NPC three-level inverter fault detection system based on image processing

Abstract

The invention discloses an NPC three-level inverter fault detection system based on image processing, and relates to the technical field of inverter fault detection, and the system comprises the following components: an image collection module which employs an industrial camera and an adjustable support to collect an NPC three-level inverter image, performing real-time evaluation on the acquired image by using an image comprehensive quality index formula, and adjusting camera parameters according to an evaluation result to obtain an image meeting subsequent processing requirements; according to the method, the image of the NPC three-level inverter is deeply analyzed from a macroscopic level and a microcosmic level by applying a multi-scale image processing technology, on the macroscopic scale, the system can quickly identify the overall fault trend and the general area of the inverter and provide a preliminary judgment basis for maintenance personnel, and on the microcosmic scale, the system can quickly identify the fault trend and the general area of the inverter and provide a preliminary judgment basis for maintenance personnel. And the system can accurately identify the positions and states of different elements, and even can capture tiny deformation or damage signs of the elements, so that maintenance personnel can quickly and accurately find fault points.

Description

Technical Field

[0001] The present invention relates to the technical field of inverter fault detection, and specifically to an NPC three-level inverter fault detection system based on image processing. Background Art

[0002] As an important device in the field of power electronics, the NPC three-level inverter plays a key role in many fields such as industrial control, new energy power generation, and electric vehicle drive. Its unique three-level output characteristic can effectively reduce the harmonic content of the output voltage, improve the efficiency and stability of the system, and thus has been widely used. The NPC three-level inverter usually consists of multiple complex electronic components, including power switching devices, capacitors, inductors, and control circuits, etc. These components play a crucial role in the operation process of the inverter.

[0003] Traditional NPC three-level inverter fault detection methods mainly rely on the monitoring and analysis of electrical parameters, such as the real-time detection and waveform analysis of current and voltage. Although these methods can reflect the operation state of the inverter to a certain extent, they have obvious limitations in fault location. First of all, the electrical parameter monitoring method usually can only provide information about the overall operation state, and it is difficult to accurately locate the specific faulty component and position, especially in the case of complex faults, the positioning accuracy is poor. Secondly, the traditional method has a low sensitivity to early potential faults and is difficult to give an early warning at the initial stage of the fault, resulting in the device being discovered only after the fault further develops, increasing the downtime and maintenance cost of the device.

[0004] In view of the above problems, it is necessary to optimize the existing NPC three-level inverter fault detection system based on image processing. By using multi-scale image processing technology, the image features are analyzed from both macroscopic and microscopic levels to achieve fast and accurate fault location. Therefore, it is of great significance to develop an NPC three-level inverter fault detection system based on image processing that can comprehensively achieve the above characteristics. Summary of the Invention

[0005] The object of the present invention is to make up for the deficiencies of the prior art and provide an NPC three-level inverter fault detection system based on image processing. It can obtain the image information of the inverter through an industrial-grade image acquisition device, and use multi-scale image processing technology to analyze the image features from both macroscopic and microscopic levels, so as to achieve fast and accurate fault location. On the macroscopic scale, the system preprocesses the low-resolution image and extracts the overall features, quickly judges whether there is a fault and the approximate location of the fault. On the microscopic scale, the system extracts local features from the high-resolution image, identifies the positions and states of different components, and analyzes the relationships between them. By combining the fault feature database and the component relationship analysis results, the system can accurately judge the root cause and influence range of the fault, provide detailed fault information for maintenance personnel, and enable them to carry out maintenance more targeted.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: An NPC three-level inverter fault detection system based on image processing, which system includes the following components:

[0007] Image acquisition module: An industrial camera and an adjustable bracket are used to acquire the image of the NPC three-level inverter, and the image comprehensive quality index formula is used to evaluate the acquired image in real time. According to the evaluation result, the camera parameters are adjusted to obtain an image that meets the requirements of subsequent processing;

[0008] Multi-scale image processing module: After preprocessing the low-resolution image, the overall features are extracted, and the macroscopic fault risk is judged through the fault risk index. If there is a risk, the high-resolution image is preprocessed. According to the extracted overall features, combined with the structure layout and working principle of the inverter, the probability of each component having a fault is determined, and the relative positions and connection relationships between different components in the image are analyzed. Check whether the connections between components are firm, whether there are looseness and short-circuit conditions, and judge whether their working states are normal by observing the electrical connection lines and mechanical connection parts between components;

[0009] Fault analysis module: Match the fault features extracted by multi-scale image processing with the standard features in the fault feature library, calculate the matching degree to judge the fault type, and evaluate the fault influence range according to the number of affected components, the importance of components and the degree of influence;

[0010] Result output module: Calculate and output the information credibility by integrating the macroscopic fault risk index, the microscopic component fault probability and the fault type matching degree for the reference and decision-making of maintenance personnel.

[0011] Further, the image acquisition module uses the image comprehensive quality index formula to evaluate the acquired image in real time, and its formula is: Among them, IQ is the comprehensive image quality index, with a value range of [0, 1]. The closer the value is to 1, the better the image quality. L is the average brightness of the currently acquired image, L min and L max are respectively the minimum and maximum values of the acceptable range of brightness, which are determined by the statistical analysis of the image brightness in the normal working environment of the inverter. ω L is the weight of brightness in the comprehensive quality. C is the color richness of the currently acquired image, which is measured by the statistical features of the color histogram. C min and C max are respectively the minimum and maximum values of the acceptable range of color richness, which are determined based on the color characteristics of normal inverter images. ω C is the weight of color richness in the comprehensive quality. S is the sharpness of the currently acquired image, which can be measured by indicators such as the edge sharpness of the image. S min and S max are respectively the minimum and maximum values of the acceptable range of sharpness. ω S is the weight of sharpness in the comprehensive quality. If the IQ value is lower than the set threshold, it indicates that the image quality is poor, and the exposure time and aperture size parameters of the camera are automatically adjusted to re-acquire the image until the IQ meets the subsequent requirements.

[0012] Furthermore, the multi-scale image processing module judges the macroscopic fault risk through the fault risk index, and its calculation formula is: Among them, R m is the macroscopic scale fault risk index. The larger the value, the higher the macroscopic fault risk. n is the number of macroscopic features extracted from the low-resolution image. α i is the weight of the i-th macroscopic feature. F i is the measured value of the i-th macroscopic feature in the current low-resolution image. F i0 is the reference value of the i-th macroscopic feature in the normal state. When R m exceeds the set threshold, it indicates that there is an abnormality in the inverter at the macroscopic level.

[0013] Furthermore, the multi-scale image processing module determines the probability of each component having a fault according to the extracted overall features, combined with the structural layout and working principle of the inverter. Its probability calculation formula is: Among them, P e is the fault probability of a certain component at the microscopic scale. m is the number of features of this component extracted from the high-resolution image. β j is the weight of the j-th component feature. d j is the degree of deviation of the j-th component feature from the normal state. When P e exceeds the set threshold, it is considered that this component has a fault.

[0014] Furthermore, the fault analysis module matches the fault features extracted by multi-scale image processing with the standard features in the fault feature library. Specifically, it collects the image features of a large number of inverters in different fault states, classifies and organizes them, and establishes a fault feature library that contains the feature descriptions and example images corresponding to various fault types. It compares the fault features extracted by the multi-scale image processing module with the features in the fault feature library, calculates their matching degrees, and when it is found that a certain fault feature is highly similar to a certain feature in the library, it determines that the inverter has the corresponding fault type. Combining the position, state, and mutual relationship of the components, it analyzes the root cause of the fault, and by tracing the association between the faulty component and other components, it finds out the fundamental reason for the fault. According to the circuit structure and working principle of the inverter, it evaluates the scope of influence of the fault on other components and the entire system, considering the electrical connection and functional dependence relationship between the faulty component and other components, and determines whether the fault will cause damage to other components or partial failure of the system functions.

[0015] Furthermore, the fault analysis module compares the fault features extracted by the multi-scale image processing module with the features in the fault feature library and calculates their matching degrees. The formula for calculating the matching degree is as follows: where M ft is the matching degree between the extracted fault feature and the t-th fault type, and its value range is [0, 1]. The closer the value is to 1, the higher the matching degree. p is the number of features used for fault type matching, γ k is the weight of the k-th fault feature, F k is the k-th actually extracted fault feature, is the k-th standard feature corresponding to the t-th fault type, is the similarity between the k-th actual fault feature and the standard feature of the t-th fault type. When M ft exceeds the set threshold, it can be determined that the inverter has the t-th fault type.

[0016] Furthermore, the fault analysis module evaluates the scope of influence of the fault on other components and the entire system according to the circuit structure and working principle of the inverter. The formula for calculating the scope of influence is as follows: where I fr is the fault influence range index. The larger the value, the wider the fault influence range. q is the number of components affected by the fault, δ s is the importance factor of the s-th affected component, and I es is the degree to which the s-th affected component is affected by the fault.

[0017] Furthermore, the result output module calculates and outputs the information credibility by integrating the macroscopic fault risk index, the microscopic component fault probability, and the fault type matching degree. The formula is as follows: where Coi is the credibility of the output information, with a value range of [0, 1]. The closer the value is to 1, the more credible the information is, ω Rm , ω Pe , ω Mft are respectively the weights of the macroscopic-scale fault risk index, the microscopic-scale component fault probability, and the fault type matching degree in the information credibility evaluation. The maintenance personnel judge whether to verify the fault information or take corresponding maintenance measures according to the value of C oi of the value to determine whether to verify the fault information or take corresponding maintenance measures.

[0018] Compared with the prior art, the NPC three-level inverter fault detection system based on image processing has the following beneficial effects:

[0019] First, by applying multi-scale image processing technology, the present invention deeply analyzes the images of the NPC three-level inverter from both macroscopic and microscopic levels. On the macroscopic scale, the system can quickly identify the overall fault trend and approximate area of the inverter, providing a preliminary judgment basis for maintenance personnel. On the microscopic scale, the system can accurately identify the positions and states of different components, and can even capture tiny deformations or damage signs of the components, enabling maintenance personnel to quickly and accurately find the fault point, thus greatly shortening the maintenance time and improving the maintenance efficiency. At the same time, the system can also combine the spatial position relationship and functional correlation relationship between components to further analyze the root cause and influence range of the fault, providing more comprehensive fault information for maintenance personnel, enabling them to perform maintenance operations more targeted and effectively reducing the maintenance cost.

[0020] Second, through image processing technology, the present invention can achieve early warning and precise positioning of inverter faults, thus effectively avoiding further damage to the equipment caused by the further development of faults. In addition, the system can also send fault information to the remote monitoring center and the mobile terminals of relevant personnel in real time, enabling managers to timely understand the fault situation of the inverter and conduct remote command and decision-making. This real-time fault monitoring and response mechanism not only improves the reliability and stability of the equipment, but also greatly reduces the downtime of the equipment and reduces the production losses caused by faults.

[0021] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic structural diagram of an NPC three-level inverter fault detection system based on image processing.

[0023] Figure 2It is a flowchart of an NPC three-level inverter fault detection system based on image processing. Specific implementation manners

[0024] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the accompanying drawings and preferred embodiments, detail the specific implementation manners, structures, features and their effects of the present invention as follows.

[0025] Embodiment 1

[0026] A large-scale electronics manufacturing enterprise widely uses NPC three-level inverters on its production line to provide stable power supply for various equipment. Due to the requirements of production line continuity and efficiency, the stable operation of the inverter is crucial. Any fault in the inverter may cause the production line to stop, resulting in huge economic losses. Therefore, it is particularly important to detect and locate the faults of the inverter in a timely and accurate manner.

[0027] Multiple high-resolution industrial cameras are installed near the inverter. These cameras are fixed by adjustable brackets and can capture the overall and key parts of the inverter from different angles. The cameras are set to perform timed image acquisition every 15 minutes. At the same time, the cameras are connected to the control system of the inverter. When the inverter starts, stops, or detects abnormal fluctuations in current and voltage, image acquisition is immediately triggered. An optical filter is installed in front of the camera lens to reduce the interference of the complex light environment in the workshop, such as strong light sources and reflected light in the workshop.

[0028] By performing real-time image quality assessment on the acquired images, using the formula where IQ is the comprehensive image quality index, with a value range of [0, 1]. The closer the value is to 1, the better the image quality. L is the average brightness of the currently acquired image, L min and L max are respectively the minimum and maximum values of the acceptable range of brightness, determined by statistical analysis of the image brightness in the normal working environment of the inverter. ω L is the weight of brightness in the comprehensive quality, satisfying 0 ≤ ω L ≤ 1, and ω L + ω C + ω S = 1. C is the color richness of the currently acquired image, which can be measured by the statistical features of the color histogram. C min and C max are respectively the minimum and maximum values of the acceptable range of color richness, determined based on the color characteristics of normal inverter images. ω C is the weight of color richness in the comprehensive quality. S is the clarity of the currently acquired image, which can be measured by indicators such as the edge sharpness of the image. S minand S max are the minimum and maximum values of the acceptable range of clarity, determined by experiments and experience. ω S is the weight of clarity in the comprehensive quality. If the image clarity is crucial for subsequent fault feature extraction, ω S can be set to 0.5. ω L and ω C are set to 0.3 and 0.2 respectively. It can also be initially set by referring to relevant image processing industry standards or experience of similar projects. Construct an experimental dataset containing images of a large number of inverters under different working conditions, divide it into a training set and a test set. Use the training set data, try different weight combinations to conduct image quality evaluation experiments, record evaluation indicators such as fault detection accuracy and recall rate. By continuously adjusting the weights, make the evaluation indicators reach the optimal. Automatically adjust the exposure time and aperture size of the camera according to the evaluation results to ensure that the collected images have good quality and meet the requirements of subsequent processing. In addition, to prevent the camera from malfunctioning and affecting image acquisition, the system also sets up a standby camera. When the main camera is abnormal, the standby camera can automatically switch and continue with the image acquisition work.

[0029] For the collected low-resolution images, first perform preprocessing operations such as removing impurities and noise, adjusting brightness and contrast, and then extract macroscopic features such as the overall brightness distribution and contour shape of the inverter. Calculate the macroscopic-scale fault risk index through the formula to determine whether there is a fault risk at the macroscopic level. Among them, R m is the macroscopic-scale fault risk index. The larger the value, the higher the macroscopic fault risk. n is the number of macroscopic features extracted from the low-resolution image. α i is the weight of the i-th macroscopic feature. F i is the measured value of the i-th macroscopic feature in the current low-resolution image. F i0 is the reference value of the i-th macroscopic feature in the normal state. If a risk is detected, perform more refined preprocessing on the high-resolution image, including further enhancing image details and clarity, identifying the positions and states of various components in the inverter, such as capacitors, inductors, power modules, etc. Calculate the microscopic-scale component fault probability through the formula to determine whether a component may have a fault. Among them, P e is the fault probability of a certain component at the microscopic scale, and the value range is [0,1]; m is the number of features of the component extracted from the high-resolution image. β j is the weight of the j-th component feature. d jis the deviation degree of the j-th component feature from the normal state, and analyze the relative positions and connection relationships between components to check for looseness, short circuits, etc. When performing image processing, the system also compares and analyzes the images collected at different time periods to more accurately judge the trend of component state changes. For example, if it is found that a certain component gradually shows color changes or shape deformations in recent images, even if the current fault probability has not reached the threshold, it will be listed as a key monitoring object and preventive measures will be taken in advance.

[0030] Match the fault features extracted by the multi-scale image processing module with the standard features in the pre-established fault feature library. Through the formula Calculate the fault type matching degree to determine the current fault type of the inverter, where M ft is the matching degree between the extracted fault feature and the t-th fault type, with a value range of [0, 1]. The closer the value is to 1, the higher the matching degree. p is the number of features used for fault type matching, and γ k is the weight of the k-th fault feature, F k is the actually extracted k-th fault feature, is the k-th standard feature corresponding to the t-th fault type, is the similarity between the k-th actual fault feature and the standard feature of the t-th fault type, with a value range of [0, 1]. At the same time, according to the number of components affected by the fault, the importance of the components in the inverter system, and the degree to which the components are affected by the fault, use the formula Evaluate the impact range of the fault on the entire inverter system, where I fr is the fault impact range index. The larger the value, the wider the fault impact range. q is the number of components affected by the fault, and δ s is the importance factor of the s-th affected component, I es is the degree to which the s-th affected component is affected by the fault, with a value range of [0, 1]. For example, if a key power component fails, by analyzing its connection relationship and functional dependency with other components, determine other components and system functions that may be affected. During the fault analysis process, the system also combines the historical fault data and operation records of the inverter to comprehensively analyze the current fault. If it is found that a certain fault type is similar to a previous fault, the system will automatically retrieve the relevant historical processing records to provide reference for maintenance personnel and help them formulate a maintenance plan faster.

[0031] The results of the fault analysis module are displayed in the monitoring room of the production line through an intuitive graphical interface, including information such as fault type, fault location, root cause of the fault, and possible scope of influence. At the same time, the fault information is transmitted to the mobile devices of maintenance personnel in real time via the network. The maintenance personnel can rush to the site for maintenance in a timely manner based on this information. The system will also evaluate the credibility of the output information according to the formula Output the credibility evaluation result of the information, where C oi is the credibility of the output information, and its value range is [0, 1]. The closer the value is to 1, the more credible the information is. are the fault risk index at the macro scale, the fault probability of components at the micro scale, and the weights of the fault type matching degree in the information credibility evaluation respectively, and decides whether it is necessary to further verify the fault information or take other measures. In addition, the result output module will also statistically analyze the fault information and generate a fault report to provide data support for the enterprise's equipment management and maintenance. For example, by analyzing the fault data over a period of time, it can be found that the fault frequency of certain components is relatively high, so as to strengthen the maintenance and replacement of these components targeted, and improve the overall reliability of the inverter.

[0032] Through the application of the present invention, the enterprise can timely detect the faults of the inverter, reduce the downtime of the production line caused by the inverter faults. The average downtime per fault has been shortened from the original 2 hours to within 30 minutes, greatly improving the production efficiency. At the same time, since the fault location and cause can be accurately located, the maintenance personnel can prepare maintenance tools and spare parts more targeted, reducing the maintenance cost. In addition, through the statistical analysis of the fault information, the enterprise can also take preventive measures in advance, further improving the reliability and stability of the inverter, and ensuring the continuous and efficient operation of the production line.

[0033] Embodiment 2

[0034] In a large-scale wind farm, there are hundreds of wind turbines distributed. Each wind turbine is equipped with an NPC three-level inverter, which is used to convert the unstable electric energy generated by the wind turbine into stable electric energy suitable for grid connection. Since wind farms are usually located in remote areas with complex and changeable environments, such as strong winds, sandstorms, high temperatures or low temperatures and other harsh conditions, the inverters operate in such an environment for a long time and are extremely prone to various faults. Once an inverter fails, it will not only cause the shutdown of that wind turbine and affect the power generation, but may also have an adverse impact on the stability of the entire power grid. Therefore, accurately and timely detecting and locating inverter faults is crucial for the efficient operation and reliable power supply of the wind farm.

[0035] Industrial cameras with good protection performance are installed around each inverter. The cameras are encapsulated in a sturdy protective housing, which can effectively resist the erosion of sand, wind, rain and extreme temperatures. The cameras are adjusted to the best shooting angle through adjustable brackets to ensure that all parts of the inverter can be comprehensively and clearly photographed, including the internal circuit boards, components and external heat dissipation devices, etc. Considering the special environment of the wind farm and the operating characteristics of the inverter, the cameras are set to perform timed image acquisition every 30 minutes. At the same time, by connecting with the monitoring system of the inverter, when the operating state of the inverter changes (such as abnormal power output, frequency fluctuation, etc.) or abnormal signals are detected (such as overheating, overcurrent alarm, etc.), image acquisition is immediately triggered. The installed optical filter can not only effectively reduce the impact of sand, dust, etc. on the image quality, but also filter out light of specific wavelengths, improving the clarity and contrast of the image. Through real-time image quality assessment, the camera parameters are automatically adjusted, such as automatically adjusting the exposure time according to the light intensity and adjusting the focal length according to the blurring degree of the image, etc., to ensure that the collected images are clear and stable. In addition, to ensure the continuity of image acquisition, the system is also equipped with an image storage and backup function. The collected images are stored in a large-capacity local hard disk and regularly backed up to a remote server for subsequent analysis and review.

[0036] Preprocess the low-resolution images to remove the noise points in the images and make the images clearer. Then extract the overall appearance features of the inverter, such as macroscopic features like the temperature distribution of the housing and changes in the overall shape, calculate the macroscopic-scale fault risk index, and judge whether there is a potential fault risk in the inverter. When a risk is found, perform a more refined preprocessing on the high-resolution images, use image enhancement techniques to further highlight the detailed parts in the images, identify the states of the internal components of the inverter, such as whether the solder joints on the circuit board are loose, whether the components show signs of overheating (judged by color change), whether the capacitors are swollen, etc. Determine the specific faulty components by calculating the microscopic-scale component fault probability, and analyze the relationships between components, including not only the electrical connection relationships, but also considering the influence of physical factors such as heat conduction on the components, and judge the fault propagation path. For example, if a faulty heating component causes an abnormal increase in temperature, judge whether the adjacent components will be affected by the high temperature and malfunction by analyzing the heat conduction path.

[0037] Match the extracted fault features with the standard features in the fault feature library, calculate the fault type matching degree through pattern recognition algorithms, and determine the fault type. For example, when the fault features detected for a certain capacitor component have a high matching degree with the features of capacitor aging faults in the library, it is judged as a capacitor aging fault. At the same time, according to the position and role of the inverter in the wind power generation system, combined with the grid topology of the wind farm, evaluate the impact scope of the fault on the entire power generation system. If the inverter is located at a critical power transmission node, its fault may cause the cascading shutdown of multiple wind turbines and even affect the stability of the entire power grid. During the fault analysis process, the system also takes into account the influence of environmental factors on the fault. For example, strong winds may damage the mechanical structure of the inverter, and dust may affect the heat dissipation effect, etc. By comprehensively analyzing these factors, more comprehensive and accurate fault information is provided for maintenance personnel to help them formulate more reasonable maintenance plans. In addition, the system also conducts data interaction with the SCADA system of the wind farm to obtain more operating parameters and environmental data, further improving the accuracy of fault analysis.

[0038] In the monitoring center of the wind farm, the fault information of each inverter is displayed in real time on a large screen, including the fault type, location, root cause, and impact scope, etc. At the same time, the fault information is sent to the remote operation and maintenance center, and the operation and maintenance personnel can conduct further analysis and processing of the fault through professional software. For fault information with low credibility, the system will automatically conduct further verification and confirmation, such as by collecting images again or obtaining more operating data for comparative analysis. The result output module also has a fault warning function. When it detects that the status of some components shows an abnormal trend but has not reached the fault threshold, the system will issue a warning signal in advance to remind the operation and maintenance personnel to take preventive measures to avoid the occurrence of faults. In addition, the system also records and statistically analyzes the fault information in detail, generating various reports and charts, such as fault frequency statistics, fault type distribution, maintenance time records, etc., providing strong data support for the equipment management and maintenance of the wind farm. Through the analysis of these data, the operation and maintenance personnel can summarize the laws of fault occurrence, optimize the equipment maintenance plan, and improve the reliability and service life of the equipment.

[0039] The application of the present invention in the wind farm effectively improves the fault detection ability of the inverter, ensures the stable operation of the wind power generation system. At the same time, through early warning and accurate fault location, the power generation loss caused by inverter faults is greatly reduced. In addition, due to the ability to carry out more targeted repairs and maintenance, the maintenance cost is reduced, bringing significant economic and social benefits to the wind farm.

[0040] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An NPC three-level inverter fault detection system based on image processing, characterized in that The system includes the following components: Image acquisition module: An industrial camera and an adjustable bracket are used to acquire images of the NPC three-level inverter. The formula for the comprehensive image quality index is used to evaluate the acquired images in real time. According to the evaluation results, the camera parameters are adjusted to obtain images that meet the requirements of subsequent processing; Multi-scale image processing module: After preprocessing the low-resolution images, the overall features are extracted. The macroscopic fault risk is judged through the fault risk index. If there is a risk, the high-resolution images are preprocessed. According to the extracted overall features, combined with the structural layout and working principle of the inverter, the probability of each component having a fault is determined, and the relative positions and connection relationships between different components in the image are analyzed. Check whether the connections between components are firm, whether there are looseness and short-circuit conditions, and judge whether their working states are normal by observing the electrical connection lines and mechanical connection parts between components; Fault analysis module: Match the fault features extracted by multi-scale image processing with the standard features in the fault feature library, calculate the matching degree to judge the fault type, and evaluate the scope of the fault impact according to the number of affected components, the importance of the components and the degree of influence; Result output module: Calculate the output information credibility by synthesizing the macroscopic fault risk index, the microscopic component fault probability and the fault type matching degree, for the reference of maintenance personnel in decision-making.

2. The NPC three-level inverter fault detection system based on image processing according to claim 1, wherein, The image acquisition module uses the image comprehensive quality index formula to evaluate the acquired images in real time. The formula is as follows: where IQ is the image comprehensive quality index, and its value range is [0, 1]. The closer the value is to 1, the better the image quality. L is the average brightness of the currently acquired image, L min and L max are the minimum and maximum values of the acceptable range of brightness, respectively, which are determined by the statistical analysis of the image brightness in the normal working environment of the inverter. ω L is the weight of brightness in the comprehensive quality. C is the color richness of the currently acquired image, which is measured by the statistical features of the color histogram. C min and C max are the minimum and maximum values of the acceptable range of color richness, respectively, which are determined based on the color characteristics of normal inverter images. ω C is the weight of color richness in the comprehensive quality. S is the sharpness of the currently acquired image, which can be measured by indicators such as the edge sharpness of the image. S min and S max are the minimum and maximum values of the acceptable range of sharpness, respectively. ω S is the weight of sharpness in the comprehensive quality. If the IQ value is lower than the set threshold, it indicates that the image quality is poor. The exposure time and aperture size parameters of the camera are automatically adjusted to re-acquire the image until the IQ meets the subsequent requirements.

3. The NPC three-level inverter fault detection system based on image processing according to claim 2, wherein The multi-scale image processing module determines the macroscopic fault risk through the fault risk index, and its calculation formula is: where R m is the macroscopic scale fault risk index, and the larger the value, the higher the macroscopic fault risk. n is the number of macroscopic features extracted from the low-resolution image, and α i is the weight of the i-th macroscopic feature, F i is the measured value of the i-th macroscopic feature in the current low-resolution image, and F i0 is the reference value of the i-th macroscopic feature under normal conditions. When R m exceeds the set threshold, it indicates that there is an abnormality in the inverter at the macroscopic level.

4. The NPC three-level inverter fault detection system based on image processing according to claim 1, wherein The multi-scale image processing module determines the probability of each component having a fault based on the extracted overall features, combined with the structural layout and working principle of the inverter. The probability calculation formula is as follows: where P e is the fault probability of a certain component at the micro scale, m is the number of features of the component extracted from the high-resolution image, and β j is the weight of the j-th component feature, d j is the deviation degree of the j-th component feature from the normal state. When P e exceeds the set threshold, it is considered that the component has a fault.

5. The NPC three-level inverter fault detection system based on image processing according to claim 4, characterized in that, The fault analysis module matches the fault features extracted by multi-scale image processing with the standard features in the fault feature library. Specifically, a large number of image features of the inverter in different fault states are collected, classified and sorted to establish a fault feature library, which contains the feature descriptions and example images corresponding to various fault types. Compare the fault features extracted by the multi-scale image processing module with the features in the fault feature library, calculate their matching degree. For a certain fault feature found to be highly similar to a certain feature in the library, it is judged that the inverter has the corresponding fault type. Combine the position, state and mutual relationship of the components to analyze the root cause of the fault. By tracing the association between the faulty component and other components, find out the root cause of the fault. According to the circuit structure and working principle of the inverter, evaluate the scope of the impact of the fault on other components and the entire system. Consider the electrical connection and functional dependence relationship between the faulty component and other components to judge whether the fault will cause damage to other components or partial system function failure.

6. The fault detection system for NPC three-level inverter based on image processing according to claim 5, characterized in that The fault analysis module compares the fault features extracted by the multi-scale image processing module with the features in the fault feature library and calculates their matching degree. The formula for calculating the matching degree is as follows: where M ft is the matching degree between the extracted fault feature and the t-th fault type, and its value range is [0, 1]. The closer the value is to 1, the higher the matching degree. p is the number of features used for fault type matching, and γ k is the weight of the k-th fault feature, F k is the k-th actually extracted fault feature, is the k-th standard feature corresponding to the t-th fault type, is the similarity between the k-th actual fault feature and the standard feature of the t-th fault type. When M ft exceeds the set threshold, it can be judged that the inverter has the t-th fault type.

7. The fault detection system for NPC three-level inverter based on image processing according to claim 5, wherein The fault analysis module evaluates the impact scope of the fault on other components and the entire system according to the circuit structure and working principle of the inverter. The calculation formula for its impact scope is as follows: Where, I fr is the fault impact scope index. The larger the value, the wider the fault impact scope. q is the number of components affected by the fault, and δ s is the importance factor of the s-th affected component, and I es is the degree to which the s-th affected component is affected by the fault.

8. The fault detection system of the NPC three-level inverter based on image processing according to claim 1, characterized in that, The result output module calculates the credibility of the output information by synthesizing the macroscopic fault risk index, the microscopic component fault probability, and the fault type matching degree. The formula is as follows: Among them, C oi is the credibility of the output information, and its value range is [0, 1]. The closer the value is to 1, the more credible the information is. ω Rm , ω Pe , ω Mft are the weights of the macroscopic scale fault risk index, the microscopic scale component fault probability, and the fault type matching degree in the evaluation of the credibility of the information respectively. The maintenance personnel judge whether it is necessary to verify the fault information or take corresponding maintenance measures according to the value of C oi .

Images