JP7319092B2 - Control device for internal combustion engine - Google Patents

Description

The present invention relates to a control device for an internal combustion engine.

In recent years, in vehicles such as automobiles, regulations on fuel consumption and exhaust gas have been strengthened. These regulations are expected to become stronger in the future. In particular, regulations on fuel consumption are a matter of great concern in connection with the recent high volatility in fuel prices, the impact on global warming, and the problem of depletion of energy resources.

Under such circumstances, for example, in the automobile industry, various technical developments are being made for the purpose of improving fuel efficiency and exhaust performance of vehicles. One example of technology aimed at improving fuel efficiency is a high compression ratio technology that increases the compression ratio of an internal combustion engine.

By the way, in the technique for increasing the compression ratio described above, the compression ratio of the internal combustion engine is increased, so the thermal efficiency is improved and the fuel consumption is improved. However, it is known that the technology for increasing the compression ratio increases the temperature in the combustion chamber, making knocking more likely to occur.

In order to solve such a problem, Japanese Patent Laid-Open No. 2002-200001 discloses that knock is accurately determined by finally determining that knock has occurred based on a plurality of primary knock determination results obtained in a plurality of knock frequency bands. It is disclosed that it can be determined.

Japanese Unexamined Patent Application Publication No. 2008-280948

However, since the vibration (knocking) of the internal combustion engine greatly changes according to the operating state of the internal combustion engine, it cannot be detected with high accuracy only by the frequency signal of the knocking sensor. Knocking can damage the engine and must be prevented. Delaying the ignition timing (retarding) is effective in suppressing the occurrence of knocking. On the other hand, retarding the ignition timing deteriorates the fuel consumption.

Furthermore, if the knocking detection accuracy is low, there is a possibility that knocking cannot be detected. Therefore, in consideration of safety, the ignition timing must be set late overall in advance, which deteriorates fuel consumption. In particular, knocking is likely to occur in the transient operating region of the internal combustion engine, and it is necessary to prevent knocking by greatly retarding the engine, which may greatly deteriorate fuel efficiency.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and an object thereof is to provide a technique capable of improving the accuracy of knocking determination.

In order to solve the above-described problems, an internal combustion engine control apparatus according to one aspect includes a frequency input section to which a detection signal including the frequency of knocking of the internal combustion engine is input, and a knocking frequency band for each knocking frequency band based on the detection signal. a frequency intensity calculation unit that calculates frequency intensity; an internal combustion engine state acquisition unit that acquires an operating state of the internal combustion engine; a knocking determination unit that stores a configured knocking determination model, and determines a knocking state by inputting the frequency intensity for each frequency band of knocking and the operating state of the internal combustion engine into the knocking determination model.

According to the present invention, it is possible to improve the accuracy of knocking determination.

1 is a configuration diagram of a control system for an internal combustion engine according to a first embodiment; FIG. FIG. 2 is a block diagram of an ECU according to the first embodiment; FIG. FIG. 2 is a block diagram of an engine operating state measuring unit according to the first embodiment; FIG. 4 is an explanatory diagram of knocking determination using a decision tree according to the first embodiment; FIG. 4 is an explanatory diagram of a knocking determination model based on a decision tree according to the first embodiment; FIG. 9 is an explanatory diagram of a method of dividing an operating region using a plurality of decision trees according to the second embodiment; FIG. 11 is an explanatory diagram of a learning range of a decision tree of divided operating regions according to the third embodiment; FIG. 11 is an explanatory diagram of a determination method for a decision tree structure according to the fourth embodiment; FIG. 12 is a flowchart of a decision tree structure determination method according to the fourth embodiment; FIG.

A control system for an internal combustion engine according to an embodiment will be described in detail below with reference to the drawings. It should be noted that this embodiment is merely an example for realizing the present invention and does not limit the technical scope of the present invention.

FIG. 1 is a configuration diagram of an internal combustion engine control system according to the first embodiment.

A control system for an internal combustion engine includes an engine 101 as an example of an "internal combustion engine" and an ECU (Electronic Control Unit) 111 as an example of a "control device". The engine 101 is, for example, a spark ignition multi-cylinder gasoline engine having four cylinders 105 . The ECU 111 includes a processor (hereinafter referred to as CPU) and memory.

Each cylinder 105 defines a combustion chamber 107 and is provided with a fuel injection device 102 for injecting fuel into the combustion chamber 107 . Each combustion chamber 107 is provided with an ignition plug to which a high-voltage ignition signal is supplied from the ignition device 103 . Combustion chamber 107 communicates with intake pipe 106 and exhaust pipe 109 . The intake pipe 106 has an airflow sensor 122 , a throttle valve 123 and a collector 120 . Air required for fuel combustion passes through an intake pipe 106, and passes through an intake valve 112 that is driven to open and close by an intake cam 104 disposed at an intake port at the downstream end of the intake pipe 106, and into each combustion chamber 107. is inhaled into A cam angle sensor 114 is arranged on the intake cam 104 .

The throttle valve 123 is controlled to an appropriate opening according to a detection signal from an accelerator operation sensor 118 that detects the amount of accelerator operation by the driver and a detection signal from a brake operation sensor 119 . Thereby, the throttle valve 123 controls the amount of intake air taken into the cylinder 107 through the collector 120 and the intake pipe 106 . The airflow sensor 122 measures this amount of intake air and outputs it to the ECU 111 .

Further, when the intake cam 104 opens the intake valve 112, the mixed air obtained by mixing the air taken into the combustion chamber 107 from the intake pipe 106 and the fuel injected from the fuel injection device 102 is ignited by the ignition device 103. It burns by As a result, an explosion occurs in the combustion chamber 107, and torque is output to the crankshaft as driving force for the vehicle.

At this time, the fuel injected into the combustion chamber 107 is injection-controlled by the ECU 111 controlling the fuel injection device 102 . For example, the ECU 111 controls the fuel injection amount and the injection timing. The injected fuel is also ignition-controlled by the ECU 111 and ignited by the ignition device 103 at the controlled timing.

Detection signals from the accelerator operation sensor 118, the brake operation sensor 119, the cam angle sensor 114, the crank angle sensor 115, the water temperature sensor 124, the knock sensor 113, and other input parameters are input to the ECU 111. The ECU 111 executes fuel injection control and ignition control of the engine 101 based on the input detection signal of the sensor.

Cam angle sensor 114 detects the angle of intake cam 104 . Crank angle sensor 115 detects the rotation speed of engine 101 . A water temperature sensor 124 is attached to the combustion chamber 107 to detect the cooling water temperature of the engine 101 . Knock sensor 113 is attached to combustion chamber 107 to detect knocking due to abnormal combustion of engine 101 .

After exploding in the combustion chamber 107, the exhaust gas passes through the exhaust pipe 109 when the exhaust valve 108 is opened, and after the harmful gas in the exhaust gas is removed by the catalyst 110, it is exhausted outside the vehicle. At this time, harmful gases in the exhaust gas are detected by an A/F sensor 116 and an O2 sensor 117 attached to the front and rear of the catalyst 110 , and the detection signals are output to the ECU 111 .

FIG. 2 is a block diagram of an ECU according to the second embodiment.

The ECU 111 includes a frequency input section 201 , a frequency intensity calculation section 202 , an engine operating state acquisition section 203 as an example of an “internal combustion engine state acquisition section”, and a knocking determination section 204 . The frequency input unit 201, the frequency intensity calculation unit 202, the engine operating state acquisition unit 203, and the knocking determination unit 204 may be configured by a CPU.

Frequency input unit 201 receives a detection signal including the frequency of knocking of engine 101 from knock sensor 113 . Frequency intensity calculation section 202 calculates the frequency intensity for each frequency band of knocking based on the detection signal of knock sensor 113 . The engine operating state acquisition unit 203 acquires the operating state of the engine 101 . Knocking determination unit 204 stores a knocking determination model (also referred to as a decision tree) constructed as learning data of the frequency intensity for each frequency band of knocking, the operating state of engine 101, and the knocking state. Knocking determination unit 204 determines the knocking state by inputting the frequency intensity for each knocking frequency band and the operating state of engine 101 into the knocking determination model.

Here, knocking determination unit 204 determines the knocking state based on the operating state of engine 101 in addition to the frequency intensity of each knocking frequency band.

FIG. 3 is a block diagram of an engine operating state measuring unit according to the first embodiment.

An engine operating state acquisition unit 203 acquires, as learning data, parameters necessary for a knocking determination unit 204 to determine a knocking state. That is, the operating state of the engine 101 measured by various sensors provided in the engine 101 may be input to the engine operating state acquisition unit 203 as learning data. Alternatively, the engine operating state acquisition unit 203 may calculate learning data based on operating states measured by various sensors provided in the engine 101 . In this way, knocking determination unit 204 requires the operating state of engine 101 or information about the external environment that influences the operating state in order to determine the knocking state.

Typical parameters of the engine 101 are ignition timing, cylinder internal pressure, cooling water temperature, intake air humidity, oil temperature, cylinder exhaust temperature, rotation speed, cylinder intake temperature, load, operating state of intake and exhaust valves, torque converter state, and clutch. information, and gradient information.

That is, the engine operating state acquisition unit 203 includes a cooling water temperature measurement unit 303, an oil temperature measurement unit 305, an engine speed measurement unit 307, an engine load measurement unit 309, a valve operation state measurement unit 310, a torque converter state measurement unit 311, a clutch An information measurement unit 312 and a gradient information measurement unit 313 are provided. Furthermore, the engine operating state acquisition unit 203 includes an ignition timing calculation unit 301 , a cylinder internal pressure calculation unit 302 , an intake air humidity calculation unit 304 , a cylinder exhaust temperature calculation unit 306 and a cylinder intake air temperature calculation unit 308 .

In this way, by adding a parameter other than the detection signal of knock sensor 113 to the selection of parameters used to determine the knocking state of engine 101, it becomes possible to detect knocking with higher accuracy.

Knocking determination unit 204 stores a knocking determination model constructed by learning, and selects parameters as necessary to determine knocking. Knocking determination section 204 may not use all of these parameters, or may add parameters in addition to these if necessary.

FIG. 4 is an explanatory diagram of knocking determination using a decision tree according to the first embodiment.

Knocking determination model 401 is constructed by decision tree learning. Knocking determination model 401 is constructed by inputting parameter values relating to the operating state of engine 101 as learning data, in addition to the frequency intensity for each frequency band of knocking calculated based on the detection signal of knock sensor 113 . .

The frequency input section 201 has a filter section 402 that removes knocking detection signals outside a specific frequency band. A specific frequency band may be set based on the learning result of knocking determination model 401 . As a result, the frequency required for knocking determination is input from the knocking determination model 401 to the frequency input unit 201 . For example, frequencies input to the frequency input unit 201 may be 7.8 kHz, 8.0 kHz, 24.2 kHz, 12.2 kHz, 12.8 kHz, 6.4 kHz, 5.0 kHz, and 25.0 kHz.

Frequency intensity calculation section 202 receives a knocking detection signal from frequency input section 201 and a frequency required for knocking determination. The frequency intensity calculator 202 converts the knocking detection signal into frequency intensity for each frequency band necessary for knocking determination.

Based on the learning result of knocking determination model 401 , parameters necessary for knocking determination are input to engine operating state acquisition unit 203 . For example, the parameter input to the engine operating state acquisition unit 203 may be at least one of ignition timing, engine speed, and load.

Knocking determination unit 204 receives the frequency intensity for each frequency band of knocking and parameter values relating to the operating state of engine 101 as inputs to knocking determination model 401 based on the determination logic according to the learning results of knocking determination model 401 . be. Accordingly, knocking determination section 204 performs decision tree determination to determine whether or not knocking has occurred.

FIG. 5 is an explanatory diagram of a knocking determination model based on a decision tree according to the first embodiment.

For example, the knocking determination model 401 determines whether or not the frequency intensity of the 7.8 kHz band is less than 0.0026 (7.8 kHz<0.0026). If the determination result is true, it is determined whether the ignition timing is less than 33° (SPT<33). If the determination result is false, it is determined that knocking has occurred.

According to this configuration, ECU 111 of engine 101 includes frequency input section 201 , frequency intensity calculation section 202 , engine operating state acquisition section 203 , and knocking determination section 204 . Frequency input unit 201 receives a detection signal including the knocking frequency of engine 101 . The frequency intensity calculator 202 calculates the frequency intensity for each knocking frequency band based on the detection signal. The engine operating state acquisition unit 203 acquires the operating state of the engine 101 . Knocking determination unit 204 stores knocking determination model 401 constructed as learning data of the frequency intensity for each knocking frequency band, the operating state of engine 101, and the knocking state. The knocking condition is determined by inputting the intensity and the operating condition of the engine 101 .

As a result, the parameters necessary for determining the knocking state, including not only the detection signal of knock sensor 113 but also the operating state of engine 101, can be arranged in order of importance and compared. Therefore, even though knocking can be detected with high precision, the decision logic can be simple and easy to calculate because of the tree structure of the YES or NO system.

Furthermore, the engine operating state acquisition unit 203 calculates learning data based on the operating state of the engine 101 measured by the sensor provided in the engine 101 . As a result, appropriate learning data can be input to the knocking determination model 401, and the accuracy of knocking determination can be improved.

Furthermore, the engine operating state acquisition unit 203 acquires a plurality of operating states of the engine 101, and the knocking determination unit 204 determines the knocking frequency band from among the plurality of operating states of the engine 101 acquired by the engine operating state acquiring unit 203. A knocking state is determined by selecting an operating state corresponding to the frequency intensity of each frequency band and inputting the selected operating state and the frequency intensity of knocking for each frequency band to the knocking determination model 401 . This makes it possible to select an operating state of engine 101 that is suitable for determining knocking.

The knocking determination model 401 is constructed by learning by any statistical machine learning other than decision tree learning, such as SVM (Support Vector Machine), least-squares probabilistic classification, Bayesian estimation, and neural network. good too. These methods can also detect knocking with high accuracy.

The input information may determine whether a plurality of predetermined knocking determinations, such as no knocking, trace knocking, or heavy knocking, may be used, and the likelihood (probability) of these multiple knocking states may be calculated. It may be something to do. When calculating the probability, the presence or absence of the occurrence of knocking may be determined based on the determination results of a plurality of times.

Knocking determination unit 204 may determine a plurality of operating states of engine 101, or may determine only the presence or absence of knocking, such as no knocking or only trace knock.

Furthermore, if necessary, decision trees may be separately prepared as a decision tree for determining a trace knock and a decision tree for determining a heavy knock.

FIG. 6 is an explanatory diagram of the operating region dividing method using a plurality of decision trees according to the second embodiment.

FIG. 6 shows an example in which five decision trees A to E are prepared according to the engine speed and load. For example, when the period rotation speed is 2000 rpm and the load is "60", decision tree C601 is selected, and knocking determination is performed using decision tree C601.

Here, when the decision tree is used to learn all the operating regions of the engine 101, a huge decision tree using a very large number of frequency bands is required in order to accurately determine knocking. On the other hand, the ECU generally limits the number of knock sensor frequency bands that can be measured at one time. Therefore, instead of creating one decision tree for all operating states of the engine 101, the operating region is divided by a certain parameter, a plurality of decision trees are created, and the decision tree to be used is determined for each operating region. You can switch to

As a result, even with a limited number of frequency bands of the knock sensor 113, it is possible to detect knocking with high accuracy in all driving ranges.

It should be noted that FIG. 6 describes an example of classification according to the engine speed and the load. However, cases may be classified according to other parameters, or cases may be classified according to one or three or more conditions.

FIG. 7 is an explanatory diagram of the learning range of the decision tree of the divided operating regions according to the third embodiment.

When preparing a plurality of decision trees, there is a concern that the determination results may differ greatly depending on the operating region where the decision trees to be used are switched. Therefore, it is desirable to learn the decision tree not only in the operating range 701 in which knocking is actually determined, but also in a wider operating range 702 that includes the portion where the decision tree in the adjacent operating range should originally determine. That is, the decision tree may be constructed by inputting the learning data of each other's driving regions to the boundary region with the other decision tree.

As a result, it is possible to suppress a step in determination of the boundary region where the operating region is switched (deviation in the result of knocking determination), and it is possible to realize linear knocking determination.

FIG. 8 is an explanatory diagram of a decision tree structure determination method according to the fourth embodiment.

If we divide the operation area, there is a problem that we have to create many decision tree logics. Therefore, when a lot of logic is created, a lot of capacity is used. On the other hand, if a master tree-structured logic is prepared as shown in FIG. can be shared and reused.

Specifically, R = 0, 1, 2, 3 represents the hierarchical structure of the decision tree. Continue this until The information of each cell includes a parameter to be compared, its threshold, the result when the parameter>threshold (=Y), and the result when the parameter≦threshold (=N). The outcome is determined by a probability of 0/1 or 1 and the determination ends, or continuation is selected. If continuation is selected, the information of the next relevant cell is read and the determination is repeated.

FIG. 9 is a flowchart of a decision tree structure determination method according to the fourth embodiment.

First, the ECU 111 sets R=0 and C=0 for the first time (S901). Next, the ECU 111 reads R and C information (parameters, their threshold values, results when parameter>threshold value (=Y), and results when parameter≦threshold value (=N)) (S902).

Next, the ECU 111 determines whether parameter>threshold (=Y) (S903). If the determination result of S903 is true, the ECU 111 determines whether or not to maintain this determination result (S904). If the determination result of S904 is true, the ECU 111 selects the next information column (S905). Next, the ECU 111 selects the next line of information (S906).

On the other hand, if the determination result of S904 is false, the ECU 111 returns the result when parameter>threshold (=Y) (S907).

Furthermore, if the determination result of S903 is false, the ECU 111 determines whether or not to maintain the result when the parameter≦threshold (=N) (S908). If the determination result of S908 is true, the ECU 111 selects the next information column (S909). If the determination result of S908 is false, the ECU 111 returns the result when parameter≦threshold (=N) (S910).

This eliminates the need to create a determination logic for each learning result, making it possible to provide a reusable control structure that can be commonly used for other controls.

Note that the data structure does not have to be a two-dimensional array configuration as shown in FIG. As long as appropriate information can be read out, a one-dimensional array may be used, or a system in which a specific data number is added to the information and selected may be used.

DESCRIPTION OF SYMBOLS 101... Engine 111... ECU 201... Frequency input part 202... Frequency intensity calculation part 203... Engine operating state acquisition part 204... Knocking determination part 401... Knocking determination model 402... Filter part 601... Operation area

Claims (7)

a frequency input unit to which a detection signal including the knocking frequency of the internal combustion engine is input;
a frequency strength calculation unit that calculates the frequency strength for each frequency band of the knocking based on the detection signal;
an internal combustion engine state acquisition unit that acquires an operating state of the internal combustion engine;
A knocking determination model constructed using the frequency intensity for each knocking frequency band, the operating state of the internal combustion engine, and the knocking state as learning data is stored, and the knocking determination model stores the frequency intensity for each of the knocking frequency bands and the knocking determination model. a knocking determination unit that determines a knocking state by inputting an operating state of the internal combustion engine ;
The knocking determination model is a control device for an internal combustion engine constructed by inputting a part of the learning data of an operating region adjacent to the knocking determining operating region among a plurality of operating regions. The internal combustion engine state acquisition unit calculates the learning data based on the operating state of the internal combustion engine measured by a sensor provided in the internal combustion engine.
The control device for an internal combustion engine according to claim 1. The internal combustion engine state acquisition unit acquires a plurality of operating states of the internal combustion engine,
The knocking determining unit selects an operating state corresponding to the frequency intensity of each frequency band of the knocking from among the plurality of operating states of the internal combustion engine acquired by the internal combustion engine state acquiring unit, and selects the operating state and Determining a knocking state by inputting the frequency intensity of each knocking frequency band into the knocking determination model;
The control device for an internal combustion engine according to claim 1. The frequency input unit has a filter unit that removes the detection signal outside a specific frequency band,
The control device for an internal combustion engine according to claim 1. The knocking determination unit
having a plurality of knocking determination models,
selecting a knocking determination model according to the operating state of the internal combustion engine from among the plurality of knocking determination models;
The control device for an internal combustion engine according to claim 4. the knocking determination model is constructed by decision tree learning;
The control device for an internal combustion engine according to claim 1. the knocking determination model is constructed by statistical machine learning;
The control device for an internal combustion engine according to claim 1.

Images