Switch cabinet circuit breaker fault detection sensor based on voiceprint

Abstract

The invention relates to the technical field of switch cabinet circuit breaker fault detection, in particular to a voiceprint-based switch cabinet circuit breaker fault detection sensor. Comprising a pickup and a data processing module, the sound pick-up is in communication connection with the data processing module; the sound pick-up is used for acquiring sound data at a switch cabinet circuit breaker; the data processing module is used for performing compressed sensing processing on the sound data to obtain low-dimensional signal data; and performing fault detection on the high-voltage switch cabinet circuit breaker according to the low-dimensional signal data. According to the invention, the compressed sensing processing is carried out on the sound data, so that the data transmission amount and the data processing amount in fault detection are reduced, the power consumption of the sensor is effectively reduced, the problem of large heating caused by the power consumption is avoided, long-time monitoring can be realized in an area with a severe environment or an area with dense power equipment, and the fault detection efficiency is improved. And meanwhile, the data processing speed is also improved.

Description

Technical Field

[0001] The present application relates to the technical field of switch cabinet circuit breaker fault detection, and in particular to a switch cabinet circuit breaker fault detection sensor based on voiceprint. Background Art

[0002] Switchgear circuit breakers play a vital role in power systems. Their main function is to quickly disconnect or connect current to protect power equipment and power systems from overload, short circuit and other faults. The normal operation of circuit breakers is directly related to the safety and stability of power systems, so their status monitoring and fault diagnosis are crucial. However, since circuit breakers are affected by high-frequency switching operations and current shocks during long-term operation, they are prone to wear, aging, loosening, jamming and other faults in their mechanical parts. These mechanical faults will not only reduce the breaking performance of the circuit breaker, but in severe cases may also cause equipment failures and even cause large-scale power outages in the power system.

[0003] Traditional circuit breaker mechanical fault detection methods are usually based on the monitoring of parameters such as temperature, current, voltage, and vibration, and use thermal imaging, vibration sensors, and current measurement to identify abnormal conditions of circuit breakers. Although these methods can provide certain fault warnings, they have high power consumption in actual applications. Monitoring methods such as temperature, vibration, and current usually require high-precision, low-noise sensors, which consume a lot of power and are not conducive to long-term monitoring at remote monitoring points of the power system, areas with harsh environments, or areas with dense power equipment.

[0004] Therefore, there is an urgent need for a switch cabinet circuit breaker fault detection sensor based on voiceprint, which can effectively reduce power consumption and realize long-term, accurate and effective switch cabinet circuit breaker mechanical fault detection. Summary of the invention

[0005] In view of this, it is necessary to provide a switch cabinet circuit breaker fault detection sensor based on voiceprint to solve the shortcomings of the switch cabinet circuit breaker fault detection sensor in the prior art, such as high power consumption, resulting in high heat and being unfavorable for long-term monitoring.

[0006] A switch cabinet circuit breaker fault detection sensor based on voiceprint comprises: a microphone and a data processing module; the microphone and the data processing module are communicatively connected; the microphone is used to obtain sound data at the switch cabinet circuit breaker; the data processing module is used to perform compressed sensing processing on the sound data to obtain low-dimensional signal data; and fault detection of the high-voltage switch cabinet circuit breaker is performed according to the low-dimensional signal data.

[0007] Preferably, the data processing module performs fault detection on the high-voltage switch cabinet circuit breaker according to the low-dimensional signal data, including:

[0008] Perform signal reconstruction according to the low-dimensional signal data to obtain original signal data;

[0009] The student model trained based on the knowledge distillation method is used to perform fault detection on the circuit breaker of the high-voltage switch cabinet.

[0010] Preferably, it further comprises a power management module, which is arranged inside the main shell and is used for supplying power and adjusting the power consumption of the pickup, the data processing module and the communication module in real time.

[0011] Preferably, it further comprises a communication module, and the communication module is used for performing data communication with an external device.

[0012] Preferably, the data processing module is further used to: receive a data sending instruction from an external device through the communication module;

[0013] The communication module sends the result of the fault detection to the external device through the communication module according to the data sending instruction.

[0014] Preferably, the power management module is used to receive a power control instruction of an external device through the communication module, and perform sleep and power supply according to the power control instruction.

[0015] Preferably, the interior of the main shell is a conical structure, and the pickup is arranged inside the conical structure.

[0016] Preferably, the pickup is arranged at the inner vertex angle of the conical structure.

[0017] Preferably, a waterproof and sound-permeable membrane is provided at the bottom of the conical structure.

[0018] Preferably, a soft magnetic attraction is provided at the bottom of the main shell, and a cavity is provided between the soft magnetic attraction and the waterproof sound-permeable membrane.

[0019] Compared with the prior art, the beneficial effect of the present application is that the present application obtains low-dimensional signal data by performing compressed sensing processing on the sound data, thereby reducing the amount of data transmission and data processing in fault detection, effectively reducing the power consumption of the sensor, and avoiding the problem of high heat caused by this. It can achieve long-term monitoring in areas with harsh environments or areas with dense power equipment, and based on the reduction of data transmission and data processing, it also improves the data processing speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic structural diagram of a switch cabinet circuit breaker fault detection sensor based on voiceprint provided in this application.

[0021] Figure 2 Schematic diagram of the knowledge distillation process.

[0022] Figure 3 Three views of the housing structure of the switch cabinet circuit breaker fault detection sensor based on voiceprint provided in an embodiment of the present application.

[0023] Figure 4 A schematic diagram of the external structure of a switch cabinet circuit breaker fault detection sensor based on voiceprint provided in an embodiment of the present application.

[0024] Figure 5 A schematic diagram of the conical structure inside the main shell of a switch cabinet circuit breaker fault detection sensor based on voiceprint provided in an embodiment of the present application.

[0025] Figure 6 The figure is a schematic diagram of the working process of the switch cabinet circuit breaker fault detection sensor based on voiceprint. DETAILED DESCRIPTION

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Traditional circuit breaker mechanical fault detection methods are usually based on the monitoring of parameters such as temperature, current, voltage, and vibration, and use thermal imaging, vibration sensors, and current measurement to identify abnormal conditions of circuit breakers. Although these methods can provide certain fault warnings, they have some significant limitations in practical applications: the equipment is complex and costly, and traditional methods usually require the configuration of a large number of sensors and auxiliary equipment. A large number of sensors and auxiliary equipment greatly increase power consumption, resulting in high heat generation, which is not conducive to long-term monitoring at remote monitoring points of the power system, areas with harsh environments, or areas with dense power equipment.

[0028] In view of this, the present application provides a switch cabinet circuit breaker fault detection sensor based on voiceprint, please refer to Figure 1 The switch cabinet circuit breaker fault detection sensor based on voiceprint includes: a pickup and a data processing module, and the pickup and the data processing module are communicatively connected.

[0029] The sound pickup is used to obtain the sound data at the switch cabinet circuit breaker; the sound pickup can be a unit such as a microphone that can capture the sound signal emitted by the switch cabinet circuit breaker and convert the sound signal into an electrical signal. Under the premise of meeting the requirements of sound pickup sampling, a low-power sound pickup can be selected as much as possible to reduce the overall power consumption.

[0030] The data processing module is used to perform compressed sensing processing on the sound data to obtain low-dimensional signal data; and to detect faults of high-voltage switch cabinet circuit breakers based on the low-dimensional signal data. The data processing module can be a CPU, a digital signal processor (DSP) and a microcontroller unit (MCU), etc. It can also be integrated with memory and peripheral interfaces, and is very suitable for designing low-power sensor nodes.

[0031] Among them, the purpose of compressed sensing (CS) is to reconstruct the original signal by collecting samples at a much lower number than the traditional Nyquist sampling rate, while ensuring the quality of reconstruction. The data processing module performs compressed sensing processing on the sound data to obtain a low-dimensional signal, which can greatly reduce the resources required for data collection and the amount of data to be processed subsequently, and improve the data processing efficiency, effectively reducing the power consumption of the data processing module.

[0032] In addition, quantization processing can also be performed after compressed sensing is executed. Under the compressed sensing framework, quantization occurs in the compression domain, that is, quantization processing is performed after the signal is compressed and sampled. Therefore, the dimension of the signal can be reduced first, and then quantized, thereby reducing the amount of quantized data, reducing the subsequent data processing volume and storage and transmission costs, and improving the overall real-time performance.

[0033] Furthermore, the data processing module performs fault detection on the high-voltage switch cabinet circuit breaker according to the low-dimensional signal data, including: performing signal reconstruction according to the low-dimensional signal data to obtain the original signal data; and performing fault detection on the high-voltage switch cabinet circuit breaker through a student model obtained by training based on the knowledge distillation method.

[0034] The original signal can be reconstructed from low-dimensional data through a sparse recovery algorithm. The purpose of reconstructing the original signal is to recover the information lost from the compressed sampling data so that further analysis can be performed based on the complete signal. In the application of mechanical fault detection of high-voltage switchgear circuit breakers, the reconstructed original signal can be used to more accurately detect and diagnose the mechanical fault state of the circuit breaker. Because compressed sensing technology allows signal acquisition at a sampling rate far below the Nyquist rate.

[0035] Specifically, the process of reconstructing the original signal includes:

[0036] Construct a sparse optimization problem: Convert the reconstruction problem into an L1 norm minimization problem. By minimizing the L1 norm of the signal, a sparse solution can be obtained, thereby achieving signal recovery.

[0037] Solving the sparse solution using the Orthogonal Matching Pursuit (OMP) method: By means of progressive approximation, at each iteration, a best basis vector is selected such that it can maximize the explanation of the remaining part of the observed signal (i.e., the residual), and the orthogonality of the basis vectors is ensured during this process.

[0038] Reconstructing the original signal: After obtaining the sparse coefficients, the original signal is restored through an inverse transform. The inverse transform is performed on the sparse coefficients using the transform matrix, thereby achieving an accurate reconstruction of the signal.

[0039] For example: An original signal x has a dimension of N, and this signal can be represented as a sparse signal in a certain transform domain. It is sampled using a random sampling matrix Φ, and the sampled data is y:

[0040] y = Φx = Φψα

[0041] where ψ is the transform matrix and α is the sparse coefficient.

[0042] y is the data after compressive sampling. Usually, its dimension M is much smaller than N, that is, the key information of the signal is obtained through a small number of sampling points (lower than the Nyquist sampling rate); Φ is the sampling matrix. To recover the signal x, it is necessary to solve from the small amount of sampled data y. Since M < N, y is an underdetermined system of equations and usually cannot be directly solved. At this time, compressive sensing uses a sparse recovery algorithm to solve the sparsest x to meet the requirements of compressive sensing. The basic process of reconstructing the original signal x from the sampled data y, that is, recovering the sparse coefficient vector α, includes:

[0043] S101: Constructing a sparse optimization problem. To reconstruct the sparse coefficient α, the reconstruction problem can be transformed into an L1 norm minimization problem because L1 minimization tends to obtain a sparse solution. The reconstruction optimization problem is as follows:

[0044]

[0045] L1 norm minimization is a commonly used method in compressive sensing. By minimizing the L1 norm of the signal, a sparse solution can be obtained, thereby achieving signal recovery.

[0046] S102: Using the Orthogonal Matching Pursuit method to solve the sparse solution. The Orthogonal Matching Pursuit method is based on the idea of progressive approximation. At each iteration, a best basis vector is selected such that it can maximize the explanation of the remaining part of the observed signal (i.e., the residual). During the process of continuously selecting basis vectors, the orthogonality of the already selected basis vectors is ensured at the same time. At the k-th iteration, calculate the correlation between each basis vector and the current residual r k-1 :

[0047]

[0048] Use the least squares method to calculate the estimated value of the current sparse solution on the support set S. Let s represent the column subset of the observation matrix on the support set S, and solve the following least squares problem:

[0049] s S =arg min||y-Φ S s S ||2

[0050] Update the elements on the support set S in the sparse representation s, and keep the other elements zero. Calculate the new residual:

[0051] r k =y-Φ S s S

[0052] Finally, the result is output. In this way, after the iteration is completed, a sparse linear combination can be obtained to reconstruct the original signal as much as possible.

[0053] S103: Reconstruct the original signal. After obtaining the sparse coefficient α, restore the original signal x through inverse transformation:

[0054] x=Ψα

[0055] By inversely transforming the sparse coefficients through the transformation matrix Ψ, the accurate reconstruction of the signal can be obtained.

[0056] Then, the student model trained based on the knowledge distillation method is used to detect faults in the high-voltage switch cabinet circuit breaker, which greatly reduces the computational complexity of subsequent processing and reduces the power consumption of the sensor.

[0057] like Figure 2 The basic idea of ​​knowledge distillation is to train a small model (student model) to imitate the behavior of a larger, trained model (teacher model) to achieve similar performance. The teacher model is usually a high-complexity deep learning model with good accuracy and generalization ability, but with high computational overhead. Through knowledge distillation, the student model can learn the characteristics of the teacher model. For example, the steps of knowledge distillation are:

[0058] S201: Training of the teacher model. First, a complex teacher model is trained on a large-scale data set. The model has high precision and generates accurate prediction results. When classifying, the teacher model outputs a probability distribution containing relative information between categories. Compared with the "hard label" (one-hot label) that only contains information about the correct category, the soft label contains the probability distribution of all categories. The output distribution of the teacher model is controlled by introducing the temperature parameter T. When the temperature T is higher, the probability distribution is smoother and the information between categories is richer.

[0059]

[0060] where z i Represents the original score of the model in the i-th category. When the temperature is high, the student model is better able to learn the subtle differences between categories.

[0061] S202: Knowledge extraction and transfer, transferring the knowledge (soft labels and features) generated by the teacher model to the student model.

[0062] S203: Training of the student model. When training the student model, cross entropy loss is used to make the output of the student model as close as possible to the soft label probability distribution of the teacher model:

[0063]

[0064] Where L CE is the cross entropy loss with the true label, KL is the Kullback-Leibler divergence, which is used to measure the difference between the teacher and student output distributions, and α is the weight.

[0065] The student model is trained by imitating the output and behavior of the teacher model, so that the student model can approach the performance of the teacher model with small-scale parameters. The student model can learn the characteristics of the teacher model, thereby significantly reducing model parameters and calculations, reducing the power consumption of the sensor, while maintaining high detection accuracy.

[0066] In addition, the operation of reconstructing the original signal is usually performed by a module or device with a certain computing power, which can be performed by the data processing module or an external device outside the sensor network, such as a server or a cloud processing center. After the low-dimensional data is obtained in the data processing module, it can be sent to an external device, server or other device with greater computing power for processing as needed, which improves the computational efficiency of the detection data, and also further reduces the amount of data processed by the sensor, reduces the power consumption of the sensor, and improves the real-time performance of the overall detection after the sensor data is collected.

[0067] Furthermore, it also includes a power management module for supplying power and adjusting the power consumption of the pickup and the data processing module in real time.

[0068] The power management module can use dynamic voltage and frequency scaling (DVFS) technology to adjust voltage and frequency in real time according to load conditions, reduce unnecessary power consumption, balance system performance and energy consumption, effectively extend battery life, reduce system temperature and improve overall efficiency. At the same time, it can also be based on a variety of preset low-power sleep modes, such as being able to quickly enter sleep mode when the device is inactive.

[0069] Furthermore, it also includes a communication module, which is used to communicate data with an external device. The communication module is used to communicate with an external device, and the communication module can send data to the external device by wired / wireless means, and can also receive data sent by the external device. When the communication module communicates with the external device by wireless means, the sensor can select the most suitable installation position as needed, so that remote monitoring and management can be achieved, which improves the flexibility and ease of use of the system.

[0070] Furthermore, the data processing module receives a data transmission instruction from an external device through the communication module; the communication module transmits the result of the fault detection to the external device through the communication module according to the data transmission instruction. In this way, the interactivity and real-time performance of the system are enhanced, so that the user can obtain the detection results and respond immediately.

[0071] Furthermore, the power management module receives the power control instruction of the external device through the communication module, and performs sleep and power supply according to the power control instruction. The power management module can receive the power control instruction of the external device through the communication module, and adjust the working state of the sensor (such as sleep or power supply) accordingly. In this way, remote power management can be achieved, and energy use can be further optimized.

[0072] Based on the above data processing module, communication module and power management module, the overall working mode and power consumption of the sensor can be finely controlled. For example, when the sensor is powered on for the first time, it will not upload data. It can receive instructions from external devices such as servers through the communication module. After waiting for 60 seconds, the sensor will enter sleep mode and wait for the next wake-up. When the sensor reaches the sending interval time, the sensor wakes up from sleep mode, collects data and uploads it, and requests configuration instructions. After the sensor uploads the request configuration instruction, it will normally receive the parameter modification instruction or parameter non-modification instruction issued by the server. The sensor will enter sleep mode immediately after executing the corresponding instruction; otherwise, it will wait for the server to issue instructions and wait for a preset time interval, such as 10 seconds, and then force it to enter sleep mode after 10 seconds. When waking up from sleep mode, the sensor will collect data according to the configured parameters, and then filter the data. After the data collection is completed, the system will sample the data and convert the data into a suitable sampling rate. When the data is collected to the specified length, the system will turn off data collection and then send the data to the server. Before sending data, you must first connect to the server. Among them, when powered on for the first time, if the sensor fails to connect to the server using the main AP within 10 seconds, the sensor will restart and connect to the server after restarting. If it still fails to connect to the server within 10 seconds, the sensor will restart again and use the main AP to connect after restarting. If it fails to connect to the server after the preset number of back and forth times, the sensor will enter sleep mode. The sleep time is consistent with the sleep time set by the system, waiting for reconnection after the next sleep wake-up. When waking up from sleep, the sensor will use the main AP connection time. If it still fails to connect to the server, it will enter sleep and wait for wake-up.

[0073] In addition, you can set the interval mode to real-time mode: external devices and the host computer need to cooperate to send commands and receive data. You can switch the sensor mode by sending commands. In the continuous acquisition mode, the host computer controls the start and stop. The sensor is normally in interval mode, waking up at fixed intervals to collect data and send the waveform to the host computer, then entering a low power state, and repeating the above operation after the next interval.

[0074] For further information, please refer to Figure 3 to Figure 4 , Figure 3 It is a three-view drawing of the shell structure of the switch cabinet circuit breaker fault detection sensor based on voiceprint provided in the embodiment of the present application. Figure 4 It is a schematic diagram of the external structure of a switch cabinet circuit breaker fault detection sensor based on voiceprint. Figure 4In the sensor, a main shell 2 is provided, which protects the internal components from the external environment. The main shell 2 is also provided with a switch button 1 of the sensor and an upper cover 3. The upper cover 3 is provided with a marking place 4 for marking sensor parameters, numbers and other information. The upper cover 3 is also provided with a reset button 5, an indicator light 6, etc. A soft magnetic attraction 7 and a waterproof sound-permeable membrane 8 are provided at the bottom of the main shell 2, and a cavity is provided between the soft magnetic attraction and the waterproof sound-permeable membrane.

[0075] Further, such as Figure 5 The schematic diagram of the conical structure inside the main shell is shown, the main shell is a conical structure inside, the microphone is arranged inside the conical structure, and the conical design can help the microphone collect sound signals more effectively, improve the sensitivity and accuracy of sound pickup, and thus improve the effect of fault detection. Specifically, the microphone is arranged at the top corner inside the conical structure, which can maximize the collection effect of sound signals and further improve the performance of the sensor.

[0076] Furthermore, a waterproof and sound-permeable membrane is provided at the bottom of the conical structure, which can prevent moisture from entering the sensor and protect the electronic components, and can also ensure that the sound signal passes smoothly without affecting the sound pickup effect.

[0077] Furthermore, a soft magnetic suction is provided at the bottom of the main shell, and a cavity is provided between the soft magnetic suction and the waterproof sound-permeable membrane. The sensor adopts a soft magnetic suction installation method, which can adapt to various surfaces (such as metal, plastic, etc.), suitable for limited space and complex surface structures. The sensor can be easily moved and reinstalled, and can adapt to circuit breaker fault detection scenarios. It is not only convenient for the installation and disassembly of the sensor, but also has a certain electromagnetic shielding effect, which can reduce external electromagnetic interference and improve signal quality. A special conical sound-receiving structure is designed on the main shell according to the characteristics of the microphone. The external soft magnetic suction and the large groove on the shell form a cavity, so that the microphone can hear the audio at the fault point to the maximum extent.

[0078] Based on the above sensors, such as Figure 6 As shown in the workflow diagram, first of all, due to the conical structure inside the main shell, the pickup can collect sound signals more effectively, improve the sensitivity and accuracy of the pickup, and thus improve the effect of fault detection. Low-power units, devices and modules are selected to make the basic power consumption of the sensor low. The bottom of the sensor adopts a soft magnetic installation method, which can adapt to various surfaces and various installation positions of the circuit breaker, improving the ease of use. In addition, the soft magnetic suction also has a certain anti-interference ability, which enhances the quality of the data. Finally, when processing the data, a low-dimensional signal representation is obtained through compressed sensing and data quantization. This representation contains the key information of the original signal. At the same time, due to sparsity and quantization processing, the amount of data is greatly reduced. In subsequent processing, the knowledge distillation technology is used to reduce the model size and calculation amount, thereby reducing power consumption.

[0079] In summary, the switch cabinet circuit breaker fault detection sensor based on voiceprint provided by the present application can effectively reduce power consumption, reduce heat generation, and increase applicable scenarios while ensuring detection accuracy. In addition, based on the greatly reduced data volume in data processing, the real-time performance of the sensor detection is improved.

Claims

1. A switch cabinet circuit breaker fault detection sensor based on voiceprint, characterized in that: include: Pickup, data processing module; The pickup is communicatively connected with the data processing module; The microphone is used to obtain sound data at the switch cabinet circuit breaker; The data processing module is used to perform compressed sensing processing on the sound data to obtain low-dimensional signal data; And perform fault detection on high-voltage switchgear circuit breakers based on low-dimensional signal data.

2. The sensor according to claim 1, characterized in that The data processing module performs fault detection on the high-voltage switch cabinet circuit breaker according to the low-dimensional signal data, including: Perform signal reconstruction according to the low-dimensional signal data to obtain original signal data; The student model trained based on the knowledge distillation method is used to perform fault detection on the circuit breaker of the high-voltage switch cabinet.

3. The sensor according to claim 1, characterized in that It also includes a power management module for supplying power and adjusting the power consumption of the pickup and the data processing module in real time.

4. The sensor according to claim 3, characterized in that It also includes a communication module, which is used for data communication with external equipment.

5. The sensor according to claim 4, characterized in that The data processing module is further used for: Receiving a data transmission instruction from an external device through the communication module; The communication module sends the result of the fault detection to the external device through the communication module according to the data sending instruction.

6. The sensor according to claim 4, characterized in that The power management module is used to receive a power control instruction of an external device through the communication module, and perform sleep and power supply according to the power control instruction.

7. The sensor according to any one of claims 1 to 6, characterized in that: It also includes a main shell, the interior of the main shell is a cone structure, and the pickup is arranged inside the cone structure.

8. The sensor according to claim 7, characterized in that The pickup is arranged at the inner vertex of the cone structure.

9. The sensor according to claim 7, characterized in that A waterproof and sound-permeable membrane is arranged at the bottom of the conical structure.

10. The sensor according to claim 9, characterized in that A soft magnetic attraction is arranged at the bottom of the main shell; a cavity is arranged between the soft magnetic attraction and the waterproof sound-permeable membrane.

Images