Electric tractor insulation fault positioning method and system based on characteristic frequency analysis

Abstract

The invention discloses an electric tractor insulation fault positioning method and system based on characteristic frequency analysis, and belongs to the technical field of insulation fault diagnosis. An insulation monitoring device is designed according to the principle of an access resistance method, the insulation monitoring device is connected to a high-voltage battery pack of the electric tractor, and an insulation state monitoring signal of a high-voltage electrical system is obtained through the insulation monitoring device. Threshold value dynamic monitoring judgment is carried out, and then whether the high-voltage electrical system of the electric tractor breaks down or not is determined; carrying out characteristic frequency analysis on the monitoring signal during the fault, and extracting a corresponding frequency domain characteristic; obtaining a parallel symmetric point image by using a parallel symmetric point diagram algorithm; the method has the advantages of being high in fault detection response speed, high in recognition precision, high in adaptability and the like, and is suitable for intelligent diagnosis of the insulation fault of the high-voltage electrical system of the electric tractor.

Description

Technical Field

[0001] This invention belongs to the field of insulation fault diagnosis technology, specifically relating to a method for locating insulation faults in electric tractors based on characteristic frequency analysis. Background Technology

[0002] With the widespread use of electric tractors in agriculture, ensuring their absolute safety and reliability at the technical level is of paramount importance. Especially when a tractor experiences an insulation failure, unlike other faults, this directly endangers the lives of the driver and passengers. Therefore, while the definition of electric tractor safety encompasses multiple aspects, electrical safety is the most critical. Because electric tractors operate under harsh conditions, their high-voltage electrical components are frequently subjected to vibration, impact, and other factors, making the insulation of the high-voltage electrical system prone to problems, thus increasing the risk of malfunctions. Therefore, insulation testing technology plays a crucial role in ensuring the safety of electric tractors. Effective insulation testing technology not only ensures the stable operation of the electrical system but also promptly identifies potential safety hazards when problems occur, thereby maximizing the protection of the driver's life.

[0003] Currently, insulation testing technology has been widely studied and applied in the field of electric vehicles, but there is still relatively little specialized research on the complex working conditions and high-voltage system characteristics of electric tractors. Based on different testing principles, existing insulation testing methods can be mainly divided into signal injection method, bridge method and voltmeter method.

[0004] The voltmeter method of insulation testing utilizes the voltage distribution law to measure the voltage change between the system terminals and ground, and then calculates the insulation resistance. Based on this principle, numerous insulation testing devices and methods have emerged and have been widely studied and developed in practical applications. For example, Indian scholars Bukya et al. proposed the FPGA-based VFF-RLS algorithm based on the voltmeter method. This algorithm uses an adaptive filter algorithm to monitor the insulation resistance change of battery terminals in real time, effectively addressing errors caused by voltage and resistance changes and quickly and accurately monitoring the system's insulation status. TI has developed an insulation testing system called TIDA-01513, which, based on the voltmeter method, measures the voltage distribution characteristics between the high-voltage terminals and ground to monitor the insulation performance of multiple high-voltage lines. This system is suitable for high-voltage systems in complex electromagnetic environments, such as those equipped with DC / DC converters and motor control inverters.

[0005] The signal injection method detects insulation faults by injecting a known signal into the system and analyzing changes in the feedback signal. Based on this principle, various insulation testing devices and methods have been proposed and studied. For example, Yang Shengbing of Wuhan University of Technology designed a battery insulation testing system that uses voltage signal injection to detect the insulation resistance. This system primarily targets the insulation of positive and negative busbars. Bende's A-Isometer iso-F1 insulation testing device assesses insulation resistance by injecting a known voltage signal into the system and measuring changes in voltage and current, thereby enabling the monitoring of the insulation status of electrical systems.

[0006] The basic principle of the bridge method is to measure the insulation resistance of a system by utilizing the balance principle of the bridge circuit. It is divided into balanced and unbalanced forms. Based on this principle, various insulation detection devices and methods have been proposed. For example, Shen Yongpeng et al. designed an online insulation resistance detection circuit based on a high-precision instrumentation amplifier for an unbalanced bridge based on the principle of the unbalanced bridge method. Experimental results show that the system can control the insulation resistance measurement accuracy to below 5%. Sun Yang et al. designed an insulation detection system based on the principle of the bridge method and combined with the signal obtained by the Hall sensor, which can determine the insulation performance of the high-voltage system of electric vehicles under different working conditions. In summary, although insulation testing technology has achieved numerous research results, mostly focused on online insulation monitoring of electric vehicles, electric tractors often operate in harsh environments such as high humidity, high dust, and complex terrain during farmland operations, accompanied by severe mechanical vibrations. Existing testing technologies have limitations when applied to electric tractors, and these methods cannot accurately pinpoint the source of faults, still requiring manual inspection. Therefore, this application proposes an insulation fault location method suitable for the high-voltage electrical system of electric tractors. This method primarily employs a data-driven approach to effectively locate insulation faulty devices within the vehicle's high-voltage electrical system, thereby improving maintenance efficiency. Summary of the Invention

[0007] Purpose of the invention: To provide a method for locating insulation faults in electric tractors based on characteristic frequency analysis. Compared with other insulation detection methods for high-voltage electrical systems, this method is adaptable to the environmental changes under complex operating conditions of electric tractors, has strong environmental adaptability and real-time response capability, and can realize intelligent location and identification of different types of insulation faults, thereby improving maintenance efficiency.

[0008] Technical solution: To achieve the above objectives, the technical solution adopted by this invention is as follows: A method for locating insulation faults in electric tractors based on characteristic frequency analysis includes the following steps: Step S1: Design an insulation monitoring device based on the principle of the access resistance method, connect the insulation monitoring device to the high-voltage battery pack of the electric tractor, and obtain the insulation status monitoring signal of the high-voltage electrical system through the insulation monitoring device.

[0009] Step S2: Based on the collected insulation status monitoring signals, perform threshold dynamic monitoring and judgment to determine whether a fault has occurred in the high-voltage electrical system of the electric tractor.

[0010] Step S3: Based on the judgment result of step S2, perform characteristic frequency analysis on the monitoring signal at the time of the fault and extract the corresponding frequency domain features.

[0011] Step S4: The frequency domain features obtained in step S3 are mapped into a two-dimensional image using a parallel symmetric point image algorithm to obtain a parallel symmetric point image.

[0012] Step S5: Use the parallel symmetric point image obtained in step S4 as the input to the convolutional neural network (CNN) model, and based on the output of the model, realize the discrimination of insulation fault location and accurate identification of faulty equipment.

[0013] Preferably, the insulation monitoring device includes a sampling resistor. Sampling resistor two Control switch 1 Control switch two Voltmeter 1, Voltmeter 2, and sampling resistor 1 Control switch 1 Sampling resistor two Control switch two The sampling resistors are connected in sequence. Control switch two The voltmeter is connected in parallel with the sampling resistor, and is connected to both ends of the high-voltage battery pack of the electric tractor. The voltmeter two is connected in parallel across the sampling resistor two. The two ends of the control switch. Sampling resistor two The connecting wires between them are grounded. The sampling resistor 1 is measured using voltmeter 1 and voltmeter 2. Sampling resistor two The voltage signal is an insulation condition monitoring signal.

[0014] Preferred method: In step S2, the method for determining whether a fault has occurred in the high-voltage electrical system of the electric tractor by performing dynamic threshold monitoring based on the collected insulation status monitoring signal: when any sampling resistor... Sampling resistor two The voltage on Or voltage two Exceeding this threshold An insulation fault can be determined in the system, as shown in the following expression:

[0015] in, For resistors The voltage on, For resistors The voltage on, For the threshold, This refers to the voltage of the high-voltage battery pack.

[0016] Preferred method: The method for performing characteristic frequency analysis and extracting corresponding frequency domain features on the monitoring signal during the fault in step S3 is as follows: Based on the determination result of step S2, the characteristic frequency analysis is performed on the insulation monitoring signal collected during the insulation grounding fault, and the signals are sorted from largest to smallest according to their amplitude, and the first 6 frequency components and their corresponding amplitudes are extracted as frequency domain features.

[0017] Preferably, step S3 is as follows: The top 6 frequency components with higher amplitudes of the voltage signal on each sampling resistor are used as features. That is, the corresponding 6 frequency components and their corresponding amplitudes are extracted, resulting in a total of 12 features as feature vectors.

[0018] Let signal The spectrum was obtained after Fast Fourier Transform. The first 6 frequency components with the largest amplitude and its amplitude As a feature, the feature vector is represented as:

[0019]

[0020]

[0021] in, The eigenvector is composed of the first six frequency components with the largest amplitudes and their magnitudes. This represents the frequency values ​​of the six frequency components with the largest amplitude (energy). Represents frequency The corresponding amplitude values, This represents the amplitude value corresponding to the i-th frequency component. Indicates frequency At this point, the complex result obtained after the signal undergoes a Fourier transform. This represents the value of the i-th frequency component. This represents the spectrum obtained after the signal undergoes a Fast Fourier Transform (FFT). Indicates frequency.

[0022] Preferred method: The method in step S4 of mapping the acquired frequency domain features into a two-dimensional image using a parallel symmetric dot plot algorithm: Based on the frequency domain features extracted in step S3, a two-dimensional visualization mapping of the features is achieved using the parallel symmetric dot plot algorithm. The image corresponding to each fault mode presents a four-petal structure, where each pair of petals represents the frequency domain feature distribution on a resistor channel, thus visually and intuitively presenting the differences between various fault modes.

[0023] First, set the sampling resistor one The voltage on The extracted features are defined as follows Sampling resistor two Voltage two The extracted features are defined as follows ,Will , Features are used as input to the parallel symmetric point map. At this stage, the parallel symmetric point map consists of four petals, with the top two petals representing... Feature vector, represented by the two petals at the bottom. The eigenvectors, where the shape parameters of the parallel symmetric point map are calculated using the following formula:

[0024]

[0025]

[0026]

[0027] in, Represents the first parallel symmetric point in the graph. Mirror rotation angle of the channel feature vectors express Channel eigenvectors eigenvalue points Polar coordinate radius in a parallel symmetric point graph express Channel eigenvectors eigenpoints Clockwise deflection angle in a parallel symmetric point diagram express Channel eigenvectors eigenpoints Counterclockwise deflection angle in a parallel symmetric point diagram express In the channel feature vector Deflection angle gain at eigenpoints Indicates the lag factor. Indicates the angular magnification factor. express The maximum value in the channel feature vector, express The minimum value in the channel eigenvectors, the eigenvectors of channels 1 and 2 are The 3rd and 4th channel feature vectors are .

[0028] Preferred method: Based on the feature visualization completed in step S4, step S5 further introduces a convolutional neural network (CNN) model to classify and discriminate the generated parallel symmetric point map, and finds the optimal model parameters during the training process, thereby achieving accurate identification and location of the equipment where the system insulation fault occurs.

[0029] Preferably, the convolutional neural network (CNN) model includes an input layer, an image feature extraction layer, a fully connected layer, and an output layer connected in sequence. The image feature extraction layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence.

[0030] Preferably: the sampling resistor is one Sampling resistor two The resistance values ​​are equal.

[0031] Another objective of this invention is to provide an insulation fault location system for electric tractors based on feature frequency analysis, comprising an insulation monitoring device, a fault determination unit, a frequency domain feature extraction unit, a parallel symmetric dot plot algorithm unit, a convolutional neural network (CNN) model unit, and an output unit, wherein: The insulation monitoring device is used to connect to the high-voltage battery pack of the electric tractor and to obtain insulation status monitoring signals of the high-voltage electrical system.

[0032] The fault determination unit is used to perform threshold dynamic monitoring and determination based on the collected insulation status monitoring signals, thereby determining whether a fault has occurred in the high-voltage electrical system of the electric tractor.

[0033] The frequency domain feature extraction unit is used to perform feature frequency analysis on the monitoring signal during a fault based on the judgment result, and extract the corresponding frequency domain features.

[0034] The parallel symmetric point map algorithm unit is used to map the acquired frequency domain features into a two-dimensional image using the parallel symmetric point map algorithm, thereby obtaining a parallel symmetric point image.

[0035] The convolutional neural network (CNN) model unit is used to take the acquired parallel symmetric point image as input to the CNN model to obtain the localization result of the parallel symmetric point image.

[0036] The output unit is used to output the positioning result.

[0037] Compared with the prior art, the present invention has the following advantages: 1. The insulation monitoring device proposed in this invention acquires voltage signals in a passive manner, without the need for an external excitation source, which significantly reduces interference and modification to the original structure of the high-voltage system and facilitates rapid integration into the high-voltage electrical system of electric tractors. 2. This invention combines rapid time-domain threshold judgment with in-depth frequency-domain feature analysis, and then uses a CNN model to classify visualized images, achieving automated identification and precise location of system insulation faults. This method significantly improves the system's intelligent diagnostic level and effectively shortens fault diagnosis time. Attached Figure Description

[0038] Figure 1 Schematic diagram of the resistance method Figure 2 This is a schematic diagram of an insulation monitoring device according to an embodiment of the present invention.

[0039] Figure 3 This is a frequency domain schematic diagram of the insulation fault monitoring signal of the drive motor part in an embodiment of the present invention.

[0040] Figure 4 This is a frequency domain schematic diagram of the insulation fault monitoring signal of the compressor motor in an embodiment of the present invention.

[0041] Figure 5 This is a schematic diagram of the insulation fault monitoring signal of the power steering motor in an embodiment of the present invention.

[0042] Figure 6 This is a frequency domain diagram of the insulation fault monitoring signal for the negative busbar and drive motor in an embodiment of the present invention.

[0043] Figure 7 This is a frequency domain diagram of the insulation fault monitoring signal for the negative busbar and compressor sections in an embodiment of the present invention.

[0044] Figure 8 Frequency domain diagram of insulation fault monitoring signals at the negative busbar and power steering motor locations in an embodiment of the present invention.

[0045] Figure 9 This is a two-dimensional image schematic diagram of feature mapping according to an embodiment of the present invention.

[0046] Figure 10 A schematic diagram of the CNN network architecture of an embodiment of the invention.

[0047] Figure 11 This is a schematic diagram of a simulation model of the high-voltage electrical system of an electric tractor according to an embodiment of the present invention.

[0048] Figure 12This is a schematic diagram of the monitoring signals of the system under normal operation according to an embodiment of the present invention.

[0049] Figure 13 This is a schematic diagram of the monitoring signal during a system insulation fault according to an embodiment of the present invention.

[0050] Figure 14 This is a schematic diagram of single-point insulation fault injection according to an embodiment of the present invention.

[0051] Figure 15 This is a schematic diagram of bus insulation fault injection according to an embodiment of the present invention.

[0052] Figure 16 This is a schematic diagram of the single-point fault threshold detection results according to an embodiment of the present invention.

[0053] Figure 17 This is a schematic diagram of the multi-point fault threshold detection results according to an embodiment of the present invention.

[0054] Figure 18 This is a schematic diagram of the training sample Parallel-SDP in an embodiment of the present invention.

[0055] Figure 19 This is a schematic diagram of the insulation fault location results according to an embodiment of the present invention. Detailed Implementation

[0056] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0057] Example 1 Numerous research achievements have been made in insulation detection technology, mostly focusing on online insulation monitoring of electric vehicles. However, electric tractors often operate in harsh environments such as high humidity, high dust, and complex terrain during farmland operations, accompanied by severe mechanical vibrations. Existing detection technologies have limitations when applied to electric tractors, and these methods cannot accurately locate the source of faults, still requiring manual inspection. Therefore, this embodiment provides an insulation fault location method for electric tractors based on characteristic frequency analysis. It mainly adopts a data-driven approach to effectively locate insulation faulty devices in the high-voltage electrical system of the vehicle, improving maintenance efficiency. First, the electrical signals during system operation are acquired using a monitoring device, and the acquired monitoring signals are analyzed and processed to construct threshold indicators as the basis for judging insulation faults, thereby achieving rapid detection of insulation faults in the system. Subsequently, frequency domain analysis is further performed on the monitoring signals to extract characteristic frequency information related to insulation faults. To achieve a visual representation of fault characteristics, symmetric point graph theory is introduced to map the extracted frequency features into a two-dimensional image form. Finally, a Convolutional Neural Network (CNN) is used as the classification model to perform deep learning and recognition of image features, thereby achieving accurate localization and discrimination of different types of insulation faults, including the following steps: Step S1: Design an insulation monitoring device based on the principle of the resistance method. Connect the insulation monitoring device to the high-voltage battery pack of the electric tractor. Obtain the insulation status monitoring signal of the high-voltage electrical system through the insulation monitoring device. This insulation monitoring device is connected in parallel with other load equipment in the system. The specific steps are as follows: Step S101: Analyze the principle of the connected resistance method. The connected resistance method is an improvement on the balanced bridge method. Its core is to connect two known measuring resistors in series between the high-voltage battery pack and ground. By measuring relevant potential parameters in the system, the insulation resistance value is calculated, thereby determining the insulation status. Specifically, two measuring resistors are connected in series with the high-voltage battery pack. and And through the control switch and The system controls the connection status of the measuring resistor by switching it on and off. voltage on and voltage on The signal is analyzed and calculated; this method assesses the insulation resistance of the positive and negative busbars by measuring voltage changes. and The detection principle of the resistance method is as follows: Figure 1 As shown, when the switch Close, switch Disconnect the circuit and measure the two resistors. and Voltage value and The equation can be obtained as follows:

[0058] When switch Close, switch Disconnect the resistors and measure the voltage across them. and The equation can be obtained as follows:

[0059] By solving the above equations, the insulation resistance of the positive and negative busbars can be calculated. and The resistance value is used to evaluate the insulation status and performance of the system.

[0060] Step S102: The insulation monitoring device is based on the principle of the connected resistance method and is mainly used in the high-voltage electrical system of electric tractors. It is connected in parallel with other high-voltage load equipment. This device does not rely on external signal injection, but works through passive voltage detection. Its specific structure is as follows: Figure 2 As shown.

[0061] This refers to the voltage of the high-voltage battery pack. and Since the two resistors are sampling resistors, their resistance values ​​are equal. Therefore, when no equipment in the electrical system experiences an insulation fault, theoretically, the voltage across the two resistors should be equal and both should be DC. However, if the load equipment in the system experiences an insulation grounding fault, the voltage across the two resistors will inevitably be equal through the ground wire. and The voltage across the two resistors has an effect, and this invention uses the voltage signals across these two resistors as insulation monitoring signals for analysis.

[0062] The insulation monitoring device proposed in this invention is based on the principle of the connected resistance method and is mainly applied to the high-voltage electrical system of electric tractors. The device is connected in parallel with other high-voltage load equipment. This device does not rely on external signal injection but operates through passive voltage detection. The device uses the voltage signals collected across two resistors as the insulation monitoring signals. The insulation monitoring device includes a sampling resistor... Sampling resistor two Control switch 1 Control switch two Voltmeter 1, Voltmeter 2, and sampling resistor 1 Sampling resistor two The resistance values ​​are equal. The sampling resistor is... Control switch 1 Sampling resistor two Control switch two The sampling resistors are connected in sequence. Control switch two The voltmeter is connected in parallel with the sampling resistor, and is connected to both ends of the high-voltage battery pack of the electric tractor. The voltmeter two is connected in parallel across the sampling resistor two. The two ends of the control switch. Sampling resistor two The connecting wires between them are grounded. The sampling resistor 1 is measured using voltmeter 1 and voltmeter 2. Sampling resistor two The voltage signal is an insulation condition monitoring signal.

[0063] Step S2: Based on the collected insulation status monitoring signals, perform threshold dynamic monitoring and judgment to determine whether a fault has occurred in the high-voltage electrical system of the electric tractor.

[0064] Based on the insulation monitoring signal obtained in step S1, a threshold index is designed from the time-domain characteristics of the signal to quickly determine whether an insulation grounding fault has occurred in the system. This index achieves dynamic monitoring of the fluctuation amplitude of the monitoring signal by setting a reasonable voltage threshold. The specific steps are as follows: Under normal operating conditions, the two sampling resistors and Since they are connected in series, the voltage division should be kept consistent, and the sum of the two should equal the voltage of the high-voltage battery pack. When an insulation fault occurs in the system, theoretically, the voltage across at least one resistor will rise, far exceeding the steady-state value under normal operating conditions. Therefore, this invention defines 50% of the battery voltage as the threshold value. When any sampling resistor is Sampling resistor two The voltage on Or voltage two Exceeding this threshold An insulation fault can be determined in the system, as shown in the following expression:

[0065] in, For resistors The voltage on, For resistors The voltage on, For the threshold, This refers to the voltage of the high-voltage battery pack.

[0066] This invention designs a time-domain threshold index as the basis for determining whether a system insulation fault has occurred. This threshold criterion judges the collected insulation monitoring signals by setting a predefined critical value. If the signal exceeds the set range, it is considered an insulation fault, thereby achieving rapid fault detection and early warning.

[0067] Step S3: Based on the judgment result of step S2, perform characteristic frequency analysis on the monitoring signal at the time of the fault and extract the corresponding frequency domain features.

[0068] Based on the detection results of step S2, characteristic frequency analysis is performed on the insulation monitoring signals collected during the insulation grounding fault, and the frequency components with higher amplitudes and their corresponding amplitudes are extracted as frequency domain features for subsequent fault location.

[0069] In step S2, a preliminary distinction between normal and fault states is achieved based on the time-domain threshold detection results. Building upon this, Fourier transforms are performed on the insulation monitoring signals acquired under various fault modes to analyze their characteristics in the frequency domain. The specific steps are as follows: Firstly, regarding single-point faults, since the voltage waveforms across the two sampling resistors are identical, this patent document only analyzes one of them. Voltage waveform across the resistor When the fault is located in the drive motor, the frequency domain image of the monitoring signal is as follows: Figure 3 As shown in the image, the fault location is in the frequency domain of the compressor motor monitoring signal. Figure 4 As shown in the image, the fault location is in the power steering motor monitoring signal frequency domain image. Figure 5 As shown.

[0070] Next, we address two-point faults, where the voltage waveforms across the two sampling resistors differ during the fault, making their frequency domain characteristics more complex than those of single-point faults. This invention performs spectral analysis on the signals from both sampling resistors. The frequency domain images of the monitored signals when the fault location is at the drive motor and the negative bus are shown below. Figure 6 As shown, the frequency domain image of the monitoring signal when the fault location is the compressor motor and the negative bus is as follows. Figure 7 As shown, the frequency domain image of the monitoring signal when the fault location is the power steering motor and the negative bus is as follows. Figure 8 As shown.

[0071] By analyzing the frequency domain of the signal, it can be seen that the specific frequency components and their amplitudes can effectively reflect the location of the insulation fault. Therefore, the first 6 frequency components with higher amplitudes of the voltage signal on each sampling resistor are used as features, that is, the corresponding 6 frequency components and their corresponding amplitudes are extracted, totaling 12 features as feature vectors.

[0072] A method for performing characteristic frequency analysis and extracting corresponding frequency domain features from monitoring signals during a fault: Based on the determination result of step S2, the characteristic frequency analysis is performed on the insulation monitoring signals collected during the insulation grounding fault, and the signals are sorted from largest to smallest according to their amplitudes. The first 6 frequency components and their corresponding amplitudes are extracted as frequency domain features.

[0073] The top 6 frequency components with higher amplitudes of the voltage signal on each sampling resistor are used as features. That is, the corresponding 6 frequency components and their corresponding amplitudes are extracted, resulting in a total of 12 features as feature vectors.

[0074] Let signal The spectrum was obtained after Fast Fourier Transform. The first 6 frequency components with the largest amplitude and its amplitude As a feature, the feature vector is represented as:

[0075]

[0076]

[0077] in, The eigenvector is composed of the first six frequency components with the largest amplitudes and their magnitudes. This represents the frequency values ​​of the six frequency components with the largest amplitude (energy). Represents frequency The corresponding amplitude values, This represents the amplitude value corresponding to the i-th frequency component. Indicates frequency At this point, the complex result obtained after the signal undergoes a Fourier transform. This represents the value of the i-th frequency component. This represents the spectrum obtained after the signal undergoes a Fast Fourier Transform (FFT). Indicates frequency.

[0078] Step S4: The frequency domain features obtained in step S3 are mapped into a two-dimensional image using a parallel symmetric point image algorithm to obtain a parallel symmetric point image, thereby enhancing the visualization and pattern recognition capabilities of the features.

[0079] The method in step S4 for mapping the acquired frequency domain features into a two-dimensional image using a parallel symmetric dot plot algorithm is as follows: Based on the frequency domain features extracted in step S3, a two-dimensional visualization mapping of the features is achieved using the parallel symmetric dot plot algorithm. The image corresponding to each fault mode presents a four-petal structure, where each pair of petals represents the frequency domain feature distribution on a resistor channel, thus visually and intuitively presenting the differences between various fault modes.

[0080] First, set the sampling resistor one The voltage on The extracted features are defined as follows Sampling resistor two Voltage two The extracted features are defined as follows ,Will , The features, as input to the parallel symmetric point map, are equivalent to the feature vectors of two channels. These features are then used as input to the parallel symmetric point map, which at this stage consists of four petals. The top two petals represent... Feature vector, represented by the two petals at the bottom. The eigenvectors, where the shape parameters of the parallel symmetric point map are calculated using the following formula:

[0081]

[0082]

[0083]

[0084] in, Represents the first parallel symmetric point in the graph. Mirror rotation angle of the channel feature vectors express Channel eigenvectors eigenpoints Polar coordinate radius in a parallel symmetric point graph express Channel eigenvectors eigenpoints Clockwise deflection angle in a parallel symmetric point diagram express Channel eigenvectors eigenpoints Counterclockwise deflection angle in a parallel symmetric point diagram express In the channel feature vector Deflection angle gain at eigenpoints Indicates the lag factor. Indicates the angular magnification factor. express The maximum value in the channel feature vector, express The minimum value in the channel eigenvectors, the eigenvectors of channels 1 and 2 are The 3rd and 4th channel feature vectors are .

[0085] To more intuitively describe the composition of the parallel symmetrical point images used in the insulation positioning stage, this study uses two sinusoidal signals as examples, which respectively represent feature vectors. and The resulting parallel symmetric point image is as follows Figure 9 As shown.

[0086] Step S5: Use the parallel symmetric dot pattern (Parallel-SDP) image obtained in step S4 as input to the convolutional neural network (CNN) model, and based on the model's output, accurately identify the location of insulation faults and the faulty equipment.

[0087] Based on the feature visualization completed in step S4, step S5 further introduces a convolutional neural network (CNN) model to classify and discriminate the generated parallel symmetric point map, and finds the optimal model parameters during the training process, thereby achieving accurate identification and location of the equipment where the system insulation fault occurs.

[0088] Using the feature visualization image from step S4, a convolutional neural network (CNN) is employed as a classifier to intelligently locate insulation faults in the high-voltage electrical system of an electric tractor. A typical CNN neural network architecture is shown below. Figure 10 As shown, the Convolutional Neural Network (CNN) model comprises an input layer, an image feature extraction layer, a fully connected layer, and an output layer connected in sequence. The image feature extraction layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence. The 2D-CNN model extracts low-level features to obtain multi-level information from different signals. Based on this, it generates a highly abstract distributed feature representation, providing effective features for the parallel symmetrical point images studied in this application. Finally, the fully connected layer of the model integrates and processes the extracted features, ultimately obtaining the recognition result at the output layer.

[0089] The input layer of a 2D-CNN model is generally used to receive two-dimensional image information. In this invention, the input layer receives a parallel symmetrical point image. After the input layer passes the image to the convolutional layer, feature extraction begins. At this time, some common image features are extracted, such as image edges, textures, and color distribution. The mathematical expression of the convolutional layer is as follows:

[0090] in the formula Indicates the first In the layer Image features, This refers to convolutional layers. It refers to the first In the layer Bias of image features This represents the convolution operation. This represents the number of feature images. This refers to the number of convolution kernels. It is a neural network activation function.

[0091] Max pooling is the next step after convolution on an image. Its main function is to extract salient features from the image, thereby reducing the dimensionality of the features. According to the max pooling method, the maximum value in the data of each pooling region is used as the output of the pooling layer. The pooling layer can be expressed as follows:

[0092] in the formula It refers to the first In the layer The weight values ​​of the images, where It is a function that performs downsampling operations on the feature image.

[0093] The function of a fully connected layer is to synthesize the above features, generate a new distribution feature, and map it to the sample label domain. The neurons are interconnected layer by layer, providing the foundation for the SoftMax classifier below. The mathematical expression for a fully connected layer is as follows:

[0094] In the formula and This refers to the dimension that represents the input vector space and the output vector space. It is the first The first in the layer and the The weight values ​​of the connections between the outputs.

[0095] The function of the output layer is to display the classification results. The most widely used classifier is SoftMax, represented as:

[0096]

[0097] In the formula This refers to the dimension of the input feature map. It is the size of the convolution kernel. It's the fill amount. It's the step length.

[0098] Based on the feature visualization completed in step S4, step S5 further introduces a convolutional neural network (CNN) model to classify and discriminate the generated parallel symmetric point map, and finds the optimal model parameters during the training process, thereby achieving accurate identification and location of the equipment where the system insulation fault occurs.

[0099] In another embodiment of the present invention, an insulation fault location system for electric tractors based on feature frequency analysis is provided, comprising an insulation monitoring device, a fault determination unit, a frequency domain feature extraction unit, a parallel symmetric dot plot algorithm unit, a convolutional neural network (CNN) model unit, and an output unit, wherein: The insulation monitoring device is used to connect to the high-voltage battery pack of the electric tractor and to obtain insulation status monitoring signals of the high-voltage electrical system.

[0100] The fault determination unit is used to perform threshold dynamic monitoring and determination based on the collected insulation status monitoring signals, thereby determining whether a fault has occurred in the high-voltage electrical system of the electric tractor.

[0101] The frequency domain feature extraction unit is used to perform feature frequency analysis on the monitoring signal during a fault based on the judgment result, and extract the corresponding frequency domain features.

[0102] The parallel symmetric point map algorithm unit is used to map the acquired frequency domain features into a two-dimensional image using the parallel symmetric point map algorithm, thereby obtaining a parallel symmetric point image.

[0103] The convolutional neural network (CNN) model unit is used to take the acquired parallel symmetric point image as input to the CNN model to obtain the localization result of the parallel symmetric point image.

[0104] The output unit is used to output the positioning result.

[0105] The embodiments of the present invention will be described below with reference to specific experimental data, using methods such as... Figure 11The simulation model of the electrical system, built using Simulink in Matlab, is shown. The left side shows the insulation monitoring device designed in this study. This device is used to collect the voltage signals across the sampling resistors of each motor in real time during operation and to analyze the insulation status of the system. The DC bus voltage of the battery pack is set to 300V. The system includes three AC motors for different purposes: a drive motor, a compressor motor, and a power steering motor, each with a different inverter output frequency. Specifically, the output frequency of the drive motor inverter is set to 60Hz, the compressor motor inverter to 70Hz, and the power steering motor inverter to 80Hz.

[0106] To verify the effectiveness of the insulation monitoring device used, when no insulation grounding fault occurred, the signal acquired by the monitoring device, namely the two sampling resistors, was tested. and The voltages obtained above are denoted as follows: and The monitoring signal results are as follows Figure 12 As shown: When according to Figure 11 When injecting a single-point insulation fault at the location marked in red on the drive motor, detailed fault injection instructions are as follows: Figure 14 As shown, the voltage across the sampling resistor and Changes such as Figure 13 As shown.

[0107] The above figures demonstrate the sensitivity of the monitoring device to insulation faults. Under fault-free conditions, the voltage values ​​of the two sampling resistors are consistent and both are DC signals, which aligns with theoretical analysis. However, when an insulation fault occurs, the monitored voltage signal changes immediately and exhibits fluctuations at a specific frequency. This indicates that the monitoring device can effectively detect the occurrence of the fault and respond to its dynamic characteristics, providing data support for the insulation fault detection and location described below.

[0108] To verify the effectiveness of the fault location method, a database of two main fault types, single-point faults and multi-point faults, was established based on the constructed simulation model. Single-point faults are injected into the three-phase lines and busbars after the inverters corresponding to each motor. Although the motors are different, the injection method on the three-phase lines is the same. Therefore, this application takes the injection of single-point faults in drive motors as an example. Figure 14 As shown, a multi-point fault is a two-point fault based on the busbar, combined with the drive motor, power steering motor, and compressor motor in pairs. The busbar insulation fault injection method is as follows: Figure 15 As shown in Table 1, the sampling rate for each mode is 10kHz, the sampling time is 1 second, 50 samples are collected for each mode, and the mode distribution is shown in Table 1. 30 samples are used for training and 20 samples are used for testing. Table 1 Insulation Fault Modes and Distribution in High Voltage Systems

[0109] This section focuses on detecting insulation faults in different locations to verify the effectiveness of the proposed threshold index. The threshold can be calculated using the formula:

[0110] In the experiment, to improve the representativeness and feasibility of the threshold detection method, this invention selected typical samples from the training samples for demonstration. Since the single-point fault modes of different motor types, namely SP1, SP2, and SP3, have high similarity in the time-domain signals, to avoid redundant display, this invention selected the compressor motor as a representative of single-point faults for demonstration. However, when multiple insulation faults occur in the system, since the time-domain signal differences between MP4, MP5, and MP6 modes are small, this application selected insulation faults in the compressor motor and the negative bus as representatives of multi-point faults for demonstration. The changing trend of the proposed threshold index is as follows: Figure 16 and Figure 17 As shown.

[0111] As can be seen from the results in the figure, the voltage signal across the sampling resistor when a single-point fault occurs... and They exhibit a consistent trend and identical waveforms, with the voltage signal significantly exceeding the set threshold. When multiple faults occur, although the shape of the voltage waveform changes, and In The value is much greater than the threshold set in this application. Therefore, regardless of whether the insulation fault occurs in a single location (single-point fault) or multiple locations (multi-point fault), the threshold method can respond quickly when the voltage signal changes to a preset critical value, fully verifying the effectiveness of this indicator under different fault types.

[0112] After threshold detection, feature extraction was performed on a total of 300 data sets across 6 fault modes. Of these, 180 sets were used for training and 120 sets for testing. Each data set contains two columns, corresponding to the voltage signals across the two sampling resistors. Therefore, the features extracted from each data set also contain two columns, and the scale of each feature was kept consistent.

[0113] Using features as input to a parallel symmetric point image, one-dimensional features are transformed into a two-dimensional image. The three main parameter values ​​of the parallel symmetric point image are set as follows: , , The final feature visualization image is a snowflake with four petals. To further demonstrate the effectiveness of Parallel-SDP image recognition across different patterns, this invention selects the parallel symmetrical point image formed by the first and fifth samples from the training samples for illustration. Figure 18 As shown.

[0114] As can be seen from the table, firstly, the features extracted from insulation faults in different locations are transformed into different shapes on the two-dimensional images. Therefore, these images can be directly input into the localization model of this study. Secondly, the parallel symmetrical point images under the same mode are almost identical, indicating that the features extracted by this invention are relatively stable within the same mode.

[0115] In this invention, images obtained from training samples are used to train a CNN model. A total of 120 parallel symmetric point images are used as test samples, with 20 images per pattern. For the 2D-CNN model, parameter settings are crucial; different parameters can lead to significant differences in recognition accuracy. The key parameters of 2D-CNN mainly include two convolutional layers. and two pooling layers Size, number of convolution kernels Sum operation step size Among these parameters, the convolutional and pooling layers are matrix parameters, which are derived from the parameter set. :

[0116] in Therefore, the combination of convolutional layers and pooling layers can be represented as:

[0117] By selecting parameters and making multiple attempts, the study found that when the step size... Total number of convolutional kernels in convolutional layers At this point, the CNN model achieves its best performance. At this time, the settings of the convolution and pooling layer parameters have little impact on the model's performance. Therefore, the final stride and number of convolution kernels in this invention are set to 2 and 12, respectively.

[0118] This study introduces batch normalization technology, which not only solves the overfitting problem of the model but also improves the stability of the model training process. First, the output size of each layer needs to be calculated according to the formula. Initially, the input image size is set to... After processing by convolutional and pooling layers, the image size is reduced. Finally, fully connected layers and classification layers are used for classification. Since the threshold index can detect the fault, it is only necessary to locate the fault. There are a total of 6 fault modes in this invention, so the number of fully connected layers is 6. The output is the unknown, which is the category label of the test sample. The parameter table of CNN is shown in Table 2. Here, the activation function is a linear rectified unit. Max pooling layer is used to reduce the dimensionality of features and retain significant local features. The classifier is SoftMax. During the training process, the optimization strategy is the Adaptive Moment Estimation (Adam) method. The maximum number of training iterations is set to 100, and the initial learning rate is set to 0.001.

[0119] Table 2 CNN Parameters

[0120] The localization results of the test samples are as follows Figure 19 As shown, out of 120 test samples, only one sample was misidentified, with the dual-point fault of the busbar and drive motor being identified as a dual-point fault of the busbar and compressor motor. All other test samples were correctly identified, and the insulation fault location accuracy rate reached 99.16%, which fully demonstrates the effectiveness of the insulation fault location method proposed in this study.

[0121] This invention has the advantages of fast fault detection response, high identification accuracy and strong adaptability, and is suitable for intelligent diagnosis of insulation faults in high-voltage electrical systems of electric tractors.

[0122] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for electric tractor insulation fault location based on characteristic frequency analysis, characterized in that, It comprises the following steps: Step S1: According to the principle of access resistance method, the insulation monitoring device is designed, the insulation monitoring device is connected to the high-voltage battery pack of the electric tractor, and the insulation state monitoring signal of the high-voltage electrical system is obtained through the insulation monitoring device; Step S2: According to the collected insulation state monitoring signal, threshold dynamic monitoring and judgment are carried out, and then it is determined whether the electric tractor high-voltage electrical system has a fault; Step S3: Based on the determination result of step S2, the feature frequency analysis of the monitoring signal at the time of fault is carried out, and the corresponding frequency domain features are extracted; Step S4: The frequency domain features obtained in step S3 are mapped into a two-dimensional image by using a parallel symmetry point algorithm, and a parallel symmetry point image is obtained; Step S5: The parallel symmetry point image obtained in step S4 is taken as the input of the convolutional neural network CNN model, and according to the output result of the model, the insulation fault position is discriminated and the fault equipment is accurately identified.

2. The method for electric tractor insulation fault location based on characteristic frequency analysis according to claim 1, characterized in that: The insulation monitoring device comprises a sampling resistor R1, a sampling resistor R2, a control switch K1, a control switch K2, a voltmeter I, and a voltmeter II, the sampling resistor R1, the control switch K1, the sampling resistor R2, and the control switch K2 are connected in sequence, the sampling resistor R1 and the control switch K2 are connected across the high-voltage battery pack of the electric tractor, the voltmeter I is connected across the sampling resistor R1 in parallel, and the voltmeter II is connected across the sampling resistor R2 in parallel; the connecting wire between the control switch K1 and the sampling resistor R2 is grounded; the voltage signals of the sampling resistor R1 and the sampling resistor R2 measured by the voltmeter I and the voltmeter II are the insulation state monitoring signals.

3. The method according to claim 2, wherein: The method for determining whether the electric tractor high-voltage electrical system fails according to the collected insulation state monitoring signal in step S2, further comprises: when the voltage V1 or the voltage V2 on the sampling resistor R1 or the sampling resistor R2 exceeds the threshold value V T It can be determined that the system has insulation failure, and the expression is as follows: wherein V1 is the voltage across resistor R1, V2 is the voltage across resistor R2, V T is a threshold value, V bat is the voltage of the high voltage battery.

4. The method according to claim 3, wherein: The method for performing feature frequency analysis on the monitoring signal at the time of fault and extracting the corresponding frequency domain features in step S3: based on the determination result of step S2, the insulation monitoring signal collected during insulation grounding fault is analyzed for feature frequency, and the frequency components and their corresponding amplitudes are sorted from large to small, and the first 6 frequency components and their corresponding amplitudes are extracted as frequency domain features.

5. The method according to claim 4, wherein: The specific steps of step S3 are as follows: The first 6 frequency components with high amplitudes of the voltage signal on each sampling resistor are taken as features, i.e. the corresponding 6 frequency components and their corresponding amplitudes are extracted, and a total of 12 features are taken as feature vectors; Let the signal x(t) be transformed by the fast Fourier transform to get the frequency spectrum X(f), the first six largest amplitude frequency components f i and their amplitudes A i As a feature, the feature vector is then given by: F=[f1,f2,...,f6,A1,A2,...,A6] A i = |X(f i )| f i = argmax f |X(f)|i = 1,2,...,6 Wherein, F represents a feature vector, which is composed of the first 6 largest amplitude frequency components and their amplitude values, f1, f2,..., f6 represent the frequency values of the first 6 largest amplitude (energy) frequency components, A1, A2,..., A6 represent the amplitude values corresponding to the frequencies f1, f2,..., f6 in turn, A i represents the amplitude value corresponding to the i-th frequency component, X(f i ) represents the complex result obtained after Fourier transform of the signal at the frequency f i This point, f i represents the value of the i-th frequency component, X(f) represents the frequency spectrum obtained after fast Fourier transform (FFT) of the signal, and f represents the frequency.

6. The method according to claim 5, wherein: The method for mapping the obtained frequency domain features into a two-dimensional image by using a parallel symmetry point algorithm in step S4: Based on the frequency domain features extracted in step S3 as the input of the parallel symmetry point algorithm, the two-dimensional visualization of the features is realized; the images corresponding to each fault mode all present a four-petal structure, wherein each two petals represent the frequency domain feature distribution on one resistor channel, so that the differences between various fault modes are visually presented; Firstly, the feature of the voltage V1 on the sampling resistor R1 is defined as F v,1 , the feature of the voltage V2 on the sampling resistor R2 is defined as F v,2 , F v,1 , F v,2 are taken as the input of the parallel symmetric point graph, and the parallel symmetric point graph is composed of four petals in this stage, the upper two petals represent the F v,1 feature vector, and the lower two petals represent the F v,2 feature vector, wherein the shape parameter calculation formula of the parallel symmetric point graph is: θ S (k) = π / 4 + 2π(k - 1) / 4 k = 1,2,3,4 r k (t) = (x k,t -x k,min ) / (x k,max -x k,min ) ψ k (t) = θ S (k) + (x k,t+a - x k,min ) · ξ / (x k,max - x k,min ) φ k (t) = θ S (k) - (x k,t+a -x k,min ) · ξ / (x k,max -x k,min ) where θ S (k) represents the mirror rotation angle of the kth channel eigenvector in the parallel symmetry point graph, r k (t) represents the tth eigenvalue point x k,t the polar coordinate radius in the parallel symmetry point graph, ψ k (t) represents the tth eigenvalue point x k,t the clockwise deflection angle in the parallel symmetry point graph, φ k (t) represents the tth eigenvalue point x k,t the counterclockwise deflection angle in the parallel symmetry point graph, x k,t+a represents the deflection angle gain of the t+a eigenvalue point in the k channel eigenvector, a represents the lag factor, ξ represents the angle amplification factor, x k,max represents the maximum value in the k channel eigenvector, x k,min represents the minimum value in the k channel eigenvector, 1, 2 channel eigenvector is F v，1 , 3, 4 channel eigenvector is F v，2 .

7. The method according to claim 6, wherein: On the basis of completing feature visualization in step S4, step S5 further introduces a convolutional neural network CNN model to classify and discriminate the generated parallel symmetry point image, and in the training process, the optimal model parameters are found, so as to realize accurate identification and positioning of the system insulation fault equipment.

8. The method according to claim 7, wherein: The convolutional neural network (CNN) model comprises an input layer, an image feature extraction layer, a full connection layer and an output layer connected in sequence, and the image feature extraction layer comprises a convolutional layer one, a pooling layer one, a convolutional layer two and a pooling layer two connected in sequence.

9. The method according to claim 8, wherein: The resistance values of the sampling resistor one R1 and the sampling resistor two R2 are equal.

10. An electric tractor insulation fault location system based on signature frequency analysis characterized in that: The method comprises an insulation monitoring device, a fault determination unit, a frequency domain feature extraction unit, a parallel symmetric point graph algorithm unit, a convolutional neural network (CNN) model unit and an output unit, wherein: The insulation monitoring device is used for connecting to a high-voltage battery pack of the electric tractor, and acquiring an insulation state monitoring signal of the high-voltage electrical system through the insulation monitoring device; The fault determination unit is used for performing threshold dynamic monitoring determination according to the collected insulation state monitoring signal, and further determining whether a fault occurs in the high-voltage electrical system of the electric tractor; The frequency domain feature extraction unit is used for performing feature frequency analysis on the monitoring signal at the time of the fault based on the determination result, and extracting corresponding frequency domain features; The parallel symmetric point graph algorithm unit is used for mapping the acquired frequency domain features into a two-dimensional image by using a parallel symmetric point graph algorithm, to obtain a parallel symmetric point image; The convolutional neural network (CNN) model unit is used for taking the acquired parallel symmetric point image as an input of the convolutional neural network (CNN) model, to obtain a positioning result of the parallel symmetric point image; The output unit is used for outputting the positioning result.

Images