Crankshaft and camshaft excitation signal identification method and device, medium and equipment

Abstract

The invention discloses an excitation signal identification method and device for a crankshaft and a camshaft, a medium and equipment. The method comprises the steps that crankshaft excitation signals and camshaft excitation signals sensed by different sensors are collected, different types of labels are marked on the excitation signals, and an excitation signal data set is obtained; constructing a precision improvement module based on a standard convolutional layer of the MobileNet and a simplified channel attention mechanism; constructing a lightweight multi-scale feature convolution module on the basis of a depth separable convolution layer of the MobileNet; a precision improvement module, a lightweight multi-scale feature convolution module and a global average pooling layer of the MobileNet are connected in sequence, and an improved MobileNet is obtained; taking the excitation signal data set as input, taking a label classification result of the excitation signal as output, and training the improved MobileNet to obtain an excitation signal identification model; and identifying the excitation signal to be identified by using the excitation signal identification model. According to the scheme, the excitation signals of the crankshaft and the camshaft can be efficiently and accurately identified through the lightweight model.

Description

Technical Field

[0001] The present invention relates to the technical field of signal recognition, and in particular to a method, device, medium and equipment for recognizing excitation signals of a crankshaft and a camshaft. Background Art

[0002] With the rapid development of urban transportation, energy, and other infrastructure, construction vehicles are playing an increasingly important role in building construction and energy development projects. However, due to the long-term operation of construction vehicles in harsh environments such as high temperatures, strong electromagnetic interference, and severe vibration, their engine electronic control units (EECUs) often malfunction. The EECU is the core electronic control unit of construction vehicles. It receives excitation signals from sensors, processes them internally, and adjusts the timing and amount of fuel injection, thereby controlling the fuel injection system and ensuring optimal and efficient engine operation.

[0003] Crankshaft and camshaft sensor signals are the core excitation signals that ensure the proper operation of the EECU. Efficient and accurate identification of these signals is crucial for EECU quality assurance and fault detection. However, there are many different types of crankshaft and camshaft sensors, and their signal characteristics are unstable and complex. They are also susceptible to harsh operating environments, such as vibration and high temperatures. These factors result in low excitation signal recognition rates, severely impacting the efficiency of EECU excitation signal recognition.

[0004] Currently, there are two main methods for identifying EECU excitation signals: signal processing-based methods and deep learning-based methods. Signal processing-based methods typically analyze sensor waveform data through time-frequency domain signal processing. While relatively simple to implement and highly interpretable, these methods still suffer from low recognition rates in scenarios with high noise levels and complex classification tasks. Deep learning-based methods, with its powerful self-learning and feature extraction capabilities, have achieved promising results in signal detection. However, deep learning models consume significant computing resources, significantly limiting their deployment and application in resource-constrained embedded devices.

[0005] In summary, existing technologies cannot guarantee that the model can efficiently identify the characteristics of the EECU crankshaft and camshaft excitation signals, and thus deploy the model into embedded devices for actual reasoning, resulting in poor real-time performance of EECU excitation signal detection. Summary of the Invention

[0006] Based on this, it is necessary to provide a crankshaft and camshaft excitation signal recognition method, device, medium and equipment to address the technical problem that the existing technology cannot ensure that the model can efficiently identify the EECU crankshaft and camshaft excitation signal characteristics, and deploy the model into the embedded device for actual reasoning, resulting in poor real-time performance of EECU excitation signal detection.

[0007] The present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying excitation signals of a crankshaft and a camshaft, the method comprising: Collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and marking the crankshaft excitation signals and camshaft excitation signals with different types of labels, respectively, to obtain crankshaft excitation signal datasets and camshaft excitation signal datasets; A precision improvement module is constructed based on the standard convolutional layer and simplified channel attention mechanism of MobileNet. A lightweight multi-scale feature convolution module is constructed based on the depthwise separable convolutional layer of MobileNet. The precision improvement module, the lightweight multi-scale feature convolution module and the global average pooling layer of MobileNet are sequentially connected to obtain an improved MobileNet. The crankshaft excitation signal dataset is used as input, and the label classification result of the crankshaft excitation signal is used as output to train the improved MobileNet to obtain a crankshaft excitation signal recognition model; the camshaft excitation signal dataset is used as input, and the label classification result of the camshaft excitation signal is used as output to train the improved MobileNet to obtain a camshaft excitation signal recognition model; the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model are used to respectively identify the crankshaft excitation signal and the camshaft excitation signal to be identified.

[0008] Furthermore, the accuracy improvement module is constructed based on the standard convolutional layer of MobileNet and the simplified channel attention mechanism, specifically including: Adding a first maximum pooling layer, a simplified channel attention mechanism, and a second maximum pooling layer in sequence to the output end of the standard convolutional layer to obtain the accuracy improvement module; The calculation formula for the simplified channel attention mechanism is: ; Among them, the input tensor X∈R T×F , T is the sequence length, F is the feature dimension, t is the bit in the sequence.

[0009] Furthermore, the lightweight multi-scale feature convolution module includes four depth-wise separable convolution layers, a Fire module, and five maximum pooling layers with a stride of 2. The lightweight multi-scale feature convolution module constructed based on the depth-wise separable convolution layer of MobileNet specifically includes: A maximum pooling layer is added after each of the four depthwise separable convolutional layers, and a Fire module is added after the fourth maximum pooling layer, and a maximum pooling layer is added after the Fire module to obtain the lightweight multi-scale feature convolution module.

[0010] Furthermore, the numbers of convolution kernels of the point convolutions of the four depthwise separable convolutional layers are 16, 24, 24, and 24, respectively.

[0011] Furthermore, the different sensors include Hall sensors and magnetoelectric sensors; the label types of the crankshaft excitation signal specifically include engine type, sensor type, total number of teeth, number of missing teeth and crankshaft type; the label types of the camshaft excitation signal specifically include engine type, sensor type, number of pulses, differential tooth interval and number of interval teeth.

[0012] In a second aspect, the present invention provides an excitation signal recognition device for a crankshaft and a camshaft, comprising: a data set construction module for collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and labeling the crankshaft excitation signals and camshaft excitation signals with different types of labels, respectively, to obtain crankshaft excitation signal data sets and camshaft excitation signal data sets; A model improvement module is used to build an accuracy improvement module based on MobileNet's standard convolutional layer and simplified channel attention mechanism; a lightweight multi-scale feature convolution module is built based on MobileNet's depthwise separable convolutional layer; and the accuracy improvement module, lightweight multi-scale feature convolution module, and MobileNet's global average pooling layer are sequentially connected to obtain an improved MobileNet. The model training and signal recognition module is used to take the crankshaft excitation signal dataset as input and the label classification result of the crankshaft excitation signal as output, train the improved MobileNet, and obtain a crankshaft excitation signal recognition model; take the camshaft excitation signal dataset as input and the label classification result of the camshaft excitation signal as output, train the improved MobileNet, and obtain a camshaft excitation signal recognition model; use the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model to respectively identify the crankshaft excitation signal and the camshaft excitation signal to be identified.

[0013] At least one technical solution adopted by the present invention can achieve the following beneficial effects: the present invention obtains a crankshaft excitation signal dataset and a camshaft excitation signal dataset by collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and respectively marks different types of labels on the crankshaft excitation signals and camshaft excitation signals, wherein the different sensors include Hall sensors and magnetoelectric sensors; an accuracy improvement module is constructed based on the standard convolution layer of MobileNet and the simplified channel attention mechanism, which can preliminarily extract features and adjust the importance of features, effectively screen out the most critical features for classification tasks, thereby enhancing the expression ability of the model; a lightweight multi-scale feature convolution module is constructed based on the depth-separable convolution layer of MobileNet, which can obtain a multi-scale feature map. The computational complexity of the model is significantly reduced. The accuracy enhancement module, lightweight multi-scale feature convolution module, and the global average pooling layer of MobileNet are sequentially connected to obtain an improved MobileNet. The crankshaft excitation signal dataset is used as input and the label classification results of the crankshaft excitation signal are used as output to train the improved MobileNet, thereby obtaining a crankshaft excitation signal recognition model. The camshaft excitation signal dataset is used as input and the label classification results of the camshaft excitation signal are used as output to train the improved MobileNet, thereby obtaining a camshaft excitation signal recognition model. The crankshaft excitation signal recognition model and the camshaft excitation signal recognition model are used to identify the crankshaft excitation signal to be identified and the camshaft excitation signal respectively. The above scheme improves the MobileNet network, uses the improved MobileNet to train the excitation signal recognition model, and further uses the excitation signal recognition model to identify excitation signals sensed by different sensors. Compared with the traditional MobileNetV1 model, the above scheme significantly reduces the number of model parameters and computational complexity while sacrificing a small amount of accuracy, enabling it to be deployed on an Arduino development board for multi-label classification of EECU excitation signals, thereby improving the efficiency of EECU excitation signal detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0015] Figure 1 A flow chart of the crankshaft and camshaft excitation signal identification method provided by the present invention; Figure 2 A schematic diagram of the structure of the improved MobileNet provided by the present invention; Figure 3 A schematic diagram of the structure of the depth-wise separable convolution provided by the present invention; Figure 4Schematic diagram of the crankshaft and camshaft excitation signal recognition device provided by the present invention; Figure 5 A schematic diagram of a computer device for implementing a crankshaft and camshaft excitation signal recognition method provided by the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] The server mentioned in the present invention can be a server installed on a business platform, or a device such as a desktop computer or laptop computer capable of executing the solution of the present invention. For ease of explanation, the following description will only use the server as the execution entity. The following, combined with the accompanying drawings, details the technical solutions provided by various embodiments of the present invention.

[0018] refer to Figure 1 The method for identifying the excitation signals of the crankshaft and camshaft in the present invention specifically comprises the following steps: S10: collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and marking the crankshaft excitation signals and camshaft excitation signals with different types of labels, respectively, to obtain a crankshaft excitation signal dataset and a camshaft excitation signal dataset.

[0019] In this embodiment, different sensors include Hall sensors and magnetoelectric sensors; the label types of the crankshaft excitation signal specifically include engine type, sensor type, total number of teeth, number of missing teeth and crankshaft type; the label types of the camshaft excitation signal specifically include engine type, sensor type, number of pulses, differential tooth interval and number of interval teeth.

[0020] Specifically, the experiment was conducted using crankshaft and camshaft sensor excitation signal data. Actual excavator vehicle data was collected from March to May 2021 in the Shichanjing mining area of ​​Shizuishan, Ningxia, China. An oscilloscope was connected to the corresponding port, the vehicle was started, and the crankshaft and camshaft sensor signals under different operating conditions were collected, and the data was saved to an external storage device. From May 2021 to June 2021, simulated data was collected based on a self-developed excitation signal generator. The simulator used LabVIEW software to generate crankshaft and camshaft sensor simulation signals and sent the signals to the EECU. After processing, the EECU generated an injector drive signal to drive the injector. After confirming that the injector was operating normally, the generated crankshaft and camshaft excitation signal data was collected using an oscilloscope and saved to an external device. The sampling points of both the actual vehicle data and the simulator data collected by the oscilloscope were 10,000. A total of 5,400 crankshaft data and 4,200 camshaft data were collected.

[0021] After data collection, the crankshaft and camshaft sensor signal data are stored as separate CSV files. To facilitate subsequent analysis, these data are combined into separate CSV files for the crankshaft and camshaft sensors, and the data are normalized to a range of [0, 1].

[0022] The pre-processed data is annotated with multiple labels. The crankshaft data is annotated with engine type, sensor type, total number of teeth, number of missing teeth, crankshaft type and other labels as shown in Table 1: Table 1 Multi-label annotation of crankshaft excitation signal data The camshaft data is labeled with engine type, sensor type, number of pulses, differential tooth spacing, number of interval teeth, etc. as shown in Table 2: Table 2 Multi-label annotation of camshaft excitation signal data To reduce the model's computational complexity and enable better model deployment and inference on embedded devices, the crankshaft and camshaft excitation signals are downsampled to 200×1. The present invention downsamples the raw input data, extracting one point for every 50 data points. This reduces the original 10,000 data points to 200. The dataset is then divided into training, validation, and test sets in a ratio of 8:1:1.

[0023] S20: Construct an accuracy improvement module based on MobileNet's standard convolutional layer and simplified channel attention mechanism; construct a lightweight multi-scale feature convolution module based on MobileNet's depthwise separable convolutional layer; connect the accuracy improvement module, lightweight multi-scale feature convolution module and MobileNet's global average pooling layer in sequence to obtain an improved MobileNet.

[0024] In this embodiment, an accuracy improvement module is constructed based on the standard convolutional layer of MobileNet and a simplified channel attention mechanism, specifically including: The first maximum pooling layer, the simplified channel attention mechanism, and the second maximum pooling layer are added to the output of the standard convolutional layer in sequence to obtain the accuracy improvement module.

[0025] The simplified calculation formula of the channel attention mechanism is: ; Where the input tensor X∈RT×F, T is the sequence length, F is the feature dimension, and t is the bit in the sequence.

[0026] Specifically, if Figure 2 As shown in the figure (), the accuracy improvement module consists of a 3×1 standard convolution, a simplified channel attention mechanism, and two max pooling layers with a stride of 2. The input crankshaft or camshaft data first undergoes a standard convolution with a small number of kernels to initially extract features. Then, simplified channel attention is combined to dynamically adjust the importance of features, thereby improving the model's classification performance. A max pooling layer with a stride of 2 is added after both the standard convolution and attention layers to reduce sequence length and computational complexity. This module primarily performs preliminary feature extraction and feature importance adjustment. This stage effectively selects the most critical features for the classification task, thereby enhancing the model's expressiveness.

[0027] In this embodiment, the lightweight multi-scale feature convolution module includes four depthwise separable convolution layers, one Fire module, and five maximum pooling layers with a stride of 2. The lightweight multi-scale feature convolution module is constructed based on the depthwise separable convolution layer of MobileNet, specifically including: A maximum pooling layer is added after each of the four depthwise separable convolutional layers, and a Fire module is added after the fourth maximum pooling layer, and a maximum pooling layer is added after the Fire module to obtain a lightweight multi-scale feature convolution module.

[0028] Specifically, refer to Figure 2In the figure (), the input data is processed by the precision enhancement module and then output to the lightweight multi-scale feature convolution module. The lightweight multi-scale feature convolution module consists of a depthwise classifiable convolution layer, a Fire module, and a max pooling layer with a stride of 2. The number of convolution kernels in the point convolution of the four depthwise separable convolution layers is 16, 24, 24, and 24, respectively.

[0029] Depthwise separable convolution Figure 3 As shown in the figure, a depth-wise convolution is first performed on each channel of the input feature map to obtain local features. Afterwards, the features of each channel are linearly fused through a 1×1 point-by-point convolution to capture cross-channel dependencies.

[0030] refer to Figure 2 The Fire module is mainly composed of the Squeeze layer and the Expand layer. The main task of the Squeeze layer is to reduce the number of channels of the feature map. It compresses the dimension of the input feature map using 1×1 standard convolution, extracts important information, and reduces computational complexity. The Expand layer is responsible for expanding the feature map output by the Squeeze layer. For one-dimensional data, Expand uses 1×1 and 3×1 convolutions to combine different receptive fields. The final output number of channels is the result of the two convolutions. After each depth-wise separable convolution layer and the Fire module, a maximum pooling layer with a stride of 2 is also followed to continuously reduce the dimension of the feature map. This design has the ability to extract multi-scale feature maps and significantly reduces the computational complexity of the model.

[0031] S30: Taking the crankshaft excitation signal dataset as input and the label classification result of the crankshaft excitation signal as output, the improved MobileNet is trained to obtain a crankshaft excitation signal recognition model; taking the camshaft excitation signal dataset as input and the label classification result of the camshaft excitation signal as output, the improved MobileNet is trained to obtain a camshaft excitation signal recognition model; using the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model, the crankshaft excitation signal to be identified and the camshaft excitation signal to be identified are respectively identified.

[0032] In this embodiment, the feature information extracted by the lightweight multi-scale feature convolution module is output to the model decision module for result reasoning, such as Figure 2 As shown in (), when performing multi-label classification on crankshaft and camshaft data, the sparse_categorical_crossentropy loss function is used to independently classify each label. When the data enters the decision module, it is first processed by the global average pooling layer and the fully connected layer. The results are then fed into five fully connected layers to perform inference on the five labels of the data.

[0033] Specifically, the processed features are input to the model decision module. In the decision module, the input features are first processed by the global average pooling layer and a fully connected layer, and then input to five fully connected layers respectively. For crankshaft data, the outputs of the five fully connected layers are 9, 2, 5, 5, and 3 respectively. For camshaft data, the outputs of the five fully connected layers are 7, 2, 2, 5, and 2 respectively, corresponding to the number of categories of each label of the crankshaft and camshaft, respectively. The loss function selects sparse_categorical_crossentropy (sparse categorical cross entropy), which is a simplified version of categorical cross entropy and is suitable for cases where the target label is in integer form. It directly uses integer indices to represent categories instead of one-hot encoding. When the number of labels is large, one-hot encoding will take up a lot of memory. Using integer labels can significantly reduce memory usage and is suitable for the embedded device scenario in this article. This can reduce memory usage and computational complexity. sparse_categorical_crossentropy is defined as:

[0034] ; in, N is the sample size, y i It is i The true category labels of samples, p i The model is i The probability distribution vector of the sample output (the result after softmax normalization). P i [ y i ] indicates that the model predicts i samples belong to the true category y i sparse_categorical_crossentropy measures the difference, or error, between the model's predictions and the true values. A smaller loss value indicates that the model's predictions are closer to the true values, and the model's performance is better.

[0035] In this embodiment, the hyperparameter settings of the model are shown in Table 3: Table 3 Hyperparameter settings of the model In this embodiment, the recognition performance of the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model is evaluated, and the specific steps are as follows: The crankshaft and camshaft test set signal data is input into the trained model in batches. After inference, the final output results are compared with the actual results, and the classification performance is evaluated using the following indicators: ; ; ; ; TP stands for True Positive, TN for True Negative, FP for False Positive, and FN for False Negative. In multi-label classification, Accuracy refers to the proportion of labels correctly predicted by the model, reflecting the model's overall classification performance. Precision refers to the proportion of labels predicted by the model as positive that are actually positive, measuring the accuracy of the model's positive predictions. Recall refers to the proportion of labels correctly predicted as positive by the model among all labels that are actually positive, reflecting the model's coverage of positive labels. F1 Score is the harmonic mean of precision and recall, providing a balanced metric for measuring the overall performance of a model in terms of precision and recall. In multi-label classification, metrics such as Accuracy, Precision, Recall, and F1 Score are calculated separately for each label. The final result is a weighted average calculated based on the number of instances in each label category. The larger the values ​​of these evaluation metrics, the better the model performance.

[0036] To evaluate model lightweighting, this article uses params, FLOPs, model size, and the Arduino's Flash Memory Usage, SRAM Usage, and inference latency of the inference data during inference when the model is deployed on an Arduino nano BLE 33Sense.

[0037] ; ; The model's parameter count is the sum of the parameters of each layer, and the model's floating-point operations are similarly the sum of the floating-point operations of each layer. Model Size refers to the size of the .h5 file generated after model training. Flash Memory Usage refers to the Flash memory used when deploying the model to the Arduino. SRAM Usage refers to the SRAM used during model inference. Inference Latency refers to the time from when inference data is input to the Arduino until inference completes. The smaller these metrics, the better.

[0038] The improved MobileNet model of the present invention is compared with SqueezeNet composed of three Fire modules, MobileNetV1 composed of five depthwise separable convolutional layers (ɑ=0.25, 0.5, 1), and MobileNetV2 composed of five residual blocks. The test results on the crankshaft excitation signal dataset are shown in Table 3: Table 3 Test results of crankshaft excitation signal dataset The test results of the camshaft excitation signal dataset are shown in Table 4: Table 4 Camshaft excitation signal dataset test results The crankshaft and camshaft .h5 models generated by the training were first converted to .tflite format and then to .h format. Each crankshaft and camshaft model was then deployed simultaneously into an Arduino Nano BLE Sense 33. Its Flash memory size is 1MB and its SRAM size is 256KB. The test results for the crankshaft and camshaft data, respectively, are shown in Table 5.

[0039] Table 5 Test results of crankshaft and camshaft data with input of 200×1 Among them, MobileNetV1 (ɑ=1) and SqueezeNet cannot be reasoned normally because the models occupy too much memory.

[0040] In summary, the engine EECU crankshaft and camshaft excitation signals are diverse and complex, and deep learning models have high resource utilization, making them difficult to deploy on embedded devices. The recognition method presented in this paper significantly reduces the number of model parameters and computational complexity, while minimizing the loss of classification performance. It also significantly reduces the model's resource utilization within the Arduino, accelerating data inference. This improves the efficiency of engine EECU excitation signal detection and ensures optimal and efficient engine operation.

[0041] After model training, the model format needs to be converted and deployed. The specific steps are as follows: Use the EECU multi-label excitation signal detection model based on lightweight multi-scale feature convolution and simplified channel attention to train on the crankshaft and camshaft multi-label excitation signal datasets, respectively, to obtain the .h5 models for the crankshaft and camshaft, respectively. The .h5 models are first converted to .tflite format and then to .h format. Both the crankshaft and camshaft models are deployed on the Arduino nanoBLE Sense for inference experiments.

[0042] based on Figure 1 The crankshaft and camshaft excitation signal recognition method shown in the figure collects crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and labels the crankshaft excitation signals and camshaft excitation signals with different types of labels respectively, thereby obtaining crankshaft excitation signal datasets and camshaft excitation signal datasets. Different sensors include Hall sensors and magnetoelectric sensors; an accuracy improvement module is constructed based on the standard convolution layer of MobileNet and the simplified channel attention mechanism, which can preliminarily extract features and adjust the importance of features, effectively screen out the most critical features for classification tasks, thereby enhancing the expression ability of the model; a lightweight multi-scale feature convolution module is constructed based on the depth-separable convolution layer of MobileNet, which can obtain multi-scale feature maps and significantly reduce The computational complexity of the model is reduced; the accuracy enhancement module, the lightweight multi-scale feature convolution module, and the global average pooling layer of MobileNet are sequentially connected to obtain an improved MobileNet; the crankshaft excitation signal dataset is used as input and the label classification results of the crankshaft excitation signal are used as output to train the improved MobileNet to obtain a crankshaft excitation signal recognition model; the camshaft excitation signal dataset is used as input and the label classification results of the camshaft excitation signal are used as output to train the improved MobileNet to obtain a camshaft excitation signal recognition model; the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model are used to identify the crankshaft excitation signal to be identified and the camshaft excitation signal recognition model, respectively. The above solution improves the MobileNet network, uses the improved MobileNet to train the excitation signal recognition model, and further uses the excitation signal recognition model to identify excitation signals sensed by different sensors. Compared with the traditional MobileNetV1 model, the model significantly reduces the number of parameters and computational complexity while sacrificing a small amount of accuracy, enabling it to be deployed on an Arduino development board for multi-label classification of EECU excitation signals, thereby improving the efficiency of EECU excitation signal detection.

[0043] When applying the crankshaft and camshaft excitation signal identification method provided by the present invention, it is not necessary to Figure 1 The steps are executed in the order shown. The specific execution order of the steps can be determined according to needs, and the present invention does not limit this.

[0044] The above is a crankshaft and camshaft excitation signal recognition method provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding crankshaft and camshaft excitation signal recognition device, such as Figure 4 Shown, including: The data set construction module is used to collect crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and mark the crankshaft excitation signals and camshaft excitation signals with different types of labels to obtain crankshaft excitation signal data sets and camshaft excitation signal data sets.

[0045] The model improvement module is used to build an accuracy improvement module based on MobileNet's standard convolutional layer and simplified channel attention mechanism; build a lightweight multi-scale feature convolution module based on MobileNet's depthwise separable convolutional layer; and sequentially connect the accuracy improvement module, lightweight multi-scale feature convolution module, and MobileNet's global average pooling layer to obtain an improved MobileNet.

[0046] The model training and signal recognition module is used to take the crankshaft excitation signal dataset as input and the label classification result of the crankshaft excitation signal as output, train the improved MobileNet, and obtain the crankshaft excitation signal recognition model; take the camshaft excitation signal dataset as input and the label classification result of the camshaft excitation signal as output, train the improved MobileNet, and obtain the camshaft excitation signal recognition model; use the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model to respectively identify the crankshaft excitation signal to be identified and the camshaft excitation signal.

[0047] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the Figure 1 A method for identifying the excitation signals of a crankshaft and a camshaft is provided.

[0048] The present invention also provides Figure 5 The structural diagram of the computer equipment shown in FIG. Figure 5 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the Figure 1 A message passing method for multi-commodity flow problems in communication networks is provided.

[0049] The specific definition of the crankshaft and camshaft excitation signal identification device can be found in the definition of the crankshaft and camshaft excitation signal identification method above, and will not be repeated here. The various modules in the crankshaft and camshaft excitation signal identification device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0050] Those skilled in the art will appreciate that all or part of the processes in the described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0051] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the 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 the present invention.

Claims

1. A method for identifying excitation signals of a crankshaft and a camshaft, characterized in that: include: Collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and marking the crankshaft excitation signals and camshaft excitation signals with different types of labels, respectively, to obtain crankshaft excitation signal datasets and camshaft excitation signal datasets; A precision improvement module is built based on MobileNet's standard convolutional layer and simplified channel attention mechanism; a lightweight multi-scale feature convolution module is built based on MobileNet's depthwise separable convolutional layer; The precision improvement module, the lightweight multi-scale feature convolution module and the global average pooling layer of MobileNet are sequentially connected to obtain an improved MobileNet; The crankshaft excitation signal dataset is used as input, and the label classification result of the crankshaft excitation signal is used as output to train the improved MobileNet to obtain a crankshaft excitation signal recognition model; the camshaft excitation signal dataset is used as input, and the label classification result of the camshaft excitation signal is used as output to train the improved MobileNet to obtain a camshaft excitation signal recognition model; the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model are used to respectively identify the crankshaft excitation signal and the camshaft excitation signal to be identified.

2. The method for identifying the excitation signal of a crankshaft and a camshaft according to claim 1, wherein: The accuracy improvement module based on MobileNet's standard convolutional layer and simplified channel attention mechanism is constructed, specifically including: Adding a first maximum pooling layer, a simplified channel attention mechanism, and a second maximum pooling layer in sequence to the output end of the standard convolutional layer to obtain the accuracy improvement module; The calculation formula for the simplified channel attention mechanism is: ； Among them, the input tensor X∈R T×F , T is the sequence length, F is the feature dimension, t is the bit in the sequence.

3. The method for identifying the excitation signal of a crankshaft and a camshaft according to claim 1, wherein: The lightweight multi-scale feature convolution module includes four depth-wise separable convolution layers, a Fire module, and five maximum pooling layers with a stride of 2. The lightweight multi-scale feature convolution module constructed based on the depth-wise separable convolution layer of MobileNet specifically includes: A maximum pooling layer is added after each of the four depthwise separable convolutional layers, and a Fire module is added after the fourth maximum pooling layer, and a maximum pooling layer is added after the Fire module to obtain the lightweight multi-scale feature convolution module.

4. The method for identifying the excitation signal of a crankshaft and a camshaft according to claim 3, wherein: The number of convolution kernels of the point convolution of the four depth-wise separable convolution layers are 16, 24, 24, and 24 respectively.

5. The method for identifying the excitation signal of a crankshaft and a camshaft according to claim 1, wherein: The different sensors include Hall sensors and magnetoelectric sensors; the label types of the crankshaft excitation signal specifically include engine type, sensor type, total number of teeth, number of missing teeth and crankshaft type; the label types of the camshaft excitation signal specifically include engine type, sensor type, number of pulses, differential tooth interval and number of interval teeth.

6. The crankshaft and camshaft excitation signal recognition device is characterized in that: include: a data set construction module for collecting crankshaft excitation signals and camshaft excitation signals sensed by different sensors, and labeling the crankshaft excitation signals and camshaft excitation signals with different types of labels to obtain crankshaft excitation signal datasets and camshaft excitation signal datasets, wherein the different sensors include a Hall sensor and a magnetoelectric sensor; The model improvement module is used to build an accuracy improvement module based on MobileNet's standard convolutional layer and simplified channel attention mechanism; and a lightweight multi-scale feature convolution module based on MobileNet's depthwise separable convolutional layer; The precision improvement module, the lightweight multi-scale feature convolution module and the global average pooling layer of MobileNet are sequentially connected to obtain an improved MobileNet; The model training and signal recognition module is used to take the crankshaft excitation signal dataset as input and the label classification result of the crankshaft excitation signal as output, train the improved MobileNet, and obtain a crankshaft excitation signal recognition model; take the camshaft excitation signal dataset as input and the label classification result of the camshaft excitation signal as output, train the improved MobileNet, and obtain a camshaft excitation signal recognition model; use the crankshaft excitation signal recognition model and the camshaft excitation signal recognition model to respectively identify the crankshaft excitation signal and the camshaft excitation signal to be identified.

7. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying excitation signals of a crankshaft and a camshaft according to any one of claims 1 to 5 is implemented.

8. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for identifying the excitation signals of the crankshaft and camshaft according to any one of claims 1 to 5 is implemented.

Images