. It is a figure which shows the regression accuracy of a torque fluctuation. It is a figure which shows the regression accuracy of a torque. It is a flowchart of the whole process. It is a figure which shows the mounting to ECU.

Explanation of symbols

  10 engine ECU, 12 memory, 14 engine.

Claims (3)

A prediction device for predicting engine output torque or output torque fluctuation,
In addition to the operating conditions of the engine, this is a parameter that takes into account the physical mechanism of the phenomenon that occurs in the operation of the engine, and the minimum flame propagation energy of the engine that is equal to the minimum ignition energy, which is the minimum energy required for ignition of the air-fuel mixture. The regression equation with the included parameters as regression factors is stored,
Input regression factors and output torque or output torque fluctuation data during actual operation of the engine, and for the regression equation, estimate the coefficient of each regression factor in the regression equation,
A prediction apparatus for predicting output torque or output torque fluctuation of the engine using a regression equation specified by a coefficient obtained by estimation. A control device that stores a regression equation that specifies a coefficient obtained in the prediction device according to claim 1,
A control device mounted on a vehicle and controlling an engine of the vehicle. A prediction program for predicting engine output torque or output torque fluctuation,
On the computer,
In addition to the operating conditions of the engine, this is a parameter that takes into account the physical mechanism of the phenomenon that occurs in the operation of the engine, and the minimum flame propagation energy of the engine that is equal to the minimum ignition energy, which is the minimum energy required for ignition of the air-fuel mixture. Store the regression equation with the included parameters as regression factors,
Input regression factors and output torque or output torque fluctuation data during actual operation of the engine, and for the regression equation, estimate the coefficient of each regression factor in the regression equation,
A prediction program for predicting output torque or output torque fluctuation of the engine using a regression equation specified by a coefficient obtained by estimation.

Images