JP2018178810A - Knocking controller, knocking adaptation method and knocking adaptation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a knocking controller, a knocking adaptation method and a knocking adaptation program, capable of improving knocking detection accuracy.SOLUTION: A knocking controller includes an acquisition unit, an image generation unit, a model generation unit and a determination unit. The acquisition unit is configured to acquire a knocking signal whose waveform changes depending on knocking occurring in an internal combustion engine. The image generation unit is configured to generate a knocking image by converting the knocking signal acquired by the acquisition unit into a frequency region for each fixed period. The model generation unit is configured to generate a knocking model indicating features of knocking, based on the knocking image generated by the image generation unit. The determination unit is configured to determine presence/absence of knocking, based on the knocking model generated by the model generation unit.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a knock control device, a knock matching method and a knock matching program.

  BACKGROUND Conventionally, there is a knock control device that detects knocking that occurs in an internal combustion engine such as an engine of a vehicle and controls the ignition timing of the engine based on the detected state of knocking. The knock control device determines the presence or absence of knocking based on a knock model obtained by modeling characteristics of the waveform shape of a knock signal indicating knocking (see, for example, Patent Document 1).

JP, 2009-203842, A

  However, in the prior art, since the process of creating the knock model is left to the worker, the matching accuracy of the generated knock model may be reduced. Therefore, when the presence or absence of knocking is determined using such a knock model, there is a possibility that the detection accuracy of knocking may decrease.

  The present invention has been made in view of the above, and it is an object of the present invention to provide a knock control device, a knock adaptation method and a knock adaptation program which can improve the detection accuracy of knocking.

  In order to solve the problems described above and achieve the object, a knock control device according to the present invention includes an acquisition unit, an image generation unit, a model generation unit, and a determination unit. The acquisition unit acquires a knock signal whose waveform changes in accordance with knocking occurring in the internal combustion engine. The image generation unit generates a knock image by converting the knock signal acquired by the acquisition unit into a frequency domain at predetermined intervals. The model generation unit generates a knock model indicating the characteristics of the knocking based on the knock image generated by the image generation unit. The determination unit determines the presence or absence of the knocking based on the knock model generated by the model generation unit.

  According to the present invention, the detection accuracy of knocking can be improved.

FIG. 1 is a diagram showing an outline of processing performed by the knock control device according to the embodiment. FIG. 2 is a block diagram showing the configuration of the knock control device according to the embodiment. FIG. 3 is a diagram showing the processing content of the model generation unit. FIG. 4 is a diagram showing the processing content of the weight value determination processing. FIG. 5 is a diagram showing the processing content of the first processing. FIG. 6 is a diagram showing the process content of the second process. FIG. 7 is an explanatory diagram of discriminator information. FIG. 8 is a flowchart showing the procedure of the adaptation process performed by the knock control device according to the embodiment. FIG. 9 is a flowchart showing the procedure of the determination process performed by the knock control device according to the embodiment. FIG. 10 is a block diagram showing the configuration of a knock control device according to a modification of the embodiment. FIG. 11 is an explanatory diagram of discriminator information according to a modification. FIG. 12 is a block diagram showing the configuration of a knock control device according to a modification of the embodiment.

  Hereinafter, embodiments of the knock control device, the knock adaptation method, and the knock adaptation program disclosed in the present application will be described in detail with reference to the attached drawings. The present invention is not limited by this embodiment.

  First, an outline of processing performed by the knock control device according to the embodiment will be described using FIG. 1. FIG. 1 is a diagram showing an outline of processing performed by knock control device 1 according to the embodiment. FIG. 1 shows an internal combustion engine 10 and a knock sensor 11 that detects knocking of the internal combustion engine 10.

  The internal combustion engine 10 is an engine such as a vehicle, for example, and is a device that obtains power by combustion gas generated by burning a fuel such as gasoline in a cylinder. Knock sensor 11 is a sensor that detects a vibration generated by driving internal combustion engine 10, and is a sensor that outputs a knock signal whose waveform changes according to the knocking generated in internal combustion engine 10. The term "knocking" refers to, for example, a state in which an abnormal vibration or an abnormal sound is produced by the abnormal burns of the fuel remaining in the cylinder after the combustion at the next ignition timing.

  Further, FIG. 1 shows knock images Pa and Pb in which the knock signal is converted into a frequency domain for each fixed cycle (for example, for each cycle). The knock image Pa is used for knock model generation processing of the knock control device 1, and is, for example, an image of a sensor signal obtained at the time of an operation test of the internal combustion engine 10.

  The knock image Pb is used in the knocking determination process of the knock control device 1 and is, for example, an image of a sensor signal detected from the internal combustion engine 10 when the vehicle actually travels on a public road or the like. In the following, the knock image Pa may be described as a learning image Pa, and the knock image Pb may be described as a determination image Pb.

  The knock control device 1 according to the embodiment uses the learning image Pa to generate a knock model indicating the characteristics of knocking, and uses the knock model to determine the presence or absence of knocking in the determination image Pb.

  Here, a conventional knock control device will be described. In the conventional knock control device, the process of creating a knock model has been entrusted to workers. Specifically, conventionally, a worker extracts the feature of the waveform shape of the knock signal from the knock image and uses it for creating the knock model.

  For this reason, conventionally, there is a possibility that the fitting accuracy of the knock model may be lowered because the characteristics of knocking extracted vary depending on the degree of skill of the operator or the like. Therefore, conventionally, when the presence or absence of knocking is determined using such a knock model, there is a possibility that the detection accuracy of knocking may decrease. In addition, since the characteristics of knocking must be extracted from each knock image, the number of steps of the worker increases.

  Therefore, the knock control device 1 according to the embodiment automatically generates a knock model from the knock image Pa (the learning image Pa) by, for example, machine learning.

  Specifically, as shown in FIG. 1, the knock control device 1 according to the embodiment first generates a knock model indicating knocking characteristics based on the knock image Pa in the learning phase (step S1). The generation of the knock model can be performed using, for example, a classifier having an algorithm of a convolutional neural network, which will be described later with reference to FIGS. 3 to 6.

  Subsequently, the knock control device 1 according to the embodiment proceeds to the identification phase after generating the knock model, and determines the presence or absence of knocking based on the knock model generated in the learning phase (step S2). Specifically, as shown in FIG. 1, the knock control device 1 first applies the generated knock model to the classifier. Subsequently, knock control device 1 determines the presence or absence of knocking in determination image Pb using such a classifier. The algorithm of the classifier used in the discrimination phase is, for example, a convolutional neural network, and uses the same one as the algorithm used in the learning phase (applying the knock model).

  That is, since the knock control device 1 according to the embodiment automatically generates the knock model by machine learning or the like, the variation in the knock model for each worker as in the prior art is eliminated, so the fitting accuracy of the knock model is improved. be able to. That is, according to the knock control device 1 according to the embodiment, the detection accuracy of knocking can be improved.

  Furthermore, since the knock control device 1 according to the embodiment extracts the characteristic of knocking by machine learning, it is possible to reduce the number of man-hours of the worker. That is, the efficiency of the fitting operation of the knock model can be achieved.

  The knock control device 1 according to the embodiment generates a weight value ω of a classifier using a convolutional neural network as a knock model, and such a point will be described later with reference to FIGS. 3 to 6.

  Further, the knock control device 1 according to the embodiment can generate a knock model for each cylinder of the internal combustion engine 10 using, for example, a cylinder pressure sensor (CPS). Regarding this point, FIGS. Will be described later.

  Next, with reference to FIG. 2, the configuration of the knock control device 1 according to the embodiment will be described in detail. FIG. 2 is a block diagram showing the configuration of knock control device 1 according to the embodiment.

  As shown in FIG. 2, the knock control device 1 according to the embodiment is connected to the knock sensor 11 and the control device 100. Although knock control device 1 and control device 100 are separately configured as two devices in FIG. 2, knock control device 1 and control device 100 are integrally configured as one device. It may be done. First, the configuration other than the knock control device 1 will be described.

  Knock sensor 11 is attached to internal combustion engine 10, and generates a knock signal based on, for example, vibrations generated in internal combustion engine 10. The knock signal is a signal whose waveform changes in accordance with knocking generated in the internal combustion engine 10. Knock sensor 11 outputs the generated knock signal to knock control device 1.

  For example, a microphone can be used as the knock sensor 11. In such a case, the microphone generates a knock sound generated as knocking as a knock signal. As a result, since a knocking model can be generated by knocking noise, knocking determination that captures human hearing can be made.

  Control device 100 calculates an engine rotational speed and an air load factor (filling efficiency) of internal combustion engine 10 based on a sensor (not shown) attached to internal combustion engine 10, and outputs the information to knock control device 1. . In FIG. 2, although the engine speed and the air load factor (filling efficiency) are output from the control device 100, for example, the detection signal of the above sensor is directly output to the knock control device 1, and the engine rotation is performed by the knock control device 1. The speed or the air load factor (filling efficiency) may be calculated.

  The control device 100 determines the ignition timing of the internal combustion engine 10 from the next time on the basis of the determination result of the presence or absence of knocking notified from the knock control device 1, and instructs the internal combustion engine 10 to ignite according to the ignition timing. Output an ignition signal.

  Next, the knock control device 1 according to the embodiment will be described. Knock control device 1 includes control unit 2 and storage unit 3. The control unit 2 includes an acquisition unit 21, an image generation unit 22, a model generation unit 23, and a determination unit 24. The storage unit 3 stores image information 31 and identifier information 32.

  Here, the knock control device 1 may be, for example, a computer having a central processing unit (CPU), a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), an input / output port, and various circuits. including.

  The CPU of the computer functions as an acquisition unit 21, an image generation unit 22, a model generation unit 23, and a determination unit 24 of the control unit 2 by reading and executing a program stored in the ROM, for example.

  In addition, at least one or all of the acquisition unit 21, the image generation unit 22, the model generation unit 23, and the determination unit 24 of the control unit 2 may be implemented as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a GPU It can also be configured by hardware such as a Graphics Processing Unit.

  The storage unit 3 corresponds to, for example, a RAM or an HDD. The RAM or HDD can store image information 31, identifier information 32, information of various programs, and the like. The knock control device 1 may acquire the above-described program and various information via another computer connected via a wired or wireless network or a portable recording medium.

  The control unit 2 uses the knock image Pa (the learning image Pa) to generate a knock model indicating the characteristics of knocking, and uses the knock model to determine the presence or absence of knocking in the knock image Pb (determination image Pb). Do.

  The acquisition unit 21 acquires a knock signal whose waveform changes in accordance with knocking generated in the internal combustion engine 10. The acquisition unit 21 outputs the acquired knock signal to the image generation unit 22.

  The image generation unit 22 generates knock images Pa and Pb by converting the knock signal acquired by the acquisition unit 21 into a frequency domain at predetermined intervals. Specifically, the image generation unit 22 generates a learning image Pa and a determination image Pb by wavelet-transforming the knock signal into the frequency domain for each cycle of the internal combustion engine 10. The learning image Pa and the determination image Pb are, for example, color images represented by three dimensions of RGB values.

  Further, the image generation unit 22 is not limited to the wavelet transform, and may use, for example, a short time Fourier transform or a mel frequency cepstrum coefficient, as long as it can convert to a frequency domain at a constant cycle.

  Further, the image generation unit 22 first generates a learning image Pa used by the model generation unit 23 described later among the learning image Pa and the determination image Pb. Then, after the model generation unit 23 generates the knock model, the image generation unit 22 generates the determination image Pb.

  The image generation unit 22 may generate the learning image Pa using, for example, a knock signal due to vibration when the internal combustion engine 10 is experimentally driven in a laboratory. At this time, a knocking signal when the internal combustion engine 10 intentionally knocks may be used. Thereby, the accuracy of the knock model generated by the model generation unit 23 described later can be enhanced.

  In addition, the image generation unit 22 may use a knock signal detected from the internal combustion engine 10 in which the vehicle is actually traveling. The image generation unit 22 stores the generated learning image Pa as the image information 31 in the storage unit 3 and outputs the determination image Pb to the determination unit 24.

  The model generation unit 23 generates a knock model indicating the characteristics of knocking based on the knock image Pa (the learning image Pa) generated by the image generation unit 22. Specifically, the model generation unit 23 generates a weight value ω (see FIG. 5) of a classifier using a convolutional neural network as a knock model. Here, the processing content of the model generation unit 23 will be specifically described using FIG. 3 and FIG. 4.

  FIG. 3 is a diagram showing the processing content of the model generation unit 23. FIG. 4 is a diagram showing the process of determining the weight value ω of the classifier. As shown in FIG. 3, when a plurality of (for example, several thousand) learning images Pa are input, the model generation unit 23 performs an averaging process and a weight value determination process, thereby a classifier using a convolutional neural network. Update the weight value ω of

  In addition, the model generation unit 23 may acquire and process all the learning images Pa of a plurality (for example, several thousand sheets) at a time, for example, acquiring a plurality of learning images Pa little by little in repetitive loop (mini-batch processing May).

  First, the averaging process will be described. The averaging process is a process of generating an average value image obtained by averaging the values of each pixel of the learning image Pa. Specifically, first, the model generation unit 23 sets an arbitrary pixel of the learning image Pa as a central pixel, and sets a small area (for example, a 3 × 3 area) including peripheral pixels in the central pixel. The model generation unit 23 plots the average value of the plurality of pixels included in the set small area as the pixel value of the central pixel on the average value image.

  Then, the model generation unit 23 generates an average value image by similarly calculating an average value for all the pixels of the learning image Pa. Then, the model generation unit 23 performs subtraction normalization on the average value image and passes it to weight value determination processing in accordance with processing in a later stage. Note that normalization of the average value image is not limited to subtraction normalization, and other normalization patterns such as division normalization and whitening may be used.

  Next, weight value determination processing will be described. In the weight value determination process, when the normalized average value image is input, the process is performed in the order of initial value setting process, forward propagation process, loss error calculation process, and back propagation process, and the determination unit 24 uses It is a process of updating the weight value.

  Specifically, the model generation unit 23 performs a process of calculating a loss error using a classifier of a convolutional neural network and determining a weight value ω that minimizes the loss error. The algorithm of the classifier is not limited to the convolutional neural network, and may be another machine learning algorithm such as a recurrent neural network. Here, the weight value determination process performed by the model generation unit 23 will be specifically described using FIGS. 4 to 6.

  FIG. 4 is a diagram showing weight value determination processing. FIG. 4 shows the flow of weight value determination processing performed by the model generation unit 23. As shown in FIG. 4, the weight value determination process is roughly divided into a first process and a second process.

  Also, as shown in FIG. 4, when the model generation unit 23 receives the average value image, first, before the first processing, the weight value ω of the classifier (see FIG. 5) and the bias b (see FIG. 5) An initial value setting process (see FIG. 3) is performed to initialize the value of.

  The model generation unit 23 may set a predetermined value as an initial value in the initial value setting process, or may determine the initial value by performing, for example, pre-training or fine tuning.

  The model generation unit 23 performs the first process after the initial value setting process. The first process is a process in which a set of a convolution layer, an activation function, and a pooling layer is one layer, and a plurality of layers (three layers in FIG. 4) are repeated. The details of the first process will be described later with reference to FIG.

  Subsequently, the model generation unit 23 performs a second process after the first process. The second process is a process including an entire bonding layer, an activation function, a total bonding layer, and a soft max function. The details of the second process will be described later with reference to FIG.

  As shown in FIG. 4, the model generation unit 23 first calculates the loss error by performing processing in order of the first processing and the second processing by forward propagation. Subsequently, based on the loss error calculated by forward propagation, the model generation unit 23 determines the weight value ω at which the loss error is minimized by performing the back propagation process.

  Here, the first process and the second process will be specifically described with reference to FIGS. 5 and 6. FIG. 5 is a diagram showing the processing content of the first processing. FIG. 6 is a diagram showing the process content of the second process.

  First, the first process will be described with reference to FIG. FIG. 5 shows the processing content of the first layer in the first processing. Further, in FIG. 5, the number of pixels of the average value image is 256 × 256. The number of pixels of the average value image, the convolutional image described later, and the pooling image shown in FIG. 5 is an example, and may be an arbitrary number of pixels.

  As shown in FIG. 5, the model generation unit 23 generates a convolutional image in which the number of pixels is compressed by inputting an average value image in order of a convolutional layer and an activation function. Specifically, the model generation unit 23 compresses the average value image using a convolution filter of a predetermined number of pixels (for example, 2 × 2), the weight value ω, and the bias b.

  For example, as illustrated in FIG. 5, the model generation unit 23 extracts feature amounts of predetermined four pixels x1 to x4 in the average value image by using a convolution filter, and weight values ω for the feature amounts of each pixel x1 to x4. Accumulate. Subsequently, the model generation unit 23 sums up the respective values of the pixels x1 to x4 on which the weight value ω is integrated, and adds the bias b to the summed value.

  The model generation unit 23 substitutes the value to which the bias b is added into the activation function, and sets the value obtained from the activation function as the value of the pixel y1 of the convolutional image (hereinafter sometimes referred to as a convolution value) decide. The activation function may be, for example, a ramp function, but is not limited thereto. For example, a sigmoid function, a max out function, or a drop out function can be used.

  Subsequently, the model generation unit 23 generates the pooling image in which the number of pixels is further compressed by inputting the generated convolutional image into the pooling layer. Specifically, the model generation unit 23 generates a pooling image by scanning the entire convolutional image using a pooling filter with a predetermined number of pixels (for example, 2 × 2).

  For example, as illustrated in FIG. 5, the model generation unit 23 extracts predetermined four pixels y1 to y4 in the convolutional image based on the pooling filter, and pulls out the largest value of the convolution value among the extracted four pixels. It is determined as the value of the pixel z1 in

  Although the model generation unit 23 sets the maximum value of the convolution value among the plurality of extracted pixels (four pixels in FIG. 5) as the value of the pixel z1 in the pooling image, for example, The average value may be the value of the pixel z1.

  Thereby, for example, the influence of the shift of the detection value of the knock sensor 11 or the shift of the frequency due to the noise superposition can be removed in the pooling image. The model generation unit 23 inputs the generated pooling image to a second convolutional layer (see FIG. 4) which is the next layer.

  Next, the second process will be described with reference to FIG. FIG. 6 is a diagram showing the second process. FIG. 6 shows a pooling image generated by the pooling layer of the last layer (the third layer in FIG. 4).

  As shown in FIG. 6, the model generation unit 23 processes the pooling image in the order of the total bonding layer, the activation function, and the total bonding layer. The entire bonding layer can use, for example, linear transformation such as affine transformation. Further, as the activation function, a ramp function, a sigmoid function, a max out function, a drop out function or the like can be used.

  The model generation unit 23 generates a combined image of a predetermined number of pixels (2 × 2 in FIG. 6) by combining the pooling images using the entire combined layer and the activation function. The number of pixels of the combined image can be arbitrarily set in accordance with the soft max function.

  Subsequently, as shown in FIG. 6, the model generation unit 23 inputs the generated combined image to the soft max function. Thus, the values of the pixels A1 to A4 of the combined image indicate the probability that the pixels A1 to A4 exhibit knocking (or the probability that they are not pixels indicating knocking), and the sum of the values of the pixels A1 to A4 is It becomes 1.

  Then, the model generation unit 23 calculates the loss error based on the values of the respective pixels A1 to A4. For example, when a learning image Pa including a feature of knocking is input, it is assumed that the feature of knocking appears in the pixel A1 shown in FIG.

  In such a case, the value of the pixel A1 should be “1” and the values of the other pixels A2 to A4 should be “0” originally. However, as shown in FIG. 6, it is assumed that the value of the pixel A1 is “0.64”.

  In such a case, the model generation unit 23 calculates “0.36” which is a value obtained by subtracting 0.64 from 1 as the loss error. Then, the model generation unit 23 performs back propagation processing based on the calculated loss error.

  Specifically, the model generation unit 23 finally learns the bias b and the weight value ω that minimize the loss error by repeatedly learning the combination of the bias b and the weight value ω by back propagation processing. Determine as a result. As the optimization function of the weight value ω in the back propagation process, for example, Adam method, stochastic gradient descent method (SGD), Momentum, Adagrad or the like can be used.

  As described above, by automatically creating a knock model using a convolutional neural network, it is possible to reduce the number of man-hours for the operator, and it is also possible to capture knocking features (for example, jump patterns etc.) that can not be covered by people. Become.

  The model generation unit 23 stores the determined bias b and weight value ω in the storage unit 3 as the discriminator information 32. Here, the discriminator information 32 will be described using FIG. 7.

  FIG. 7 is an explanatory diagram of the discriminator information 32. As shown in FIG. As shown in FIG. 7, the discriminator information 32 includes items such as “engine type”, “sensor type”, “engine rotational speed”, “air load factor (filling efficiency)”, “weight value”, and “bias”. Including.

  The “engine type” is information on the type of the internal combustion engine 10, and is classified, for example, by vehicle type. “Sensor type” indicates the type of knock sensor 11. Specifically, the “knock sensor” is a sensor that outputs the vibration itself of the internal combustion engine 10 as a knock signal. The “microphone” is a sensor that outputs vibration noise or knocking noise of the internal combustion engine 10 as a knock signal. “CPS” is a sensor that outputs the combustion pressure in the internal combustion engine 10 as a knock signal.

  “Engine rotational speed” indicates the rotational speed of the internal combustion engine 10. “Air load factor (charging efficiency)” indicates the air charging efficiency of the internal combustion engine 10. “Weight value” indicates the weight value ω of the classifier used by the determination unit 24. The “bias” indicates the bias b of the discriminator used by the determination unit 24.

  That is, the model generation unit 23 determines the engine rotational speed, the weight value ω for each air load rate (filling efficiency), and the bias b for each sensor type. Thereby, the determination accuracy of knocking can be improved by the determination unit 24 described later.

  Returning to FIG. 2, the determination unit 24 will be described. The determination unit 24 determines the presence or absence of knocking based on the knock model generated by the model generation unit 23. Specifically, the determination unit 24 performs forward propagation processing on the knock image Pb (the determination image Pb) generated by the image generation unit 22, and calculates the probability indicating knocking.

  The determination unit 24 calculates the probability of knocking by using the weight value ω and the bias b of the identifier information 32 in the forward propagation process (the first process and the second process). When the probability indicating knocking is equal to or greater than a predetermined threshold value, determination unit 24 determines that knocking has occurred, and notifies control device 100 of the determination result. Thus, the control device 100 can set an ignition timing in consideration of knocking in the subsequent cycles.

  Next, with reference to FIG. 8, a processing procedure of the adaptation processing performed by the knock control device 1 according to the embodiment will be described. FIG. 8 is a flowchart showing the procedure of the adaptation process performed by knock control device 1 according to the embodiment.

  As shown in FIG. 8, first, the acquisition unit 21 acquires a knock signal whose waveform changes according to knocking generated in the internal combustion engine 10 (step S101). Subsequently, the image generation unit 22 generates the knock image Pa (the learning image Pa) by converting the knock signal acquired by the acquisition unit 21 into the frequency domain at predetermined intervals (step S102).

  Subsequently, the model generation unit 23 generates a knock model indicating the characteristics of knocking based on the learning image Pa generated by the image generation unit 22 (step S103).

  Subsequently, the model generation unit 23 is a weight value included in the classifier information 32 stored in the storage unit 3 based on the generated knock model, and updates the weight value of the classifier used by the determination unit 24 (see FIG. Step S104), the process ends.

  Next, with reference to FIG. 9, the procedure of the determination process performed by the knock control device 1 according to the embodiment will be described. FIG. 9 is a flowchart showing the procedure of the determination process performed by knock control device 1 according to the embodiment.

  As shown in FIG. 9, first, the acquisition unit 21 acquires a knock signal whose waveform changes in accordance with knocking generated in the internal combustion engine 10 (step S201). Subsequently, the image generation unit 22 generates a knock image Pb (determination image Pb) by converting the knock signal acquired by the acquisition unit 21 into a frequency domain at predetermined intervals (step S202).

  Subsequently, the determination unit 24 determines the presence or absence of knocking based on the knock model generated by the model generation unit 23 (step S203). Subsequently, the determination unit 24 notifies the control device 100 of the determination result (step S204), and ends the process.

  As described above, the knock control device 1 according to the embodiment includes the acquisition unit 21, the image generation unit 22, the model generation unit 23, and the determination unit 24. The acquisition unit 21 acquires a knock signal whose waveform changes in accordance with knocking generated in the internal combustion engine 10. The image generation unit 22 generates knock images Pa and Pb by converting the knock signal acquired by the acquisition unit 21 into a frequency domain at predetermined intervals. The model generation unit 23 generates a knock model indicating knocking characteristics based on the knock images Pa and Pb generated by the image generation unit 22. The determination unit 24 determines the presence or absence of knocking based on the knock model generated by the model generation unit 23. As a result, since the fitting accuracy of the knock model can be improved, the knocking detection accuracy can be improved.

  In the above-described embodiment, it is determined whether knocking has occurred in the internal combustion engine 10. However, for example, it may be determined in which cylinder of the internal combustion engine 10 knocking has occurred. This point will be described using FIG. 10 and FIG.

  FIG. 10 is a block diagram showing the configuration of a knock control device 1 according to a modification of the embodiment. FIG. 11 is an explanatory diagram of the discriminator information 32 according to the modification. In the following, the same reference numerals are given to the same configuration as the above embodiment, and the description is omitted.

  As shown in FIG. 10, four CPSs 11a to 11d are provided in each cylinder of the internal combustion engine 10 (for example, four cylinders), and measure the combustion pressure in the cylinders. The acquisition unit 21 acquires a plurality of (four in FIG. 10) knock signals for each cylinder based on the detection values of the CPS sensors 11a to 11d.

  The image generation unit 22 generates a plurality of (four in FIG. 10) knock images Pa and Pb by converting each of the plurality of knock signals into a frequency domain at predetermined intervals. The model generation unit 23 generates a plurality of knock models based on the plurality of knock images Pa and Pb, and stores the plurality of knock models as the classifier information 32.

  As shown in FIG. 11, the identifier information 32 includes “engine type”, “cylinder ID”, “sensor ID”, “engine rotation speed”, “air load factor (filling efficiency)”, “weight value”, “ Include items such as “bias”. “Cylinder ID” is information for identifying each cylinder. “Sensor ID” is information for identifying the CPS sensors 11 a to 11 d provided in each cylinder. The items other than the “cylinder ID” and the “sensor ID” have been described above and thus will not be described.

  That is, the determination unit 24 determines the presence or absence of knocking for each cylinder of the internal combustion engine 10 based on a plurality of knock models. Thereby, it can be detected in which cylinder of the internal combustion engine 10 knocking has occurred. Further, the fuel consumption can be improved by the control device 100 controlling the ignition timing for each cylinder.

  Further, in the above-described embodiment, the image generation unit 22 generates the knock images Pa and Pb using the noise signal as it is, but for example, the noise signal is allowed to pass through a band pass filter and knocked using the signal after passing. Images Pa and Pb may be generated. This point will be described with reference to FIG.

  FIG. 12 is a block diagram showing the configuration of a knock control device 1 according to a modification of the embodiment. In the following, the same reference numerals are given to the same configuration as the above embodiment, and the description is omitted.

  As shown in FIG. 12, the control unit 2 of the knock control device 1 according to the modification further includes a band pass filter 25 (BPF 25). The BPF 25 passes only a specific knock frequency band among the knock signals input from the acquisition unit 21, and outputs the signal after the pass to the image generation unit 22.

  Specifically, in the BPF 25, for example, a knock frequency band (for example, 5 to 20 kHz) is set so as to attenuate only a vibration component at the normal time of the internal combustion engine 10 (see FIG. 2) among the knock signals.

  That is, the BPF 25 removes the vibration component of the noise signal in the normal state of the internal combustion engine 10 (see FIG. 2) as noise, and passes only the knock frequency band of the component corresponding to knocking.

  Thereby, the image generation unit 22 can generate the knock images Pa and Pb from the signal including only the component corresponding to the knocking. That is, it is possible to realize generation of a high-accuracy knock model by the model generation unit 23 in the latter stage and improvement in the determination accuracy of knocking by the determination unit 24.

  In the embodiment described above, the determination unit 24 performs only the knocking determination using the classifier of the convolutional neural network, but, for example, the weight value ω is updated again based on the loss error calculated at the time of the knocking determination. It is also good. That is, the determination unit 24 may always perform the determination process using the latest classifier by continuously learning the classifier (sequential learning or online learning).

  Further effects and modifications can be easily derived by those skilled in the art. Thus, the broader aspects of the invention are not limited to the specific details and representative embodiments represented and described above. Accordingly, various modifications may be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and their equivalents.

Reference Signs List 1 knock control device 2 control unit 3 storage unit 10 internal combustion engine 11 knock sensor 21 acquisition unit 22 image generation unit 23 model generation unit 24 determination unit 31 image information 32 identifier information 100 control device Pa, Pb knock image

Claims (7)

An acquisition unit for acquiring a knock signal whose waveform changes in accordance with knocking occurring in the internal combustion engine;
An image generation unit configured to generate a knock image by converting the knock signal acquired by the acquisition unit into a frequency domain at predetermined intervals;
A model generation unit that generates a knock model indicating the characteristics of the knocking based on the knock image generated by the image generation unit;
A determination unit that determines the presence or absence of the knocking based on the knock model generated by the model generation unit. The model generation unit
Generating a weight value of a classifier using a convolutional neural network as the knock model,
The determination unit is
The knock control device according to claim 1, wherein the presence or absence of the knocking is determined based on the identifier using the weight value. The acquisition unit
Acquiring the knock signal for each cylinder of the internal combustion engine;
The determination unit is
The knock control device according to claim 1, wherein the presence or absence of the knocking is determined for each of the cylinders. The acquisition unit
The knock control device according to any one of claims 1 to 3, wherein a knocking noise generated as a result of the knocking is acquired as the knocking signal. And a band pass filter for selectively passing a signal of a knock frequency band corresponding to the knocking among the knock signals acquired by the acquisition unit.
The image generation unit
The knock control device according to any one of claims 1 to 4, wherein the knock image is generated by converting the signal in the knock frequency band that has passed through the band pass filter. An acquiring step of acquiring a knock signal whose waveform changes in accordance with knocking generated in the internal combustion engine;
An image generation step of generating a knock image by converting the knock signal acquired in the acquisition step into a frequency domain at predetermined intervals;
A model generation step of generating a knock model indicating the characteristics of the knocking based on the knock image generated by the image generation step;
And K. updating the weight value of the classifier used to determine the knocking on the basis of the knock model generated in the model generation process. An acquisition procedure for acquiring a knock signal whose waveform changes in accordance with knocking occurring in an internal combustion engine;
An image generation procedure for generating a knock image by converting the knock signal acquired by the acquisition procedure into a frequency domain every fixed cycle;
A model generation procedure for generating a knock model indicating the characteristics of the knocking based on the knock image generated by the image generation procedure;
An update procedure for updating the weight value of the classifier used for the determination of the knocking based on the knock model generated by the model generation procedure.