Cylinder pressure virtual measurement method based on physical information neural network

Abstract

The invention discloses a cylinder pressure virtual measurement method based on a physical information neural network. The method comprises the following steps: acquiring a cylinder cover vibration signal and a cylinder pressure signal of a certain cylinder in a specific time period under different loads; constructing a physical information neural network, taking a cylinder cover vibration signal as an input, and taking a corresponding cylinder pressure signal as an output; determining a transfer function of the cylinder cover system under different loads as a physical constraint, and embedding the physical constraint into the neural network in a form of constructing a loss function; training a physical information neural network by using cylinder cover vibration signals and cylinder pressure signals of the cylinder in specific time periods under different loads, wherein the cylinder cover vibration signals and the cylinder pressure signals under different loads use physical constraints under corresponding loads; and inputting the cylinder cover vibration signal in the specific time period into the trained physical information neural network to obtain a corresponding cylinder pressure signal. The cylinder pressure is predicted through the cylinder cover vibration signal by using the neural network, and the transfer function of the cylinder cover system is introduced, so that the cylinder pressure can be accurately obtained.

Description

Technical Field

[0001] The invention relates to the technical field of engine state monitoring, and in particular to a cylinder pressure virtual measurement method based on physical information neural network. Background Art

[0002] The engine cylinder pressure is an important parameter in the combustion process in the cylinder. It reflects the combustion cycle changes in the cylinder, the combustion heat release rate, the intake and exhaust process, etc. It is a key indicator for evaluating the working process in the cylinder. By analyzing the cylinder pressure, we can not only judge the changes in the engine operating state, but also diagnose the abnormal conditions of the engine, such as abnormal combustion, misfire, etc. Therefore, cylinder pressure plays an important role in engine status monitoring and fault diagnosis.

[0003] Using neural network for cylinder pressure virtual measurement is an effective technical solution. At present, traditional neural network is relatively mature in both network theory and performance. Its outstanding advantages are strong nonlinear mapping ability and flexible network structure. The number of intermediate layers and the number of neurons in each layer of the network can be set arbitrarily according to the specific situation, and its performance varies with the difference of structure. However, there are the following defects when using traditional neural network for cylinder pressure virtual measurement: (1) Traditional neural networks are black box models, and their internal working mechanisms and decision-making processes are not explainable. For example, there is no clear theoretical basis for the nonlinear mapping process between neurons and the selection of weights.

[0004] (2) When the input signal of the neural network fluctuates, the network will have a deviation in the distribution of the input data and will have poor immunity to unexpected noise or abnormal signals. The network cannot accurately capture the input features, resulting in inaccurate output results.

[0005] Due to the above defects of traditional neural networks, they need to be further optimized when performing virtual measurement of cylinder pressure. Summary of the invention

[0006] The purpose of the present invention is to provide a virtual cylinder pressure measurement method based on a physical information neural network, combining data-driven neural networks with physical laws, so that the training process not only relies on limited data, but also ensures that it follows physical laws, so that the network model has clear interpretability and enhances the trust of the results. Due to the integration of physical laws, even if there are certain fluctuations or interference components in the data, physical laws can help the model provide robust results and improve the robustness of the model.

[0007] To achieve the above object, according to a first aspect of the present invention, a cylinder pressure virtual measurement method based on a physical information neural network is provided, the method comprising: Under different loads, obtain the cylinder head vibration signal and cylinder pressure signal of a certain cylinder at a specific time period; Constructing a physical information neural network, which takes the cylinder head vibration signal of the cylinder in a specific period as input and the corresponding cylinder pressure signal as output; Determine the transfer function of the cylinder head system under different loads and use the transfer function of the cylinder head system under different loads as a physical constraint, which is in the form of:

[0008] In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal; Embed physical constraints into the neural network by constructing a loss function; The physical information neural network is trained using the cylinder head vibration signal and cylinder pressure signal of the cylinder at a specific time period under different loads, and the cylinder head vibration signal and cylinder pressure signal under different loads use the physical constraints under the corresponding loads; The cylinder head vibration signal of a specific period of time is input into the trained physical information neural network to obtain the corresponding cylinder pressure signal.

[0009] In the above scheme, the specific time period includes the combustion period, that is, the time period when the crank angle is at specific angles before and after the compression top dead center.

[0010] In the above solution, the specific angle is 60°.

[0011] In the above scheme, the cylinder head vibration signal and cylinder pressure signal of a certain cylinder combustion section are obtained, including: One of the marker and the Hall sensor is set on the flywheel of the cylinder crankshaft, and the other of the marker and the Hall sensor is fixedly set outside the flywheel. When the marker sweeps across the end face of the Hall sensor, the Hall sensor generates an induced voltage, which is processed and outputs a square wave signal. The first time corresponding to the rising edge or falling edge of two adjacent square wave signals is the time for the crankshaft to rotate one circle. Acquire a cylinder head vibration data sequence and a cylinder pressure data sequence, and intercept the cylinder head vibration data sequence and the cylinder pressure data sequence corresponding to the first time, that is, the cylinder head vibration data sequence and the cylinder pressure data sequence of one crankshaft rotation; Interpolate the cylinder head vibration data sequence and cylinder pressure data sequence corresponding to the first time to 360×N points, each point corresponds to 1 / N degree of crankshaft rotation; Determine the crank angle from the rising edge or falling edge of the square wave signal to the compression top dead center, combine the crank angle range corresponding to the combustion section, and intercept the cylinder head vibration signal and cylinder pressure signal of the combustion section from the cylinder head vibration data sequence and cylinder pressure data sequence corresponding to the first time.

[0012] In the above scheme, a vibration sensor is used to obtain a cylinder head vibration signal, and the vibration sensor is located on the cylinder head surface close to the exhaust valve side; a pressure sensor is used to obtain a cylinder pressure signal, and the pressure sensor is located at a hole in the cylinder head.

[0013] In the above scheme, the method further comprises: Data cleaning of cylinder head vibration signal and cylinder pressure signal, including data outlier processing and denoising: Detect abnormal values ​​based on Fourier transform, judge and eliminate cylinder head vibration signals and cylinder pressure signals with abnormal values; The cylinder head vibration signal and cylinder pressure signal are denoised based on periodic averaging and low-pass filtering.

[0014] In the above scheme, the transfer function of the cylinder head system under different loads is determined, including: In the time domain, the transfer function describes the time domain response of the system to the input signal, that is, given the input signal , the output signal of the system By comparing the input signal to the system's transfer function Perform a convolution to compute:

[0015] In the frequency domain, the transfer function describes the response of the system to input signals of different frequencies, that is, the spectrum of the input signal The frequency response of the system Multiply them to get the spectrum of the output signal:

[0016] When calculating the transfer function of the cylinder head system under a certain load, it is necessary to first calculate the frequency spectrum of the cylinder head vibration signal and the cylinder pressure signal under the load, and then obtain the transfer function of the cylinder head system under the load from the following formula:

[0017] In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal.

[0018] In the above scheme, the total loss function of the physical information neural network includes the data-driven loss function and the physical constraint loss function, as shown below:

[0019] In the formula, is the total loss function; is the data-driven loss function, which is the difference between the network output value and the true value; is the physical constraint loss function; and is the weight coefficient; Among them, the physical constraint loss function for:

[0020] In the formula, is the cylinder head vibration signal, is the frequency spectrum of the cylinder head vibration signal, is the transfer function of the cylinder head system under the corresponding load, is the corresponding cylinder pressure signal, is the number of samples, is the Fourier transform, is the inverse Fourier transform.

[0021] According to the second aspect of the present invention, an electronic device is provided, which includes: a storage device for storing executable instructions; a processing device for executing the executable instructions stored in the storage device to implement the cylinder pressure virtual measurement method based on physical information neural network described in any of the above technical solutions.

[0022] According to a third aspect of the present invention, a storage medium is provided, in which a computer program is stored. When the computer program is executed, the cylinder pressure virtual measurement method based on physical information neural network described in any of the above technical solutions is implemented.

[0023] Compared with the prior art, the present invention has the following advantages and beneficial effects: The present invention proposes a virtual cylinder pressure measurement method based on a physical information neural network, which uses a neural network to predict cylinder pressure through a cylinder head vibration signal, and introduces the transfer function of the cylinder head system into the neural network to guide the learning process of the model. The output result not only matches the real data, but also must satisfy the physical equation, so that the behavior of the network can be explained from a physical perspective, rather than just relying on the fitting of the data; and due to the constraint of the physical equation, the network can still rely on physical priors to correct the output result when there is abnormal fluctuation or noise interference in the input signal, which fundamentally enhances the robustness and generalization ability of the model. In addition, when the neural network is trained, the cylinder head vibration signal and cylinder pressure signal under different loads use the physical constraints under the corresponding load, so that the neural network can be applied to different load situations, and the neural network only needs to obtain load information during training, and no load information is required during actual prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the cylinder pressure virtual measurement method based on physical information neural network; Figure 2It is the cylinder head vibration signal and cylinder pressure signal diagram of the whole cycle; Figure 3 It is the combustion section cylinder head vibration signal and cylinder pressure signal diagram; Figure 4 This is a schematic diagram of the physical information neural network structure; Figure 5 It is a comparison chart between the virtual cylinder pressure measured by the physical information neural network and the actual cylinder pressure; Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0026] The present invention provides a virtual cylinder pressure measurement method based on physical information neural network, such as Figure 1 As shown, it mainly includes the following steps: S1. When the engine is running, the top dead center, cylinder head vibration and cylinder pressure signals are collected to build an engine operation data set to provide data support for establishing a physical information neural network model.

[0027] S2. Preprocessing the collected data samples, including whole cycle signal interception and combustion segment interception.

[0028] S21 uses the top dead center signal to perform full-cycle interception on the collected original signal. The top dead center signal collected by the Hall sensor is a square wave signal. One working cycle of a four-stroke diesel engine corresponds to two crankshaft rotations. Two crankshaft rotations produce three rising edges. The sequence indexes of the three rising edges of the square wave signal are recorded, and the points corresponding to the three rising edge sequence indexes are selected from the collected cylinder pressure and cylinder head vibration data sequences to form new cylinder pressure data sequences and vibration data sequences to complete the full-cycle interception.

[0029] S22 intercepts the combustion segment of the vibration signal and the cylinder pressure signal according to the crank angle of each working cycle.

[0030] S3. Clean the preprocessed data, including outlier processing and denoising.

[0031] S31 determines and removes cylinder head vibration signals with abnormal values ​​based on time-frequency domain analysis.

[0032] S32 denoises the combustion section signal based on wavelet threshold denoising.

[0033] S4. Constructing physical constraints The transfer function of the cylinder head system takes into account the mechanical damping, stiffness and resonance characteristics of the system, and can accurately describe the relationship between the dynamic load of the in-cylinder gas pressure impact and the vibration response of the cylinder head. Embedding the transfer function as a physical constraint in the neural network can ensure that the learning process of the model conforms to the dynamic behavior of the physical system.

[0034] In the frequency domain, the transfer function describes the response of the system to input signals of different frequencies, namely:

[0035] In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal.

[0036] Substituting the calculated frequency spectrum of the cylinder head vibration signal and cylinder pressure signal after preprocessing and cleaning into the above formula, the transfer function of the cylinder head system can be obtained.

[0037] S5. Construct and train physical information neural network S51 divides the data processed in steps S2 and S3 into a training set and a test set, embeds the transfer function as a physical constraint into the training process of the neural network, and determines the structural parameters of the physical information neural network through multiple iterations.

[0038] The training process of the S52 physical information neural network model mainly includes the following steps: (1) Forward propagation. The output of the network is calculated through forward propagation, and the output value of the unit is calculated based on the current parameters and input data.

[0039] (2) Calculate the loss. Use the root mean square error as the loss function to calculate the difference between the model output value and the true value to obtain the loss value.

[0040] (3) Back propagation: Through the back propagation algorithm, the network parameters are adjusted according to the loss value, and the weights and biases are updated to reduce the loss value.

[0041] (4) Parameter update: Based on the gradient calculated by back-propagation, the Adam optimization algorithm is used to update the network parameters.

[0042] (5) Repeat training. Repeat the above steps until the model converges or reaches the set number of training rounds.

[0043] S6. After obtaining a model with better performance through S5, it can be put into use.

[0044] Specifically, the physical information neural network cylinder pressure virtual measurement method of the embodiment of the present invention has the following specific steps: S1. Collect the top dead center signal, cylinder head vibration signal and cylinder pressure signal when the engine is running.

[0045] S11 The vibration signal will attenuate during the transmission process. The attenuation of the vibration signal determines how much of the state information of the vibration source can be recognized by the sensor. Therefore, in order to ensure that the sensor can fully obtain the information contained in the vibration source, the measurement point of the vibration signal is required to be as close to the vibration source as possible to reduce the consumption of the vibration signal during the transmission process. In addition, when selecting the measurement point, the layout of the engine pipeline (cooling water channel, lubricating oil channel, etc.) and the interference of auxiliary equipment such as water pumps and fans must also be considered. Combined with the actual situation of the test model, the vibration signal measurement point is selected on the cylinder head surface close to the exhaust valve side.

[0046] S12 Drill a hole on the cylinder head and install the cylinder pressure sensor.

[0047] S13 A bracket is set on one side of the camshaft to install the Hall sensor, and the installation clearance is 1.5mm. The Hall sensor is used to obtain the top dead center signal.

[0048] S2. Preprocess the collected data samples, including whole cycle interception and combustion segment interception.

[0049] S21 Install a marker on the flywheel of the crankshaft. When the marker sweeps across the end face of the sensor, the semiconductor at the top of the sensor generates an induced voltage, which is processed and output as a square wave signal. The sampling time corresponding to the three rising edges of the square wave signal is the sampling time when the crankshaft rotates two times. The pulse signal of the crankshaft rotating two times is obtained by capturing the rising edges of the three square wave signals. The sequence indexes of the three rising edges of the square wave signal are recorded, and the points corresponding to the three rising edge sequence indexes are selected from the collected cylinder pressure data and cylinder head vibration data sequences to form a new cylinder pressure data sequence and vibration data sequence. The length of this data sequence is interpolated to 720 points (one point corresponds to one degree) to complete the whole cycle interception and equal crank angle processing, such as Figure 2 shown.

[0050] It should be noted that the falling edge can also be used to intercept the entire cycle. The time corresponding to the rising edge or falling edge of two adjacent square wave signals is the time for the crankshaft to rotate one circle. In addition, the number of data sequence length interpolation is not limited and can be selected according to actual conditions. For the convenience of calculation, a multiple of 360 can be selected, such as 180, 360, etc.

[0051] S22 The combustion section of the cylinder head vibration signal and cylinder pressure signal is the main stage where energy is released from the fuel and converted into mechanical work. It reflects key information such as combustion efficiency and maximum explosion pressure, and provides an important basis for engine status monitoring and fault diagnosis. Therefore, the cylinder head vibration signal and cylinder pressure signal of the combustion section (60° before and after the compression top dead center, a total of 120 data) are intercepted according to the crank angle, such as Figure 3 Specifically, the crank angle from the rising edge of the square wave signal to the compression top dead center is determined, and the crank angle range corresponding to the combustion section is combined to intercept the cylinder head vibration signal and cylinder pressure signal of the combustion section from the interpolated cylinder head vibration data sequence and cylinder pressure data sequence.

[0052] Similarly, if the falling edge is used for full-cycle interception, it is sufficient to determine the crank angle from the falling edge of the square wave signal to the compression top dead center.

[0053] S3. Clean the preprocessed data, including outlier processing and denoising.

[0054] S31 detects abnormal values ​​based on Fourier transform, and determines and eliminates cylinder head vibration signals with abnormal values.

[0055] S32 denoises the cylinder head vibration signal of the combustion section based on period averaging and low-pass filtering.

[0056] S4. Constructing physical constraints in physical information neural networks In the time domain, the transfer function describes the time domain response of the system to the input signal, that is, given the input signal , the output signal of the system The input signal can be compared with the system's transfer function Perform a convolution to compute:

[0057] In the frequency domain, the transfer function describes the response of a system to input signals of different frequencies. It converts the frequency spectrum of the input signal The frequency response of the system Multiply them to get the spectrum of the output signal:

[0058] In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal.

[0059] From the above formula, the transfer function of the cylinder head system can be obtained as:

[0060] Since load changes will affect the dynamic characteristics of the system, the transfer functions under different loads are calculated to accurately characterize the relationship between cylinder pressure and cylinder head vibration under each load. The transfer function of the cylinder head system can be obtained by inserting the preprocessed and cleaned cylinder head vibration signal and the cylinder pressure signal into the above formula.

[0061] S5. Constructing a physical information neural network S51 Data Preparation The preprocessed and cleaned data set contains 0%, 25%, 50%, and 75% loads, 60 samples under each load, and each load corresponds to a transfer function. The transfer function corresponding to the load is selected during training. 75% of them are used as training data sets, and 25% are used as test data sets.

[0062] S52 Physical Constraints The transfer function of the cylinder head system is used as the physical constraint, which is in the form of:

[0063] Physical constraints are loaded into the neural network by constructing a loss function. That is, the spectrum of the cylinder head vibration signal is substituted into the transfer function to obtain the spectrum of the cylinder pressure signal. The cylinder pressure signal in the time domain is obtained by inverse Fourier transform. The cylinder pressure signal corresponding to the cylinder head vibration signal forms a root mean square error, which is used as the physical loss. The physical loss and the data-driven loss together constitute the total loss function. If the network output completely satisfies the physical equation, the total loss function should be close to zero. If it does not approach zero, the model parameters are optimized by minimizing the total loss function to guide the iterative update of the network. As the total loss function decreases, the output value of the network gradually approaches the solution of the physical equation, thereby embedding the physical constraints into the neural network.

[0064] S53 Network Structure like Figure 4 As shown in the figure, the physical information neural network structure mainly includes input layer, hidden layer, loss function, gradient optimization and output layer.

[0065] The input layer passes the dataset to the hidden layer for processing.

[0066] Each neuron in the hidden layer performs a linear transformation on the data in the input layer in the following form:

[0067] The result of the linear transformation is passed to a nonlinear activation function. The function of the activation function is to introduce nonlinearity so that the network can represent complex nonlinear mappings. The output after the activation function is processed It will be passed to the next layer and become the input of the next layer of neurons. If the current hidden layer is the last layer of the network, it will be directly passed to the output layer for the final output. The hidden layer captures the complex features of the physical problem through this nonlinear mapping and ensures that the output results conform to the laws of physics through physical constraints.

[0068] The hidden layer adopts a fully connected structure, that is, each neuron is connected to all neurons in the previous layer. The activation function uses the ReLU function. The number of hidden layers is 20, and the number of neurons in each layer is 150.

[0069] The loss function includes data-driven loss and physical constraint loss. The data-driven loss measures the difference between the model output value and the true value to ensure that the network can accurately fit the data. The physical constraint loss ensures that the output value of the model satisfies the laws of physics. The total loss function is the weighted sum of the data-driven loss and the physical constraint loss:

[0070] In the formula, , is the weight coefficient of the loss term. By adjusting the weight coefficient, a balance is achieved between data-driven and physical constraints.

[0071] Physical Constraint Loss Function The root mean square error of the cylinder pressure signal is selected:

[0072] In the formula, is the cylinder head vibration signal, is the frequency spectrum of the cylinder head vibration signal, is the transfer function of the cylinder head system under the corresponding load, is the corresponding cylinder pressure signal, is the number of samples, is the Fourier transform, is the inverse Fourier transform.

[0073] Gradient optimization uses the gradient descent optimizer Adam to calculate the gradient of the loss function relative to the network parameters (such as weights and biases), and updates the network parameters according to the gradient to minimize the loss function. The iteration and update of network parameters are achieved by minimizing the loss function.

[0074] S54 Model Training 75% of the data set processed by steps S2 and S3 is used as the training data set, and 25% is used as the test data set. The structural parameters of the physical information neural network are determined through multiple training iterations. The number of hidden layers is 20, the number of neurons in each layer is 150, the number of iterations is 1500, the minimum step size is 32, and the initial learning rate is 0.001. , is 1. After training, save the model and input the combustion section of the cylinder head vibration signal collected in real time into the model to obtain the corresponding cylinder pressure signal. Figure 5 This is a comparison diagram between the cylinder pressure virtually measured by the physical information neural network and the actually measured cylinder pressure. It can be seen from the figure that the cylinder pressure obtained by the present invention is basically consistent with the actually measured cylinder pressure.

[0075] Of course, the present invention is also applicable to any time period outside the combustion stage, or even the entire cycle.

[0076] An embodiment of the present invention further provides an electronic device, Figure 6 1 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as a smart phone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server or cabinet server (including an independent server or a server cluster composed of multiple servers) that can execute programs. The computer electronic device 20 of this embodiment includes at least but is not limited to: a memory 21 and a processor 22 that can communicate with each other via a system bus, such as Figure 6 It should be pointed out that Figure 6 Only an electronic device 20 having components 21 - 22 is shown, but it should be understood that implementing all of the components shown is not a requirement, and more or fewer components may alternatively be implemented.

[0077] In this embodiment, the memory 21 (i.e., readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM). The memory 21 may also be an external storage device of the electronic device 20, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 20. Of course, the memory 21 may also include both the internal storage unit of the electronic device 20 and its external storage device. In this embodiment, the memory 21 is generally used to store the operating system and various application software installed in the electronic device 20, such as the program code of the cylinder pressure virtual measurement method based on the physical information neural network in the method embodiment. In addition, the memory 21 may also be used to temporarily store various data that have been output or are to be output.

[0078] The processor 22 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 22 is generally used to control the overall operation of the electronic device 20. In this embodiment, the processor 22 is used to run the program code stored in the memory 21 or process data, such as running a device storing the program code of the cylinder pressure virtual measurement method based on the physical information neural network, so as to implement the cylinder pressure virtual measurement method based on the physical information neural network in the method embodiment.

[0079] The present invention also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App application store, etc., on which a computer program is stored, and the program realizes corresponding functions when executed by a processor. The computer-readable storage medium of this embodiment is used to store the program code of the cylinder pressure virtual measurement method based on the physical information neural network, and when executed by the processor, the cylinder pressure virtual measurement method based on the physical information neural network of the method embodiment is realized.

[0080] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] In summary, the present invention provides a cylinder pressure virtual measurement method, device and medium based on physical information neural network. By introducing physical equations into the neural network to guide the learning process of the model, the output result not only matches the real data, but also must satisfy the physical equations, so that the behavior of the network can be explained from a physical perspective, rather than just relying on the fitting of the data. And due to the constraint of the physical equations, the network can still rely on physical priors to correct the output results when there are abnormal fluctuations or noise interference in the input signal, which fundamentally enhances the robustness and generalization ability of the model.

[0085] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0086] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0087] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A virtual cylinder pressure measurement method based on physical information neural network, characterized in that: The method includes: Under different loads, obtain the cylinder head vibration signal and cylinder pressure signal of a certain cylinder at a specific time period; Constructing a physical information neural network, which takes the cylinder head vibration signal of the cylinder in a specific period as input and the corresponding cylinder pressure signal as output; Determine the transfer function of the cylinder head system under different loads and use the transfer function of the cylinder head system under different loads as a physical constraint, which is in the form of: In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal; Embed physical constraints into the neural network by constructing a loss function; The physical information neural network is trained using the cylinder head vibration signal and cylinder pressure signal of the cylinder at a specific time period under different loads, and the cylinder head vibration signal and cylinder pressure signal under different loads use the physical constraints under the corresponding loads; The cylinder head vibration signal of a specific period of time is input into the trained physical information neural network to obtain the corresponding cylinder pressure signal.

2. The cylinder pressure virtual measurement method based on physical information neural network according to claim 1 is characterized in that: The specific period includes the combustion period, that is, the period when the crank angle is at specific angles before and after the compression top dead center.

3. The cylinder pressure virtual measurement method based on physical information neural network according to claim 2 is characterized in that: The specific angle is 60°.

4. The cylinder pressure virtual measurement method based on physical information neural network according to claim 2 is characterized in that: Obtain the cylinder head vibration signal and cylinder pressure signal of a cylinder combustion section, including: One of the marker and the Hall sensor is set on the flywheel of the cylinder crankshaft, and the other of the marker and the Hall sensor is fixedly set outside the flywheel. When the marker sweeps across the end face of the Hall sensor, the Hall sensor generates an induced voltage, which is processed and outputs a square wave signal. The first time corresponding to the rising edge or falling edge of two adjacent square wave signals is the time for the crankshaft to rotate one circle. Acquire a cylinder head vibration data sequence and a cylinder pressure data sequence, and intercept the cylinder head vibration data sequence and the cylinder pressure data sequence corresponding to the first time, that is, the cylinder head vibration data sequence and the cylinder pressure data sequence of one crankshaft rotation; Interpolate the cylinder head vibration data sequence and cylinder pressure data sequence corresponding to the first time to 360×N points, each point corresponds to 1 / N degree of crankshaft rotation; Determine the crank angle from the rising edge or falling edge of the square wave signal to the compression top dead center, combine the crank angle range corresponding to the combustion section, and intercept the cylinder head vibration signal and cylinder pressure signal of the combustion section from the cylinder head vibration data sequence and cylinder pressure data sequence corresponding to the first time.

5. The cylinder pressure virtual measurement method based on physical information neural network according to claim 1 is characterized in that: A vibration sensor is used to obtain the cylinder head vibration signal, and the vibration sensor is located on the cylinder head surface close to the exhaust valve side; a pressure sensor is used to obtain the cylinder pressure signal, and the pressure sensor is located at the hole in the cylinder head.

6. The cylinder pressure virtual measurement method based on physical information neural network according to claim 1 is characterized in that: The method further includes: Data cleaning of cylinder head vibration signal and cylinder pressure signal, including data outlier processing and denoising: Detect abnormal values ​​based on Fourier transform, judge and eliminate cylinder head vibration signals and cylinder pressure signals with abnormal values; The cylinder head vibration signal and cylinder pressure signal are denoised based on periodic averaging and low-pass filtering.

7. The cylinder pressure virtual measurement method based on physical information neural network according to claim 1 is characterized in that: Determine the transfer function of the cylinder head system under different loads, including: In the time domain, the transfer function describes the time domain response of the system to the input signal, that is, given the input signal , the output signal of the system By comparing the input signal to the system's transfer function Perform a convolution to compute: In the frequency domain, the transfer function describes the response of the system to input signals of different frequencies, that is, the spectrum of the input signal The frequency response of the system Multiply them to get the spectrum of the output signal: When calculating the transfer function of the cylinder head system under a certain load, it is necessary to first calculate the frequency spectrum of the cylinder head vibration signal and the cylinder pressure signal under the load, and then obtain the transfer function of the cylinder head system under the load from the following formula: In the formula, is the spectrum of the transfer function, is the frequency spectrum of the cylinder head vibration signal, is the frequency spectrum of the cylinder pressure signal.

8. The cylinder pressure virtual measurement method based on physical information neural network according to claim 1 is characterized in that: The total loss function of the physical information neural network includes the data-driven loss function and the physical constraint loss function, as shown below: In the formula, is the total loss function; is the data-driven loss function, which is the difference between the network output value and the true value; is the physical constraint loss function; and is the weight coefficient; Among them, the physical constraint loss function for: In the formula, is the cylinder head vibration signal, is the frequency spectrum of the cylinder head vibration signal, is the transfer function of the cylinder head system under the corresponding load, is the corresponding cylinder pressure signal, is the number of samples, is the Fourier transform, is the inverse Fourier transform.

9. An electronic device, characterized in that: The electronic device comprises: A storage device for storing executable instructions; A processing device is used to execute the executable instructions stored in the storage device to implement the cylinder pressure virtual measurement method based on physical information neural network described in any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed, the cylinder pressure virtual measurement method based on physical information neural network described in any one of claims 1 to 8 is implemented.

Images