JPH0491348A - Automobile control device - Google Patents

Abstract

PURPOSE:To always attain adequate control by inputting signals, related to an engine speed, knocking, exhaust temperature, air-fuel ratio, cooling water temperature and lubricating oil temperature, to a story type nerve circuit model, and outputting information for showing possibility of generating burning in an engine cylinder wall surface. CONSTITUTION:A predetermined arithmetic process is executed to create a control signal by fetching necessary signals from detectors of engine speed detector 3, O2 sensor 4, air flow meter 5, knocking sensor 6, cooling water temperature sensor 7, exhaust temperature sensor 8, oil temperature sensor 9, etc., to an engine main unit 1. In an engine control circuit 2, a nerve circuit model 15 is additionally provided. A signal D for showing possibility of generating burning in a cylinder wall surface is generated from each signal for respectively showing cooling water temperature, knocking generation, oxygen concentration, exhaust gas temperature, engine speed and lubricating oil temperature. A rotational speed of a cooling fan 13 is increased by the engine control circuit 2 in accordance with this signal D, and generation of seizure of the engine 1 is prevented by increasing heat radiating power of a radiator 22 to decrease the cooling water temperature.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an automobile control device equipped with a signal processing function based on a hierarchical neural circuit model, and is particularly suitable for an automobile powered by a gasoline engine. Regarding.

[Conventional technology]

In recent years, signal processing methods based on hierarchical neural circuit models have been attracting attention because they do not require complicated programs and can be applied with high versatility. Devices that apply this method are now being seen; an example is the Japanese Patent Application Laid-Open No. 1-30
The description in Japanese Patent No. 1946 can be mentioned.

In the technique described in this publication, various combustion states of the engine are predicted by detecting the pressure within the engine cylinder combustion chamber and inputting it to a hierarchical neural circuit model.

[! @Problems that Akira is trying to solve]

The conventional technology described in the above publication does not take into consideration the diversification of information necessary for controlling automobiles, and the information obtained is limited to information related to combustion pressure such as engine knocking and misfire. There was a problem.

The purpose of the present invention is to provide various information necessary for controlling a vehicle.
It is an object of the present invention to provide a vehicle control device that can arbitrarily and easily obtain signals from general sensors that are often conventionally installed in engines, and that can always perform accurate control.

[Means for solving problems].

The present invention is capable of detecting various abnormal phenomena that appear in the engine or body of an automobile only by comprehensively determining various types of signals.
This was done with a focus on the fact that detection was possible for the first time.
Therefore, in the present invention, in order to achieve the above object, a wide variety of data such as engine speed, knock, exhaust temperature, oxygen concentration in exhaust gas, lubricating oil temperature, and cooling water temperature are used as input signals. The car is controlled by inputting the input into a hierarchical neural circuit model, multiplying it by a weight constant, adding it, and using the final output value obtained by nonlinear transformation as the set value. This method was adopted.

First, in the first aspect of the present invention, the input layer of the hierarchical neural circuit model includes an engine rotational speed signal, a knock signal, an exhaust temperature signal, an air-fuel ratio signal, a cooling water temperature signal, and a lubricating oil temperature signal. A means for inputting a seed signal is provided.

Similarly, in the second invention, the input layer of the hierarchical neural circuit model is provided with means for inputting four types of signals: an engine rotational speed fluctuation signal, an exhaust gas temperature signal, an air-fuel ratio signal, and a cooling water temperature signal. It is.

Further, the third invention also includes means for inputting two types of signals, an engine rotational speed fluctuation signal and an automobile body vibration signal, to the input layer of the hierarchical neural circuit model, and a back property of the hierarchical neural circuit model. The system is provided with means for inputting an allowable stability signal set by the driver as training data necessary for the steering system.

Similarly, in the fourth invention, means is provided for inputting two types of signals, an engine rotational speed fluctuation signal and a crankshaft torsion signal, to the input layer of the hierarchical neural circuit model.

Also, in the fifth invention, means is provided in the input layer of the hierarchical neural circuit model for panning three types of signals: an engine exhaust gas treatment catalyst temperature signal, an exhaust temperature signal, and an air-fuel ratio signal. be.

Furthermore, in the sixth invention, the input layer of the hierarchical neural circuit model is provided with means for converting the engine rotational speed signal into a predetermined number of time-series signals at predetermined intervals and manually inputting the signals.

[Operation] The hierarchical neural circuit model learns weight constants by backpropagation using training data.

Therefore, by appropriately selecting the input signal and the teacher data, the final output value obtained from each hierarchical neural circuit model becomes the required predetermined control signal.

An individual explanation of this is as follows.

In the first invention, six types of signals are input into a hierarchical neural circuit model: an engine rotational speed signal, a knock signal, an exhaust temperature signal, an air-fuel ratio signal, a cooling water temperature signal, and a lubricating oil temperature signal. Therefore, information representing the possibility of burnout occurring on the engine cylinder wall surface is extracted from the output layer.

In the second invention, four types of signals, namely, an engine speed fluctuation signal, an exhaust temperature signal, an air-fuel ratio signal, and a cooling water temperature signal, are input to the input layer of the hierarchical neural circuit model, so that the output layer The information representing the actual air-fuel ratio of the engine is extracted from the .

In the third invention, two types of signals, the engine rotational speed fluctuation signal and the car body vibration signal, are input to the input layer of the hierarchical neural circuit model, and the signals necessary for backpropagation of the hierarchical neural circuit model are inputted into the input layer of the hierarchical neural circuit model. Since the permissible stability signal set by the driver is input as training data, the output layer provides information representing the stability felt by the passenger when the engine's drive power transmission system is in an overtop state. Work to be taken out.

In the fourth invention, since two types of signals, the engine rotational speed fluctuation signal and the crankshaft torsion signal, are input to the input layer of the hierarchical neural circuit model, information representing the output torque of the engine is extracted from the output layer. Work so that you can

In the fifth invention, three types of signals, the engine exhaust gas processing catalyst temperature signal, the exhaust temperature signal, and the air-fuel ratio signal, are input to the input layer of the hierarchical neural circuit model. The information representing the NOx exhaust flow rate at is retrieved.

In the sixth invention, the engine rotational speed signal is converted into a predetermined number of time-series signals at predetermined intervals and input into the input layer of the hierarchical neural circuit model, so that the misfiring cylinder of the engine can be detected from the output layer. It works so that the information it represents is retrieved.

〔Example〕

DESCRIPTION OF THE PREFERRED EMBODIMENTS The automobile control device according to the present invention will be explained in detail below with reference to illustrated embodiments.

FIG. 1 is an overall configuration diagram of an embodiment of the present invention. In the figure, 1 is an engine main body, which is controlled by an engine control circuit 2. As shown in FIG.

The engine body l has a rotation speed detector 3.0. sensor 4,
Various detectors such as an air flow meter (intake flow meter) 5, a knock sensor 6, a cooling water temperature sensor 7, an exhaust temperature sensor 8, and an oil temperature sensor 9 are provided, and the engine control circuit 2 receives information from these detectors. Required signals are taken in, predetermined arithmetic processing is performed to create various required control signals, and the ignition signals are supplied to the ignition coil 10 and the spark plug 11
performs a predetermined ignition operation, supplies a fuel injection signal to the injector 12 to control the fuel supply amount, and controls the rotational speed of the cooling fan 13 to maintain the engine temperature at a predetermined value.

In addition, in this figure, 20 is an intake pipe, 21 is an exhaust pipe,
22 is a radiator.

By the way, the above is the same as a general engine control system, but in this embodiment, a hierarchical neural circuit model 15 is further provided, and the six types of signals from the various detectors described above are transmitted to this model. Each buffer amplifier 16a
-16f. The final output signal of this hierarchical neural circuit model 15 is supplied to the engine control circuit 2 and used for engine control.

First, a signal Tw representing the coolant temperature of the engine 1 detected by the coolant temperature sensor 7 is inputted via the buffer amplifier 16a.

Next, a signal K detected by the knock sensor 6 that does not indicate the occurrence of knock is inputted via the buffer amplifier 16b.
Similarly, a signal 02 representing the oxygen concentration in the exhaust gas detected by the 02 sensor 4, and a signal T representing the temperature of the exhaust gas detected by the exhaust temperature sensor 8. , a signal N representing the engine rotational speed detected by the rotational speed detector 3, and a signal T representing the lubricating oil temperature of the engine detected by the oil temperature sensor 9. However, the respective buffer amplifiers 16c and 16d
, 16e, 16f.

As a result, the hierarchical neural circuit model 15 generates a signal representing the possibility of burnout of the cylinder wall of the engine 1 as its final output, and inputs this signal to the engine control circuit 2.

Therefore, in response to this signal, the engine control circuit 2 increases the rotational speed of the cooling fan 13, increases the heat dissipation capacity of the radiator 22, lowers the cooling water temperature, and performs control to prevent the engine 1 from seizing. That's what I do.

Next, the signal processing operation using this hierarchical neural circuit model will be explained with reference to FIG.

This figure 2 shows the relationship between the input and output of the synalapse i of the hierarchical neural circuit model 23' (15).The output signals from 1j and fl are applied to the previous synalapse with Ok and Oj%On, respectively. When inputting these output signals, first, a synalapse weight Wi is added to each input signal and input as signals Wik, Wij, and WiQ.

Then, the output Oi of synalapse i is expressed by the following equation.

Of =f (net)-e=ge(ΣWij 0j)-
〇) --(1) Here, θ: Threshold value net i: Total sum of inputs Figure 3 is an example of this threshold value O, and the total sum of inputs net
In the case where there is a stepwise change in response to a change in i, when the sum of inputs net i becomes greater than θ, the output becomes “1′”.

FIG. 4 shows a case where the function is a predetermined function, and in this case, the output fi is as shown in the following equation.

fi=□・・・・・・(2)
1+e''' Next, the backpropagation process for learning each synalapse weight of this hierarchical neural circuit model 23 (15) will be explained with reference to FIG. Focusing on the output signal y, this is intended to match the teacher signal y.

First, the error e between the output signal y and the teacher signal y is given by the following equation.

e= yt -ys (3) Next, let d be the degree of influence of error on the operating level U of the unit, then d, = e - f,' (U.)... ...(4) becomes.

However, here, f″(U) is as shown in the following equation.

fdl f' (U) = 7 tō = au (1 y. 4-...(5) Therefore, 5 in the output layer of this hierarchical neural circuit model
The synalapse weight W, which is the second input part.

The correction amount ΔW, 2 j (N+1) of 2 j is expressed by the following equation.

△W,,2j(N+1)=ηds 3' t 4...
...(6) This shows the correction procedure for synalapse weighting. Here, N: Symbol representing the previous time η: Learning constant yIJ: j-th output of the intermediate layer However, in order to achieve stable convergence, , without using the correction amount obtained by this equation (6) as is, by using the following equation,
A new synalapse weight W−r2J (N+1) is modified and used.

W-+ 2 J (N+ 1); W-r 2 J (
N)+ΔW,, 2j(N+1)+αΔW−+ 2 J
(N) (7) Here, the synalapse weight at the input part of the output layer is modified as follows: α: stabilization constant.

Next, we will proceed to explain the case of the middle layer.
In the figure, synalapse weights Wj,, 1 at the i-th input section of the third unit of the intermediate layer will be explained.

First, let d and j be the influence of an error on the operation level U1 of the unit, which can be expressed by the following equation.

dy=”y(U-i)d, W,, 2 j ・
...(8) Therefore, i of the first unit in the middle layer
The correction amount ΔW−Js 1 + (N+1) of the synalapse weights W, .

Δw,,, 1 + (N+1) = ηLJ y++
......(9) Here, 'l + t' is the output signal of the i-th input layer. However, in order to obtain stable convergence also here, a new synalapse weight W1r + i l
(N+1) is determined using the following formula.

W, 4.1 l(N+1) = W-ts 1 t (
N) ten ΔW, J, 1, (N+1) ten αΔw*,,
1 + (N + I) (10) Therefore, by repeating the calculation process of these equations (3) to (lO), it is possible to specify the synalapse weight necessary to minimize the error e. As in the embodiment shown in FIG.
A signal 01 represents the oxygen concentration in the gas, and a signal T represents the temperature of the exhaust gas. , a signal N representing the rotational speed of the engine,
In addition, there is a signal T that indicates the engine lubricating oil temperature. By inputting this, the output layer of the hierarchical neural circuit model 15 can provide information representing the possibility of burnout occurring on the cylinder wall surface of the engine.

Here, the relationship between the input signal and the output signal of the hierarchical neural circuit model 15 in the first embodiment of the present invention described above is schematically shown in FIG. 6.

Since there are six types of input signals as described above, there are six input layer synalapses of the hierarchical neural circuit model 15, but five in the intermediate layer and one in the output layer.

Note that it is sufficient that the number in the intermediate layer is two or more, but the measurement accuracy increases as the number increases. Therefore, an arbitrary number may be determined in consideration of the resulting complexity of the circuit configuration and increase in learning time.

In the embodiment shown in FIG. 1, the engine control circuit 2 increases the blowing capacity of the cooling fan 13 in response to information indicating the possibility of burnout on the cylinder wall of the engine, thereby increasing the cooling water temperature. However, instead of or in addition to this, the capacity of an oil cooler (not shown) may be controlled to lower the lubricating oil temperature.

Therefore, according to the embodiment described in FIG. 1 and FIG. 6, the possibility of burnout due to a rise in the temperature of the cylinder wall surface is predicted based on the information, and the control of the cooling water temperature and lubricating oil temperature is thereby performed. This eliminates the need for prior art techniques, such as making the air-fuel ratio too rich in advance in anticipation of a risk of temperature rise on the cylinder wall during high loads, and improves exhaust gas purification. It is possible to obtain sufficient improvement and reduction in fuel consumption.

FIG. 7 shows another embodiment of the present invention, in which the hierarchical neural circuit model 15 in FIG. Input four types of signals: signal 02 representing the oxygen concentration, and signal Tw representing the cooling water temperature,
Accordingly, a configuration is used in which information B representing the actual air-fuel ratio within the cylinder of the engine is outputted from the output layer.

Therefore, in the embodiment shown in FIG. 7, the input layer of the hierarchical neural circuit model 15 has four synalapses, but four are also used as the intermediate layer, making one output layer. be.

The real air-fuel ratio information B output from this hierarchical neural circuit model 15 is input to the engine control circuit 2, and when the engine is under a predetermined operating condition such as at a low temperature, 0. It is used for air-fuel ratio feedback control in place of the signal O1 from sensor 4, and as a result, according to this embodiment, the air-fuel ratio of the engine is effectively prevented from becoming excessively rich at low temperatures, etc. Regardless of the operating conditions, highly accurate air-fuel ratio control can be easily maintained at all times.

FIG. 8 also shows an embodiment of the present invention, in which information C representing the driving stability felt by the passenger when the vehicle driving force transmission system is set to over-top is obtained from the hierarchical neural circuit model 15. Therefore, as an input signal, a signal dN/indicating the fluctuation of the engine rotational speed is used.
The hierarchical neural circuit model 15 described above uses two types of signals: dt and a signal V representing the vibration of the car body.
As training data necessary for backpropagation of
The system uses a signal Se representing the permissible stability, which is input and set by the operation of the driver (passenger). Although the signal V representing the vibration of the vehicle body is not shown in FIG. 1, it may be obtained by installing a predetermined vibration detection sensor in the vehicle. For setting, an input device such as a digital switch or a push-button switch is provided, and the driver inputs in advance the stability that he or she feels when the car is running over the top by operating the input device described above. It is supposed to be done.

Therefore, in the embodiment shown in FIG. 8, the synapses weighting of the hierarchical neural circuit model 15 is changed using the signal Se as teacher data, and is set so as to obtain a predetermined output. The information C representing this is input to the engine control circuit 2, and based on the level, the driver knows the limit of stability that the driver can tolerate, and when this level exceeds a predetermined value, the use of the overtop is stopped. It controls a transmission, which is not shown in the figure.

It should be noted that the synalapse weighting state of the hierarchical neural circuit model 15 at this time can be saved in a predetermined volatile memory, etc., and it can be set to that state from the beginning when the occupants of the car are the same. , for the same passenger (driver), the time required for backpropagation is no longer necessary from the second time onwards.

Therefore, according to this embodiment, it is possible to prevent loss of driving stability of the automobile due to excessive use of the overtop.
A comfortable driving condition can be easily obtained.

Moreover, on the contrary, according to the embodiment shown in FIG.
Since you can definitely know the limit of overtop use without compromising stability, you can safely ride overtop up to the limit where ride comfort deteriorates.
Its range of use can be widened and good fuel efficiency can be easily obtained.

FIG. 9 also shows an embodiment of the present invention. In this embodiment, a hierarchical neural circuit model 15 is provided with two signals: a signal dN/dt representing the engine rotational speed fluctuation, and a signal X representing the amount of twist of the engine crankshaft. The hierarchical neural circuit model 15 is configured to input a seed signal and thereby obtain information representing the output torque of the engine from the output thereof.

At this time, data from a torque meter (dynamometer, dynamometer) temporarily attached to the engine is used as the training data necessary for the backpropagation described above, and the synalapse of the hierarchical neural circuit model 15 is used. Once you finish setting the weight, this torque meter is no longer needed.

Next, in the embodiment shown in FIG. 10, the input signals are a signal Tc representing the temperature of the exhaust gas processing catalytic converter of the engine and a signal T representing the temperature of the exhaust gas. , and the signal O1 representing the oxygen concentration in the exhaust gas, information E representing the NOx emission flow rate in the engine exhaust can be obtained from the output layer of the hierarchical neural circuit model 15. This is how it was constructed.

Then, the engine control circuit 2 (FIG. 1) receives information E representing the NOx exhaust flow rate in the engine exhaust, and
This allows the NOx exhaust flow rate to be known and the engine EGR amount to be controlled as necessary.

Therefore, according to this embodiment, NO in the engine exhaust
Since the NOx exhaust flow rate can be accurately known in real time, the NOx exhaust flow rate can be reliably reduced without sacrificing engine performance.

The embodiment shown in FIG. 11 is an embodiment of the present invention that enables misfiring cylinder discrimination in a multi-cylinder engine.
0 is a sample-and-hold circuit that inputs a signal dN/dt representing engine speed fluctuation, samples it at a predetermined frequency, and holds it for a predetermined period of time (for a 4-cycle engine, the crankshaft rotates twice). , converts the signal dN/dt that appears sequentially and time-sequentially into a spatial signal and inputs it to the hierarchical neural circuit model 15.

On the other hand, in this embodiment, the hierarchical neural circuit model 15 is
The number of synalapses in the input layer is determined by the sample and hold circuit 3.
The number of synalapses in the output layer is equal to the number n of cylinders in the engine.

Now, looking at the spatial signal output from the sample and hold circuit 30, it exhibits a unique rotational fluctuation pattern due to the upper torque fluctuation for each stroke of the engine and the presence of a plurality of cylinders.

Therefore, if backpropagation of the hierarchical neural circuit model 15 is performed using the signal pattern when no misfire occurs in any cylinder as training data,
From the pattern change when a misfire appears, it is possible to specify the cylinder in which the misfire occurred and obtain information F1 to F for determining it. Therefore, according to this embodiment, a special misfire detection Engine operating conditions can always be known reliably and in real time without using sensors.

〔Effect of the invention〕

The present invention makes it possible to easily obtain information necessary for truly controlling the engine, which was not possible with conventional technology, in real time without the need for special sensors. This makes it possible to sufficiently improve performance and ride comfort.

[Brief explanation of the drawing]

FIG. 1 is an overall configuration diagram showing an embodiment of an automobile control device according to the present invention, FIG. 2 is an explanatory diagram of a hierarchical neural circuit model,
Figures 3 and 4 are explanatory diagrams of the synalapse output characteristics, respectively. Figure 5 is an explanatory diagram of the backpropagation operation of the hierarchical neural circuit model. Figures 6, 7, 8, and 9
10 and 11 are explanatory diagrams each showing an embodiment of the present invention. 1...Engine, 2...Engine control circuit, 3...°...Rotation speed detector, 4...
・01 sensor, 5...air flow meter (intake flow meter), 6...knock sensor, 7...
Cooling water temperature sensor, 8...Exhaust temperature sensor, 9...
...Oil temperature sensor, 15...Hierarchical neural circuit model.

Claims (6)

[Claims] 1. In a vehicle control device that obtains the information necessary for vehicle operation control from a hierarchical neural circuit model, the input layer of the hierarchical neural circuit model is an engine rotational speed signal, a knock signal, an exhaust temperature signal, and an air-fuel ratio. A means is provided for inputting six types of signals: a signal, a cooling water temperature signal, and a lubricating oil temperature signal, thereby extracting information representing the possibility of burnout of the engine cylinder wall surface from the output layer of the hierarchical neural circuit model. An automobile control device characterized in that it is configured to 2. In a vehicle control device that obtains information necessary for vehicle operation control from a hierarchical neural circuit model, the input layer of the hierarchical neural circuit model includes an engine rotational speed fluctuation signal, an exhaust temperature signal, an air-fuel ratio signal, An automobile characterized in that it is provided with means for inputting four types of signals, including a cooling water temperature signal, so that information representing the actual air-fuel ratio of the engine is extracted from the output layer of the hierarchical neural circuit model. Control device. 3. In a vehicle control device that obtains information necessary for vehicle operation control from a hierarchical neural circuit model, two types of signals, an engine rotational speed fluctuation signal and a vehicle body vibration signal, are input to the input layer of the hierarchical neural circuit model. A means for inputting a signal and a means for inputting an allowable stability signal set by the driver as training data necessary for backpropagation of the hierarchical neural circuit model are provided. An automobile control device characterized in that information representing the stability felt by a passenger when the driving force transmission system of the engine is in an overtop state is extracted from the output layer. 4. In a vehicle control device that obtains information necessary for vehicle operation control from a hierarchical neural circuit model, two types of signals, an engine rotational speed fluctuation signal and a crankshaft torsion signal, are input to the input layer of the hierarchical neural circuit model. 1. An automobile control device, comprising means for inputting the information such that the information representing the output torque of the engine is extracted from the output layer of the hierarchical neural circuit model. 5. In a vehicle control device that obtains the information necessary for vehicle operation control from a hierarchical neural circuit model, the input layer of the hierarchical neural circuit model includes an engine exhaust gas processing catalyst temperature signal, an exhaust temperature signal, and an air-fuel ratio. A means for inputting three types of signals is provided, whereby NO in the engine exhaust is detected from the output layer of the hierarchical neural circuit model.
_x An automobile control device characterized in that it is configured to extract information representing an exhaust flow rate. 6. In a vehicle control device that obtains information necessary for vehicle operation control from a hierarchical neural circuit model, a predetermined number of time-series engine rotational speed signals are input to the input layer of the hierarchical neural circuit model at predetermined intervals. An automobile control device characterized in that it is provided with means for converting into a signal and inputting it, so that information representing a misfiring cylinder of an engine is extracted from the output layer of the hierarchical neural circuit model.