A multi-channel acquisition and control system based on FPGA

Abstract

This invention discloses a multi-channel acquisition and control system based on FPGA, including a multi-channel strain measurement subsystem, a loading control system, an FPGA-based multi-channel acquisition system, an inverter open-circuit fault diagnosis subsystem, an FPGA core control board, and a host computer for the measurement and control system. The multi-channel strain measurement subsystem measures the torque of the loading spindle; the loading control system controls the rotation of the loading spindle; the FPGA-based multi-channel acquisition system detects the current and voltage on the grid side and motor side of the wind turbine converter; and the inverter open-circuit fault diagnosis subsystem diagnoses and analyzes the current acquired by the FPGA-based multi-channel acquisition system. This invention can be used for the measurement and control of the mechanical and electrical components of a test bench to simulate test data of wind turbine generators under different wind conditions and grid faults, thereby optimizing and adjusting the design of the wind turbine generator.

Description

Technical Field

[0001] This invention relates to the field of wind power testing technology, specifically to a multi-channel acquisition and control system based on FPGA. Background Technology

[0002] With the continuous increase in installed wind power capacity, the operational reliability of wind turbine generators has an increasingly significant impact on the stability of power supply. In recent years, more in-depth research has been conducted on the loads experienced by wind turbine generators, and the reliability of wind power grid connection has been greatly improved. Research on wind turbine loads is fundamental to component-level testing of wind turbine drivetrains and is widely used in hardware-in-the-loop component testing of wind turbine generators; however, there is currently no system on the market for acquiring and controlling data during the testing process of wind turbine drivetrain components. Therefore, we propose an improvement, namely a multi-channel acquisition and control system based on FPGA. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0004] This invention discloses a multi-channel acquisition and control system based on FPGA, comprising a multi-channel strain measurement subsystem, a loading control system, an FPGA-based multi-channel acquisition system, an inverter open-circuit fault diagnosis subsystem, an FPGA core control board, and a host computer for the measurement and control system. The multi-channel strain measurement subsystem measures the torque of the loading spindle; the loading control system controls the rotation of the loading spindle; the FPGA-based multi-channel acquisition system detects the current and voltage on the grid side and motor side of the wind power converter; the inverter open-circuit fault diagnosis subsystem diagnoses and analyzes the current acquired by the FPGA-based multi-channel acquisition system; the multi-channel strain measurement subsystem, the FPGA-based multi-channel acquisition system, and the loading control system are all electrically connected to the FPGA core control board; the inverter open-circuit fault diagnosis subsystem is electrically connected to the host computer for the measurement and control system; and the FPGA core control board is electrically connected to the host computer for the measurement and control system.

[0005] The FPGA core control board is used to receive the torque measurement results of the multi-channel strain measurement subsystem. The FPGA core control board can control the operation of the loading control system based on the received torque measurement results.

[0006] As a preferred embodiment of the present invention, the multi-channel strain measurement subsystem includes a torque measuring device with a pressure sensor. The torque measuring device is mounted on the loading spindle and measures the torque of the loading spindle. The torque measuring device is equipped with a microprocessor, Bluetooth, and 4G transmission modules. The pressure sensor is electrically connected to the microprocessor via a first AD signal acquisition circuit, and the microprocessor is communicatively connected to the host computer of the measurement and control system via the Bluetooth and 4G transmission modules.

[0007] As a preferred embodiment of the present invention, the loading control system includes a loading system with a loading spindle. The loading system is equipped with a drive motor that drives the loading spindle to rotate. The drive motor is controlled by a frequency converter. The frequency converter is electrically connected to the FPGA core control board. The FPGA core control board controls the drive motor to perform frequency conversion operation by controlling the frequency converter.

[0008] As a preferred embodiment of the present invention, the FPGA-based multi-channel acquisition system includes an FPGA data processing terminal and current and voltage sensors for detecting the current and voltage on the grid side and the motor side of the wind power converter. The FPGA data processing terminal is electrically connected to the current and voltage sensors via a second AD signal acquisition module. The FPGA data processing terminal is electrically connected to the FPGA core control board and the inverter open-circuit fault diagnosis subsystem via Ethernet.

[0009] As a preferred technical solution of the present invention, the voltage measurement results of the wind power converter grid side and motor side acquired by the FPGA-based multi-channel acquisition system are calibrated. The calibration method is to use a multimeter to simultaneously measure the voltage value, and then perform a linear fitting of the voltage value obtained by the multimeter synchronously with the voltage value acquired by the FPGA-based multi-channel acquisition system using a first-order polynomial method to obtain the fitted function relationship.

[0010] As a preferred embodiment of the present invention, the diagnostic method of the inverter open-circuit fault diagnostic subsystem includes the following steps.

[0011] Step 1: Preprocess the acquired current signal to obtain a current signal with noise interference removed;

[0012] Step 2: Then convert the noise-free current signal into a threshold-free recursive graph;

[0013] Step 3: Implement self-learning of nonlinear features in the thresholdless recursive graph of the current signal through a convolutional neural network, and establish a classification model for the current signal;

[0014] Step 4: Use the classification model of current signals to identify inverter faults in subsequent current signals.

[0015] As a preferred technical solution of the present invention, a simulation model of an NPC-type three-level inverter is built before step 1. The simulation model mainly includes three-phase bridge arms A, B, and C and an IGBT switching controller. Each phase bridge arm consists of four IGBT power switches and two clamping diodes. The IGBTs in the bridge arm are turned on and off by using sinusoidal pulse width modulation to realize the DC to AC inversion process. By controlling the on and off of the IGBTs, the output is a sinusoidal pulse width, and the area of ​​the output digital signal pulse voltage is equal to the area of ​​the standard grid sine wave within the same time. Different open-circuit fault states are simulated by controlling the trigger signal of the IGBTs, thereby measuring different current signals.

[0016] As a preferred embodiment of the present invention, the method for simulating different open-circuit fault states by controlling the trigger signal of the IGBT is as follows: taking the open-circuit fault of phase A of a three-level inverter as an example, SS1 is created as the conduction module of switch S1, and a constant value of 0 is created as the open-circuit fault module of switch S1.

[0017] When the SS1 conduction module is connected to switch S1, the output is an SPWM pulse; when the constant value 0 is connected to switch S1 as an open circuit fault module, the output is a constant value 0, which is equivalent to not providing a pulse, and is used to simulate the open circuit fault of the IGBT high power switching transistor.

[0018] As a preferred embodiment of the present invention, the simulation model of the NPC-type three-level inverter classifies faults, including the following situations.

[0019] a. All IGBT power switches are operating normally without any faults.

[0020] b. Only one IGBT power switch failed;

[0021] c. Two IGBT power switches on different half-bridges on the same bridge arm fail simultaneously;

[0022] d. Two IGBT power switches on the same half-bridge of the same bridge arm fail simultaneously;

[0023] e. Two IGBT power switches on the same half-bridge on different bridge arms fail simultaneously;

[0024] f. Cross connection means that two IGBT power switches on different bridge arms and different half bridges fail at the same time.

[0025] The IGBT power switch faults in b to d all occur on the same bridge arm and only affect the single-phase current in the three-phase current, so b to d are defined as simple faults; the IGBT power switch faults in e to f occur on different bridge arms and affect the two-phase current in the three-phase current, so e to f are defined as complex faults; different open-circuit fault states are simulated by controlling the IGBT trigger signals, and the waveforms of the current signals under each open-circuit fault state are obtained.

[0026] As a preferred technical solution of the present invention, the method for self-learning nonlinear features in a thresholdless recursive graph of a current signal through a convolutional neural network first introduces a residual network structure into a traditional convolutional neural network with convolutional layers, pooling layers and average pooling layers to form an improved convolutional neural network. In the residual network structure, an identity mapping is added through shortcut connections, and the difference between the target value H(X) and x is used as the learning target.

[0027] In step 3, the collected current signals are divided into training and test sets, and these sets are input into a classification model for the current signals to obtain classification results. The macro-average is used as the evaluation index for classification performance, and the relevant formulas for calculating average recall and average precision are as follows:

[0028]

[0029]

[0030] In the formula, Macro_R represents the average recall rate and Macro_R represents the average precision rate.

[0031] The beneficial effects of this invention are:

[0032] 1. This FPGA-based multi-channel acquisition and control system consists of a multi-channel strain measurement subsystem, a loading control system, an FPGA-based multi-channel acquisition system, an inverter open-circuit fault diagnosis subsystem, an FPGA core control board, and a measurement and control system host. The multi-channel strain measurement subsystem measures the torque of the loading spindle. The loading control system controls the rotation of the loading spindle. The FPGA-based multi-channel acquisition system detects the current and voltage on the grid side and motor side of the wind turbine converter. The inverter open-circuit fault diagnosis subsystem diagnoses and analyzes the current acquired by the FPGA-based multi-channel acquisition system. It can monitor and control variables in the transmission chain mathematical model of the wind power test bench, and can be used for the measurement and control of the mechanical and electrical parts of the test bench. This allows for the simulation of test data from wind turbine generators under different wind conditions and grid faults, thereby optimizing the design of wind turbine generators and promoting their development towards higher power and higher reliability.

[0033] 2. The diagnostic method of the inverter open-circuit fault diagnosis subsystem in this FPGA-based multi-channel acquisition and control system involves converting the current signal waveform into a thresholdless recursive graph. This maps the nonlinear characteristics of the current signal to a two-dimensional image. By not setting a threshold during the conversion process, the nonlinear information in the original signal waveform can be better preserved. Subsequently, a convolutional neural network is constructed to train the thresholdless recursive graph, establishing a classification model for inverter open-circuit fault diagnosis. In this invention, the model, through convolution and stacked network layers, can self-learn more detailed features from the thresholdless recursive graph, thus achieving better fault diagnosis results. Attached Figure Description

[0034] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0035] Figure 1 This is a control system diagram of a multi-channel acquisition and control system based on FPGA according to the present invention;

[0036] Figure 2 This is a flowchart illustrating the diagnostic method of the inverter open-circuit fault diagnosis subsystem of the present invention.

[0037] Figure 3 This is a schematic diagram of the structure of a simulation model of a multi-channel acquisition and control system based on FPGA according to the present invention;

[0038] Figure 4 This is a schematic diagram of the residual network structure of a multi-channel acquisition and control system based on FPGA according to the present invention;

[0039] Figure 5 This is a schematic diagram of the structure of a convolutional neural network for a multi-channel acquisition and control system based on FPGA according to the present invention;

[0040] Figure 6 This is a classification result diagram of different algorithms for a multi-channel acquisition and control system based on FPGA according to the present invention;

[0041] Figure 7 This is a result diagram of the confusion matrix plus wavelet feature of a multi-channel acquisition and control system based on FPGA according to the present invention;

[0042] Figure 8 This is a result diagram of a random forest thresholdless recursive graph and convolutional neural network for a multi-channel acquisition and control system based on FPGA of the present invention;

[0043] Figure 9 This is a schematic diagram of the structure of a single-phase bridge arm of an NPC-type three-level inverter based on an FPGA-based multi-channel acquisition and control system according to the present invention.

[0044] Figure 10 This invention relates to an IGBT switching state diagram of a multi-channel acquisition and control system based on FPGA.

[0045] Figure 11 This is a waveform diagram of the voltage waveform of an NPC-type three-level inverter in a fault-free state, which is based on an FPGA-based multi-channel acquisition and control system according to the present invention.

[0046] Figure 12 This is a waveform diagram of the voltage waveform of an NPC-type three-level inverter in a multi-channel acquisition and control system based on FPGA of the present invention when it is faulty.

[0047] Figure 13 This is a normal state diagram of an IGBT in a multi-channel acquisition and control system based on FPGA according to the present invention;

[0048] Figure 14 This invention relates to an IGBT open-circuit state diagram of a multi-channel acquisition and control system based on FPGA.

[0049] Figure 15 This invention relates to an open-circuit fault current waveform of a single IGBT in a multi-channel acquisition and control system based on FPGA.

[0050] Figure 16 This invention relates to the fault current waveform of two IGBTs simultaneously open-circuited in a multi-channel acquisition and control system based on FPGA.

[0051] In the diagram: 1. Multi-channel strain measurement subsystem; 2. Loading control system; 3. FPGA-based multi-channel acquisition system; 4. Inverter open-circuit fault diagnosis subsystem; 5. FPGA core control board; 6. Measurement and control system host computer; 7. Pressure sensor; 8. Torque measurement device; 9. Microprocessor; 10. Bluetooth and 4G transmission module; 11. First AD signal acquisition circuit; 12. FPGA data processing terminal; 13. Current and voltage sensor; 14. Second AD signal acquisition module; 16. Loading spindle; 17. Loading system; 18. Drive motor; 19. Control inverter. Detailed Implementation

[0052] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0053] Example: Figure 1-16 As shown, the present invention provides a multi-channel acquisition and control system based on FPGA, such as... Figure 1 As shown, the system includes a multi-channel strain measurement subsystem 1, a loading control system 2, an FPGA-based multi-channel acquisition system 3, an inverter open-circuit fault diagnosis subsystem 4, an FPGA core control board 5, and a measurement and control system host computer 6. The multi-channel strain measurement subsystem 1 measures the torque of the loading spindle; the loading control system 2 controls the rotation of the loading spindle; the FPGA-based multi-channel acquisition system 3 detects the current and voltage on the grid side and motor side of the wind power converter; the inverter open-circuit fault diagnosis subsystem 4 diagnoses and analyzes the current acquired by the FPGA-based multi-channel acquisition system 2; the multi-channel strain measurement subsystem 1, the FPGA-based multi-channel acquisition system 3, and the loading control system 2 are all electrically connected to the FPGA core control board 5; the inverter open-circuit fault diagnosis subsystem 4 is electrically connected to the measurement and control system host computer 6; and the FPGA core control board 5 is electrically connected to the measurement and control system host computer 6.

[0054] The FPGA core control board 5 is used to receive the torque measurement results from the multi-channel strain measurement subsystem 1. Based on the received torque measurement results, the FPGA core control board 5 can control the operation of the loading control system 2. The most significant feature of the FPGA core control board 5 is its excellent real-time performance, enabling true parallel execution. FPGAs have the advantage of parallel computing, making them highly suitable for parallel acquisition of high-speed signals and real-time data processing. FPGA processors possess excellent data parallel processing capabilities, thus exhibiting superior real-time performance. The hardware data processing will employ parallel processing methods, which can significantly improve the calculation speed.

[0055] The multi-channel strain measurement subsystem 1 includes a torque measuring device 8 with a pressure sensor 7. The torque measuring device 8 is mounted on the loading spindle and measures the torque of the loading spindle. The torque measuring device 8 is equipped with a microprocessor 9, a Bluetooth and 4G transmission module 10. The pressure sensor 7 is electrically connected to the microprocessor 9 via a first AD signal acquisition circuit 11. The microprocessor 9 is communicatively connected to the host computer 6 of the measurement and control system via the Bluetooth and 4G transmission module 10.

[0056] The loading control system 2 includes a loading system 17 with a loading spindle 16. The loading system 17 is equipped with a drive motor 18 that drives the loading spindle 16 to rotate. The drive motor 18 is controlled by a frequency converter 19. The frequency converter 19 is electrically connected to the FPGA core control board 5. The FPGA core control board 5 controls the drive motor 18 to perform frequency conversion operation by controlling the frequency converter 19.

[0057] The FPGA-based multi-channel acquisition system 3 includes an FPGA data processing terminal 12 and a current and voltage sensor 13 for detecting the current and voltage on the grid side and the motor side of the wind power converter. The FPGA data processing terminal 12 is electrically connected to the current and voltage sensor 13 via a second AD signal acquisition module 14. The FPGA data processing terminal is electrically connected to the FPGA core control board 5 and the inverter open circuit fault diagnosis subsystem 4 via Ethernet.

[0058] The voltage measurement results of the wind power converter on the grid side and the motor side acquired by the FPGA-based multi-channel acquisition system are calibrated. The calibration method is to use a multimeter to measure the voltage value simultaneously, and then perform a linear fit between the voltage value measured by the multimeter and the voltage value acquired by the FPGA-based multi-channel acquisition system using a first-order polynomial method to obtain the fitted function relationship.

[0059] This invention relates to a current signal analysis method based on thresholdless recursive graphs and convolutional neural networks, comprising the following steps, such as... Figure 2 As shown,

[0060] Step 1: Preprocess the acquired current signal to obtain a current signal with noise interference removed;

[0061] Step 2: Then convert the noise-free current signal into a threshold-free recursive graph;

[0062] Step 3: Implement self-learning of nonlinear features in the thresholdless recursive graph of the current signal through a convolutional neural network, and establish a classification model for the current signal;

[0063] Step 4: Use the classification model of current signals to identify inverter faults in subsequent current signals.

[0064] Taking a single-phase bridge arm of an NPC (diode neutral clamp) three-level inverter as the research object, the current path within it is analyzed, as shown in the schematic diagram below. Figure 9 As shown. A three-level inverter refers to an inverter whose output voltage has three voltage levels: +Udc / 2, 0, and -Udc / 2. The inversion process is achieved through these three voltage levels. In an NPC-type three-level inverter, these three voltage levels are achieved by the switching combinations of four IGBTs in a single bridge arm. Each phase bridge arm has three operating states: P (high level), 0 (medium level), and N (low level). The corresponding IGBT switching states are shown below. Figure 9 As shown.

[0065] A simulation model of an NPC-type three-level inverter is built between steps 1, such as... Figure 3 As shown, the simulation model mainly includes three-phase bridge arms (A, B, and C) and an IGBT switch controller. Each phase bridge arm consists of four IGBT power switches and two clamping diodes. The IGBTs in the bridge arm are controlled to turn on and off using sinusoidal pulse width modulation (PWM) to achieve the DC-to-AC inversion process. By controlling the IGBTs' on / off state, the output is a sinusoidal pulse width, and the area of ​​the output digital signal pulse voltage is equal to the area of ​​the standard power grid sine wave within the same time interval. Different open-circuit fault states are simulated by controlling the IGBT trigger signals, thereby measuring different current signals.

[0066] The method of simulating different open-circuit fault states by controlling the IGBT trigger signal is illustrated by taking an open-circuit fault in phase A of a three-level inverter as an example. Figure 12 and Figure 13 Create SS1 as the conduction module for switch S1, and create a constant value of 0 as the open-circuit fault module for switch S1.

[0067] When the SS1 conduction module is connected to switch S1, the output is an SPWM pulse; when the constant value 0 is connected to switch S1 as an open circuit fault module, the output is a constant value 0, which is equivalent to not providing a pulse, and is used to simulate the open circuit fault of the IGBT high power switching transistor.

[0068] The simulation model of the NPC-type three-level inverter is classified into the following fault types:

[0069] a. All IGBT power switches are operating normally without any faults.

[0070] b. Only one IGBT power switch failed;

[0071] c. Two IGBT power switches on different half-bridges on the same bridge arm fail simultaneously;

[0072] d. Two IGBT power switches on the same half-bridge of the same bridge arm fail simultaneously;

[0073] e. Two IGBT power switches on the same half-bridge on different bridge arms fail simultaneously;

[0074] f. Cross connection means that two IGBT power switches on different bridge arms and different half bridges fail at the same time.

[0075] like Figure 11 This is a waveform diagram of the voltage waveform of an NPC-type three-level inverter when there is no fault. Figure 11 This is a waveform diagram of the voltage waveform when an NPC-type three-level inverter malfunctions.

[0076] like Figure 15 The diagram shows the current waveform of a single IGBT open-circuit fault. When a single IGBT experiences an open-circuit fault, the amplitude or phase of the inverter's output current waveform will change differently. When a single power switch S1 or S4 experiences an open-circuit fault, the output sine wave will shift towards the negative or positive half-axis because one bridge arm (upper / lower) is open-circuited while the other bridge arm (upper / lower) is operating normally. When a single power switch S2 or S3 experiences an open-circuit fault, the output sine wave will only have the lower or upper half-cycle because the clamping diodes directly truncate the negative or positive half-axis waveform. As shown in Figure 15, when two switching transistors S1 and S2, and S2 and S4 have open-circuit faults, the waveforms are all below the x-axis; when S1 and S3, and S3 and S4 have faults at the same time, the waveforms are all above the x-axis; when S2 and S3 have open-circuit faults at the same time, the output current is close to 0 and there is no waveform output; when S1 and S4 have open-circuit faults at the same time, the fault waveforms change little compared to normal operation, the most obvious change is the reduction in amplitude, and careful observation of the waveform changes reveals a slight distortion at the intersection of the positive and negative half-axis.

[0077] By comparison Figure 15 and 16 It can be observed that the inverter output current waveform when S2 is open is consistent with the waveform when S1 and S2 are open simultaneously, and the inverter output current waveform when S3 is open is consistent with the waveform when S3 and S4 are open simultaneously. Therefore, these two are classified as the same type of fault. Accurate fault identification requires the use of other fault diagnosis methods, typically achieved by measuring the voltage of the upper and lower bridge arms. The dynamic changes in these voltage values ​​determine whether it is a single IGBT open circuit or both open simultaneously. Since the amplitude of the current waveform changes with the output power of the wind turbine, the biggest difference in the current waveform corresponding to an inverter open circuit fault, besides the difference in shape, lies in the change in amplitude. For example… Figure 16The main difference between the output current waveforms when S1 and S4 are simultaneously open and when S3 and S4 are simultaneously open lies in their amplitude. Apart from this, the waveform changes little when both S1 and S4 are simultaneously open compared to when they are not.

[0078] Among them, the IGBT power switch faults in b to d all occur on the same bridge arm and only affect the single-phase current in the three-phase current, so b to d are defined as simple faults; the IGBT power switch faults in e to f occur on different bridge arms and affect the two-phase current in the three-phase current, so e to f are defined as complex faults; different open-circuit fault states are simulated by controlling the trigger signal of the IGBT, and the waveform diagram of the current signal under each open-circuit fault state is obtained.

[0079] like Figure 4 and Figure 5 As shown, the method for self-learning nonlinear features in a thresholdless recursive graph of a current signal using a convolutional neural network first introduces a residual network structure into a traditional convolutional neural network with convolutional layers, pooling layers, and average pooling layers to form an improved convolutional neural network. In the residual network structure, an identity mapping is added through shortcut connections, and the difference between the target value H(X) and x is used as the learning target.

[0080] In step 3, the collected current signals are divided into training and test sets. When training the classification model, the input samples are typically data with large amplitude variations and obvious characteristics. After an open-circuit fault occurs in the inverter, the amplitude of the collected current signals will fluctuate, and may even experience severe distortion. Based on the established simulation model and considering the differences between doubly-fed wind turbine generators of different power ratings, different load powers are set. Specific sample data includes current signal waveforms at 690V / 100KW, 690V / 200KW, 690V / 300KW, 690V / 400KW, 690V / 700KW, 690V / 1MW, 690V / 1.2MW, 690V / 1.5MW, and 690V / 2MW. The training and test sets are then input into the current signal classification model to obtain the classification results. The macro-average is used as the evaluation index for classification performance, and the relevant formulas for calculating the average recall and average precision are as follows:

[0081]

[0082]

[0083] In the formula, Macro_R is the average recall rate and Macro_P is the average precision rate.

[0084] To verify the effectiveness of the thresholdless recursive graph and deep learning algorithm used in this chapter for inverter open-circuit fault diagnosis, a comparative experiment is conducted between the proposed fault diagnosis algorithm and commonly used classification algorithms. The comparison focuses on a wavelet feature + random forest classification model. This model first requires a four-level wavelet packet decomposition of the waveform data, and then uses the obtained feature values ​​for model training and prediction. The comparison results are as follows: Figure 6 As shown, compared to the wavelet feature + random forest classification method, the thresholdless recursive graph combined with ResNet50 achieves better classification results, with an accuracy of 96.67%. Its average precision and average recall are also higher than the wavelet feature + random forest classification model, by 8.09% and 11.86%, respectively. To more intuitively illustrate the difference in classification performance between the two methods...

[0085] Depend on Figure 7 and Figure 8 It can be seen that, compared with the random forest algorithm, the thresholdless recurrent graph + ResNet50 classification model leads in the recognition accuracy of all eight types of current signals, improving by 20%, 10%, 8%, 12%, 13%, 18%, 18%, and 12% respectively, indicating that this classification model has better classification performance. Among them, the improvement is most significant for Normal, S1_S3, and S1_S4 open-circuit faults, greatly reducing the fault diagnosis error rate. Therefore, the thresholdless recurrent graph + convolutional neural network classification model can effectively improve the accuracy of fault diagnosis and has better fault location diagnosis performance. Figure 8 It can be seen that in the classification model using a thresholdless recursive graph + ResNet50, compared with the classification results of other fault types, its prediction accuracy for the Normal and S1_S4 states is slightly lower, at 93% and 92% respectively, but still higher than the wavelet feature + random forest algorithm. The main reason for the identification error is that the waveforms of the two states are quite similar, both exhibiting the characteristics of standard sine waves, making it difficult for the classifier to accurately determine their fault type, thus leading to incorrect judgment.

[0086] This invention transforms a current signal waveform into a thresholdless recursive graph, mapping the nonlinear characteristics of the current signal onto a two-dimensional image. By not setting a threshold during the transformation, the nonlinear information in the original signal waveform can be better preserved. A convolutional neural network is then constructed to train the thresholdless recursive graph, establishing a classification model for inverter open-circuit fault diagnosis. In this invention, the model, through convolution and stacked network layers, can self-learn more detailed features from the thresholdless recursive graph, thus achieving better fault diagnosis results.

[0087] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A FPGA-based multi-channel acquisition and control system; characterized in that: The application relates to a wind power converter open circuit fault diagnosis system which comprises a multichannel strain measurement subsystem (1), a loading control system (2), a multichannel FPGA-based acquisition system (3), an inverter open circuit fault diagnosis subsystem (4), an FPGA core control board (5) and a measurement and control system host computer (6); the multichannel strain measurement subsystem (1) is used for measuring the torque of a loading spindle; the loading control system (2) is used for controlling the rotation of the loading spindle; the multichannel FPGA-based acquisition system (3) is used for detecting the current and voltage of the grid side and the motor side of a wind power converter; the inverter open circuit fault diagnosis subsystem (4) is used for diagnosing and processing the current collected by the multichannel FPGA-based acquisition system (3); the multichannel strain measurement subsystem (1), the multichannel FPGA-based acquisition system (3) and the loading control system (2) are electrically connected with the FPGA core control board (5); the inverter open circuit fault diagnosis subsystem (4) is electrically connected with the measurement and control system host computer (6); the FPGA core control board (5) is electrically connected with the measurement and control system host computer (6). The FPGA core control board (5) is used for receiving the torque measurement results of the multichannel strain measurement subsystem (1), and the FPGA core control board (5) can control the loading control system (2) to work according to the received torque measurement results.

2. The FPGA-based multi-channel acquisition and control system of claim 1, wherein, The multichannel strain measurement subsystem (1) comprises a torque measurement device (8) provided with a pressure sensor (7), the torque measurement device (8) is installed on the loading spindle and is used for measuring the torque of the loading spindle, and a microprocessor (9), a Bluetooth and a 4G transmission module (10) are arranged in the torque measurement device (8); the pressure sensor (7) is electrically connected with the microprocessor (9) through a first AD signal acquisition circuit (11), and the microprocessor (9) is in communication connection with the measurement and control system host computer (6) through the Bluetooth and the 4G transmission module (10).

3. The FPGA-based multi-channel acquisition and control system of claim 1, wherein, The loading control system (2) comprises a loading system (17) provided with a loading spindle (16), a driving motor (18) for driving the loading spindle (16) to rotate is arranged on the loading system (17), the driving motor (18) is controlled through a frequency converter (19), the frequency converter (19) is electrically connected with the FPGA core control board (5), and the FPGA core control board (5) controls the frequency converter (19) to control the driving motor (18) to work in a frequency conversion mode.

4. The FPGA-based multi-channel acquisition and control system of claim 1, wherein, The multichannel FPGA-based acquisition system (3) comprises an FPGA data processing terminal (12) and a current and voltage sensor (13) for detecting the current and voltage of the grid side and the motor side of a wind power converter, the FPGA data processing terminal (12) is electrically connected with the current and voltage sensor (13) through a second AD signal acquisition module (14); and the FPGA data processing terminal (12) is electrically connected with the FPGA core control board (5) and the inverter open circuit fault diagnosis subsystem (4) through Ethernet.

5. The FPGA-based multi-channel acquisition and control system of claim 1, wherein, The voltage measurement results of the grid side and motor side of the wind power converter collected by the FPGA-based multi-channel acquisition system are calibrated, and the calibration method is that the voltage values measured by the multimeter are linearly fitted by a polynomial method with the voltage values collected by the FPGA-based multi-channel acquisition system, so as to obtain a fitting function relationship.

6. The FPGA-based multi-channel acquisition and control system of claim 1, wherein, The diagnosis method of the inverter open circuit fault diagnosis subsystem comprises the following steps, Step 1, the collected current signal is preprocessed to obtain a current signal free of noise interference; Step 2, the current signal free of noise interference is converted into a threshold-free recurrence graph; Step 3, the convolutional neural network is used to realize the self-learning of the nonlinear features in the current signal threshold-free recurrence graph, and a classification model of the current signal is established; Step 4, the classification model of the current signal is used to identify the subsequent current signal to identify the inverter fault.

7. The FPGA-based multi-channel acquisition and control system of claim 6, wherein, Before the step 1, a simulation model of the NPC three-level inverter is built, and the simulation model mainly comprises A, B and C three-phase bridge arms and IGBT switch tube controllers. Each phase bridge arm is composed of 4 IGBT power switches and 2 clamping diodes. The opening and closing of the IGBT in the bridge arm are controlled by the method of sine wave pulse width modulation, so as to realize the inverter process from direct current to alternating current. By controlling the on-off of the IGBT, the pulse width of the sinusoidal rule is output, and the area of the digital signal pulse voltage output in the same time is equal to the area of the standard grid sine wave. The trigger signal of the IGBT is controlled to simulate different open circuit fault states, so as to measure different current signals.

8. The FPGA-based multi-channel acquisition and control system of claim 7, wherein, The method for simulating different open circuit fault states by controlling the trigger signal of the IGBT is that, taking the three-level inverter A phase open circuit fault as an example, an SS1 is created as the conduction module of switch S1, and a constant value 0 is created as the open circuit fault module of switch S1, When the SS1 conduction module is connected to the switch S1, the SPWM pulse is output; when the constant value 0 is connected to the switch S1 as the open circuit fault module, the constant value 0 is output, which is equivalent to not providing a pulse, and is used to simulate the open circuit fault of the IGBT high power switch.

9. The FPGA-based multi-channel acquisition and control system of claim 7, wherein, The simulation model of the NPC three-level inverter is classified according to the fault, including the following cases, a. All IGBT power switches are in normal operation without fault; b. Only one IGBT power switch is faulty; c. Two IGBT power switches of different half-bridges in the same bridge arm are simultaneously faulty; d. Two IGBT power switches of the same half-bridge in the same bridge arm are simultaneously faulty; e. Two IGBT power switches of the same half-bridge in different bridge arms are simultaneously faulty; f. Two IGBT power switches of different half-bridges in different bridge arms are simultaneously faulty; Wherein the IGBT power switch fault in b~d occurs in the same bridge arm, only has influence on single-phase current in three-phase current, so b~d is defined as simple fault; Wherein the IGBT power switch fault in e~f occurs in different bridge arms, has influence on two-phase current in three-phase current, so e~f is defined as complex fault; Different open circuit fault states are simulated by controlling the trigger signal of IGBT, and the waveform diagram of current signal under each open circuit fault state is obtained.

10. The FPGA-based multi-channel acquisition and control system of claim 6, wherein, The method for realizing self-learning of nonlinear characteristics in the current signal threshold-free recurrent graph through the convolutional neural network firstly introduces a residual network structure into a traditional convolutional neural network with a convolutional layer, a pooling layer and an average pooling layer to form an improved convolutional neural network; in the residual network structure, an identity mapping is added through a shortcut connection, and a difference between a target value H(X) and x is taken as a learning target; In the step 3, the collected current signals are divided into a training set and a test set, the training set and the test set are input into a classification model of the current signals, a classification result is obtained, and macro average is taken as an evaluation index of the classification effect, and the related average recall rate and average precision calculation formula are, In the formula, Macro_R is the average recall rate, and Macro_P is the average precision.

Images