JP2023116987A - Knocking estimation device - Google Patents

Abstract

To improve the accuracy in estimating knocking intensity.SOLUTION: A control device stores mapping data for defining mapping that outputs an output variable indicating knocking intensity by being inputted with an input variable. The control device executes correction processing of offsetting the values indicating frequency intensities of a plurality of bands so that the smallest value among the values indicating the intensities of frequencies of the plurality of bands becomes zero. The control device executes normalization processing in which the values indicating the frequency intensities of the plurality of bands offset by the correction processing are multiplied by the same coefficient to convert the values into values in a predetermined range. The control device executes acquisition processing that is the processing of acquiring, as input variables, the values indicating the frequency intensities of the plurality of bands converted by the normalization processing. The control device executes calculation processing of outputting the value of an output variable by inputting the input variable acquired through the acquisition processing into the mapping.SELECTED DRAWING: Figure 2

Description

The present invention relates to a knocking estimation device.

A vehicle described in Patent Document 1 includes an internal combustion engine, a knocking sensor, and a knocking estimation device. The knocking sensor detects vibrations generated in the internal combustion engine. The knocking estimation device has an execution device and a storage device. The storage device stores map data defining a map for outputting an output variable indicating the knocking intensity of the internal combustion engine when the input variable is input. The execution device includes an acquisition process, which is a process of acquiring values indicating the intensity of frequencies in a plurality of bands of vibration detected by the knocking sensor, and an output variable by inputting the input variables acquired by the acquisition process into the mapping. A calculation process for outputting a value is executed.

Japanese Patent No. 6683282

The vibrations detected by the knocking sensor disclosed in Patent Document 1 include noise generated by factors other than knocking, in addition to vibrations associated with knocking. In this case, a knocking estimation device such as that disclosed in Patent Document 1 may estimate the knocking intensity to be small due to the noise described above, even though the vibration caused by knocking is large. Further, due to the above noise, the knocking intensity may be estimated to be large even though the vibration caused by knocking is small. As a result, a knocking estimation device such as that disclosed in Japanese Patent Laid-Open No. 2002-200312 may have a lower accuracy in estimating the knocking intensity.

A knocking estimation apparatus for solving the above-described problems is applied to a vehicle provided with an internal combustion engine and a knocking sensor for detecting vibration generated in the internal combustion engine, and the vibration detected by the knocking sensor is divided into a plurality of frequency bands. A knocking estimation device for estimating the knocking intensity of the internal combustion engine based on the intensity, comprising an execution device and a storage device, the storage device producing an output indicating the knocking intensity when an input variable is input. mapping data defining a mapping for outputting a variable is stored, and the executing device maps the frequencies of the plurality of bands such that the smallest value among values indicating the intensity of the frequencies of the plurality of bands becomes zero. Correction processing for offsetting the value indicating the intensity, and normalization for converting the values indicating the intensity of the frequencies of the plurality of bands offset by the correction processing by the same coefficient to convert them into values within a predetermined range. a process, an acquisition process which is a process of acquiring, as the input variables, the values indicating the intensity of the frequencies of the plurality of bands transformed by the normalization process; and inputting the input variables acquired by the acquisition process into a mapping. and a calculation process for outputting the value of the output variable.

According to the above configuration, by executing the correction process and the normalization process, compared to the case where only the normalization process is executed, the magnitude of the value indicating the intensity of the frequency of the plurality of bands converted by the normalization process is reduced. The difference is noticeable. The difference between vibration and noise caused by knocking is also noticeable. As a result, it is possible to prevent noise from estimating the knocking intensity to be small even though the vibration caused by knocking is large, or from estimating the knocking intensity to be large even though the vibration caused by knocking is small. be done. As a result, the accuracy of estimating the knocking intensity can be improved.

1 is a schematic diagram showing a schematic configuration of a vehicle; FIG. 4 is a flowchart showing ignition timing control; FIG. 4 is an explanatory diagram showing values before and after correction processing; FIG. 5 is an explanatory diagram showing values after normalization processing and comparison values;

<Outline configuration of vehicle>
An embodiment will be described below with reference to FIGS. 1 to 4. FIG. First, a schematic configuration of the vehicle 100 will be described.

As shown in FIG. 1, a vehicle 100 includes an internal combustion engine 10, an automatic transmission 30, a differential mechanism 40, and a plurality of drive wheels 50.
The internal combustion engine 10 includes a plurality of cylinders 11, a crankshaft 12, an intake passage 21, a throttle valve 22, a plurality of fuel injection valves 23, a plurality of ignition devices 24, an exhaust passage 26, a three-way catalyst 27, and a filter 28. there is

The cylinder 11 is a space for burning a mixture of fuel and intake air. The internal combustion engine 10 has four cylinders 11 . The intake passage 21 is connected to the cylinder 11 . A portion of the intake passage 21 including the downstream end is branched into four. Each branched passage is connected to each cylinder 11 . The intake passage 21 introduces intake air into each cylinder 11 from the outside of the internal combustion engine 10 . The throttle valve 22 is positioned upstream of the branched portion of the intake passage 21 . The throttle valve 22 adjusts the amount of intake air flowing through the intake passage 21 .

The fuel injection valve 23 is positioned near the downstream end of the intake passage 21 . The internal combustion engine 10 has four fuel injection valves 23 corresponding to the four cylinders 11 . The fuel injection valve 23 injects fuel supplied from a fuel tank (not shown) into the intake passage 21 . The ignition device 24 is located in the cylinder 11 . The internal combustion engine 10 has four ignition devices 24 corresponding to the four cylinders 11 . The ignition device 24 ignites the mixture of fuel and intake air by spark discharge.

The exhaust passage 26 is connected to the cylinder 11 . A portion of the exhaust passage 26 including the upstream end is branched into four. Each branched passage is connected to each cylinder 11 . The exhaust passage 26 discharges the exhaust from each cylinder 11 to the outside of the internal combustion engine 10 .

The three-way catalyst 27 is located downstream of the branched portion of the exhaust passage 26 . The three-way catalyst 27 purifies exhaust gas flowing through the exhaust passage 26 . The filter 28 is located downstream of the three-way catalyst 27 in the exhaust passage 26 . Filter 28 collects particulate matter contained in exhaust gas flowing through exhaust passage 26 .

The crankshaft 12 is connected to a piston (not shown) positioned inside each cylinder 11 . The crankshaft 12 rotates by receiving combustion torque. The crankshaft 12 is also connected to left and right drive wheels 50 via an automatic transmission 30 and a differential mechanism 40 . Automatic transmission 30 selects one of a plurality of gear stages according to the running state of vehicle 100 . The automatic transmission 30 shifts and outputs the input torque at a gear ratio corresponding to the selected gear stage. The differential mechanism 40 allows the left and right driving wheels 50 to have different rotation speeds.

As shown in FIG. 1 , vehicle 100 includes crank angle sensor 71 , accelerator operation amount sensor 72 , vehicle speed sensor 73 , and knocking sensor 74 . A crank angle sensor 71 detects a crank angle SC, which is the angular position of the crankshaft 12 . An accelerator operation amount sensor 72 detects an accelerator operation amount ACC, which is an operation amount of an accelerator pedal operated by a driver. Vehicle speed sensor 73 detects vehicle speed SP, which is the speed of vehicle 100 . Knocking sensor 74 detects vibration V generated in internal combustion engine 10 . Here, the knocking sensor 74 is, for example, a resonant knocking sensor having a vibrating body, a non-resonant knocking sensor which is a type of acceleration sensor, or the like. The knocking sensor 74 is attached to a cylinder block or the like of the internal combustion engine 10 .

Vehicle 100 includes a control device 90 . Control device 90 acquires a signal indicating crank angle SC from crank angle sensor 71 . The control device 90 calculates an engine rotation speed NE, which is the rotation speed of the crankshaft 12, based on the crank angle SC. The control device 90 acquires a signal indicating the accelerator operation amount ACC from the accelerator operation amount sensor 72 . Control device 90 acquires a signal indicating vehicle speed SP from vehicle speed sensor 73 . Control device 90 acquires a signal indicating vibration V from knock sensor 74 . Then, the control device 90 subjects the vibration V to bandpass filtering. Specifically, the control device 90 performs 11 types of bandpass filtering. Each band-pass filter process is a process for extracting only vibrations in a predetermined frequency band from the vibrations V. FIG. The frequency bands extracted by these 11 types of bandpass filter processing are obtained by dividing the frequency band of 5 kHz to 25 kHz into 11 so as not to overlap. The control device 90 acquires the intensity of each vibration extracted by the 11 types of band-pass filtering as a first intensity VA1 to an eleventh intensity VA11 in ascending order of frequency band. That is, the control device 90 acquires the first intensity VA1 to the eleventh intensity VA11 as the intensity of the frequencies of the vibration V in a plurality of bands.

The control device 90 includes a CPU 91 , a peripheral circuit 92 , a ROM 93 , a storage device 94 and a bus 95 . A bus 95 connects the CPU 91, the peripheral circuit 92, the ROM 93, and the storage device 94 so that they can communicate with each other. The ROM 93 stores in advance various programs for the CPU 91 to execute various controls. The storage device 94 preliminarily stores mapping data 94A. A map M defined by the map data 94A outputs an output variable indicating a knocking intensity NS, which is the intensity of knocking occurring in the internal combustion engine 10, when an input variable is input. A specific description of the map M will be given later.

The storage device 94 stores data including the first intensity VA1 to the eleventh intensity VA11 for a certain period of time. Peripheral circuit 92 includes a circuit that generates a clock signal that defines internal operations, a power supply circuit, a reset circuit, and the like. In this embodiment, the CPU 91 and ROM 93 are execution devices. Also, the storage device 94 is a storage device. In this embodiment, control device 90 functions as a knocking estimation device that estimates knocking intensity NS.

The CPU 91 controls the internal combustion engine 10, the automatic transmission 30, etc. by executing various programs stored in the ROM 93. FIG. Specifically, the CPU 91 calculates a vehicle required output, which is a required output value required for the vehicle 100 to travel, based on the accelerator operation amount ACC and the vehicle speed SP. The CPU 91 controls the internal combustion engine 10 and the automatic transmission 30 based on the vehicle output power demand. In controlling the internal combustion engine 10, the CPU 91 controls the opening of the throttle valve 22, the fuel injection amount from the fuel injection valve 23, the ignition timing of the ignition device 24, and the like.

<Ignition timing control>
Next, ignition timing control performed by the CPU 91 will be described. In ignition timing control, the CPU 91 controls the ignition timing based on the presence or absence of knocking of the internal combustion engine 10 . The ROM 93 prestores a program for executing ignition timing control. The CPU 91 repeatedly executes ignition timing control from when the internal combustion engine 10 is requested to start until it stops.

As shown in FIG. 2, when starting the ignition timing control, the CPU 91 advances the process of step S11. In step S11, the CPU 91 executes a correction process of offsetting the first intensity VA1 to the eleventh intensity VA11 so that the smallest value among the first intensity VA1 to the eleventh intensity VA11 becomes zero. For example, as indicated by the chain double-dashed line in FIG. 3, the value of the fourth strength VA4 is assumed to be the smallest among the first strength VA1 to the eleventh strength VA11. At this time, the CPU 91 offsets all of the first intensity VA1 to the eleventh intensity VA11 by subtracting the value of the fourth intensity VA4 from each of the first intensity VA1 to the eleventh intensity VA11. As a result, as indicated by the solid line in FIG. 3, the CPU 91 calculates the first correction strength VB1 to the eleventh correction strength VB11 as values offset by the correction processing. Also, due to this offset, the value of the fourth correction strength VB4 becomes zero. Thereafter, as shown in FIG. 2, the CPU 91 advances the process to step S12.

As shown in FIG. 2, in step S12, the CPU 91 performs a normalization process of multiplying all of the first correction strength VB1 to the eleventh correction strength VB11 by the same coefficient and converting them into values within a predetermined range. Execute. In the present embodiment, the CPU 91 multiplies the first correction strength VB1 to the eleventh correction strength VB11 by the same coefficient to obtain the first normalization strength VC1 to the eleventh normalization strength VC1 to the eleventh correction strength VB11 as indicated by the solid lines in FIG. A strength VC11 is calculated. At this time, the CPU 91 adjusts the coefficient value according to the first correction strength VB1 to the eleventh correction strength VB11 so that the sum of the first normalization strength VC1 to the eleventh normalization strength VC11 becomes 1. That is, the CPU 91 converts the first corrected strength VB1 to the eleventh corrected strength VB11 into a range in which the sum of the first normalized strength VC1 to the eleventh normalized strength VC11 becomes one. Note that the first normalized strength VC1 to the eleventh normalized strength VC11 are values converted by normalization processing. Thereafter, as shown in FIG. 2, the CPU 91 advances the process to step S13.

As shown in FIG. 2, in step S13, the CPU 91 executes acquisition processing for acquiring the first normalized strength VC1 to the eleventh normalized strength VC11 as values to be used as input variables. Specifically, the CPU 91 performs non-negative matrix factorization on the first normalized strength VC1 to the eleventh normalized strength VC11 to obtain a plurality of Z values to be used as input variables. Here, non-negative matrix factorization is a mathematical algorithm that decomposes a non-negative matrix into the matrix product of two other non-negative matrices. Note that the non-negative matrix factorization is a well-known technique used for separating signals related to vibration and sound, as described in Japanese Patent Application Laid-Open No. 2020-126021 and Japanese Patent Application Laid-Open No. 2009-128906, for example. After that, the CPU 91 advances the process to step S14.

In step S14, the CPU 91 generates the Z values obtained in step S13 as input variables x(1) to x(Z) to the map M for estimating the knocking intensity NS. Specifically, the CPU 91 sets the values of the components of the first row and first column among the Z values included in the matrix acquired in step S13 to the input variable x(1), the first row and second column Substitute the values of the eye components to the input variable x(2) one by one, and so on. After that, the CPU 91 advances the process to step S15.

In step S15, the CPU 91 inputs the input variables x(1) to x(Z) generated in the process of step S14 and the input variable x(0) as the bias parameter to the mapping M, thereby obtaining the output variable y( Calculate the value of i). After that, the CPU 91 advances the process to step S21.

An example of the mapping M defined by the mapping data 94A is a function approximator, which is a fully-connected forward-propagating neural network with one intermediate layer. Specifically, in the map M, the input variables x(1) to x(Z) and the input variable x(0) as the bias parameter are the coefficients wFjk (j=1 to m, k=0 to Z) Each of the 'm' values transformed by the linear mapping defined by is substituted into the activation function f. As a result, the value of the intermediate layer node is determined. Also, the output variable y(1) is determined by substituting each of the values obtained by transforming the node values of the intermediate layer by the linear mapping defined by the coefficient wSij (i=1) into the activation function g. Here, the output variable y(1) is a value indicating the knocking intensity NS. In this embodiment, an example of the activation function f is the ReLU function. Also, an example of the activation function g is the sigmoid function. Therefore, the output variable y(1) is calculated as a value between 0 and 1 inclusive. Note that the processing of steps S14 and S15 is the calculation processing.

In this embodiment, the map M defined by the map data 94A is generated, for example, as follows. First, an internal combustion engine 10 for testing is prepared. This internal combustion engine 10 for testing has a pressure sensor that detects the pressure in the cylinder 11 . Therefore, in this test internal combustion engine 10, the actual knocking intensity NS can be measured based on the pressure fluctuation detected by the pressure sensor. Then, the internal combustion engine 10 is driven under various conditions including a first state in which the load of the test internal combustion engine 10 is small and a second state in which the load is large. At this time, the first to eleventh strengths VA1 to VA11 are obtained in the same manner as the vehicle 100 described above, and the knocking strength NS is obtained by the pressure sensor. Further, according to steps S11 to S14, input variables x(1) to x(Z) as training data are generated based on the first intensity VA1 to eleventh intensity VA11. Further, the knocking intensity NS is set as an output variable y(1) as teacher data. The input variables x(1) to x(Z) as training data and the output variable y(1) as teacher data are input to the map M, and the map M is learned by machine learning.

In step S21, the CPU 91 determines whether or not the output variable y(1) is less than a predetermined threshold. In step S21, when the CPU 91 determines that the output variable y(1) is less than the threshold value (S21: YES), the process proceeds to step S31.

In step S31, the CPU 91 determines that knocking of the internal combustion engine 10 has not occurred. After that, the CPU 91 advances the process to step S32. In step S32, the CPU 91 calculates the reference ignition timing based on the engine speed NE and the like. Then, the CPU 91 sets the reference ignition timing as the current ignition timing. Thereafter, the CPU 91 terminates the current ignition timing control, and advances the process to step S11 again.

On the other hand, in step S21, CPU91 advances a process to step S41, when determining with output variable y (1) being more than a threshold value (S21:NO).
In step S41, the CPU 91 determines that the internal combustion engine 10 is knocking. After that, the CPU 91 advances the process to step S42. In step S42, the CPU 91 calculates the reference ignition timing based on the engine speed NE and the like. Then, the CPU 91 sets a timing retarded by a predetermined value from the reference ignition timing as the current ignition timing. Thereafter, the CPU 91 terminates the current ignition timing control, and advances the process to step S11 again.

<Action of this embodiment>
In this embodiment, the vibration V generated in the internal combustion engine 10 is detected by the knocking sensor 74 . By inputting a value based on the vibration V as an input variable to the learned map M, an output variable indicating the knocking intensity NS can be obtained.

<Effects of this embodiment>
(1) Suppose that only the normalization process among the correction process and the normalization process is executed in calculating the knocking intensity NS. In this case, for example, as indicated by the chain double-dashed line in FIG. 4, the fourth normalized strength VC4, which is the minimum value of the first normalized strength VC1 to the eleventh normalized strength VC11, is greater than zero. As described above, the sum of the first normalized strength VC1 to the eleventh normalized strength VC11 in the normalization process is one. Therefore, due to the fact that the fourth normalized strength VC4, which is the minimum value of the first normalized strength VC1 to the eleventh normalized strength VC11, becomes a large value, the first normalized strength VC1 to the eleventh normalized strength VC11 The seventh normalized intensity VC7, which is the maximum value of , becomes a relatively small value. As a result, the difference in magnitude between the first normalized intensity VC1 to the eleventh normalized intensity VC11 tends to be small.

On the other hand, in the present embodiment, by executing the correction process and the normalization process, for example, as indicated by the solid line in FIG. , the fourth normalized intensity VC4 becomes zero. Therefore, the seventh normalized strength VC7, which is the maximum value of the first normalized strength VC1 to the eleventh normalized strength VC11, has a relatively large value. That is, the difference between the first normalized strength VC1 to the eleventh normalized strength VC11 increases. Therefore, for example, even if the vibration V detected by the knocking sensor 74 includes noise caused by factors other than knocking in addition to the vibration caused by knocking, the difference between the vibration caused by knocking and the noise is conspicuously visible. As a result, due to noise, the knocking intensity NS may be estimated to be small even though the vibration caused by knocking is large, or the knocking intensity NS may be estimated to be large even though the vibration caused by knocking is small. is suppressed. As a result, the accuracy of estimating the knocking intensity NS can be improved.

(2) The magnitude of the vibration V generated in the internal combustion engine 10 by the knocking sensor 74 may vary due to manufacturing errors of the knocking sensor 74 or the like. As a result, the values of the first intensity VA1 to the eleventh intensity VA11 may increase, and the values of the first intensity VA1 to the eleventh intensity VA11 may decrease. In this regard, in the present embodiment, in the correction process, all of the first intensity VA1 to the eleventh intensity VA11 are offset so that the smallest value among the first intensity VA1 to the eleventh intensity VA11 becomes zero. 1st correction strength VB1 to 11th correction strength VB11 are calculated. As a result, even if the magnitude of the vibration V varies due to manufacturing errors of the knocking sensor 74 or the like, variations in the values of the first to eleventh correction strengths VB1 to VB11 due to the variations described above can be suppressed.

(3) In this embodiment, non-negative matrix factorization is used to obtain input variables to the mapping M for estimating the knocking intensity NS. Therefore, it is expected that the calculation load of the CPU 91 will be reduced as compared with the case where the input variables are obtained by using the fast Fourier transform, for example.

<Change example>
This embodiment can be implemented with the following modifications. This embodiment and the following modified examples can be implemented in combination with each other within a technically consistent range.

- In the above embodiment, the normalization process may be changed. For example, in step S12, the CPU 91 multiplies the first correction strength VB1 to the eleventh correction strength VB11 by the same coefficient, so that each value of the first normalization strength VC1 to the eleventh normalization strength VC11 is 0 to 1. may be converted to a value within the range of Also in this case, the minimum value of the first normalized intensity VC1 to the eleventh normalized intensity VC11 becomes zero by executing the correction process and the normalization process. Therefore, for example, even if the maximum values of the first normalized strength VC1 to the eleventh normalized strength VC11 are the same, the difference between the first normalized strength VC1 to the eleventh normalized strength VC11 becomes large.

- In the above embodiment, the acquisition process may be changed. For example, in step S13, the CPU 91 may acquire the values of the first normalized strength VC1 to the eleventh normalized strength VC11 as variables to be used as input variables without using non-negative matrix factorization. Even in this case, the map M can be used to estimate the knocking intensity NS.

- In the above embodiment, the input variables of the mapping M may be changed. For example, as the input variables of the map M, in addition to the values indicating the intensity of frequencies in a plurality of bands based on the vibration V, a value based on the engine rotation speed NE may be added. In this case, the accuracy of estimating the knocking intensity NS can be improved by taking into consideration the operating state of the internal combustion engine 10 . That is, as the input variables for the mapping M, in addition to variables indicating the intensity of frequencies in a plurality of bands based on the vibration V, other variables can be used.

- In the above-described embodiment, the number of values obtained as the intensity of frequencies in a plurality of bands may be changed. As a specific example, a value indicating the strength of vibration for each frequency in 2 to 10 bands may be obtained by bandpass filtering, or a value indicating the strength for each frequency in 12 or more bands may be obtained. may

- In the above embodiment, the configuration for acquiring the intensity of frequencies in a plurality of bands may be changed. For example, CPU91 may acquire the intensity|strength of the frequency of several bands by carrying out the fast Fourier transform of the vibration V. FIG. Even with this configuration, when only the normalization process is executed out of the correction process and the normalization process for the frequency intensities of a plurality of bands, the minimum value indicating the frequency intensities of the plurality of bands converted by the normalization process is Values can deviate from zero. Therefore, it is more effective to perform the correction process and the normalization process than to perform only the normalization process among the correction process and the normalization process.

- In the above embodiment, the activation function of the mapping M is an example, and is not limited to the example of the above embodiment. For example, as the activation function f of the map M, a sigmoid function or the like may be used.
- In the above-described embodiment, a neural network having one intermediate layer was exemplified as a neural network, but the number of intermediate layers may be two or more.

- In the above-described embodiment, the fully-connected forward propagation neural network was exemplified as the neural network, but the neural network is not limited to this. For example, a recursive neural network may be employed as the neural network.

- In the above embodiment, the function approximator as the mapping M is not limited to a neural network. For example, it may be a regression equation without an intermediate layer. That is, the map M used in step S15 is an example, and various configurations can be adopted as described above. Therefore, the Z values obtained by non-negative matrix factorization in step S13 can be adopted as input variables for another mapping M.

- In the above-described embodiment, the execution device is not limited to one that includes the CPU 91 and the ROM 93 and executes software processing. As a specific example, a dedicated hardware circuit such as an ASIC may be provided that performs hardware processing at least part of what is software processed in the above embodiments. That is, the execution device may have any one of the following configurations (a) to (c). (a) A processing device that executes all of the above processes according to a program, and a program storage device such as a ROM that stores the program. (b) A processing device and a program storage device for executing part of the above processing according to a program, and a dedicated hardware circuit for executing the remaining processing. (c) provide dedicated hardware circuitry to perform all of the above processing; Here, there may be a plurality of software execution devices provided with a processing device and a program storage device, or a plurality of dedicated hardware circuits.

- In the above embodiment, other configurations of the vehicle 100 may be changed.
For example, vehicle 100 may include a motor generator in addition to internal combustion engine 10 as a drive source.

NS... Knocking intensity V... Vibration 10... Internal combustion engine 11... Cylinder 12... Crankshaft 21... Intake passage 22... Throttle valve 23... Fuel injection valve 24... Ignition device 26... Exhaust passage 30... Automatic transmission 40... Differential mechanism 50 ... Drive wheel 71 ... Crank angle sensor 72 ... Accelerator operation amount sensor 73 ... Vehicle speed sensor 74 ... Knocking sensor 90 ... Control device 91 ... CPU
92... Peripheral circuit 93... ROM
94 Storage device 94A Mapping data 95 Bus 100 Vehicle

Claims (1)

Applied to a vehicle equipped with an internal combustion engine and a knocking sensor for detecting vibration generated in the internal combustion engine, and estimating the knocking intensity of the internal combustion engine based on the intensity of frequencies in a plurality of bands of the vibration detected by the knocking sensor A knocking estimation device that
comprising an execution device and a storage device,
The storage device stores mapping data defining a mapping for outputting an output variable indicating the knocking intensity when an input variable is input,
The execution device is
A correction process of offsetting the values indicating the intensity of the frequencies in the plurality of bands so that the smallest value among the values indicating the intensity of the frequencies in the plurality of bands is zero;
a normalization process of multiplying the values indicating the intensity of the frequencies of the plurality of bands offset by the correction process by the same coefficient to convert them into values within a predetermined range;
an acquisition process, which is a process of acquiring, as the input variable, a value indicating the frequency intensity of the plurality of bands converted by the normalization process;
a calculation process of outputting the value of the output variable by inputting the input variable acquired by the acquisition process into a mapping.

Images