Tractor fault diagnosis method and system based on GRU recurrent neural network

Abstract

The invention relates to a tractor fault diagnosis method and system based on a GRU recurrent neural network. According to the method, a gate cycle unit (GRU) is utilized to effectively solve the problem of gradient disappearance or gradient explosion occurring when long-sequence data is processed through an optimized version of a recurrent neural network (RNN), so that the long-term dependency relationship in the data is effectively learned; relevant data such as the rotating speed, the water temperature and the engine oil pressure of a tractor engine are collected in real time, so that the data show time sequence characteristics, the GRU can dynamically screen, memorize and update information changing along with time through a gating mechanism, and then feature modes in normal operation and fault states are learned; and powerful support is provided for tractor fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to a tractor fault diagnosis method and system based on a GRU recurrent neural network. Background Technology

[0002] Tractors are one of the most important agricultural implements in agricultural production, and they are used frequently, basically covering all aspects of agricultural production, including plowing, planting, managing, and harvesting. However, due to the high frequency of use, major components of tractors are easily worn out, and improper use and maintenance can easily reduce the working efficiency and service life of tractors.

[0003] Therefore, in practical applications, tractors need to be monitored and warned of malfunctions in real time to improve the overall efficiency and lifespan of the tractor. Summary of the Invention

[0004] In order to address the aforementioned problems and shortcomings in the existing technology, and to ensure that the tractor is in a fault-free working state during use, improve the overall efficiency of the tractor, and achieve high-efficiency and high-quality operation of the tractor, this invention proposes a tractor fault diagnosis method and system based on GRU recurrent neural network.

[0005] The specific plan includes the following:

[0006] The tractor fault diagnosis method based on GRU recurrent neural network performs the following steps to obtain an ideal model, which is used to predict tractor data during actual production operation and outputs a prediction result indicating whether a fault has occurred; the method includes:

[0007] Long-term operational data during tractor operation is collected and preprocessed to create a dataset; the dataset includes operational data and corresponding fault labels.

[0008] The GRU model is trained iteratively using the dataset, enabling the model to learn the mapping relationship between the running data and the corresponding fault labels. The loss function is used to measure the difference between the model's predicted label and the true label, and then the gradient is calculated through backpropagation to adjust the model parameters to obtain the ideal model.

[0009] The actual tractor operating data is preprocessed and then input into the ideal model. The model then makes a judgment and prediction on the unknown operating data and outputs the result of whether or not there is a fault.

[0010] The long-term operational data includes, but is not limited to, data related to engine speed, water temperature, and oil pressure during normal tractor operation and under fault conditions, as well as classification labels for whether the operation status of each data time segment is faulty or not.

[0011] The dataset includes a training set, a validation set, and a test set; the training set is used for training, the validation set is used to verify the model's accuracy, and the test set is used for testing.

[0012] The data preprocessing requires segmenting the collected raw data and normalizing it to map the data to the interval [-1, 1], as shown in the formula: ,in For the normalized data, It is the raw data. It is the maximum value of this set of data. It is the minimum value of this set of data.

[0013] The GRU model includes an input layer, a GRU layer, a fully connected layer, and an output layer;

[0014] The input layer receives preprocessed time-series data segments;

[0015] The GRU layer consists of multiple GRU cells. Each GRU cell interacts with other cells through update and reset gates. The computation process of a GRU cell at time t is as follows:

[0016] 1) Calculate the reset door : ,in It is a weight matrix. It is the bias term, where σ is the Sigmoid function. This indicates that the hidden state from the previous moment will be restored. and current input To splice;

[0017] 2) Calculate the update gate : ,in It is a weight matrix. It is a bias term;

[0018] 3) Calculate the temporary hidden state : ,in It is a weight matrix. It is the bias term, and tanh is the hyperbolic tangent excitation function;

[0019] 4) Calculate the hidden state : ;

[0020] The fully connected layer flattens the hidden state output by the GRU layer and performs feature mapping, adjusting the data dimensions to fit the output layer;

[0021] The number of nodes in the output layer is the same as the number of fault categories. The Softmax activation function is used to convert the output of the fully connected layer into the predicted probability of each fault category, as shown in the formula: ,in For the sample belong Predicted probability of the category It is the first The input value for each node, where C is the total number of categories.

[0022] The model training uses the cross-entropy loss function to measure the difference between the model's predictions and the true labels, with the following formula: Where N is the number of samples and C is the total number of categories. It is a sample belong The true label of the category, sample belong The probability of the category.

[0023] During the iterative training process, the optimizer updates the model's weight parameters. After each training cycle, the model is evaluated using a validation set. The model's hyperparameters are adjusted based on the loss value and accuracy on the validation set to prevent overfitting. The training process is repeated until the model's performance on the validation set reaches its optimal level.

[0024] The model uses the Adams optimizer to update the model parameters, adaptively adjusting the learning rate to accelerate model convergence. Accuracy, recall, and F1 score are used as evaluation metrics to assess the model. When the model's performance on the test set meets the actual requirements, the trained GRU model is deployed in the tractor fault diagnosis system to achieve real-time diagnosis and early warning of tractor faults.

[0025] The accuracy rate: Recall rate: F1 value: ;in, To accurately predict faults, The correct prediction is normal. The error was incorrectly predicted as a fault. The incorrect prediction was correct.

[0026] A tractor fault diagnosis system based on a GRU recurrent neural network includes: multiple sensors installed on the tractor and a host computer. The multiple sensors include, but are not limited to, speed sensors, water temperature sensors, and pressure sensors. The host computer includes a control backend and a frontend interface. The frontend interface is used for human-computer interaction to collect model parameters and control commands input by the user, and to display the result data. The control backend is equipped with a memory and a processor. The memory stores the program, and when the processor loads the program, it executes the method steps described above, controls the multiple sensors to collect data according to the user commands, iteratively trains the model, predicts the tractor's operation in actual production, and outputs a prediction result indicating whether a fault has occurred.

[0027] The present invention has the following beneficial effects and advantages:

[0028] 1. This invention provides a tractor fault diagnosis method based on GRU recurrent neural network, which fully leverages the advantages of GRU in processing time-series data. Through scientific data acquisition, preprocessing, and rigorous model construction and training, it can achieve efficient and accurate diagnosis of tractor faults, providing reliable protection for the stable operation of agricultural production and helping agricultural mechanization move towards intelligence.

[0029] 2. This invention uses speed sensors, water temperature sensors, pressure sensors, etc., to collect relevant data such as engine speed, water temperature, and oil pressure of tractor engines in real time, and performs data segmentation and other processing on these data. The model is then trained using a GRU recurrent neural network model. Once the model performance meets the actual requirements, it is deployed to the tractor fault diagnosis system.

[0030] 3. During actual operation, the tractor's operating data is collected in real time, preprocessed, and then input into the trained GRU model. The category with the highest predicted probability output by the model is the type of fault that the tractor may currently have, thus realizing real-time diagnosis and early warning of tractor faults.

[0031] 4. This invention can ensure that the tractor is in a fault-free working state during use, realize the tractor's high-efficiency and high-quality operation, and improve the tractor's efficiency and lifespan. Attached Figure Description

[0032] Figure 1 This is a flowchart of the tractor fault diagnosis method based on GRU recurrent neural network of the present invention.

[0033] Figure 2 This is a schematic diagram of the network structure of the GRU cyclic unit of the present invention.

[0034] Figure 3 This is a schematic diagram of the model architecture based on the GRU recurrent neural network of the present invention. Detailed Implementation

[0035] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] 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 in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0037] like Figure 1 The flowchart of the tractor fault diagnosis method based on GRU recurrent neural network of the present invention is shown. As can be seen from the figure, the tractor fault diagnosis method based on GRU recurrent neural network mainly includes the following steps: data acquisition, data preprocessing, construction of GRU model network, model training and model evaluation.

[0038] First, real-time data such as engine speed, coolant temperature, and oil pressure of the tractor are collected using sensors including speed sensors, coolant temperature sensors, and pressure sensors. This data must cover both normal tractor operation and fault conditions, and the corresponding operating status label for each data segment must be accurately recorded to construct a complete dataset.

[0039] Then, the continuously acquired time-series data is segmented and normalized through data preprocessing. Data preprocessing requires segmenting the raw acquired data and normalizing it to map the data to the interval [-1, 1], as shown in the formula: ,in For the normalized data, It is the raw data. It is the maximum value of this set of data. It is the minimum value of this set of data.

[0040] The processed time-series data is then input into the GRU model for training. The cross-entropy loss function is used to measure the difference between the model's predictions and the actual values. The optimizer is then used to update the model parameters, adaptively adjusting the learning rate and accelerating model convergence.

[0041] Finally, the model is evaluated using accuracy, recall, and F1 score as evaluation metrics. When the model's performance on the test set meets the actual requirements, the trained GRU model is deployed in the tractor fault diagnosis system to achieve real-time diagnosis and early warning of tractor faults.

[0042] like Figure 2 The diagram shows the network structure of the GRU recurrent unit of the present invention. As can be seen from the diagram, the GRU recurrent unit network mainly includes: the hidden state of the previous time step. Current input Reset door Update door Temporary hidden state and hidden state .

[0043] Among them, the reset door It determines how new input information is combined with past memories; it does so by controlling the hidden state of the previous moment. How much information can be forgotten in order to address the problem of long-term dependence?

[0044] ,in It is a weight matrix. It is the bias term, where σ is the Sigmoid function. This indicates that the hidden state from the previous moment will be restored. and current input Then, the parts are assembled.

[0045] Update Gate Used to control the current hidden state. To what extent is there a hidden state from the previous moment? and the current temporary hidden state This determines how much the model retains past information and how much it accepts new information.

[0046] ,in It is a weight matrix. It is a bias term

[0047] Temporary hidden state Based on the current input and the hidden state of the previous moment, combined with the output of the reset gate, a temporary hidden state is calculated to update the hidden state of the current moment.

[0048] ,in It is a weight matrix. It is the bias term, and tanh is the hyperbolic tangent excitation function.

[0049] Hidden state The output of GRU combines past and current input information, serving as an abstract representation of the sequence data. The hidden states are continuously updated as time steps progress and can be used as input for the next time step, thus enabling long-term dependency modeling of the sequence data.

[0050]

[0051] like Figure 3 As shown in the figure, the model architecture based on the GRU recurrent neural network of the present invention is illustrated. As can be seen from the figure, the GRU recurrent neural network model architecture mainly consists of an input layer, a GRU neural network layer, a fully connected layer, and an output layer.

[0052] The input layer is the starting point for the model to receive data, responsible for receiving preprocessed raw data. This data typically has a specific format and dimensions, providing foundational information for subsequent model processing.

[0053] The GRU neural network layer, composed of multiple GRU units, extracts features and memorizes information from the input time-series data through mechanisms such as update gates and reset gates, effectively processing data with time-series characteristics.

[0054] In a fully connected layer, each neuron is connected to all neurons in the previous layer. Its main function is to integrate and transform the features output by the GRU layer. By remapping the features and adjusting the data dimensions to fit the input requirements of the output layer, it further refines the data features to support the final output decision.

[0055] The output layer is the final stage of the model. The number of nodes in the output layer is usually related to the task objective. The activation function converts the output of the fully connected layer into a specific prediction result, thereby completing the entire model's processing and response to the input data.

[0056] The model training uses the cross-entropy loss function to measure the difference between the model's predictions and the true labels. The formula is: Where N is the number of samples and C is the total number of categories. It is a sample The first of the real labels The values ​​that a class can take. sample belong The probability of the category.

[0057] The above formula is used for multi-class classification tasks, where the total number of classes C > 3. One-hot encoding is typically used when... Belongs to the kind, The rest For example, tractor malfunctions are categorized into three types: "engine malfunction," "transmission malfunction," and "braking system malfunction." When the sample... When it falls under the category of "engine malfunction", When the sample When it falls under the category of "braking system failure" When the sample When it falls under the category of "transmission failure", .

[0058] During training, the preprocessed dataset is divided into training, validation, and test sets. During training, the training set data is input into the GRU model, the loss between the model output and the true label is calculated, the gradient is calculated using backpropagation, and the optimizer updates the model's weight parameters. After each training epoch, the model is evaluated using the validation set. Based on the loss and accuracy on the validation set, the model's hyperparameters are adjusted to prevent overfitting. The training process is repeated until the model's performance on the validation set reaches its optimal level.

[0059] Model evaluation measures the model's fault diagnosis performance using metrics such as accuracy, recall, and F1 score.

[0060] Accuracy:

[0061] Recall rate:

[0062] F1 value:

[0063] in, To accurately predict faults, The correct prediction is normal. The error was incorrectly predicted as a fault. The incorrect prediction was correct.

[0064] This invention also provides another example: a tractor fault diagnosis system based on a GRU recurrent neural network, comprising: multiple sensors installed on the tractor and a host computer; the multiple sensors include, but are not limited to, speed sensors, water temperature sensors, and pressure sensors; the host computer includes a control backend and a front-end interface; the front-end interface is used for human-computer interaction to collect model parameters and control commands input by the user; the control backend is equipped with a memory and a processor; the memory stores a program; when the processor loads the program, it executes the above-described steps, controls the multiple sensors to collect data according to user commands, iteratively trains the model, predicts the tractor's operation during actual production, and outputs a prediction result indicating whether a fault has occurred.

[0065] This invention relates to a tractor fault diagnosis method and system based on GRU recurrent neural networks. It fully leverages the advantages of GRU in processing time-series data and, through a scientific and standardized data acquisition and preprocessing process, coupled with a rigorous and meticulous model building and training strategy, can achieve high-speed and accurate diagnosis of tractor faults. This will help lay a solid foundation for the stable operation and intelligent development of agricultural production in my country.

[0066] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should be considered within the scope of protection of the present invention.

Claims

1. A tractor fault diagnosis method based on a GRU recurrent neural network, characterized in that, The following steps are performed to obtain an ideal model for predicting tractor data during actual production operation and outputting a prediction result of whether a fault occurs; the method comprises: Collecting long-time sequence operation data during tractor operation and preprocessing to make a data set; the data set comprises operation data and corresponding fault labels; Using the data set data to iteratively train a GRU model, so that the model learns the mapping relationship between the operation data and the corresponding fault labels, and combines a loss function to measure the difference between the model prediction label result and the true value label, thereby reversely propagating the gradient to adjust the model parameters to obtain an ideal model; The actual collected tractor operation data is input into the ideal model after preprocessing, and unknown operation data is predicted to output a result of whether a fault occurs.

2. The GRU recurrent neural network-based tractor fault diagnosis method according to claim 1, characterized in that, The long-time sequence operation data includes but is not limited to the related data of the engine speed, water temperature, and oil pressure of the tractor during normal operation and under fault conditions, and the classification labels of the corresponding operation state fault or not for each data time slice. 3.The GRU recurrent neural network-based tractor fault diagnosis method according to claim 1, characterized in that, The data set includes a training set, a validation set, and a test set; the training set is used for training, the validation set is used for verifying the model accuracy, and the test set is used for testing. 4.The GRU recurrent neural network-based tractor fault diagnosis method according to claim 1, characterized in that, The data preprocessing needs to segment the original data collected and map the data to the interval [-1, 1] through normalization processing, the formula is: , wherein is the normalized data, is the original data, is the maximum value of the data, is the minimum value of the data. 5.The GRU recurrent neural network based tractor fault diagnosis method of claim 1, wherein, The GRU model comprises an input layer, a GRU layer, a full connection layer, and an output layer; The input layer receives the preprocessed time sequence data slice; The GRU layer is provided with a plurality of GRU units, each GRU unit interacts with other units through an update gate and a reset gate, and the calculation process of the GRU unit at time t is as follows: 1) Compute reset gate : where is a weight matrix, is a bias term, and σ is a Sigmoid function, denotes concatenation of the hidden state at the previous time step and the current input ; 2) Compute update gates : where is a weight matrix, is a bias term; 3) Compute temporary hidden state : where is a weight matrix, is a bias term, and tanh is the hyperbolic tangent activation function. 4) Compute hidden state : ; The full connection layer flattens and maps the hidden state output by the GRU layer, adjusts the data dimension to adapt to the output layer; The number of output layer nodes is consistent with the number of fault categories, and a Softmax activation function is used to convert the output of the full connection layer into a prediction probability of each fault category, and the formula is: wherein is the prediction probability of the sample belonging to the category, is the input value of the th node, and C is the total number of categories. 6.The GRU recurrent neural network based tractor fault diagnosis method according to claim 1, characterized in that, The model training uses a cross-entropy loss function to measure the difference between the model prediction result and the true label, and the formula is: Wherein, N is the number of samples, C is the total number of categories, is the true label of the sample belongs to category, The probability that the sample belongs to category.

7. The GRU recurrent neural network-based tractor fault diagnosis method according to claim 1, characterized in that, In the iterative training process, the optimizer is used to update the weight parameters of the model, and after each training period, the model is evaluated using the validation set, the model hyperparameters are adjusted according to the loss value and accuracy on the validation set to prevent model overfitting, and the training process is repeated until the performance of the model on the validation set reaches the optimal. 8.The GRU recurrent neural network based tractor fault diagnosis method of claim 1, wherein, The Adams optimizer is used to update the model parameters to realize adaptive adjustment of the learning rate and accelerate the convergence of the model; the accuracy, recall rate, and F1 value are used as evaluation indicators to evaluate the model; when the performance of the model on the test set meets the actual demand, the trained GRU model is deployed in the tractor fault diagnosis system to realize real-time diagnosis and early warning of tractor faults. 9.The GRU recurrent neural network based tractor fault diagnosis method of claim 8, wherein, The accuracy: , recall: , F1 value: ; wherein, correctly predicted failure, correctly predicted normal, incorrectly predicted failure, incorrectly predicted normal.

10. A tractor fault diagnosis system based on a GRU recurrent neural network, characterized by, It comprises: A plurality of sensors and an upper computer are arranged on the tractor; the plurality of sensors include but are not limited to a speed sensor, a water temperature sensor, and a pressure sensor; the upper computer comprises a control background and a front-end interface; the front-end interface is used for human-computer interaction to collect user input model parameters and control instructions, and display result data; the control background is provided with a storage and a processor; the storage stores a program; the processor loads the program to execute the method steps of any one of claims 1-9; the control background controls the plurality of sensors to collect data according to user instructions, iteratively trains the model, predicts the tractor during actual production operation, and outputs a prediction result of whether a fault occurs.

Images