Deep Learning-Based Internal Combustion Engine Calibration System and Method

Abstract

This invention provides a deep learning-based internal combustion engine calibration system and method. The system includes a data acquisition module, a feedforward execution module, and a remote neural network training module. The data acquisition module collects data from the internal combustion engine and converts the data into input parameters. The feedforward execution module includes a neural network model and an actuator. The neural network model processes the input parameters to obtain training data. The remote neural network training module selects a neural network algorithm for the calibration target and trains the model based on the algorithm to obtain updated model parameters. The actuator updates the neural network model based on the updated parameters and performs calibration on the calibration target using the updated model. This invention, based on a remote neural network training module that uploads and downloads data in a time-sharing / real-time manner, achieves accurate prediction of emission indicators for internal combustion engines under different operating conditions.

Description

Technical Field

[0001] This invention relates to the field of internal combustion engine technology, and in particular to an internal combustion engine calibration system based on deep learning. Background Technology

[0002] Internal combustion engine calibration is the process of adjusting and optimizing the control system of an internal combustion engine to achieve optimal performance or emissions. It is an important part of the application of electronic control systems in the development of internal combustion engines, the final stage of the entire control system development, and also the stage with the largest workload and the most difficult.

[0003] Traditional test calibration methods are the simplest, easiest to implement, and relatively intuitive. However, for increasingly diverse internal combustion engine electronic control systems, the greater degree of control freedom and more flexible calibration make performance calibration and optimization more difficult, and the workload increases dramatically with the number of calibration parameters. Traditional test calibration methods generally employ a full factorial rotation approach, evaluating the engine performance under each set of operating parameter configurations based on the optimized target parameters, and then selecting the optimal parameter configuration across all experimental states as the final calibration result. This method has two drawbacks. First, because it uses a full factorial rotation test method, the workload required for calibration is related to the number of parameters to be calibrated and the number of test levels. When the number of operating parameters to be calibrated is m, and the number of test levels for each parameter is n, the total number of tests required is m. n Therefore, as the operating parameters increase, the workload of calibration will increase exponentially. Furthermore, the level of refinement in calibration is limited by the number of experimental levels, and the optimal combination of calibration parameters should be selected from the finally determined best operating parameter configuration. Thus, the final optimal calibration parameters are directly influenced by the experimental plan formulated by the designers, and have a certain degree of subjectivity; the final calibration result is often not the globally optimal value.

[0004] Taking emissions calibration as an example, its characteristics include complex optimization objectives, a high dimensionality of the optimization space, and a relatively high proportion of fuzzy and uncertain relationships between control parameters. Therefore, emissions are extremely dependent on engineers' experience. Introducing artificial intelligence technology can effectively shorten the development cycle of internal combustion engines, reduce the workload of matching and optimization experiments, improve test accuracy, reduce development costs, and alleviate the problems of insufficient manpower and equipment.

[0005] Deep learning possesses characteristics such as massively parallel processing capabilities, strong learning ability, high nonlinear processing capability, and adaptability. Applying deep learning to internal combustion engine calibration allows for the construction of a data-driven emission calibration and optimization method. This enables nonlinear combustion parameter estimation based on multi-sensor information fusion, addressing the challenges posed by complex optimization objectives and high-dimensionality optimization spaces. Furthermore, it significantly aids in the estimation and feedback of nonlinear combustion parameters based on multi-sensor information fusion. This holds promise for laying the foundation for the development and realization of future intelligent, efficient, and clean combustion engines. Summary of the Invention

[0006] Therefore, the purpose of this invention is to provide a deep learning-based internal combustion engine calibration system and method to at least address the shortcomings of the aforementioned technologies.

[0007] This invention proposes a deep learning-based internal combustion engine calibration system, comprising a data acquisition module, a feedforward execution module, and a remote neural network training module:

[0008] The data acquisition module is used to acquire data from the internal combustion engine and convert the data into input parameters for the feedforward execution module.

[0009] The feedforward execution module includes at least a neural network model and an executor. The neural network model is used to process the input parameters to obtain the training data of the remote neural network training module.

[0010] The remote neural network training module is used to select the corresponding neural network algorithm for the calibration target, and train the training data according to the neural network algorithm to obtain the updated model parameters;

[0011] The actuator is used to update the neural network model according to the updated model parameters, and to perform calibration on the calibration target using the updated neural network model.

[0012] Furthermore, the internal combustion engine calibration system also includes a test bench assembly module, which is used to assemble an internal combustion engine test bench based on the internal combustion engine, control equipment, and host computer.

[0013] Furthermore, the data acquisition module includes multiple sensors and transmitters, all of which are mounted on the internal combustion engine test bench.

[0014] Furthermore, the neural network model is specifically used for:

[0015] The input parameters are received, and a feedforward operation is performed on the input parameters to obtain a calibrated predicted value;

[0016] The calibrated predicted values ​​and the input parameters are packaged and integrated to obtain the training data for the remote neural network training module.

[0017] Furthermore, the internal combustion engine calibration system also includes a control parameter input module, which is used to optimize the data of the internal combustion engine when the prediction accuracy of the neural network model reaches a preset threshold, so as to generate a control signal as the input parameter of the feedforward execution module.

[0018] Furthermore, the control device is a PXIe device, which is communicatively connected to the host computer, and the data transmission between the internal combustion engine test bench and the data acquisition module is realized through the PXIe device.

[0019] Furthermore, the neural network algorithm includes an input layer, an LSTM network layer, a fully connected layer, and a linear output layer. The neural network algorithm uses the Adam optimization method to perform grid training on the training data.

[0020] Furthermore, the control parameter input model is specifically used for:

[0021] When the prediction accuracy of the neural network model reaches a preset threshold, a genetic algorithm is used to encode the data of the internal combustion engine in binary form, and the data of the internal combustion engine is controlled to change randomly within a preset range.

[0022] The optimal data range is obtained by using a fitness function for decoding and evaluation, and then the data range is converted into the corresponding control signal.

[0023] Compared with existing technologies, the beneficial effects of this invention are as follows: By setting up a data acquisition module, a feedforward execution module, and a remote neural network training module, the invention solves the problem that the optimal calibration parameters obtained by the calibration method in existing technologies are directly affected by the experimental scheme formulated by the designers, resulting in a certain degree of subjectivity and the final calibration result not being the globally optimal value. Furthermore, by processing the internal combustion engine data through a neural network model and actuator, and based on the remote neural network training module that uploads and downloads data in a time-sharing / real-time manner, the invention achieves accurate prediction of emission indicators under different operating conditions of the internal combustion engine, reducing the investment of manpower and experimental resources in the internal combustion engine calibration process, and achieving the technical effect of cost reduction and efficiency improvement.

[0024] This invention also proposes an internal combustion engine calibration method, applied to the aforementioned internal combustion engine calibration system, the internal combustion engine calibration method comprising:

[0025] S1: Collect the operating data of the internal combustion engine under various operating conditions at a preset acquisition frequency, and convert the operating data into input parameters;

[0026] S2: Obtain the input parameters and perform low-pass filtering and PCA dimensionality reduction on the input parameters to obtain the corresponding training data;

[0027] S3: Use the training data to train the neural network model to obtain the operating conditions where the prediction effect is lower than a preset threshold;

[0028] S4: Repeat steps S1 to S3 above a preset number of times at the operating point until the prediction effect is greater than or equal to the preset threshold, and obtain the model parameters in this state;

[0029] S5: The internal combustion engine is calibrated using the updated neural network model.

[0030] Furthermore, the method also includes:

[0031] When the prediction accuracy of the neural network model in processing the input parameters reaches a preset threshold, the operating data of the internal combustion engine is optimized to generate a control signal as the input parameter. Attached Figure Description

[0032] Figure 1 This is a structural block diagram of the deep learning-based internal combustion engine calibration system in the first embodiment of the present invention;

[0033] Figure 2 This is a structural block diagram of the data acquisition module in the first embodiment of the present invention;

[0034] Figure 3 This is a structural block diagram of the feedforward execution module in the first embodiment of the present invention;

[0035] Figure 4 This is a structural block diagram of the internal combustion engine test bench in the first embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram of the internal combustion engine calibration system in the first embodiment of the present invention;

[0037] Figures 6 to 7 This is a diagram showing the prediction results of the neural network in the first embodiment of the present invention;

[0038] Figure 8 This is a flowchart of the PXI prediction program in the first embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the neural network arranged on the FPGA in the first embodiment of the present invention;

[0040] Figure 10 This is a diagram showing the real-time prediction results of the FPGA in the first embodiment of the present invention;

[0041] Figure 11This is a flowchart of the internal combustion engine calibration method in the second embodiment of the present invention.

[0042] Explanation of key component symbols:

[0043] Frame assembly module 10 sensor 21 Data acquisition module 20 Transmitter 22 Feedforward execution module 30 Neural network model 31 Remote Neural Network Training Module 40 Actuator 32 Control parameter input module 50

[0044] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0045] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0046] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0048] Example 1

[0049] Please see Figures 1 to 3 The figure shows a deep learning-based internal combustion engine calibration system according to the first embodiment of the present invention, including a data acquisition module 20, a feedforward execution module 30, and a remote neural network training module 40:

[0050] The data acquisition module 20 is used to acquire data from the internal combustion engine and convert the data into input parameters for the feedforward execution module 30.

[0051] The feedforward execution module 30 includes at least a neural network model 31 and an executor 32. The neural network model 31 is used to process the input parameters to obtain the training data of the remote neural network training module 40.

[0052] The remote neural network training module 40 is used to select a corresponding neural network algorithm for the calibration target, and train the training data according to the neural network algorithm to obtain updated model parameters.

[0053] The actuator 32 is used to update the neural network model 31 according to the updated model parameters, and to perform calibration on the calibration target using the updated neural network model.

[0054] Understandably, this embodiment addresses the problem that the optimal calibration parameters obtained by the calibration method in the prior art are directly affected by the experimental scheme formulated by the designer, resulting in a certain degree of subjectivity and the final calibration result not being the globally optimal value. Furthermore, by setting up a data acquisition module 20, a feedforward execution module 30, and a remote neural network training module 40, the data of the internal combustion engine is processed through a neural network model 31 and an actuator 32. Based on the data being uploaded to and downloaded to the remote neural network training module 40 in a time-sharing / real-time manner, the emission indicators of the internal combustion engine under different operating conditions are accurately predicted, reducing the human and experimental resource input in the internal combustion engine calibration process and achieving the technical effect of cost reduction and efficiency improvement.

[0055] For details, please refer to Figure 4 The internal combustion engine calibration system also includes a test bench assembly module 10, which is used to assemble an internal combustion engine test bench based on the internal combustion engine, control equipment and host computer. The data acquisition module 20 includes multiple sensors 21 and transmitters 22, and the sensors 21 and the transmitters 22 are all mounted on the internal combustion engine test bench.

[0056] In this embodiment, the control device is a PXIe device, which is connected to the host computer via a network cable. Data transmission between the host computer, the DAQ device, and the experimental platform can be achieved using LabVIEW, the programming software provided by the parent company of the PXIe device.

[0057] The internal combustion engine uses Geely's 1.4T China VI standard PFI engine. The engine is equipped with intake and exhaust pressure and temperature sensors, ABZ phase encoders, five-gas emission instruments and oxygen sensors. The ignition signal is taken out from the ECU.

[0058] Furthermore, the neural network model 31 is specifically used to receive the input parameters and perform feedforward operations on the input parameters to obtain a calibrated predicted value;

[0059] The calibrated predicted values ​​and the input parameters are packaged and integrated to obtain the training data of the remote neural network training module 40.

[0060] In practice, the signal is transmitted to the PXIe device, which includes an RT system and an FPGA acquisition card. The FPGA acquires the voltage signal at a frequency of 40MHz, and the RT system reads the voltage signal acquired by the FPGA and transmits the data back to the host computer. The PXIe device is controlled by LabVIEW programming, and the LabVIEW program is compiled into the PXIe on the host computer. After the data is transmitted to the host computer, it is processed and imported into an LSTM neural network for training. Once the neural network training is complete, the model parameters with the optimal loss rate are extracted and input into the network model set on the PXIe. In this way, the values ​​predicted by the neural network on the PXIe replace the actual engine emission values, realizing the use of PXIe to replace the real engine output, thereby reducing the number of experiments in the engine calibration process and reducing the consumption of manpower and resources.

[0061] In this embodiment, the internal combustion engine calibration system further includes a control parameter input module 50. The control parameter input module 50 is used to optimize the data of the internal combustion engine when the prediction accuracy of the neural network model 31 reaches a preset threshold, so as to generate a control signal as the input parameter of the feedforward execution module 30.

[0062] Specifically, the control parameter input model is used to: when the prediction accuracy of the neural network model 31 reaches a preset threshold, use a genetic algorithm to encode the data of the internal combustion engine in binary form, and control the data of the internal combustion engine to change randomly within a preset range;

[0063] The optimal data range is obtained by using a fitness function for decoding and evaluation, and then the data range is converted into the corresponding control signal.

[0064] In practice, the data is first processed for noise reduction using low-pass filtering. Secondly, correlation analysis is performed on multi-dimensional parameters to remove redundant parameters. The trends of input and output data are compared, and feature crossing is performed appropriately. The neural network prediction model is built using the PyTorch and TensorFlow frameworks, constructing a Dataset dataset. The main components of the neural network are an input layer, an LSTM network layer, a fully connected layer, and a linear output layer. The Adam optimization method is used for network training, with an adaptive learning rate.

[0065] The FPGA prediction model inputs the neural network parameter matrix trained in software into the PXIE, and uses LabVIEW programming to implement the software model structure on the FPGA, constructing a sub-VI for each network layer, which can be called when needed. It also leverages the powerful forward matrix operation capabilities of the FPGA to achieve rapid calculation and prediction.

[0066] Finding the optimal parameters is a multidimensional parameter optimization problem. A genetic algorithm is used to encode the parameters in binary form, allowing them to vary randomly within a specified range. A fitness function is then used to decode and evaluate the parameters, leaving the optimal parameter range.

[0067] Please see Figure 5 The diagram shown is a structural schematic of the internal combustion engine calibration system in this embodiment. The following are the experimental steps for the internal combustion engine calibration system:

[0068] 1. Discretize the throttle opening and engine speed, and combine different combinations of "throttle opening and engine speed" as a single operating point.

[0069] 2. Allow the internal combustion engine to run in steady state for several seconds at different operating conditions, and collect the required parameters at a sampling frequency of 1Hz.

[0070] 3. Use the collected data to train a neural network to find operating conditions with poor prediction performance.

[0071] 4. Repeat steps 2 and 3 several times at the operating points where the prediction effect is poor.

[0072] 5. A NO prediction network with good accuracy was obtained.

[0073] Please see Figures 6 to 7 The image shows the prediction results from the neural network. Please refer to [link / reference]. Figure 8 The diagram shown is a framework diagram of the PXI prediction program.

[0074] First, the network parameters trained on the computer are exported and placed into a parameter matrix in LabVIEW. Engine data, acquired from the FPGA and processed, is then fed into the pre-deployed neural network for forward propagation to obtain predicted emissions values. Figure 9 This is a schematic diagram of a neural network deployed on an FPGA. Compared to a fully connected layer, a traditional RNN recurrent layer only requires adding a feedback node to store timing information on top of the fully connected layer to complete the deployment of the recurrent layer on LabVIEW. Figure 10 This is a graph showing the real-time prediction results of the FPGA. As you can see, under steady-state conditions, although the FPGA prediction values ​​fluctuate, they are very close to the actual values.

[0075] In summary, the deep learning-based internal combustion engine calibration system in this embodiment, by setting up a data acquisition module, a feedforward execution module, and a remote neural network training module, solves the problem that the optimal calibration parameters obtained by existing calibration methods are directly affected by the experimental scheme formulated by the designers, resulting in a certain degree of subjectivity and the final calibration result not being the globally optimal value. Furthermore, by processing the internal combustion engine data through a neural network model and actuators, and based on the remote neural network training module that uploads and downloads data in a time-sharing / real-time manner, it achieves accurate prediction of emission indicators of the internal combustion engine under different operating conditions, reducing the human and experimental resource input in the internal combustion engine calibration process, and achieving the technical effect of cost reduction and efficiency improvement.

[0076] Example 2

[0077] Please see Figure 11 The diagram shows a second embodiment of the internal combustion engine calibration method, applied to the aforementioned internal combustion engine calibration system. The internal combustion engine calibration method specifically includes steps S11 to S5:

[0078] S1, collect the operating data of the internal combustion engine under various working conditions at a preset acquisition frequency, and convert the operating data into input parameters;

[0079] S2, obtain the input parameters, and perform low-pass filtering and PCA dimensionality reduction on the input parameters to obtain the corresponding training data;

[0080] S3, use the training data to train the neural network model to obtain the operating conditions where the prediction effect is lower than a preset threshold;

[0081] S4, repeat steps S1 to S3 above a preset number of times at the operating point until the prediction effect is greater than or equal to the preset threshold, and obtain the model parameters in this state;

[0082] S5, the internal combustion engine is calibrated using the updated neural network model.

[0083] In some alternative embodiments, the method further includes:

[0084] When the prediction accuracy of the neural network model in processing the input parameters reaches a preset threshold, the operating data of the internal combustion engine is optimized to generate a control signal as the input parameter.

[0085] The internal combustion engine calibration method provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned system embodiment. For the sake of brevity, any parts not mentioned in the method embodiment can be referred to the corresponding content in the aforementioned system embodiment.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A deep learning-based internal combustion engine calibration system, characterized by, The system comprises a data acquisition module, a feedforward execution module and a remote neural network training module. The data acquisition module is configured to acquire data of the internal combustion engine and convert the data into input parameters of the feedforward execution module. The feedforward execution module comprises at least a neural network model and an executor, wherein the neural network model is configured to process the input parameters to obtain training data of the remote neural network training module. The remote neural network training module is configured to select a corresponding neural network algorithm for a calibration target, train the training data according to the neural network algorithm, and obtain updated model parameters, wherein the neural network algorithm comprises an input layer, an LSTM network layer, a fully connected layer and a linear output layer, and the neural network algorithm adopts an Adam optimization method to perform grid training on the training data. The executor is configured to update the neural network model according to the updated model parameters, and perform calibration on the calibration target by using the updated neural network model.

2. The deep learning-based internal combustion engine calibration system according to claim 1, characterized by, The internal combustion engine calibration system further comprises a test bench building module configured to build an internal combustion engine test bench according to the internal combustion engine, a control device and an upper computer.

3. The deep learning-based internal combustion engine calibration system of claim 2, wherein, The data acquisition module comprises a plurality of sensors and transducers, which are mounted on the internal combustion engine test bench.

4. The deep learning-based internal combustion engine calibration system of claim 2, wherein, The neural network model is specifically configured to: receive the input parameters and perform feedforward operation on the input parameters to obtain a predicted value of a calibration quantity; pack and integrate the predicted value of the calibration quantity and the input parameters to obtain the training data of the remote neural network training module.

5. The deep learning-based internal combustion engine calibration system of claim 4, wherein, The internal combustion engine calibration system further comprises a control parameter input module configured to perform optimization on the data of the internal combustion engine when the prediction accuracy of the neural network model reaches a preset threshold to generate a control signal as the input parameter of the feedforward execution module.

6. The deep learning-based internal combustion engine calibration system of claim 2, wherein, The control device adopts a PXIe device, which is in communication connection with the upper computer and realizes data transmission between the internal combustion engine test bench and the data acquisition module through the PXIe device.

7. The deep learning-based internal combustion engine calibration system of claim 5, wherein, The control parameter input model is specifically configured to: when the prediction accuracy of the neural network model reaches the preset threshold, use a genetic algorithm to perform binary coding on the data of the internal combustion engine, control the data of the internal combustion engine to randomly change within a preset range, decode and judge by using a fitness function to obtain an optimal data interval, and convert the data interval into a corresponding control signal. The internal combustion engine calibration method comprises:

8. An internal combustion engine calibration method applied to the internal combustion engine calibration system according to any one of claims 1 to 7, characterized by, S1: acquiring running data of the internal combustion engine under multiple working conditions at a preset acquisition frequency, and converting the running data into input parameters; S2: obtaining the input parameters, and performing low-pass filtering and PCA dimensionality reduction processing on the input parameters to obtain corresponding training data; S3: training the neural network model by using the training data to obtain a working condition point with a prediction effect lower than a preset threshold; ​ S4: repeating the above steps S1-S3 a preset number of times at the operating point until the prediction effect is greater than or equal to the preset threshold, and obtaining the model parameters in this state; S5: calibrating the internal combustion engine through the updated neural network model.

9. The internal combustion engine calibration method according to claim 8, characterized by, The method further comprises: When the prediction accuracy of the neural network model in data processing of the input parameters reaches a preset threshold, the operation data of the internal combustion engine is optimized to generate a control signal as an input parameter.

Images