JP2013083239A - Control device - Google Patents

Abstract

The accuracy of control using a control parameter is improved without increasing the number of learning points of the control parameter.
A delay time td associated with a pressure P is stored in a map M, and learning is performed by updating the value of the delay time td1 in the map based on detected values PK and tdK of the pressure and the delay time. Assumption. Then, a delay time corresponding to the pressure detection value PK used for learning is calculated by linear interpolation from a plurality of delay times td1 ′ and td3 in the map, and the calculated value tdKα and the detection value tdK of the delay time are calculated. An interpolation error ΔtdK that is an error is learned in advance. Then, the delay time tdJα corresponding to the current pressure PJ is calculated by linear interpolation from a plurality of delay times td1 ′ and td3 in the map, and the calculated delay time tdJα is set to the learned interpolation error ΔtdK. Correct based on. The fuel injection valve is controlled based on the corrected delay time tdJ.
[Selection] Figure 4

Description

  The present invention relates to a control device that calculates a control parameter according to the current environment by interpolating from a learned control parameter, and controls a control target based on the calculated control parameter.

  As an example of this type of control parameter, there is an injection start delay time td from when an injection start command is output to the fuel injection valve of the internal combustion engine until actual injection is performed (see, for example, Patent Document 1). In Patent Document 1, the delay time td is measured by detecting the time when the fuel pressure starts to drop with the start of fuel injection using a fuel pressure sensor mounted on the fuel injection valve. Then, the delay time td is learned based on the measured value, and the timing for outputting the injection start command is controlled based on the learned delay time td.

  The delay time td becomes a different value depending on the fuel pressure (fuel pressure at the start of injection) supplied to the fuel injection valve. Therefore, the present inventor studied to learn as follows by associating the delay time td (control parameter) with the fuel pressure (environment variable).

  That is, the delay times td (50), td (100), and td (150) corresponding to specific values of fuel pressure (50 MPa, 100 MPa, and 150 MPa in the example of FIG. 11) are stored and updated in the map as respective learning values. To go. For example, when the measured delay time td (120act) is at a pressure of 120 MPa, the specific delay time td (100) closest to the measured pressure is updated as a learning target. Specifically, a point td obtained by linearly interpolating td (50) and td (120act), that is, a point where a straight line L1 connecting the learned value td (50) and the measured value td (120act) intersects 100 MPa. The learning value td (100) is updated to (100α).

  When the current pressure is 130 MPa, the two learning values td (100α) and td (150) are linearly interpolated to calculate a delay time value td (130) corresponding to 130 MPa, and the calculated value td ( 130) is used to control the timing of outputting the injection start command.

  However, the characteristic line representing the relationship between the delay time td and the pressure P is a curve as shown by the dotted line L2 and lacks linearity. Therefore, an interpolation error Δtd (130) occurs between the value td (130) obtained by linear interpolation as described above and the actual value td (130act). As a result, it is impaired to control the injection start timing with high accuracy.

JP 2009-57924 A

  To solve this problem, if the number of divisions of the specific value is increased on the pressure axis of the map to increase the number of learning points of the delay time td, the interpolation error Δtd (130) can be reduced and the control accuracy can be improved. However, this type of control device is also required to reduce the required memory capacity by reducing the number of learning points as much as possible. For this reason, simply increasing the number of learning points cannot achieve both reduction in the required memory capacity and improvement in control accuracy.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a control device that can improve the accuracy of control using control parameters without increasing the number of learning points of the control parameters. It is to provide.

  Hereinafter, means for solving the above-described problems and the operation and effects thereof will be described. In addition, the code | symbol in the bracket | parenthesis in Claim 1, 2 illustrates the correspondence with FIG. 4, The code | symbol in the bracket | parenthesis in Claim 3 illustrates the correspondence with FIGS. The reference numerals in parentheses in claim 7 exemplify the correspondence with FIG. 12, and the invention is not limited to the illustrated embodiment.

  In the first aspect of the present invention, storage means for storing the control parameter (td) in association with a specific value of the predetermined environment variable (P), detection means for detecting the environment variable and the control parameter, and the detection means Control parameter learning means for learning by updating the value of the control parameter (td1) stored in the storage means on the basis of the detected value (PK, tdK) by the control, and the environment used for learning of the control parameter learning means A control parameter corresponding to the detected value (PK) of the variable is calculated by linear interpolation from a plurality of control parameters (td1 ′, td3) stored in the storage means, and the calculated value (tdKα) and the control parameter Interpolation error learning means for learning an interpolation error (ΔtdK) that is an error from the detected value (tdK), and control parameters corresponding to the current environment variable (PJ). Current control parameter interpolation means for calculating data (tdJα) by linear interpolation from a plurality of control parameters (td1 ′, td3) stored in the storage means, and the interpolation error learned by the interpolation error learning means Correction means for correcting the control parameter (tdJα) calculated by the current control parameter interpolation means based on (ΔtdK), control means for controlling the control object based on the control parameter (tdJ) corrected by the correction means, It is characterized by providing.

  In short, when the value (PK, tdK) used for learning the control parameter (td1) is detected, the value of td1 is updated based on the detected value (tdK) and learning is performed, and the control corresponding to the detected value (PK) is performed. Parameters are calculated by linear interpolation from a plurality of control parameters (td1 ′, td3), and an error (interpolation error (ΔtdK)) between the calculated value (tdKα) and the detected value (tdK) is learned. Then, in calculating the control parameter (tdJα) corresponding to the current environment variable (PJ) by linear interpolation from the plurality of control parameters (td1 ′, td3), the learned interpolation error (ΔtdK) is taken into account. To calculate.

  Therefore, according to the above invention, the value of the control parameter (tdJ) used for the control is increased without increasing the number of divisions of the specific value of the environment variable set in the storage means and increasing the learning point of the control parameter. It can be close to the actual value (value indicated by the dotted line L2 in FIG. 11), and the controlled object can be controlled with high accuracy. Therefore, it is possible to achieve both reduction in the required storage capacity of the storage means and improvement in control accuracy.

  In the invention according to claim 2, as the difference between the detected value (PK) of the environmental variable used in the interpolation error learning means and the current environmental variable (PJ) used in the current control parameter interpolation means increases, The correction means reduces the degree to which the interpolation error (ΔtdK) is reflected in the correction.

  Here, in the following description, as illustrated in FIG. 4, the actual value (detected value) among the control parameters corresponding to the detected value (PK) of the environment variable is tdK, and the value obtained by linear interpolation is tdKα. Call it. In addition, a value obtained by linear interpolation among control parameters corresponding to the current environment variable (PJ) is referred to as tdJα, and an actual value is referred to as tdJ. The difference between tdK and tdKα (interpolation error (ΔtdK)) and the difference between tdJ and tdJα are more likely to become different values as the difference between PK and PJ increases.

  In view of this point, in the above invention, the greater the difference between PK and PJ, the smaller the degree to which the interpolation error (ΔtdK) is reflected in the correction, so that the value of the control parameter (tdJ) used for the control is brought closer to the actual value. It is possible to promote the improvement of control accuracy.

  According to a third aspect of the present invention, a control parameter (td) is stored in the storage means in association with a plurality of environment variables (P, Q), and the plurality of environment variables (P, Q) and the control are stored. The present invention relates to a learning method by control parameter learning means in spatial coordinates using the parameter (td) as a reference axis. The control parameter learning means according to the present invention is based on the control parameter (tdKβ) of the detection reference point (Kβ) described below, and the value of the control parameter (tdG1) of the update target learning point (G1) described below. It is characterized by updating and learning.

Various terms defined in the present invention, such as the update target learning point (G1) and the detection reference point (Kβ), are as follows.
Detection point (K): A point representing a detection value (PK, QK, tdK) by the detection means.
Learning points (G1 to G9): points representing learning values of the control parameters.
Update target learning point (G1): A learning point corresponding to the environment variable closest to the environment variable of the detected value among the plurality of learning points.
-Learning point for interpolation (G2): A learning point at a position where the update target learning point (G1) is interpolated with the detection point (K) among the plurality of learning points.
Learning surface (SG): a surface including a plurality of the learning points.
First point (K1): among the learning plane (SG), the first environment variable (PK) at the detection point (K) and the second environment variable (QG2) at the learning point for interpolation (G2) Corresponding point.
Second point (K2): A point corresponding to the first environment variable (PG2) of the interpolation learning point and the second environment variable (QK) of the detection point.
Detection reference plane (SK): a plane including at least one of the first point (K1) and the second point (K2), the interpolation learning point (G2), and the detection point (K).
Detection reference point (Kβ): The same point as the environment variables (PG1, QG1) of the update target learning point (G1) in the detection reference plane (SK).

  By the way, the present inventor also examined a method of updating the update target learning point (G1) as follows, contrary to the above invention. That is, the value obtained by linear interpolation between the control parameter (tdG2) of the interpolation learning point (G2) and the control parameter (tdK) of the detection point (K) is used as the control parameter of the update target learning point (G1). This is a method of setting an updated value of (tdG1).

  However, on a plane including two points of the interpolation learning point (G2) and the detection point (K) and parallel to the axis of the control parameter (td) (see the one-dot chain line L3 in FIG. 8). The update target learning point (G1) is not necessarily located. Therefore, in the above method, when performing linear interpolation between tdG2 and tdK, it is difficult to perform linear interpolation in consideration of the shift amount between the planes L3 and G1. In short, the shift amount of the environment variables (P, Q) between the detection point (K) and the update target learning point (G1) on the PQ plane in FIG. 8 (shift amount between K and G1 on the PQ plane in FIG. 8). Accordingly, it is required to calculate the value of tdG to tdG1 with high accuracy. However, it is difficult to increase the calculation accuracy with the above method.

  In the above invention in view of the above points, a point corresponding to (PK, QG2) in the learning plane (SG) is a first point (K1), and a point corresponding to (PG2, QK) is a second point (K2). In the detection reference plane (SK) including at least one of the first point (K1) and the second point (K2), the interpolation learning point (G2), and the detection point (K), Based on the detection reference point (Kβ) corresponding to the update target learning point (G1), the value of the update target learning point (G1) is updated.

  Therefore, among the deviation amounts of the environment variables (P, Q) between the detection point (K) and the update target learning point (G1), the error of the control parameter due to the deviation amount of the first environment variable (P) is This is solved by including the first point (K1) in the detection reference plane (SK). Alternatively, the error of the control parameter due to the amount of deviation of the second environment variable (Q) is eliminated by including the second point (K2) in the detection reference plane (SK).

  Therefore, even if the update target learning point (G1) is not located on the plane L3 described above, depending on the shift amount of the environment variable between the detection point (K) and the update target learning point (G1). , TdK to tdG1 can be calculated with high accuracy. Therefore, the accuracy of learning by the control parameter learning means can be improved, and as a result, improvement of the accuracy of control can be promoted.

  According to a fourth aspect of the present invention, the control parameter learning unit approximates the control parameter stored in the storage unit and the value obtained by linear interpolation from the detected value to the value of the control parameter before update by a predetermined ratio. And an annealing processing means that uses the corrected value as a learning update value.

  According to the above invention, the value obtained by linear interpolation from the detected value is not used as the update value of the control parameter as it is, but is updated by performing the annealing process, so that the influence of the detection error included in the detected value is reduced. Accordingly, it is possible to suppress the update value of the control parameter from becoming hunted and unstable.

  In the invention according to claim 5, the control object is a fuel injection valve that injects fuel to be used for combustion of an internal combustion engine, and the fuel injection valve is equipped with a fuel pressure sensor that detects fuel pressure, Based on the detected value of the fuel pressure sensor, a fuel pressure waveform detecting means for detecting a change in fuel pressure caused by injection as a fuel pressure waveform, and on the basis of the detected fuel pressure waveform, an injection rate waveform corresponding to the fuel pressure waveform is specified. Injection rate parameter calculation means for calculating an injection rate parameter required for the control parameter, and the detected value of the control parameter is the injection rate parameter calculated by the injection rate parameter calculation means.

  Specific examples of the injection rate parameter include an injection start delay time td described below. That is, since the detected pressure of the fuel pressure sensor starts to decrease with the start of injection, the actual injection start timing can be detected by detecting the decrease start timing. Therefore, it is possible to detect the delay time td from when the injection start command signal is output to the fuel injection valve until when the actual injection is started. However, since the delay time td changes depending on the fuel pressure and the injection amount at the injection start time at that time, the delay time td (control parameter) is learned in association with the fuel pressure (environment variable) and the injection amount (environment variable). The output timing of the injection command signal is controlled based on the learned delay time td.

The figure which shows the outline of the fuel-injection system with which the control apparatus concerning 1st Embodiment is applied. The figure which shows the change of the injection rate, fuel pressure, and differential value corresponding to an injection command signal. FIG. 3 is a block diagram showing an overview of injection rate parameter learning, injection command signal setting, and the like in the first embodiment. The figure explaining the update method of a learning value in the injection rate parameter map of 1st Embodiment. 6 is a flowchart illustrating a procedure for learning an injection rate parameter and an interpolation error in the first embodiment. The flowchart which shows the procedure which correct | amends the injection rate parameter using an interpolation error in 1st Embodiment. The figure explaining the update method of a learning value in the injection rate parameter map of 2nd Embodiment. The figure which planarly viewed FIG. 7 from the direction of the injection rate parameter. The flowchart which shows the procedure which correct | amends the injection rate parameter using interpolation error in 2nd Embodiment. The figure which planarly viewed FIG. 9 from the direction of the injection rate parameter. The figure explaining the subject in case correction | amendment by an interpolation error is not implemented. The figure explaining the update method of a learning value in the injection rate parameter map of 3rd Embodiment. The figure which planarly viewed FIG. 12 from the direction of the injection rate parameter. The figure explaining the update method of a learning value, and the calculation method of an interpolation error in other embodiment.

  Hereinafter, each embodiment which actualized the control device concerning the present invention is described based on a drawing. In the following embodiments, parts that are the same or equivalent to each other are denoted by the same reference numerals in the drawings, and the description of the same reference numerals is used. Further, the control device described below is mounted on a vehicle engine (internal combustion engine), in which high pressure fuel is injected into a plurality of cylinders # 1 to # 4 to perform compression auto-ignition combustion. A diesel engine is assumed.

(First embodiment)
FIG. 1 is a schematic diagram showing a fuel injection valve 10 mounted on each cylinder of the engine, a fuel pressure sensor 20 mounted on each fuel injection valve 10, an ECU 30 that is an electronic control device mounted on a vehicle, and the like. is there. First, an engine fuel injection system including a fuel injection valve 10 will be described with reference to FIG.

  The fuel in the fuel tank 40 is pumped and stored in the common rail 42 (pressure accumulating container) by the fuel pump 41, and distributed and supplied to the fuel injection valves 10 (# 1 to # 4) of each cylinder. The plurality of fuel injection valves 10 (# 1 to # 4) sequentially inject fuel in a preset order. In addition, since the plunger pump is used for the fuel pump 41, fuel is pumped in synchronism with the reciprocating motion of the plunger.

  The fuel injection valve 10 (control target) includes a body 11, a needle-shaped valve body 12, an actuator 13, and the like described below. The body 11 forms a high-pressure passage 11a inside and a nozzle hole 11b for injecting fuel. The valve body 12 is accommodated in the body 11 and opens and closes the nozzle hole 11b.

  A back pressure chamber 11c for applying a back pressure to the valve body 12 is formed in the body 11, and the high pressure passage 11a and the low pressure passage 11d are connected to the back pressure chamber 11c. The communication state between the high pressure passage 11a and the low pressure passage 11d and the back pressure chamber 11c is switched by the control valve 14, and the actuator 13 such as an electromagnetic coil or a piezoelectric element is energized to push the control valve 14 downward in FIG. As a result, the back pressure chamber 11c communicates with the low pressure passage 11d and the fuel pressure in the back pressure chamber 11c decreases. As a result, the back pressure applied to the valve body 12 is lowered and the valve body 12 is lifted up (opening operation). On the other hand, when the power supply to the actuator 13 is turned off and the control valve 14 is operated upward in FIG. 1, the back pressure chamber 11c communicates with the high pressure passage 11a and the fuel pressure in the back pressure chamber 11c increases. As a result, the back pressure applied to the valve body 12 increases and the valve body 12 is lifted down (closed valve operation).

  Therefore, the ECU 30 controls the energization of the actuator 13 so that the opening / closing operation of the valve body 12 is controlled. Thereby, the high-pressure fuel supplied from the common rail 42 to the high-pressure passage 11 a is injected from the injection hole 11 b according to the opening / closing operation of the valve body 12.

  The fuel pressure sensor 20 includes a stem 21 (distortion body), a pressure sensor element 22, a mold IC 23, and the like described below. The stem 21 is attached to the body 11, and the diaphragm portion 21a formed on the stem 21 is elastically deformed by receiving the pressure of the high-pressure fuel flowing through the high-pressure passage 11a. The pressure sensor element 22 is attached to the diaphragm portion 21a, and outputs a pressure detection signal in accordance with the amount of elastic deformation generated in the diaphragm portion 21a.

  The mold IC 23 is formed by resin molding electronic components such as an amplification circuit that amplifies the pressure detection signal output from the pressure sensor element 22 and a transmission circuit that transmits the pressure detection signal. 10 is installed. A connector 15 is provided on the upper portion of the body 11, and the mold IC 23, the actuator 13, and the ECU 30 are electrically connected by a harness 16 (signal line) connected to the connector 15. The amplified pressure detection signal is transmitted to the ECU 30 and received by a receiving circuit included in the ECU 30. This communication process for transmission / reception is performed for each fuel pressure sensor 20 of each cylinder.

  The ECU 30 calculates the target injection state (for example, the number of injection stages, the injection start timing, the injection end timing, the injection amount, etc.) based on the engine operation state such as the accelerator pedal operation amount, the engine load, and the engine rotational speed NE. For example, the optimal injection state corresponding to the engine load and the engine rotation speed is stored in the memory 30a as an injection state map. Based on the current engine load and engine speed, the target injection state is calculated with reference to the injection state map. Then, injection command signals t1, t2, and Tq (see FIG. 2A) corresponding to the calculated target injection state are based on injection rate parameters td, te, Rα, Rβ, and Rmax (control parameters) described in detail later. The operation of the fuel injection valve 10 is controlled by setting and outputting to the fuel injection valve 10.

  Next, a method of injection control when fuel is injected from the fuel injection valve 10 will be described below with reference to FIGS. 2 and 3.

  The ECU 30 (fuel pressure waveform detecting means), for example, when fuel is injected by the fuel injection valve 10 of the # 1 cylinder, is based on the detection value of the fuel pressure sensor 20 (# 1) mounted on the fuel injection valve 10 and accompanying the injection. The change in the generated fuel pressure is detected as a fuel pressure waveform (see FIG. 2C). Then, the injection state is detected by calculating an injection rate waveform (see FIG. 2B) representing a change in the fuel injection amount per unit time based on the detected fuel pressure waveform. Then, while learning the injection rate parameters Rα, Rβ, and Rmax that specify the detected injection rate waveform (injection state), the correlation between the injection command signals (pulse-on timing t1, pulse-off timing t2, and pulse-on period Tq) and the injection state. The injection rate parameters td and te for specifying

  Specifically, in the fuel pressure waveform, a descending approximation line that approximates a descending waveform from the inflection point P1 at which the fuel pressure drop starts at the start of injection to the inflection point P2 at which the descent ends by a least square method or the like. Lα is calculated. Then, a time (a crossing time LBα between Lα and Bα) that is the reference value Bα in the descending approximate straight line Lα is calculated. Focusing on the fact that the intersection time LBα and the injection start time R1 are highly correlated, the injection start time R1 is calculated based on the intersection time LBα. For example, a timing that is a predetermined delay time Cα before the intersection timing LBα may be calculated as the injection start timing R1.

  Further, among the fuel pressure waveforms, an ascending approximate line Lβ that approximates the rising waveform from the inflection point P3 at which the fuel pressure rises at the end of the injection to the inflection point P5 at which the fuel pressure rise ends by a least square method or the like is obtained. calculate. Then, a time (intersection time LBβ between Lβ and Bβ) that is the reference value Bβ in the rising approximate straight line Lβ is calculated. Focusing on the fact that the intersection timing LBβ and the injection end timing R4 are highly correlated, the injection end timing R4 is calculated based on the intersection timing LBβ. For example, a timing that is a predetermined delay time Cβ before the intersection timing LBβ may be calculated as the injection end timing R4.

  Next, paying attention to the fact that the slope of the descending approximate line Lα and the slope of the injection rate increase are highly correlated, the slope of the straight line Rα indicating the increase in the injection rate waveform shown in FIG. Calculation is based on the slope of Lα. For example, the slope of Rα may be calculated by multiplying the slope of Lα by a predetermined coefficient. Similarly, since the slope of the rising approximate line Lβ and the slope of the injection rate decrease are highly correlated, the slope of the straight line Rβ indicating the decrease in injection in the injection rate waveform is calculated based on the slope of the rising approximate line Lβ.

  Next, based on the straight lines Rα and Rβ of the injection rate waveform, a timing (valve closing operation start timing R23) at which the valve body 12 starts lift-down in response to the command to end injection is calculated. Specifically, the intersection of both straight lines Rα and Rβ is calculated, and the intersection timing is calculated as the valve closing operation start timing R23. Further, a delay time (injection start delay time td) with respect to the injection start command timing t1 of the injection start timing R1 is calculated. Further, a delay time (valve closing start delay time te) with respect to the injection end command timing t2 of the valve closing operation start timing R23 is calculated.

  The maximum injection rate Rmax changes as the fuel injection valve 10 changes over time. For example, when aged deterioration such as deposits or the like deposits on the nozzle holes 11b and the injection amount decreases, the pressure drop amount ΔP shown in FIG. 2C decreases. Further, when the aging deterioration such that the seat surface 12a wears and the injection amount increases, the pressure drop amount ΔP increases. Therefore, in the present embodiment, focusing on the fact that the maximum injection rate Rmax and the pressure drop amount ΔP are highly correlated, the learning value of the maximum injection rate Rmax is calculated from the detection result of the pressure drop amount ΔP.

  Note that the pressure drop amount ΔP is the amount of decrease in the detected pressure caused by the increase in the injection rate. For example, the pressure drop amount from the reference pressure Pbase to the inflection point P2 or the change from the inflection point P1. It is the amount of pressure drop to the bend point P2. The reference pressure Pbase is a pressure immediately before the time point P1 at which the pressure starts to drop with the start of injection. For example, an average value of the detected pressures in a predetermined period immediately before the time point P1 is calculated as the reference pressure Pbase. To do.

  As described above, the ECU 30 (injection rate parameter calculation means) calculates the injection rate parameters td, te, Rα, Rβ, and Rmax from the fuel pressure waveform. Based on the learned values of the injection rate parameters td, te, Rα, Rβ, and Rmax, an injection rate waveform (see FIG. 2B) corresponding to the injection command signal (see FIG. 2A) is calculated. be able to. Since the area of the injection rate waveform calculated in this way (see halftone dot hatching in FIG. 2B) corresponds to the injection amount, the injection amount can also be calculated based on the injection rate parameter.

  FIG. 3 is a block diagram showing an outline of learning of these injection rate parameters, setting of an injection command signal, and the like. Each means 31, 32, 33 functioning by the ECU 30 will be described below. The injection rate parameter calculation means 31 (detection means) calculates injection rate parameters td, te, Rα, Rβ, Rmax based on the fuel pressure waveform detected by the fuel pressure sensor 20.

  The learning means 32 (control parameter learning means) learns by updating the calculated injection rate parameter in the memory 30a of the ECU 30. Since the injection rate parameter varies depending on the supply fuel pressure (pressure in the common rail 42) at that time, it is desirable to learn in association with the supply fuel pressure or a reference pressure Pbase described later. Further, the injection rate parameter excluding the maximum injection rate Rmax may be learned in association with the injection amount. In the example of FIG. 3, the injection rate parameter value corresponding to the fuel pressure P (environment variable) is stored in the injection rate parameter map M.

  The setting means 33 acquires an injection rate parameter (learned value) corresponding to the current pressure PJ from the injection rate parameter map M. And based on the acquired injection rate parameter, the injection command signals t1, t2, and Tq corresponding to the target injection state are set. The fuel pressure sensor 20 detects the fuel pressure waveform when the fuel injection valve 10 is operated in accordance with the injection command signal set in this way, and the injection rate parameter calculation means 31 based on the detected fuel pressure waveform, the injection rate parameter td, te, Rα, Rβ, Rmax are calculated.

  In short, an actual injection state (that is, injection rate parameters td, te, Rα, Rβ, Rmax) with respect to the injection command signal is detected and learned, and an injection command signal corresponding to the target injection state is set based on the learned value. . Therefore, the injection command signal is feedback-controlled based on the actual injection state, and the fuel injection state can be controlled with high accuracy so that the actual injection state coincides with the target injection state even when the above-described aging deterioration proceeds.

  Next, the learning method of the injection rate parameter by the learning unit 32 will be described below taking the injection start delay time td as an example.

  FIG. 4A shows an injection rate parameter map M in which the delay time td is stored in association with the pressure P. The two-dimensional map has the delay time td on the vertical axis and the pressure P on the horizontal axis. The horizontal axis (pressure P) of the map M is divided by a plurality of specific values P1, P2, and P3 as indicated by a one-dot chain line in the figure. Then, the delay time td corresponding to these specific values is stored on the map M as a learning value. Therefore, as the number of divisions based on this specific value is set, the number of learning points for td increases, and the capacity required for the memory 30a (storage means) increases.

  FIG. 5 is a flowchart showing a learning process procedure for updating the learning value (delay time td) stored in the injection rate parameter map M by the microcomputer included in the ECU 30. This process is repeatedly executed at a predetermined cycle during engine operation. In the following description, the delay time td will be taken as an example of the injection rate parameter and will be described with reference to FIG.

  First, in step S10 of FIG. 5, it is determined whether or not the calculation (detection) of the delay time td by the injection rate parameter calculation means 31 has been newly executed in accordance with the fuel injection from the fuel injection valve 10. If it is determined that there is detection (S10: YES), the process proceeds to step S11, and the value of the reference pressure Pbase at the time of injection at the delay time td is acquired as the detected value PK of the environment variable P. Further, the value of the detected delay time td is acquired as the detected value tdK of the control parameter corresponding to PK.

  In addition, the code | symbol K on the map M shown to Fig.4 (a) is corresponded to the detection point showing these acquired detection values (PK, tdK). The symbols G1, G2, and G3 correspond to learning points that represent td values learned in association with the specific values P1, P2, and P3.

  In the subsequent step S12, as shown in FIG. 4A, a learning point having a value closest to the detected value PK among the plurality of specific values P1, P2, and P3 is determined as the update target learning point G1. In the subsequent step S13, a learning point at a position where the update target learning point G1 is interpolated with the detection point K among the plurality of learning points G1, G2, G3 is selected as an interpolation learning point G2. Then, the interpolation learning point G2 and the detection point K are linearly interpolated to calculate the update value td1 'of the learning point G1. As a specific example of interpolation, there is a calculation method using a known multidimensional spline function.

  Note that the linearly interpolated value td1 ′ is replaced with the updated value as it is, and the value td1 ″ that has been subjected to the annealing process as represented by the equation td1 ″ = td1 + (td1′−td1) × C. May be an updated value. That is, the learning unit 32 (smoothing processing unit) corrects the pre-update value td1 by a predetermined ratio C, and uses the corrected value td1 '' as a learning update value.

  In subsequent step S14, learning is performed by updating the learning value td1 of the update target learning point G1 to the update value td1 'calculated in step S13. In addition, the code | symbol G1 'on the map M shown to Fig.4 (a) is equivalent to the learning point G1 after an update. In subsequent step S15, learning for interpolating a learning point at a position where the detection point K is interpolated with the update target learning point G1 (updated learning point G1 ′) among the plurality of learning points G1, G2, G3. Select as point G3. Then, the updated learning point G1 'and the interpolation learning point G3 are linearly interpolated to calculate a delay time tdKα corresponding to the detected value PK. A symbol Kα on the map M shown in FIG. 4A corresponds to an interpolation calculation point representing a value tdKα and a detection value PK obtained by linear interpolation with G1 ′ and G3.

  In the subsequent step S16, an interpolation error ΔtdK that is an error between the delay time tdKα applied to the interpolation calculation point Kα and the detected delay time tdK is calculated. In the subsequent step S17 (interpolation error learning means), the interpolation error ΔtdK calculated in step S16 is learned in association with the detected value PK. Note that the learning value of the interpolation error ΔtdK is stored in the memory 30a in the same manner as the learning points G1 to G3.

  FIG. 6 is a flowchart showing a procedure for calculating the delay time tdJ corresponding to the current pressure PJ using the learning value (delay time td) stored in the injection rate parameter map M by the microcomputer of the ECU 30. is there. This process is repeatedly executed at a predetermined cycle during engine operation. The delay time tdJ calculated in this procedure is used for calculation of the injection command signals t1, t2, and Tq by the setting means 33.

  First, in step S20 of FIG. 6, it is determined whether or not there is a request for calculating the injection command signals t1, t2, and Tq for the next injection. If it is determined that there is a request (S20: YES), the process proceeds to step S21, and the current pressure detection value PJ is acquired. In the subsequent step S22, learning points at positions where the detected value PJ is interpolated among the plurality of learning points G1, G2, G3 are selected as learning points G1, G3 for interpolation (see FIG. 4B). In subsequent step S23 (current control parameter interpolation means), the two interpolation learning points G1 and G3 are linearly complemented to calculate a delay time tdJα corresponding to the detected value PJ.

  In the subsequent step S24, the difference between the pressure PK value corresponding to the interpolation error ΔtdK learned in step S17 and the current pressure PJ acquired in step S21 is calculated. Based on this difference, the value of a reflection coefficient R described later is calculated. Specifically, as shown in FIG. 4C, the reflection coefficient R is set to a smaller value as the difference is larger. The reflection coefficient R is set to a value between 0 and 1.

  In the subsequent step S25, a value obtained by multiplying the interpolation error ΔtdK by the reflection coefficient R calculated in step S24 is calculated as a correction amount D (D = ΔtdK × R). In subsequent step S26 (correction means), a value obtained by adding the correction amount D calculated in step S25 to the delay time tdJα calculated in step S23 is calculated as the delay time tdJ corresponding to the current pressure PJ. (TdJ = tdJα + D). Then, the setting means 33 (control means) calculates the injection command signal t1 for the next injection control using the delay time tdJ calculated in this way.

  In the example of FIG. 4C, when the current pressure PJ exists within a predetermined range M1 centered on the pressure PK corresponding to the interpolation error ΔtdK, the reflection coefficient R is set to a value greater than zero. . That is, the interpolation error ΔtdK is reflected in the delay time tdJ used for control. Incidentally, the size of the predetermined range M1 is set to the same size as the intervals of the specific values P1, P2, P3 of the map M.

  As described above, according to the present embodiment, the current pressure PJ is increased without increasing the number of data points of the learning points G1 to G3 by increasing the number of divisions of the pressures P1 to P3 set in the injection rate parameter map M. The value of the corresponding injection rate parameter tdJ can be calculated with high accuracy. Therefore, the fuel injection state can be controlled with high accuracy so that the change in the injection rate becomes a desired state. Therefore, it is possible to achieve both reduction in the required storage capacity of the memory 30a and improvement in the accuracy of injection control.

  Further, as the difference between the pressure PK corresponding to the interpolation error ΔtdK and the current pressure PJ is larger, the reflection coefficient R is set to a smaller value to reduce the degree of reflecting the interpolation error ΔtdK in the delay time tdJ used for control. ing. For this reason, it is possible to facilitate calculating the value of the injection rate parameter tdJ corresponding to the current pressure PJ with high accuracy.

(Second Embodiment)
In the first embodiment, various injection rate parameters are learned in association with one environmental variable (pressure P). However, in the present embodiment, in association with two environmental variables (pressure P and injection amount Q). Various injection rate parameters are learned.

  FIG. 7 is an injection rate parameter map according to the present embodiment, which is a three-dimensional map (spatial coordinates) of P, Q, and td. The axis of the pressure P and the axis of the injection amount Q in the map are divided by a plurality of specific values as indicated by dotted lines in the figure. The delay time td corresponding to these specific values is stored on the map as a learning value. The black circles in the figure are learning points G1 to G9 indicating the values of td corresponding to the specific values of P and Q.

<Learning point update method>
First, a method for updating the learning points G1 to G9 when the detection point K (PK, QK, tdK) is detected will be described.

  Of the plurality of learning points G1 to G9, the learning point in the region closest to the pressure PK and the injection amount QK at the detection point K is determined as the update target learning point G1. For example, as indicated by the dotted line in FIG. 8, regions of the learning points G1 to G9 are set, and the learning point in the region where the detection point K is located is determined as the update target learning point G1.

  Then, based on the other learning points G2, G3, G4, G5, G6 excluding the update target learning point G1 and the detection point K, a detection reference plane SK described below is calculated, and the update target learning point in the td-axis direction is calculated. The distance between G1 and the detection reference plane SK is set as a learning amount. In other words, the delay time td (learned value) of the update target learning point G1 is the delay time td of the point (detection reference point Kβ) corresponding to the environment variables (PG1, QG1) of the update target learning point G1 in the detection reference plane SK. Update to (update value) and learn.

  The calculation method of the detection reference plane SK will be described. First, among the plurality of learning points G1 to G9, a learning point at a position where the update target learning point G1 is interpolated with the detection point K is used as an interpolation learning point G2. decide. Next, a point corresponding to the injection amount QG2 of the interpolation learning point G2 and the pressure PK of the detection point K among the surfaces including the plurality of learning points G1 to G9 (learning surface SG) is determined as the first point K1. To do. For example, the first point K1 may be calculated by interpolating the learning point G3 and the learning point G4. Note that “interpolation” means that a calculation target point (first point K1) is located between two reference points (learning points G3, G4), and these reference points are linearly interpolated to calculate the calculation target points. Is to calculate the value of.

  Similarly, for the second point K2, the point corresponding to the pressure PG2 of the interpolation learning point G2 and the injection amount QK of the detection point K in the learning surface SG is determined as the second point K2. As described above, the detection reference plane SK including at least one of the first point K1 and the second point K2, the update target learning point G1, and the detection point K is specified.

  FIG. 8 is a PQ plan view for explaining a specific example of the procedure for calculating the update value from the detection reference plane SK. First, after determining the first point K1 as described above, the injection amount QG1 of the update target learning point G1 is shown. A point corresponding to the pressure PK at the detection point K (interpolation point K1a) is calculated by interpolation between the first point K1 and the detection point K. Next, a point (detection reference point Kβ) corresponding to the injection amount QG1 and the pressure PG1 at the update target learning point G1 is calculated by interpolation between the interpolation point K1a and the learning point G5. Then, the delay time tdKβ of the detection reference point Kβ is set as an update value of the delay time tdG1 of the update target learning point G1.

  The detection reference point Kβ can also be calculated by the following procedure. That is, first, after determining the second point K2 as described above, the point (interpolation point K2a) corresponding to the pressure PG1 of the update target learning point G1 and the injection amount QK of the detection point K is detected as the second point K2. Interpolated with point K and calculated. Next, a point (detection reference point Kβ) corresponding to the injection amount QG1 and pressure QG1 of the update target learning point G1 is calculated by interpolation between the interpolation point K2a and the learning point G3. Then, the delay time tdKβ of the detection reference point Kβ is set as an update value of the delay time tdG1 of the update target learning point G1.

  As described above, the detection reference point Kβ calculated based on the first point K1, the interpolation point K1a, and the learning point G5, and the detection reference point Kβ calculated based on the second point K2, the interpolation point K2a, and the learning point G3. Have the same value.

<Calculation method of interpolation error>
Next, an interpolation error calculation method will be described with reference to FIG. First, a point (interpolation calculation point Kα) corresponding to the pressure PK and the injection amount QK at the detection point K is calculated by interpolating at a plurality of learning points G1, G7, G8, and G9. As an example of the interpolation, a point corresponding to (PK, QK) among the planes including these learning points G1, G7, G8, G9 can be calculated as the interpolation calculation point Kα. Of these learning points G1, G7, G8, and G9, three points that interpolate the interpolation calculation point Kα are selected, and the point corresponding to (PK, QK) among the planes including the three points is selected as the interpolation calculation point. It may be calculated as Kα.

  Then, the difference between the calculated delay time tdKα of the interpolation calculation point Kα and the delay time tdK of the detection point K is calculated as an interpolation error. Note that a value obtained by the difference may be calculated as an interpolation error.

<Delay time calculation method used for control>
Next, a method for calculating a delay time used for control will be described with reference to FIGS. 9 and 10. First, the current pressure PJ and command injection amount QJ are acquired. Next, the delay time tdJα corresponding to these values (PJ, QJ) is calculated by interpolating at four learning points G1, G7, G8, G9. As an example of the interpolation, a point corresponding to (PJ, QJ) among the planes including these learning points G1, G7, G8, G9 can be calculated as the delay time tdJα. Of these learning points G1, G7, G8, and G9, three points that interpolate the delay time tdJα are selected, and a point corresponding to (PJ, QJ) among the planes including the three points is set as the delay time tdJα. It may be calculated.

  Next, a distance LJK between the pressure PJ and injection amount QJ at the current point Jα and the pressure PKα and injection amount QKα at the interpolation calculation point Kα is calculated. A value obtained by multiplying the interpolation error (tdKα−tdK) by the reflection coefficient R corresponding to the distance LJK is calculated as the correction amount D.

  Next, a value obtained by adding the correction amount D to the delay time tdJα is calculated as a delay time tdJ corresponding to the current pressure PJ and injection amount QJ (tdJ = tdJα + D). Then, using the delay time tdJ calculated in this way, the setting means 33 calculates an injection command signal t1 related to the next injection control.

  As described above, even when various injection rate parameters are learned in association with two environmental variables (pressure P and injection amount Q), the data of the learning points G1 to G9 is the same as in the first embodiment. Without increasing the score, the value of the injection rate parameter tdJ corresponding to the current pressure PJ and injection amount QJ can be calculated with high accuracy. Therefore, it is possible to achieve both reduction in the required storage capacity of the memory 30a and improvement in the accuracy of injection control.

  In particular, the detection reference point Kβ is calculated using the first point K1 or the second point K2, the learning point G2 for interpolation, and the detection point K, and the update target learning point G1 is converted to the detection reference point Kβ. Therefore, the learning accuracy of the update target learning point G1 can be improved as compared with the case where the detection reference point Kβ is calculated using two points of the interpolation learning point G2 and the detection point K.

(Third embodiment)
In the second embodiment, the case where the detection point K (PK, QK, tdK) exists within the setting range of the injection rate parameter map has been described. However, in this embodiment, the detection point K is outside the setting range of the injection rate parameter map. A case where there are detection points (PK, QK, tdK) will be described.

  FIG. 12 is an injection rate parameter map according to the present embodiment, which is basically the same as the injection rate parameter map of FIG. However, FIG. 12 differs from FIG. 7 in that the detection point K (PK, QK, tdK) exists outside the map range. In the present embodiment, a case where the detected value of the injection amount among a plurality of environmental variables (pressure, injection amount) is larger than the upper limit value of the map setting range will be described as an example.

<How to update learning points>
A method for updating the learning point when the detection point K exists outside the map range will be described. First, among the plurality of learning points G1 to G12, a learning point in a region closest to the pressure PK and the injection amount QK at the detection point K is determined as the update target learning point G11. For example, as indicated by the dotted line in FIG. 12, regions for the learning points G1 to G12 are set, and the learning point in the region where the detection point K is located is determined as the update target learning point G11.

  Subsequently, a virtual point is set as a point in which an environment variable and a control parameter are associated with each other in an area outside the map setting range and including the detection point K. Specifically, first, an axis (virtual axis) extending in the axial direction of the environmental variable is created from a learning point located at the boundary of the map among a plurality of learning points included in the map, toward the outside of the map range. At this time, by providing a virtual axis in a direction parallel to the axis of the environment variable, each point on the virtual axis has a learning value of a learning point located at the boundary of the map as a control parameter (delay time). Like that. For example, when the detected value of the injection amount exists outside the map range, as indicated by a two-dot chain line in FIG. 12, each of the learning points extends from the learning points G10 to G12 in the axial direction of the injection amount. The virtual axes Lm10 to Lm12 having the learning values are created. Then, virtual points are respectively set at positions on the axis that exceed the injection amount of the detection point K. In this embodiment, as shown in FIG. 12, the virtual point V10 is set to the coordinates (PV10, QV10, tdG10), the virtual point V11 is set to the coordinates (PV11, QV11, tdG11), and the coordinates (PV12, QV12, A virtual point V12 is set to tdG12). Note that the axis of the pressure P and the axis of the injection amount Q in the map are divided at equal intervals by a plurality of specific values as indicated by dotted lines in the figure. In FIG. 12, the learning points are indicated by black circles, and the virtual points are indicated by white circles. In FIG. 12, a virtual surface SV is formed by a surface including a plurality of virtual points V11 to V12.

  Based on the plurality of learning points G8 to G12, the virtual point V12, and the detection point K located near the detection point K, a detection reference plane SK described below is calculated, and the update target learning point G11 in the td axis direction is calculated. The distance from the detection reference plane SK is calculated as a learning amount. That is, the delay time td (learned value) of the update target learning point G11 is set to the delay time td of the point (detection reference point Kβ) corresponding to the environment variable (PG11, QG11) of the update target learning point G11 in the detection reference plane SK. Update to (update value) and learn.

  The calculation method of the detection reference plane SK will be described. First, among the plurality of learning points G1 to G12, a learning point at a position where the update target learning point G11 is interpolated with the detection point K is set as an interpolation learning point G9. decide. Next, a point corresponding to the injection amount QG9 of the interpolation learning point G9 and the pressure PK of the detection point K among the surfaces including the plurality of learning points G1 to G12 (learning surface SG) is determined as the first point K1. To do. For example, the first point K1 is specified by calculating the delay time of the first point K1 by interpolating the delay time of the learning point G6 and the delay time of the learning point G8. Further, a point corresponding to the pressure PG9 of the interpolation learning point G9 and the injection amount QK of the detection point K in the learning surface SG is determined as the second point K2. For example, the second point K2 is specified by calculating the delay time of the second point K2 by interpolating the delay time of the learning point G12 and the delay time of the virtual point V12. As the delay time of the virtual point V12, the same value as the delay time of the learning point G12 is used. In this way, the detection reference plane SK including at least one of the first point K1 and the second point K2, the update target learning point G1, and the detection point K is specified.

  Subsequently, an update value is calculated from the detection reference plane SK. Specifically, referring to FIG. 13, first, after determining the first point K1 as described above, points corresponding to the injection amount QG11 of the update target learning point G11 and the pressure PK of the detection point K (interpolation). The point K1a) is determined by interpolation between the first point K1 and the detection point K. Here, the interpolation point K1a is specified by interpolating the delay time of the first point K1 and the delay time of the detection point K to calculate the delay time of the interpolation point K1a. Next, a point (detection reference point Kβ) corresponding to the injection amount QG11 and the pressure QG11 of the update target learning point G11 is determined by interpolation between the interpolation point K1a and the learning point G12. Here, the detection reference point Kβ is identified by interpolating the delay time of the interpolation point K1a and the delay time of the learning point G12 to calculate the delay time tdKβ of the detection reference point Kβ. Then, the delay time tdKβ of the detection reference point Kβ is set as the update value of the delay time tdG11 of the update target learning point G11.

  Note that the delay time of the detection reference point Kβ is also used when a point (interpolation point K2a) corresponding to the pressure PG11 of the update target learning point G11 and the injection amount QK of the detection point K is used instead of the interpolation point K1a. tdKβ can be calculated. That is, first, the interpolation point K2a is calculated by interpolating between the second point K2 and the detection point K. Next, a point (detection reference point Kβ) corresponding to the injection amount QG11 and the pressure QG11 of the update target learning point G11 is calculated by interpolation between the interpolation point K2a and the learning point G8. Then, the delay time tdKβ of the detection reference point Kβ is set as the update value of the delay time tdG11 of the update target learning point G11.

<Calculation method of interpolation error>
The interpolation error can be calculated by performing basically the same process as when the detection point K exists within the set range of the map, except that a virtual point is used (see FIG. 8). That is, first, a point (interpolation calculation point Kα) corresponding to the pressure PK and the injection amount QK at the detection point K is calculated by interpolating at a plurality of points surrounding the detection point K. In the case of FIG. 13, the calculation is performed by interpolating at the learning points G10 and G11 and the virtual points V10 and V11. However, among these four points (G10, G11, V10, V11), three points that interpolate the interpolation calculation point Kα are selected, and the point corresponding to (PK, QK) among the planes including the three points is selected. It may be calculated as an interpolation calculation point Kα. Then, the difference between the calculated delay time tdKα of the interpolation calculation point Kα and the delay time tdK of the detection point K is calculated as an interpolation error. Note that a value obtained by the difference may be calculated as an interpolation error.

<Delay time calculation method used for control>
The delay time used for the control can also be calculated by performing a process basically similar to the case where the detection point K exists within the set range of the map by using the virtual point. That is, first, the current pressure PJ and the command injection amount QJ are acquired, and then the delay time tdJα corresponding to the current point Jα (PJ, QJ) is interpolated and calculated at four points surrounding the current point Jα. . At this time, if the current point Jα is outside the setting range of the map, the learning point and the virtual point are used as four points surrounding the current point Jα, and the delay time tdJα is calculated by interpolation at these points. For example, when the current point Jα exists in a region surrounded by the learning points G10 and G11 and the virtual points V10 and V11, these four points (the learning points G10 and G11 and the virtual points V10 and V11) are used. Of the four points surrounding the current point Jα, three points that interpolate the delay time tdJα are selected, and a point corresponding to (PJ, QJ) among the planes including the three points is calculated as the delay time tdJα. May be.

  Next, a distance LJK between the pressure PJ and injection amount QJ at the current point Jα and the pressure PKα and injection amount QKα at the interpolation calculation point Kα is calculated. A value obtained by multiplying the interpolation error (tdKα−tdK) by the reflection coefficient R corresponding to the distance LJK is calculated as the correction amount D. Further, a value obtained by adding the correction amount D to the delay time tdJα is calculated as a delay time tdJ corresponding to the current pressure PJ and injection amount QJ (tdJ = tdJα + D). Then, using the delay time tdJ calculated in this way, the setting means 33 calculates an injection command signal t1 related to the next injection control.

  As described above, even if the detection point (PK, QK, tdK) is outside the setting range of the injection rate parameter map, the virtual point is outside the map setting range and includes the detection point K. By setting V10 to V12, the value of the injection rate parameter tdJ corresponding to the current pressure PJ and the injection amount QJ without increasing the number of data points of the learning points G1 to G12, as in the second embodiment. Can be calculated with high accuracy. Therefore, it is possible to achieve both reduction in the required storage capacity of the memory 30a and improvement in the accuracy of injection control.

(Other embodiments)
The present invention is not limited to the description of the above embodiment, and may be modified as follows. Moreover, you may make it combine the characteristic structure of each embodiment arbitrarily, respectively.

  In the above embodiment, only one interpolation error ΔtdK is stored in the memory 30a for learning, but the specific value region on the injection rate parameter map M is divided into a plurality of regions, and the interpolation error ΔtdK is set for each region. You may learn. According to this, the correction accuracy using the interpolation error ΔtdK can be improved, but as a contradiction, the required storage capacity of the memory 30a increases by the increase in the number of storage points of the interpolation error ΔtdK.

  The reflection coefficient R used in each of the above embodiments is abolished, and the interpolation error ΔtdK is uniformly corrected D regardless of the difference between the pressure PK value corresponding to the interpolation error ΔtdK and the current pressure PJ. May be set. In this case, it is unnecessary to learn the Δ interpolation error tdK in association with the pressure PK.

  In the second embodiment, the injection rate parameter is learned in association with two environmental variables. However, the Δ interpolation error tdK is similarly learned in the case of learning in association with three or more environmental variables. Thus, the correction amount D may be set.

  In the above embodiment shown in FIG. 1, the fuel pressure sensor 20 is mounted on the fuel injection valve 10, but the fuel pressure sensor according to the present invention is in the fuel supply path from the discharge port 42a of the common rail 42 to the injection hole 11b. Any fuel pressure sensor may be used so long as it detects the fuel pressure. Therefore, for example, a fuel pressure sensor may be mounted on the high-pressure pipe 42 b that connects the common rail 42 and the fuel injection valve 10.

  In the third embodiment, the case where the detected value of the injection amount among the plurality of environmental variables (pressure, injection amount) is larger than the upper limit value of the setting range of the map has been described as an example. When the detected value of the environment variable is smaller than the lower limit value of the map setting range, a virtual point is set in an area outside the map setting range and including the detection point K, and the set virtual point is used. Learning points may be updated and interpolation errors may be calculated. In addition, the above configuration may be applied when the detected pressure value out of the plurality of environment variables is outside the set range of the map.

  In the third embodiment, a case has been described in which various injection rate parameters are learned in association with two environmental variables (pressure P and injection amount Q), but various types are associated with one environmental variable (for example, pressure P). In the configuration in which the injection rate parameter is learned, when the detection point K exists outside the set range of the map, control using virtual points may be performed as in the third embodiment. For example, a case where the injection delay time as the injection rate parameter is learned in association with the pressure P as the environmental variable will be described with reference to FIG. In FIG. 14, when the detected pressure value PK exceeds the upper limit value P3 of the map setting range, first, a learning point at a position closest to the detected value PK is determined as an update target learning point. Here, the learning point G3 closest to the detected value PK is selected as the update target learning point. Further, a learning point at a position where the update target learning point G3 is interpolated with the detection point K is selected as an interpolation learning point (here, the learning point G1). Then, the interpolation learning point G1 and the detection point K are linearly interpolated to calculate an updated value of the learning point G3. Subsequently, an interpolation error is calculated. Specifically, first, a virtual point V3 is set on an axis extending in the axial direction of the environmental variable (pressure P) from the learning point G3 at the boundary of the map setting range. At this time, the virtual point V3 is set at a position exceeding the detection value PK, and the virtual point V3 is set at coordinates (PV3, td3) in FIG. P1 to P3 and PV3 are set at equal intervals on the pressure axis. Next, the updated learning point G3 'and virtual point V3 are linearly interpolated to calculate a delay time tdKα with respect to the detected value PK. Then, an interpolation error ΔtdK is calculated from the difference between the calculated delay time tdKα and the detected value tdK.

30a ... Memory (storage means), 31 ... Injection rate parameter calculation means (detection means), 32 ... Learning means (control parameter learning means, smoothing processing means), 33 ... Setting means (control means), S17 ... Interpolation error learning Means, S23: Current control parameter interpolation means, S26: Correction means.

Claims (7)

Storage means for storing a control parameter (td) in association with a specific value of a predetermined environment variable (P);
Detecting means for detecting the environment variable and the control parameter;
Control parameter learning means for updating and learning the value of the control parameter (td1) stored in the storage means based on the detection values (PK, tdK) by the detection means;
A control parameter corresponding to the detected value (PK) of the environment variable used for learning by the control parameter learning unit is calculated by linear interpolation from a plurality of control parameters (td1 ′, td3) stored in the storage unit. Interpolation error learning means for learning an interpolation error (ΔtdK) that is an error between the calculated value (tdKα) and the detected value (tdK) of the control parameter;
Current control parameter interpolation means for linearly interpolating a control parameter (tdJα) corresponding to the current environment variable (PJ) from a plurality of control parameters (td1 ′, td3) stored in the storage means;
Correction means for correcting the control parameter (tdJα) calculated by the current control parameter interpolation means based on the interpolation error (ΔtdK) learned by the interpolation error learning means;
Control means for controlling the control object based on the control parameter (tdJ) corrected by the correction means;
A control device comprising: The interpolation error learning means learns the interpolation error (ΔtdK) in association with the detected value (PK) of the environment variable,
The correction means corrects the interpolation error (ΔtdK) as the difference between the detected value (PK) of the associated environment variable and the current environment variable (PJ) used in the current control parameter interpolation means increases. The control device according to claim 1, wherein a degree of reflection is reduced. In the case where a control parameter (td) is stored in the storage means in association with a plurality of environment variables (P, Q),
A point representing a detection value (PK, QK, tdK) by the detection means in a spatial coordinate with the plurality of environment variables (P, Q) and the control parameter (td) as a reference axis is called a detection point (K). ,
In the spatial coordinates, the points representing the learning values of the control parameters are called learning points (G1 to G9),
Among the plurality of learning points, a learning point corresponding to the environment variable closest to the detected environment variable is called an update target learning point (G1),
Among the plurality of learning points, a learning point at a position where the update target learning point (G1) is interpolated with the detection point (K) is referred to as an interpolation learning point (G2).
A surface including a plurality of learning points is called a learning surface (SG),
Of the learning plane (SG), points corresponding to the first environment variable (PK) of the detection point (K) and the second environment variable (QG2) of the learning point for interpolation (G2) are defined as the first point ( K1), a point corresponding to the first environment variable (PG2) of the learning point for interpolation and the second environment variable (QK) of the detection point is called a second point (K2),
A surface including at least one of the first point (K1) and the second point (K2), the learning point for interpolation (G2), and the detection point (K) is referred to as a detection reference surface (SK). ,
In the detection reference plane (SK), the same point as the environment variable (PG1, QG1) of the update target learning point (G1) is referred to as a detection reference point (Kβ).
The control parameter learning means updates and learns the value of the control parameter (tdG1) of the update target learning point (G1) based on the control parameter (tdKβ) of the detection reference point (Kβ). The control device according to claim 1 or 2. The control parameter learning means includes
The control parameter stored in the storage means and the value obtained by linear interpolation from the detected value are corrected so as to approach the value of the control parameter before the update by a predetermined ratio, and the corrected value is not used as the learning update value. The control apparatus according to claim 1, further comprising a mash processing unit. The control object is a fuel injection valve that injects fuel to be used for combustion of an internal combustion engine,
The fuel injection valve is equipped with a fuel pressure sensor for detecting fuel pressure,
A fuel pressure waveform detecting means for detecting a change in fuel pressure caused by injection as a fuel pressure waveform based on a detection value of the fuel pressure sensor;
An injection rate parameter calculation means for calculating an injection rate parameter required to specify an injection rate waveform corresponding to the detected fuel pressure waveform based on the detected fuel pressure waveform;
With
The control device according to any one of claims 1 to 4, wherein the detected value of the control parameter is the injection rate parameter calculated by the injection rate parameter calculation means. When the environment variable detected by the detection unit is out of the set range of the environment variable stored in the storage unit, the area is outside the set range and includes the detection value by the detection unit Setting means for setting a virtual point having the learning value of the boundary value of the setting range stored in the storage means as the control parameter at a position farther from the setting range than the detected value of the environment variable. Prepared,
The interpolation error learning means uses the virtual point as one of a plurality of control parameters stored in the storage means when the environmental variable detected by the detection means is out of the set range. The control apparatus according to claim 1, wherein an error is learned. When the environmental variable detected by the detection means is out of the setting range, the control parameter (td) is stored in the storage means in association with a plurality of environmental variables (P, Q),
The setting unit sets a plurality of virtual points (V10 to V12) in which at least one of the plurality of environment variables is different from each other;
A point representing a detection value (PK, QK, tdK) by the detection means in a spatial coordinate with the plurality of environment variables (P, Q) and the control parameter (td) as a reference axis is called a detection point (K). ,
In the spatial coordinates, a point representing the learning value of the control parameter is called a learning point (G1 to G12),
Among the plurality of learning points, a learning point corresponding to the environment variable closest to the detected environment variable is referred to as an update target learning point (G11),
Among the plurality of learning points, a learning point at a position where the update target learning point (G11) is interpolated with the detection point (K) is referred to as an interpolation learning point (G9).
A surface including a plurality of learning points is called a learning surface (SG),
A surface including a plurality of the virtual points is called a virtual surface (SV),
Of the learning surface (SG) and the virtual surface (SV), corresponding to the first environment variable (PK) of the detection point (K) and the second environment variable (QG9) of the learning point for interpolation (G9). The point corresponding to the first point (K1), the point corresponding to the first environment variable (PG9) of the learning point for interpolation and the second environment variable (QK) of the detection point is referred to as the second point (K2),
A surface including at least one of the first point (K1) and the second point (K2), the learning point for interpolation (G9), and the detection point (K) is referred to as a detection reference surface (SK). ,
In the detection reference plane (SK), when the same point as the environment variable (PG11, QG11) of the update target learning point (G11) is called a detection reference point (Kβ),
The control parameter learning means updates and learns the value of the control parameter of the update target learning point (G11) based on the control parameter (tdKβ) of the detection reference point (Kβ). The control device described in 1.

Images