阅读说明：本技术 用于监测开关设备的装置 (Device for monitoring a switchgear ) 是由 萨巴纳塔拉简·萨比亚 拉尔夫·吉策尔 贝内迪克特·施密特 于 2020-03-02 设计创作，主要内容包括：本发明涉及一种用于监测开关设备的装置。该装置包括输入单元、处理单元和输出单元。输入单元被配置为向处理单元提供开关设备的监测红外图像。处理单元被配置为实现机器学习分类器算法以分析监测红外图像并且确定开关设备中是否存在一个或多个异常热点。机器学习分类器算法已经基于多个不同训练图像被训练,其中多个训练图像包括由图像处理算法生成的多个合成红外图像。输出单元被配置为输出与一个或多个异常热点相关的信息。(The invention relates to a device for monitoring a switchgear. The device comprises an input unit, a processing unit and an output unit. The input unit is configured to provide the processing unit with a monitoring infrared image of the switching device. The processing unit is configured to implement a machine learning classifier algorithm to analyze the monitored infrared images and determine whether one or more anomalous hotspots exist in the switchgear. The machine learning classifier algorithm has been trained based on a plurality of different training images, wherein the plurality of training images includes a plurality of synthetic infrared images generated by an image processing algorithm. The output unit is configured to output information related to the one or more anomalous hotspots.)

1. An apparatus for monitoring a switchgear, the apparatus comprising:

-an input unit;

-a processing unit; and

-an output unit;

wherein the input unit is configured to provide the processing unit with a monitored infrared image of a switching device;

wherein the processing unit is configured to implement a machine learning classifier algorithm to analyze the monitored infrared images and determine whether one or more anomalous hotspots exist in the switchgear;

wherein the machine learning classifier algorithm has been trained based on a plurality of different training images, wherein the plurality of training images includes a plurality of synthetic infrared images generated by an image processing algorithm; and

wherein the output unit is configured to output information related to the one or more anomalous hotspots.

2. The apparatus of claim 1, wherein the image processing algorithm is a conditional countermeasure network.

3. The apparatus of any of claims 1-2, wherein the plurality of composite infrared images are generated based on at least one computer-aided design drawing, or sketch of a switching device.

4. The apparatus of claim 3, wherein the at least computer-aided design drawing, or sketch comprises a plurality of drawings comprising drawing data for a plurality of different switching devices.

5. The apparatus of any of claims 3 to 4, wherein the at least computer-aided design drawing, or sketch comprises drawing data of at least one circuit breaker.

6. The apparatus of claim 5, wherein the at least computer-aided design drawing, or sketch comprises a plurality of drawings comprising drawing data for a plurality of circuit breakers having different internal structures.

7. The apparatus of any of claims 3-6, wherein two or more of the plurality of synthetic infrared images are generated based on one computer-aided design drawing, or sketch of a switching device.

8. The apparatus of any of claims 3-6, wherein the plurality of composite infrared images are generated based on a corresponding number of computer-aided design drawings, plots, or sketches of switching devices.

9. The apparatus of any of claims 3 to 8, wherein the generation of one or more of the plurality of synthetic infrared images comprises: addition of at least one hotspot to the one or more composite infrared images.

10. The apparatus of any of claims 3-9, wherein the generation of one or more of the plurality of synthetic infrared images comprises: the addition of at least one hotspot to the one or more computer-aided design drawings, or sketches of the switchgear.

11. The apparatus of any of claims 9 to 10, wherein the addition of the at least one hotspot is performed manually.

12. The apparatus of any one of claims 1 to 11, wherein the at least one computer-aided design drawing, drawing or sketch of a switchgear device comprises drawing data of at least one circuit breaker.

13. The apparatus of any of claims 1-12, wherein the image processing algorithm uses a style migration algorithm.

14. The apparatus of any of claims 1-13, wherein the image processing algorithm uses a pix2pix algorithm.

15. The apparatus of any one of claims 1 to 14, wherein the monitoring infrared image comprises image data of at least one circuit breaker.

16. The apparatus of any of claims 1-15, wherein the machine learning classifier algorithm is a neural network.

17. The apparatus of claim 16, wherein the neural network is a convolutional neural network.

18. The apparatus of any of claims 1-17, wherein the processing unit is configured to update the training of the machine learning classifier algorithm, the training comprising the use of the monitoring infrared images.

19. The apparatus of claim 18, wherein the training update comprises: a manual indication that the monitored infrared image does not include an abnormal hotspot, or a manual indication that the monitored infrared image includes one or more abnormal hotspots.

20. The apparatus of claim 19, wherein the manual indication that the monitoring infrared image includes one or more anomalous hotspots comprises: manual indication of one or more locations in the monitored infrared image of the one or more anomalous hotspots.

21. A system for monitoring a switchgear, the system comprising:

-an infrared camera; and

-an arrangement for monitoring a switchgear according to any of claims 1 to 20; and is

Wherein the infrared camera is configured to acquire the monitoring infrared image of the switching device.

Technical Field

The invention relates to a device and a system for monitoring a switchgear.

Background

Switching device failure due to high temperature hot spots can have serious consequences such as arcing/flashover, almost like an explosion. There is a significant need for a method and related system for monitoring and warning prior to the occurrence of such events, which is currently not available in an affordable form, which can be installed as a standard in every switchgear and provide sufficient information about the health of the switchgear. Furthermore, there is no method or system available for capturing and transmitting images of switchgear for processing elsewhere in order to provide such monitoring and early warning. Currently, in the utility industry, measurement of temperature changes is one of the most common measures of structural health of equipment and components. Corroded connections, faulty contacts, damaged components, etc. can all lead to hot spots. Currently, it is common practice to capture thermal variation images using an infrared camera and manually analyze them to find hotspots and then perform maintenance operations. Currently, the use of IR sensors to detect hot spots in circuit breakers, switchgear and other electrical equipment requires extensive, very precise calibration to accurately measure the temperature at the correct location. A further related problem is to identify the correct area to be monitored in the IR image. Due to the different types and geometries, a universal solution for all switchgear and all circuit breakers within such switchgear is not possible.

There is a need to address these problems.

Disclosure of Invention

Therefore, it would be advantageous to have an improved capability for monitoring a switchgear.

The object of the invention is solved by the subject matter of the independent claims, wherein further embodiments are incorporated in the dependent claims.

In a first aspect, there is provided an apparatus for monitoring a switchgear, the apparatus comprising:

-an input unit;

-a processing unit; and

-an output unit.

The input unit is configured to provide the processing unit with a monitoring infrared image of the switching device. The processing unit is configured to implement a machine learning classifier algorithm to analyze the monitored infrared images and determine whether one or more anomalous hotspots exist in the switchgear. Machine learning classifier algorithms have been trained based on a number of different training images. The plurality of training images includes a plurality of composite infrared images generated by an image processing algorithm. The output unit is configured to output information related to the one or more anomalous hotspots.

In this manner, the apparatus may more accurately determine whether hotspots exist in switching devices and other electrical components, as a large training set may be generated to improve the ability to determine whether hotspots exist in a wider range of situations for different switching devices without human intervention.

The apparatus is specified for a switchgear, but the apparatus may be used to monitor other electrical devices where hot spots may occur.

Thus, the image processing algorithm can be considered to act as a composite image generator.

In an example, the image processing algorithm is a conditional countermeasure network.

In an example, the plurality of composite infrared images are generated based on at least one computer-aided design drawing, or sketch of the switchgear.

Thus, as described above, the drawing or the like may also be a drawing of other electrical devices in which hot spots may occur.

In an example, the at least one computer-aided design drawing, or sketch includes a plurality of drawings including drawing data for a plurality of different switching devices.

In an example, at least the computer-aided design drawing, drawing or sketch includes drawing data for at least one circuit breaker.

In an example, at least the computer-aided design drawing, or sketch includes a plurality of drawings including drawing data for a plurality of circuit breakers having different internal structures.

In an example, two or more of the plurality of composite infrared images are generated based on one computer-aided design drawing, or sketch of the switching device.

In an example, the plurality of composite infrared images are generated based on a corresponding number of computer-aided design drawings, or sketches of the switching device.

In an example, the generation of the one or more of the plurality of synthetic infrared images includes an addition of at least one hotspot to the one or more synthetic infrared images.

In an example, the generation of the one or more of the plurality of synthetic infrared images includes an addition of the at least one hotspot to one or more computer-aided design drawings, or sketches of the switchgear.

In an example, the adding of the at least one hotspot is performed manually.

In an example, the at least one computer-aided design drawing, drawing or sketch of the switchgear device comprises drawing data of the at least one circuit breaker.

In an example, the image processing algorithm uses a style migration algorithm.

In an example, the image processing algorithm uses a pix2pix algorithm.

In an example, the monitoring infrared image includes image data of at least one circuit breaker.

In an example, the machine learning classifier algorithm is a neural network.

In an example, the neural network is a convolutional neural network.

In an example, the processing unit is configured to update training of the machine learning classifier algorithm, the training including monitoring usage of the infrared images.

In an example, the training update includes a manual indication that the monitoring infrared image does not include an abnormal hotspot or a manual indication that the monitoring infrared image includes one or more abnormal hotspots.

In an example, the manual indication that the monitor infrared image includes one or more anomalous hotspots includes a manual indication of one or more locations in the monitor infrared image of the one or more anomalous hotspots.

In a second aspect, there is provided a system for monitoring a switchgear, the system comprising:

-an infrared camera; and

-an apparatus for monitoring a switchgear according to the first aspect.

The infrared camera is configured to acquire a monitoring infrared image of the switching device.

The aspects and examples described above will become apparent from and elucidated with reference to the embodiments described hereinafter.

Drawings

Exemplary embodiments will be described below with reference to the following drawings:

FIG. 1 shows a schematic example of sketch transformation into an infrared image; and

fig. 2 shows an example of a flow of generating a composite image.

Detailed Description

The apparatus and system enable detection of temperature hotspots in switchgear (e.g., in circuit breakers and in other electrical equipment) by identifying hotspots in infrared images of the switchgear. This is achieved by using a machine learning algorithm that is image-trained, wherein at least some of the images are generated synthetically, so that the training set is important enough for the machine learning algorithm to be able to identify hot spots in different device types, from different vantage points, and in different situations. Accordingly, synthetically created infrared images of electrical asset components or subsystems having several variations are used with machine learning methods to detect hotspots without having to use infrared images or infrared cameras, or at least reduce the need to use such images and cameras, to develop machine learning classifiers that can identify and locate hotspots based on infrared images.

This is thus achieved by an apparatus comprising an input unit, a processing unit and an output unit. The input unit is configured to provide the processing unit with a monitoring infrared image of the switching device. The processing unit is configured to implement a machine learning classifier algorithm to analyze the monitored infrared images and determine whether one or more anomalous hotspots exist in the switchgear. Machine learning classifier algorithms have been trained based on a number of different training images. The plurality of training images includes a plurality of composite infrared images generated by an image processing algorithm. The output unit is configured to output information related to the one or more anomalous hotspots.

According to an example, the image processing algorithm is a conditional countermeasure network.

According to an example, the plurality of composite infrared images are generated based on at least one computer-aided design drawing, or sketch of the switchgear.

According to an example, at least the computer-aided design drawing, drawing or sketch comprises a plurality of drawings, the plurality of drawings comprising drawing data for a plurality of different switching devices.

According to an example, at least the computer-aided design drawing, drawing or sketch comprises drawing data of at least one circuit breaker.

According to an example, at least the computer-aided design drawing, drawing or sketch comprises a plurality of drawings comprising drawing data for a plurality of circuit breakers having different internal structures.

According to an example, two or more of the plurality of composite infrared images are generated based on one computer-aided design drawing, or sketch of the switching device.

According to an example, the plurality of composite infrared images are generated based on a corresponding number of computer-aided design drawings, or sketches of the switching device.

According to an example, the generating of the one or more of the plurality of synthetic infrared images includes an addition of at least one hotspot to the one or more synthetic infrared images.

According to an example, the generating of the one or more of the plurality of synthetic infrared images includes addition of the at least one hotspot to one or more computer-aided design drawings, or sketches of the switchgear.

According to an example, the adding of the at least one hotspot is performed manually.

According to an example, the at least one computer-aided design drawing, drawing or sketch of the switchgear device comprises drawing data of the at least one circuit breaker.

According to an example, the image processing algorithm uses a style migration algorithm.

According to an example, the image processing algorithm uses the pix2pix algorithm.

According to an example, the monitoring infrared image includes image data of the at least one circuit breaker.

According to an example, the machine learning classifier algorithm is a neural network.

According to an example, the neural network is a convolutional neural network.

According to an example, the processing unit is configured to update training of the machine learning classifier algorithm, the training including monitoring usage of the infrared images.

According to an example, the training update includes a manual indication that the monitor infrared image does not include an abnormal hotspot or a manual indication that the monitor infrared image includes one or more abnormal hotspots.

According to an example, the manual indication that the monitor infrared image includes one or more anomalous hotspots includes a manual indication of one or more locations in the monitor infrared image of the one or more anomalous hotspots.

Thus, as the user indicates where in the image there are one or more anomalous hotspots.

As mentioned above, the apparatus may be part of a system with a camera that acquires the monitoring infrared image and provides it to the processing unit via the input unit.

The apparatus and system are explained in more detail with reference to a circuit breaker in a switchgear, but this is just one example, and the apparatus and system can be used in other parts of switchgear and many other types of electrical equipment where hot spots may occur and are problematic.

Therefore, setting up a scene is convenient. There is currently a great interest in using infrared data to assess the health of circuit breakers, where hot spots indicate problems and these are easily identified and interpreted by humans. However, the use of skilled technicians and engineers in this manner is very expensive. Here, a machine learning algorithm is used to achieve this goal without human intervention. However, machine learning algorithms require a large training data set of relevant data, which is difficult and expensive to provide. The apparatus and system described herein address this situation.

Thus, the described apparatus and system provide a way to analyze infrared images in their entirety regardless of camera angle or circuit breaker configuration. It takes into account in particular the fact that there are different breaker geometries. For this purpose, a machine learning algorithm (e.g., a convolutional neural network) is used. The network is trained with artificial data generated by the conditional opposition network based on a mapping of switchgear including circuit breakers and other electrical components. Thus, expensive manual intervention for calibration or zone definition is eliminated from the process.

Thus, the structural health of a component (such as a circuit breaker) in an electrical system may be detected by a hot spot created in the component. An infrared camera focused on such components captures thermal variations to detect hot spots. Automating this process using neural network classifier algorithms requires training them with correctly labeled images and requires a large number of problematic and non-problematic such training images to automatically detect such hotspots. There are several variations of circuit breakers and the internal structure is very different, so using data driven methods such as machine learning will be challenging because the number of relevant images is not large, and acquiring such images requires a long wait to acquire them for the various circuit breakers, for example when a hot spot occurs. It is current practice for thermal images to be manually inspected by domain experts from time to detect any damage in a component, and may be harmful if the elapsed monitoring period is long enough to have damage occurring at the same time.

The apparatus and system described herein address these issues. This is essentially achieved by using a conditional countermeasure network to perform the following operations:

(1) comprehensively training the required infrared images by CAD or sketches of different types of circuit breakers (different internal structures);

(2) individual pre-trained machine learning models are generated from (1) for different circuit breakers so that they can be used to detect hot spots (if any) (or one pre-trained machine learning model is generated for different circuit breakers with sufficient robustness to handle differences between circuit breakers), so monitoring of switchgear and circuit breakers can begin from the date of installation.

In other words, the apparatus and system enable the pre-trained model to be applied with potentially acceptable accuracy from the first day of installation, which may improve over time as more training data is provided. There are several device variants, such as switchgear and related circuit breakers, which are structurally different and require long observation times, otherwise several infrared cameras need to be installed to create samples to build reliable data driven models to detect hot spots. This will take a long time to detect any hot spots in practice using the model. However, by creating or generating relevant data for classifier training, the apparatus and system addresses two issues related to data collection for asset management. The first problem is the lack of infrared data for devices that train several variations of the mechanical structure of the model, and the second problem is the lack of analytical systems and agility methods that use these infrared images to automatically detect hotspots rather than manually, or to provide a semi-automatic detection mechanism where the operator is made aware of the need for further investigation.

The technical result of this solution enables maintenance and service engineers to indicate any anomalies in the circuit breaker variants through a machine learning model-based detection mechanism and reduces the lengthy waiting time for acquiring data (infrared images) during anomalous situations.

Continuing the situation before the development of the described devices and systems, some existing solutions require very precise calibration of the sensors to observe the correct position. This means that the engineering costs are high. Solutions that cover a larger area still require a human to determine the correct area to view.

For any general solution that does not require the identification of the correct area (e.g. based on a convolutional neural network), this needs to be robust to differences in breaker geometry and function to be useful. However, if the training network has never experienced certain conditions, it is unlikely that they will be correctly classified. The traditional solution of the last point is to use a large amount of training data from different circuit breakers. Like humans, neural networks learn from experience and are more adept at judging the same or similar situation as seen before. One way to address this problem is to use a large amount of training data from different circuit breakers. However, this is expensive and time consuming.

Thus, a surprising solution for use in the apparatus and system described herein involves obtaining additional data that is a synthetically generated of real data, but is based on a plot. With this additional data, the neural network can handle a wider range of situations.

A surprising solution uses a conditional countermeasure network, such as the pix2pix algorithm, which can turn the sketch into a photograph. Thus, in the present case, the conditional countermeasure network transforms the CAD or sketch of the internal components of the circuit breaker, for example, into equivalent infrared image-style transmission elements. Thus, CAD or sketches of various circuit breakers are provided, in which areas (made with the help of domain experts) where hot spots may occur are marked, and a conditional countermeasure network is used to convert these into a composite infrared image containing thermally varying features. This is shown in figure 1.

A classification model generator is then used. This element takes an image generated by the previous element "CAD-to-IR style migration element" and is trained to generate a machine-learned classification model for each variant of the circuit breaker. This pre-trained model may then identify structural damage. The machine learning model corresponding to the circuit breaker variant can detect whether the circuit breaker is suffering any structural damage and provide technical results by classifying infrared images generated in the field.

Fig. 2 shows a schematic diagram of a flow for implementing the solution. Generic CAD images of several variants of circuit breakers are stored in a database and are first presented to a domain expert who uses a template to mark locations where hot spots may occur. Then, based on the domain experts and the input of the conditional countermeasure network, several composite infrared images with variations are created for each breaker variant. The synthesized infrared image is then provided to a suitable classifier (machine learning model) along with other synthesized infrared images that do not contain hotspots to train the classifier in preparation for a pre-trained model. The pre-trained model is then used to identify and detect hotspots in the respective circuit breaker to acquire current infrared monitoring images from the camera.

Thus, in summary, the apparatus and system are capable of:

1. for hotspot mark templates: the device/system contains features that enable a domain expert to mark components that may have hot spots.

2. Automatically generating an infrared image from a CAD or sketch: the apparatus/system may automatically generate an infrared image from a CAD drawing or sketch using a conditional opposition network.

3. Generating a pre-training model: the apparatus/system may generate respective pre-trained models based on synthetically generated infrared images of circuit breaker variants to automatically classify images as inputs to a real-time system with an infrared camera to detect hotspots.