Method and system for monitoring a winding process, method for manufacturing a transformer winding, and method and system for manufacturing a transformer

Abstract

A computer vision system (40) processes at least one image (42) showing a transformer winding (20) captured during a winding process to monitor the winding process. A computer vision system (40) detects a difference between a result of the object classification and design data (49) of the transformer winding (20).

Description

Technical Field

[0001] Embodiments of the present invention relate to the manufacture of windings of electrical power system components. Embodiments of the present invention particularly relate to a quality control method and system operable to detect errors during a winding process performed on a winding device. Embodiments of the present invention also relate to a manufacturing system and method including the quality control method or system according to an embodiment. Background Art

[0002] Transformers are important components of electrical power systems. Transformers can have various configurations, depending on their intended use. Due to the specific requirements imposed on such transformers, manufacturing transformers designed for power generation, power transmission, and / or power distribution is a very complex task. This is particularly applicable to the manufacture of transformer windings.

[0003] The design of transformer windings is optimized considering their performance for a given amount of copper used. The design of transformer windings is also optimized to occupy a small amount of space while still maintaining high reliability as well as efficient cooling and insulation. This complicates the manufacturing process.

[0004] Fortunately, errors during the winding process of forming transformer windings are rare, but any failure can have a serious negative impact on power supply, operational safety, and labor safety.

[0005] During the winding process of forming windings of other electrical power system components, such as reactors, similar challenges exist.

[0006] Therefore, quality control of transformer windings or windings of other electrical power system components is an important issue for reducing the risk of failures that can have catastrophic consequences.

[0007] Unfortunately, it is not easy to perform such quality control using conventional techniques. Given the complexity of the winding process, it is difficult to detect errors. Many errors, such as those related to the position of crossovers or incorrect deviations from design data, are hidden and invisible in transformer windings. The challenges in error detection are particularly evident when the winding device provides both rotational freedom and translational freedom, such that during manufacturing the transformer winding rotates around an axis and is displaced translationally.

[0008] In view of the above, there is a need for a method and system that can mitigate the risk of using electrical power system component windings in a transformer or transformer core or another electrical power system component that are not formed according to their specifications. In particular, there is a need for a method and system that allows for the detection of quality issues while the winding process of forming the windings is in progress. Summary of the Invention

[0009] In accordance with aspects of the present invention, methods and systems are provided as recited in the independent claims. The dependent claims define preferred embodiments.

[0010] In accordance with one aspect of the present invention, a method of monitoring a winding process of forming a winding of a power system component, such as a transformer winding, on a winding device is provided. The method includes: processing, by a computer vision system, at least one image captured during the winding process and showing at least a portion of the winding (e.g., a transformer winding), wherein processing the at least one image includes performing object classification. The method further includes: detecting, by the computer vision system, a difference between the result of the object classification and the design data of the winding (e.g., a transformer winding).

[0011] The method of monitoring the winding process allows for detecting errors while the winding (e.g., a transformer winding) is still being formed. Thus, mitigation and / or corrective actions can be taken during the manufacture of the transformer winding.

[0012] The method of monitoring the winding process utilizes a computer vision system in combination with design data to detect differences therebetween. The differences are indicators that an error has occurred or may have occurred. Thus, the computer vision system provides assistance during the winding process of forming the transformer winding.

[0013] The method may include: implementing, by the computer vision system, at least one action in response to the detected difference.

[0014] Thus, when there is a difference detected by the computer vision system, one or several actions can be performed automatically or semi-automatically.

[0015] The at least one action may include: causing, by the computer vision system, a human-machine interface (HMI) to output information depending on the detected difference.

[0016] Thus, the detected difference can be used as a trigger for an output action that can warn an operator of the winding device that an error has occurred or may have occurred during the winding process.

[0017] The information may include one, several, or all of the following: an alarm and / or a warning; information indicating the location of the detected difference; information indicating the time when the detected difference occurred; information about the root cause of the detected difference; instructions for correcting the detected difference.

[0018] Thus, the detected difference can be used as a trigger for an output action that can warn an operator of the winding device and can optionally provide additional information that assists the operator in identifying the root cause of the error or assists the operator in taking appropriate or corrective mitigation actions to correct the error in the winding process.

[0019] At least one action may include a mitigation action and / or a correction action.

[0020] Accordingly, the detected difference can be used as a trigger for an action that can be performed automatically or semi-automatically to mitigate and / or correct an error identified by the computer vision system.

[0021] The action may include stopping the rotation and / or translational displacement of the winding device.

[0022] Accordingly, visual inspection or correction of the winding process can be facilitated.

[0023] The winding device may include a winding support on which a winding (e.g., a transformer winding) is formed. The winding device may include one or several motors for driving the winding support. The winding device may include: a first motor that rotates the winding support; and a second motor that translates the winding support, preferably along the axis of rotation of the winding support.

[0024] Accordingly, the manufacture of the winding (e.g., a transformer winding) is facilitated. Additionally, such a configuration of the winding device is suitable for a monitoring method using a computer vision system. An image acquisition device (such as a camera) can be fixedly mounted because the winding device is displaced when forming a transformer winding so that its top remains at substantially the same height.

[0025] The method may further include: determining at least one image acquisition time of at least one image during the winding process.

[0026] Accordingly, the computer vision system can determine at which image acquisition times during the winding process at least one image provides particularly good information to facilitate error detection. By way of illustration, one or several image acquisition times may depend on how a shield covering the conductors and / or fillers of the winding (e.g., a transformer winding) is positioned relative to the image acquisition device.

[0027] The method may further include: performing image acquisition device control and / or image selection based on the determined at least one image acquisition time.

[0028] Accordingly, image acquisition can be controlled in such a way that at least one image provides particularly good information to facilitate error detection. Alternatively or additionally, an image particularly suitable for detecting errors occurring during the winding process can be selected from the image stream provided by the image acquisition device (e.g., because the shield covering the conductors or fillers is suitably positioned relative to the image acquisition device).

[0029] At least one image acquisition time can be determined based on at least one position of a winding device. At least one image acquisition time can be determined based on the rotational position and / or translational position of the winding device. Design data can be used in combination with the (multiple) positions of the winding device to determine at least one image acquisition time. For illustration, the design data can be used to determine which rotational and / or translational positions of the winding device shield, conductor, filler, and / or crossover and / or other winding components or structural features are properly positioned within the field of view of the image acquisition device.

[0030] At least one image acquisition time can be determined based on an image stream acquired by at least one image acquisition device. A computer vision system can use image processing techniques (such as edge detection techniques, Fourier-based techniques, and / or contrast enhancement techniques) to identify the images in the stream that are most likely to provide information about whether an error has occurred during the winding process.

[0031] A computer vision system can use a first machine learning (ML) model to determine at least one image acquisition time. The first ML model can have a first input layer that receives the image pixels of the stream. The first ML model can have a first output layer that outputs a value indicating whether further analysis is to be performed on any of the images. Alternatively or additionally, the first input layer can receive the winding device position, such as the rotational and / or translational position of the winding support on which the transformer winding is formed.

[0032] Processing at least one image can include identifying a region of interest (ROI) in which object classification is to be performed.

[0033] Thereby, object classification can be focused on regions where differences (if any) are likely to be visible.

[0034] The boundary of the ROI can include the edge of the shield of the transformer winding.

[0035] Thereby, object classification can be focused on the region between adjacent shields interposed between consecutive disks of the transformer winding.

[0036] Alternatively or additionally, the boundary of the ROI can include the inner circumferential edge and the outer circumferential edge of the layer formed by the conductor and the filler.

[0037] Thereby, object classification can be focused on regions where winding errors are likely to be encountered.

[0038] Identifying the ROI can include performing edge detection.

[0039] Thereby, the ROI can be efficiently identified.

[0040] Identifying the ROI may include applying at least one ROI classifier to at least one image. Several ROI classifiers may be applied to at least one image. Each of one or more ROI classifiers may indicate for each pixel whether the corresponding pixel is classified as being within the ROI or outside the ROI.

[0041] One or several ROI classifiers may include at least one second ML model. The at least one second ML model may have a second input layer that receives pixels of the at least one image. The at least one second ML model may have a second output layer that outputs for each pixel a value indicating whether the corresponding pixel is within the ROI.

[0042] The at least one second ML model may include a deep learning ML model.

[0043] Thus, the object classification task can be reliably performed.

[0044] The at least one second ML model may include a U-Net ML model.

[0045] Thus, the ROI identification task can be reliably performed.

[0046] Object classification may include executing at least one third ML model. The at least one third ML model may have a third input layer that receives pixels located within the ROI. The at least one third ML model may have a third output layer that outputs an object classifier indicating to which of several object categories the corresponding pixel in the ROI is assigned. The third output layer may indicate the probability that the object belongs to a given object category (such as a filler, a conductor, a shield, and / or a crossover).

[0047] The third ML model may include a random forest (RF) model, but is not limited thereto.

[0048] Thus, the object classification task can be performed in a manner that provides particularly good results during inference.

[0049] Object classification may classify at least conductors, fillers, and shields in at least one image.

[0050] Alternatively or additionally, object classification may classify crossovers. Object classification may classify inner crossovers and outer crossovers.

[0051] Alternatively or additionally, object classification may classify other components of the transformer winding and / or other structural features.

[0052] Thus, object categories suitable for comparison with design data are achieved.

[0053] Several classifiers can be used, each classifier being associated with a given object class (such as, for example, a filler, a conductor, a shield, and / or a crossover) and trained to identify the given object class.

[0054] Object classification can define a sequence of objects that includes at least a conductor and a filler arranged along a radial direction (e.g., from the inside to the outside or from the outside to the inside), as identified in the ROI of an image.

[0055] Processing at least one image can include using one or several position measurements of a winding device. The one or several position measurements can include rotational and / or translational position measurements. The one or several position measurements can be received from a sensor. Alternatively, the one or several position measurements can be received from a controller of the winding device.

[0056] The one or several position measurements can be used to determine which images are to be further analyzed, for example by performing ROI determination and object classification. Alternatively or additionally, the one or several position measurements can be used to assist in ROI determination.

[0057] Processing at least one image can include executing at least one ML model having an input layer that receives at least a portion of the at least one image and / or a winding device position, wherein the at least one ML model has an output layer, and wherein the output layer provides at least one of the following: at least one image acquisition time defining the at least one image to be processed; the ROI in which object classification is to be performed; and / or an object classifier indicating to which of several predefined object classes an object is assigned, respectively.

[0058] The method can further include: determining a target sequence of objects along a radial direction of a transformer winding based on design data.

[0059] Thereby, the design data is converted into a format that can be easily compared with the results of object classification.

[0060] The design data can include at least one crossover diagram.

[0061] The design data can define the winding as a three-dimensional volume component. The design data can have various formats, such as one or several crossover diagrams, but is not limited thereto.

[0062] Determining the target sequence of objects can include generating a machine-readable representation indicating the target sequence based on at least one crossover diagram. The machine-readable representation can include an XML representation, but is not limited thereto.

[0063] Detecting a difference can include comparing the target sequence with the results of object classification.

[0064] The target sequence can be automatically determined by a computer vision system or a processing system communicatively coupled thereto.

[0065] A winding (e.g., a transformer winding) may include several layers (also referred to as disks in the art). When forming a winding on a winding device, at least one image may show at least one of the several layers on the upper surface of the winding. An image of the uppermost layer on the upper surface of the winding may be processed in this method to detect differences.

[0066] During the winding process, processing at least one image and comparing the result of classifying the object with the design data may be repeated on an ongoing basis (i.e., continuously).

[0067] When the method uses one or several ML models (such as one, several, or all of the first, second, and third ML models mentioned above), labeled image data may be used to train the ML models. The labeled image data may include synthetically generated images that are generated from real-world images showing errors during the winding process. The generation of the synthetically generated images may include one or several image modification operations, such as rotation, translation, mirroring, adding distortion, etc.

[0068] One or several ML models may be trained using supervised training or semi-supervised training.

[0069] The method may further include: using the labeled image data to train the ML models.

[0070] The training may be performed by a computer system that is different and separate from the computer vision system, or may be performed by the computer vision system.

[0071] Training the ML models may include synthetically generating images from real-world images showing errors during the winding process. The generation of the synthetically generated images may include one or several image modification operations, such as rotation, translation, mirroring, adding distortion, etc.

[0072] The synthetically generated images allow for providing a more balanced training set, thus improving the results when the ML models are subsequently used for inference during the on-site use of the computer vision system.

[0073] The ML models may be trained to perform at least one classification task. The at least one classification task may be or may include one or several of the following: classifying an image as an image to be processed; classifying a pixel as belonging to a ROI; performing object classification.

[0074] The ML models may include at least one of the following: a deep learning model, such as a U-Net model; a random forest model; a recurrent neural network (RNN); a convolutional neural network (CNN).

[0075] Synthetically generating images may include using an adversarial ML model to generate synthetic images.

[0076] The transformer winding can be the following transformer windings: power generation and transmission systems (such as high-voltage, medium-voltage transmission systems), distribution systems, railway transformers, or reactance or inductance systems, or shunt reactors (such as variable reactors or other shunt reactors). According to another aspect of the present invention, a method for training at least one ML model is provided, where at least one ML model is used to monitor the winding process of forming a winding (such as a transformer winding) of a power system component on a winding device.

[0077] The method may include: using image data showing the markings of these windings when forming the windings (such as transformer windings) to train one or several ML models (such as one, several, or all of the first, second, and third ML models mentioned above).

[0078] The training method may include: synthetically generating at least some training images from real-world images showing errors during the winding process. Generating the synthetically generated images may include one or several image modification operations, such as rotation, translation, mirroring, adding distortion, etc.

[0079] One or several ML models can be trained using supervised training or semi-supervised training.

[0080] The training can be performed by a computer system different and separate from the computer vision system, or can be performed by the computer vision system.

[0081] The synthetically generated images allow for providing a more balanced training set, thus improving the results when the ML model is subsequently used for inference during the on-site use of the computer vision system.

[0082] The training method may include training at least one ML model to perform at least one classification task. At least one classification task can be or can include one or several of the following: classifying an image as an image to be processed; classifying a pixel as belonging to an ROI; performing object classification.

[0083] At least one ML model may include at least one of the following: a deep learning model, such as a Unet model; a random forest model; a recurrent neural network (RNN); a convolutional neural network (CNN).

[0084] Synthetically generating images may include using an adversarial ML model to generate synthetic images.

[0085] The training method can be performed by the computer vision system that executes the trained ML model(s) during inference, or by a computing system separate from the computer vision system.

[0086] The training method may include deploying the trained ML model(s) to a computer vision system for monitoring the winding process of forming a transformer winding on a winding device.

[0087] According to another aspect of the present invention, there is provided a method of manufacturing a winding of a power system component (such as a transformer winding). The method includes: controlling, by at least one control device, a winding device on which a winding (such as a transformer winding) is formed. The method includes: performing a method of monitoring the winding process according to any one of the embodiments while forming the transformer winding.

[0088] The method may further include: correcting at least one error that occurs during the winding process in response to a detected difference.

[0089] The method may further include: discarding the transformer winding in response to a detected difference.

[0090] The method may be automatically performed by a system including a winding device and a computer vision system, the computer vision system being operable to perform the monitoring method discussed in detail herein. The computer vision system may be operable if it is coupled to a controller of the winding device.

[0091] According to another aspect of the present invention, there is provided a method of manufacturing a transformer core. The method includes: assembling a transformer core including at least one transformer winding manufactured using the method of manufacturing a transformer winding according to an embodiment.

[0092] Assembling the transformer core may include assembling at least one transformer winding with a yoke.

[0093] The yoke may include a laminate including several layers. This allows for efficient transfer of magnetic flux in the transformer core.

[0094] The transformer core may be a transformer core for a power transmission system (such as a high-voltage, medium-voltage power transmission system), a power distribution system, a railway transformer, or a reactance or inductance system.

[0095] According to another aspect of the present invention, there is provided a method of manufacturing a transformer. The method includes: assembling a transformer including at least one transformer winding manufactured using the method of manufacturing a transformer winding according to an embodiment.

[0096] Assembling the transformer may include assembling at least one transformer winding with a yoke.

[0097] The yoke may include a laminate including several layers. This allows for efficient transfer of magnetic flux in the transformer core.

[0098] Assembling a transformer can include arranging at least one transformer winding in a transformer tank.

[0099] Assembling a transformer can include filling the interior of the transformer tank with an insulating fluid (such as, insulating oil).

[0100] The transformer can be a transformer of a power transmission system (such as, a high-voltage, medium-voltage power transmission system), a power distribution system, or a railway system, or a reactor (such as, a variable reactor or a shunt reactor).

[0101] According to another aspect of the present invention, there is provided a system configured to monitor a winding process of a winding (such as, a transformer winding) forming a power system component. The system includes a computer vision system configured to: process at least one image captured during the winding process and showing at least a part of the winding (e.g., a transformer winding) to perform object classification; detect a difference between the result of the object classification and the design data of the winding; and perform an action in response to the detected difference.

[0102] The system for monitoring the winding process allows detecting errors while the winding of the power system component is still being formed. This enables taking mitigation and / or correction actions during the manufacture of the transformer winding.

[0103] The system for monitoring the winding process uses a computer vision system in combination with design data to detect the difference therebetween. The difference is an indicator that an error has occurred or may have occurred. Thus, the computer vision system provides assistance during the winding process of the winding (e.g., a transformer winding) forming a power system component.

[0104] The system may further include a human-machine interface (HMI). The computer vision system is configured to control the HMI in response to the detected difference.

[0105] The system may be configured such that the information can include one, several, or all of the following: an alarm and / or a warning; information indicating the location of the detected difference; information indicating the time when the detected difference occurred; information about the root cause of the detected difference; instructions for correcting the detected difference.

[0106] The system may be configured such that at least one action can include a mitigation action and / or a correction action. The system may be configured to generate and output a control signal or a control command for a winding device or the HMI.

[0107] The system may be configured such that the action can include stopping the rotation and / or translational displacement of the winding device.

[0108] The system may further include a winding device. The winding device may include a winding support on which a winding (e.g., a transformer winding) is formed. The winding device may include one or several motors for driving the winding support. The winding device may include: a first motor for rotating the winding support; and a second motor for translating the winding support, preferably along the rotation axis of the winding support.

[0109] The system may be configured to determine at least one image acquisition time of at least one image during the winding process.

[0110] The system may be configured to perform image acquisition device control and / or image selection based on the determined at least one image acquisition time.

[0111] The system may be configured such that at least one image acquisition time can be determined based on at least one position of the winding device. At least one image acquisition time can be determined based on the rotational and / or translational position of the winding device. Design data may be used in combination with the (multiple) positions of the winding device to determine at least one image acquisition time. By way of illustration, design data may be used to evaluate which rotational and / or translational positions of winding device shields, conductors, fillers, and / or crossings and / or other components or structural features are suitably positioned within the field of view of the image acquisition device.

[0112] The system may be configured such that at least one image acquisition time can be determined based on an image stream acquired by at least one image acquisition device. A computer vision system may use image processing techniques (such as edge detection techniques, Fourier-based techniques, and / or contrast enhancement techniques) to identify the images in the stream that are most likely to provide information about whether an error has occurred during the winding process.

[0113] The system may be configured such that the computer vision system can use a first machine learning (ML) model to determine at least one image acquisition time. The first ML model may have a first input layer that receives the image pixels of the stream. The first ML model may have a first output layer that outputs a value indicating whether any of the images is to be further analyzed and (if so) which one of the images is to be further analyzed. Alternatively or additionally, the first input layer may receive the winding device position, such as the rotational and / or translational position of the winding support on which the transformer winding is formed.

[0114] The system may be configured to identify a region of interest (ROI) in which object classification is to be performed.

[0115] The system may be configured such that the boundary of the ROI may include the edge of the shield of the transformer winding.

[0116] Alternatively or additionally, the system can be configured such that the boundaries of the ROI can include the inner circumferential edge and the outer circumferential edge of the layer formed by the conductor and the filler.

[0117] The system can be configured to perform edge detection to detect the ROI.

[0118] The system can be configured to apply at least one ROI classifier to at least one image. Several ROI classifiers can be applied to at least one image. Each of the one or more ROI classifiers can indicate for each pixel whether the corresponding pixel is classified as being within the ROI or outside the ROI.

[0119] One or several ROI classifiers can include at least one second ML model. The at least one second ML model can have a second input layer that receives the pixels of the at least one image. The at least one second ML model can have a second output layer that outputs for each pixel a value indicating whether the corresponding pixel is within the ROI.

[0120] The at least one second ML model can include a deep learning ML model.

[0121] The at least one second ML model can include a U-Net ML model.

[0122] The system can be configured such that object classification can include performing at least one third ML model. The at least one third ML model can have a third input layer that receives the pixels located within the ROI. The at least one third ML model can have a third output layer that outputs an object classifier indicating to which of several object categories the corresponding pixel in the ROI is assigned.

[0123] The third ML model can include a Random Forest (RF) model, but is not limited thereto.

[0124] The system can be configured such that object classification can classify at least the conductor, the filler, and the shield in at least one image; and / or other components and / or structural features of the winding.

[0125] Alternatively or additionally, the system can be configured such that object classification can classify crossings. Object classification can classify inner crossings and outer crossings.

[0126] Object classification can define an object sequence that includes at least the conductor and the filler arranged along a radial direction (e.g., from the inside to the outside or from the outside to the inside), as identified in the ROI of the image.

[0127] The system can be configured such that processing at least one image can include using one or several position measurements of a winding device. One or several position measurements can include rotational and / or translational position measurements. One or several position measurements can be received from a sensor. Alternatively, one or several position measurements can be received from a controller of the winding device.

[0128] The system can be configured to use one or several position measurements to determine which images are to be further analyzed, for example by performing ROI determination and object classification. Alternatively or additionally, the system can be configured to use one or several position measurements to assist in ROI determination.

[0129] The system can be configured such that processing at least one image can include executing at least one ML model having an input layer that receives at least a portion of at least one image and / or a winding device position, wherein the at least one ML model has an output layer, and wherein the output layer provides at least one of the following: at least one image acquisition time defining the at least one image to be processed; an ROI in which object classification is to be performed; and / or an object classifier defining one of several predefined object classes.

[0130] The system can be configured to determine a target sequence of objects along a radial direction of a transformer winding based on design data.

[0131] The design data can include at least one cross-sectional view.

[0132] The design data can define a winding (e.g., a transformer winding) as a three-dimensional volumetric component. The design data can have various formats, such as one or several cross-sectional views, but is not limited thereto.

[0133] The system can be configured such that determining the target sequence of objects can include generating a machine-readable representation indicative of the target sequence based on at least one cross-sectional view. The machine-readable representation can include an XML representation, but is not limited thereto.

[0134] The system can be configured such that detecting a difference can include comparing the target sequence and the result of object classification.

[0135] A computer vision system can be configured to determine a target sequence. Alternatively, the system can further include a processing system communicatively coupled to the computer vision system and configured to determine the target sequence and provide the target sequence to the computer vision system.

[0136] The system can be configured to monitor the winding process of windings for a power system, such as a transformer winding, which includes several layers (also known as disks in the art). When forming a transformer winding on a winding device, at least one image can show at least one of the upper layers on the upper surface of the transformer winding. Images of the uppermost layer on the upper surface of the transformer winding can be processed in the method to detect differences.

[0137] The system can be configured to repeatedly process at least one image on an ongoing basis (i.e., continuously) during the winding process and compare the results of object classification with design data.

[0138] When the system is configured to use one or several ML models (such as one, several, or all of the first, second, and third ML models mentioned above), labeled image data can be used to train the ML models. The labeled image data can include synthetically generated images, which are generated from real-world images showing errors during the winding process. The generation of synthetically generated images can include one or several image modification operations, such as rotation, mirroring, adding distortion, etc.

[0139] The ML model can include at least one of the following: a deep learning model, such as a U-Net model; a random forest model; a recurrent neural network (RNN); a convolutional neural network (CNN).

[0140] The transformer winding can be a transformer winding of the following: a power transmission system (such as a high-voltage, medium-voltage power transmission system), a power distribution system, a railway transformer, or a reactance or inductance system (such as a shunt reactor and / or a variable reactor).

[0141] According to another aspect of the present invention, a manufacturing system for manufacturing a transformer winding, a transformer core, or a transformer is provided. The system includes: a winding device; and a system configured to monitor the winding process of forming a transformer winding according to an embodiment.

[0142] According to a further embodiment, machine-readable instruction code is provided, which when executed by at least one programmable circuit causes the programmable circuit to execute the method according to the embodiment.

[0143] According to a further embodiment, a non-transitory storage medium storing machine-readable instruction code is provided, which when executed by at least one programmable circuit causes the programmable circuit to execute the method according to the embodiment.

[0144] According to another aspect of the present invention, a transformer winding, a transformer core, or a transformer manufactured using the method according to the embodiment is provided.

[0145] The effects achieved by the system according to the embodiments correspond to the effects disclosed in detail in association with the method according to the embodiments.

[0146] Embodiments of the present invention are disclosed by the following list of embodiments:

[0147] Embodiment 1: A method for monitoring a winding process of forming a transformer winding on a winding device, the method comprising: processing, by a computer vision system, at least one image captured during the winding process and showing at least a part of the transformer winding, wherein processing the at least one image includes performing object classification; detecting, by the computer vision system, a difference between the result of the object classification and the design data of the transformer winding; and implementing, by the computer vision system, at least one action in response to the detected difference.

[0148] Embodiment 2: The method according to Embodiment 1, wherein the at least one action includes: causing, by the computer vision system, a human-machine interface HMI to output information depending on the detected difference.

[0149] Embodiment 3: The method according to Embodiment 2, wherein the information includes one, several, or all of the following: an alarm and / or a warning; information indicating the location of the detected difference; information indicating the time when the detected difference occurs; information about the root cause of the detected difference; instructions for correcting the detected difference.

[0150] Embodiment 4: The method according to any one of the preceding embodiments, wherein the at least one action includes one or both of the following: a mitigation action; a correction action.

[0151] Embodiment 5: The method according to any one of the preceding embodiments, wherein the action includes stopping the rotation and / or translational displacement of the winding device.

[0152] Embodiment 6: The method according to any one of the preceding embodiments, further comprising: determining at least one image acquisition time of the at least one image during the winding process; and performing image acquisition device control and / or image selection based on the determined at least one image acquisition time.

[0153] Embodiment 7: The method according to any one of the preceding embodiments, wherein processing the at least one image includes identifying a region of interest ROI in which object classification is to be performed, and optionally the boundary of the ROI may include the edge of a shield of the transformer winding.

[0154] Example 8: The method according to any one of the preceding examples, wherein processing the at least one image includes: using one or several position measurement values of a winding device; and / or executing at least one machine learning ML model having: an input layer that receives at least a portion of the at least one image; and an output layer, wherein the output layer provides a region of interest ROI in which object classification is to be performed, and / or an object classifier.

[0155] Example 9: The method according to any one of the preceding examples, wherein detecting a difference includes: determining a target sequence of objects along a radial direction of a transformer winding based on design data; and comparing the target sequence with the result of object classification.

[0156] Example 10: The method according to any one of the preceding examples, wherein object classification classifies at least conductors, fillers, and shields in the at least one image; and / or crossings; and / or other components of the transformer winding; and / or other structural features of the transformer winding.

[0157] Example 11: A method of manufacturing a transformer winding, the method including: controlling, by at least one control device, a winding device on which a transformer winding is formed; and executing the method according to any one of the preceding examples while forming the transformer winding.

[0158] Example 12: A method of manufacturing a transformer core or a transformer, the method including: manufacturing at least one transformer winding by the method according to Example 11; and assembling a transformer core or a transformer including the at least one transformer winding.

[0159] Example 13: A system configured to monitor a winding process of forming a transformer winding, the system including: a computer vision system configured to: process at least one image captured during the winding process and showing at least a portion of the transformer winding to perform object classification; detect a difference between the result of object classification and design data of the transformer winding; and perform an action in response to the detected difference.

[0160] Example 14: The system according to Example 13, further including: a human-machine interface HMI, wherein the computer vision system is configured to control the HMI in response to the detected difference.

[0161] Example 15: A manufacturing system for manufacturing a transformer winding, a transformer core, or a transformer, the system including: a winding device; and the system according to Example 13 or Example 14.

[0162] The various effects and advantages are achieved by embodiments of the present invention. These systems and methods allow for the detection of errors while the winding process is ongoing. This allows for corrective and / or mitigation actions to be taken during the winding process. Even if the error is irreparable, these systems and methods provide an advantage as the winding process can be terminated early (i.e., before the transformer winding is completed), thereby reducing the underlying materials.

[0163] If the winding process is completed, these systems and methods are also capable of detecting errors that will be located inside the transformer winding. This provides a significant advantage compared to techniques that analyze the transformer winding after the winding process is completed.

[0164] These systems and methods are configured such that they can automatically monitor the winding process using images acquired by one or several image acquisition devices (such as cameras).

[0165] These systems form components of transformers (such as transformers in power transmission systems or distribution systems) with complex configurations and structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0166] Embodiments of the present invention will be described with reference to the accompanying drawings, in which like or identical reference numerals designate elements having like or identical configurations and / or functions.

[0167] Figure 1 A system including a monitoring system is shown.

[0168] Figure 2 A monitoring system is shown.

[0169] Figure 3 A flowchart is shown.

[0170] Figure 4 A flowchart is shown.

[0171] Figure 5 A monitoring system is shown.

[0172] Figure 6 An image illustrating ROI detection and object classification is shown.

[0173] Figure 7 An image illustrating ROI detection is shown.

[0174] Figure 8 A block diagram of a monitoring system is shown.

[0175] Figure 9 A flowchart is shown.

[0176] Figure 10 A flowchart is shown.

[0177] Figure 11 A transformer is shown.

[0178] Figure 12 shows an assembly system.

[0179] Figure 13 shows a flowchart. Detailed implementation

[0180] Embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, like or identical reference numerals denote elements having like or identical configurations and / or functions.

[0181] Although the embodiments will be described in connection with a transformer winding configured to be installed in a transformer in a power transmission or distribution system, the embodiments are not limited thereto. Although the embodiments will be described in connection with certain winding configurations, such as a transformer winding including several disks with spacers disposed between the disks, the embodiments are not limited thereto.

[0182] Unless otherwise explicitly stated, the features of the embodiments may be combined with each other.

[0183] Figure 1 Shows system 10 according to an embodiment. System 10 is configured to perform a winding process for forming a transformer winding. System 10 is configured to monitor the winding process in which the transformer winding 20 is formed. System 10 includes a computer vision system 40 for monitoring the winding process to detect errors that may occur during the winding process.

[0184] System 10 includes a winding device 30. The winding device 30 includes a winding support on which the transformer winding is supported during the winding process of the transformer winding. The winding device 30 may be configured to rotate the transformer winding 20 about an axis 31 while forming the transformer winding 20. The winding device 30 may be configured to shift the already formed portion of the transformer winding 20 along the axis 31 in a translational manner. The shift along the axis 31 may be performed respectively after each layer (or disk) of the transformer winding 20 has been completed. The distance by which the already formed portion of the transformer winding 20 is shifted may correspond to the height of each disk formed during one full rotation of the winding device, possibly including spacers interposed between adjacent layers.

[0185] The winding device 30 may include a first motor 32 configured to rotate the transformer winding during the winding process.

[0186] The winding device 30 may include a second motor 33 configured to shift the transformer winding 20 in a translational manner during the winding process. The translational shift may be achieved intermittently, for example, in an incremental, step-by-step manner. The translational shift may be achieved after a complete revolution about the axis 31. As will be described in more detail below, the translational shift along the axis 31 also provides the effect that during the winding process, the images captured by the image acquisition device 41 for visual quality monitoring remain at a substantially same height above the topmost surface 21 of the partially wound transformer coil 20.

[0187] The winding device 30 may include one or several position sensors 34. The one or several position sensors 34 may be configured to measure the rotational and / or translational position of the winding device 30. The image processing system 50 of the computer vision system 40 may use these position measurements to determine, for example, which images need further processing and / or perform ROI detection.

[0188] As a supplement or alternative to receiving position measurements from the (one or more) position sensors 34, the computer vision system may receive position information from the winding device controller 35.

[0189] The computer vision system includes an image processing system 50. The image processing system 50 will be described in more detail below.

[0190] The image processing system 50 is generally configured to process the images 42 captured by one or several image acquisition devices 41. The one or several image acquisition devices 41 may be mounted in a fixed manner relative to the frame of the winding device 30. The translational shift achieved by the winding device along the axis 31 prevents excessive variation in the angle between the optical axis of the image acquisition device 40 and the topmost layer 21 of the transformer winding 20 being wound.

[0191] The image processing system 50 is configured to perform at least object classification. The object classification may specify a series of different objects, such as conductors and fillers seen along the radial direction from the radially inner edge to the radially outer edge of the topmost layer 21 of the formed part of the transformer winding 20.

[0192] The image processing system 50 may optionally be configured to perform additional processing on the images 42. By way of illustration, the image processing system 50 may be configured to perform preprocessing to achieve one or more of the following: adjusting perspective effects; enhancing contrast; identifying periodic features; eliminating light reflections, etc.

[0193] Alternatively or additionally, the image processing system 50 may be configured to perform region of interest (ROI) determination. The ROI determination may be implemented as a classifier. The classifier may provide a binary output for any pixel indicating whether the corresponding pixel is included in the ROI. Then, object classification may be restricted to the ROI. When the transformer winding 20 includes several layers with shields arranged therebetween, the ROI classifier may operate to distinguish pixels representing conductors or fillers from pixels representing shields.

[0194] The image processing system 50 uses the result of object classification and the design data 49 specifying the target design of the transformer coil 20 to detect a difference that indicates an error occurring during the winding process. The image processing system 50 or a separate computing system coupled thereto may process the design data 49 into a format that can be compared with the object classification result. By way of illustration, design data 49 including one or several cross diagrams may be automatically converted into a list that specifies for the desired rotational and translational positions of the winding device 30 which objects are expected to be present in any ROI within the field of view of the image acquisition device 41.

[0195] When the image processing system 50 detects a difference between the result of object classification and the result expected according to the design data, the image processing system 50 may trigger an action. The action may include an output via the human-machine interface (HMI) 37. The output may indicate that a difference has been detected, but may also include additional information such as information indicating the location where the difference was detected and / or suggestions on how the identified difference may be corrected, if such correction is possible. Additionally or alternatively, the output may affect the operation of the winding device 30. By way of illustration, the image processing system 50 may trigger the winding device controller 35 to stop rotating and / or translating in response to the detected difference. This facilitates visual inspection by a human operator and possible corrective actions.

[0196] Additional features of the image processing system according to an embodiment will be described in more detail hereinafter.

[0197] Figure 2 The image processing system 50 is shown. The image processing system 50 has at least one first interface 51 to receive an image 42. The received image 42 shows at least a portion of the top face 21 of the transformer winding 20 during various stages of the winding process. The at least one first interface 51 may be a data interface configured to receive the image 42 as digital data. The at least one first interface 51 may also be configured to receive design data 49 specifying the desired target configuration of the transformer winding 20.

[0198] The result of the processing performed by the image processing system 50 can be output via at least one first interface 51. Alternatively, the image processing system 50 can have at least one second interface 52, which is configured to be communicatively coupled to the winding device controller 35 and / or the HMI 37. The image processing system 50 can generate and output signals or commands to trigger actions, such as providing information about the detected differences by the HMI 37 and / or causing the winding device controller 35 to react to the detected differences.

[0199] The image processing system 50 can include a storage system 53. Images (such as an image stream received from the (multiple) image acquisition devices 41) can be temporarily stored in the storage system 53 for processing and further analysis. The storage system 53 can also include parameters for processing the image 42, such as the parameters of an ROI classifier that has been trained to identify ROIs in the image and / or the parameters of an object classifier that has been trained to distinguish different objects (such as conductors, fillers, shields, and / or crossings, and / or other components or structural features of the winding).

[0200] The image processing system 50 can include one or several processing circuits 60. The one or several processing circuits 60 can be programmable circuits. The one or several processing circuits can include any one or any combination of integrated circuits, integrated semiconductor circuits, processors, controllers, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), (multiple) circuits including qubits and / or quantum gates, but not limited thereto.

[0201] The image processing system 50 can include an image time selector 61. The image time selector 61 can receive the (multiple) winding device positions and / or the image stream included in the image 42 as inputs. Then, the image time selector 61 can determine which images are to be further processed to perform object classification. The image time selector 61 can select the images that have already been captured from the image stream for more detailed processing. Alternatively or additionally, the image time selector 61 can predict at which image acquisition times the images are to be taken, which is particularly useful for detecting potential differences from the design data. The image processing system 50 may cause the (multiple) image acquisition devices 41 to capture images according to the output of the image time selector 61.

[0202] The image processing system 50 may include an ROI detector 62. The ROI detector 62 may include a first classifier, which is also referred to herein as the ROI classifier. The ROI detector 62 receives the pixels of the image, which may be the pixels after image preprocessing. The ROI detector 62 may classify each pixel as a pixel belonging to an ROI or a pixel not belonging to any ROI. The ROI detector 62 may use edge detection techniques. Alternatively or additionally, the ROI detector 62 may include one or several trained ML models suitable for performing image segmentation tasks. A deep learning model is an example of such an ML model, but the implementation of the ROI detector 62 is not limited thereto.

[0203] The image processing system 50 may include an object classifier 63. The object classifier 63 is configured to classify each pixel in at least one ROI as belonging to one of several object categories, which represent different physical component types included in the transformer winding 20. By way of illustration, these object categories may include object categories for: conductors (it should be understood that the conductors may be covered with dielectric materials), fillers between conductors that may be arranged within the layers (also referred to as discs) of the transformer winding, or shields that may be arranged between adjacent layers of the transformer winding. This list is not exhaustive. Additional or alternative object categories may be provided to reflect the components and / or structural features of the winding.

[0204] More than one object classifier 63 may be used. By way of illustration, there may be provided: an object classifier that is specifically adapted to distinguish between conductors, fillers, and shields; and another object classifier that is specifically adapted to identify crossings and / or distinguish between internal crossings and sequential crossings.

[0205] The image processing system 50 may include a difference detector 64, which is configured to detect the differences between the objects classified by the object classifier 63 and the expected object sequence, such as the differences in the radial direction along the topmost layer 21 of the transformer winding according to the design data 49.

[0206] The image processing system 50 may include an output generator 65. The output generator 65 may be configured to generate a command or signal for triggering an action in response to the output of the difference detector 64. The action may include providing information related to the detected difference (such as a warning or an alarm), and / or a mitigation or correction action.

[0207] Figure 3 is a flowchart of method 80. Method 80 may be automatically executed by the computer vision system 40 or the image processing system 50 of the computer vision system 40.

[0208] At process block 81, image preprocessing is performed. The image preprocessing may include correcting perspective effects (e.g., by applying a transformation matrix to the image), edge enhancement, contrast enhancement, etc. The image preprocessing may include performing at least one transformation of the image, depending on the intrinsic and / or extrinsic camera parameters of the camera of the image acquisition device. The intrinsic and / or extrinsic camera parameters may be known (e.g., provided by the manufacturer of the camera) or determined in a calibration routine.

[0209] At process block 82, the ROI is detected. Detecting the ROI may include edge detection. The ROI may be detected such that it is bounded by the edges of the shield on the topmost layer 21 and by the radial inner and outer circumferences of the topmost layer 21.

[0210] At process block 83, object classification is performed. The object classification may distinguish different physical components present in the transformer winding 20. The object classification may determine in what sequence the conductors and the filler are arranged along the radial direction of the transformer winding 20. The object classification may additionally or alternatively distinguish different structural features (such as crossings) present in the transformer winding 20.

[0211] At process block 84, possible differences between the results of the object classification and the design data are detected. In response to detecting a difference, an action may be performed.

[0212] Sensor signals may be used at various stages of the process. The sensor signals may indicate the position of the winding device. The sensor signals may indicate the translational and / or rotational position of the winding device.

[0213] The winding device position may be used in various ways. By way of illustration, the sensor signals may determine which image preprocessing is to be applied to accommodate the corresponding height difference between the image acquisition device 41 and the topmost layer 21. The winding device position may also be used when performing ROI detection.

[0214] The winding device position may also be used when determining which object sequence is expected to be present in the image (e.g., in the ROI) at a given winding device position based on the design data.

[0215] Figure 4 is a flowchart of method 90. Method 90 may be automatically performed by the computer vision system 40 or the image processing system 50 of the computer vision system 40.

[0216] At process block 91, the (multiple) winding device positions are obtained. The (multiple) winding device positions may be obtained from the (multiple) sensors 34 or the winding device controller 35.

[0217] At process block 92, an object sequence is determined based on one or several images 42 (e.g., in the topmost layer 21). The object sequence can be determined along a radial direction from the inner circumference to the outer circumference of the topmost layer 21. Object sequence determination can use, for example, ROI detection and object classification as previously described.

[0218] At process block 93, it is determined whether the determined object sequence and the design data are consistent with each other. If there is consistency (as is usually the case), the process returns to process block 91. The various process blocks can be repeated for multiple rotational winding device positions and (if any) multiple translational winding device positions.

[0219] At process block 94, if there is an inconsistency between the detected object sequences in the design data, an action is triggered. As previously described, the action can include controlling the HMI and / or triggering correction and / or mitigation actions.

[0220] Various classifiers that can be used in computer vision system 40, and more specifically in image processing system 50, can be implemented using techniques such as: edge detection (for example, to detect ROIs or detect the boundaries between different objects within an ROI), detection of periodicity (for example, to detect the periodic arrangement of layers at the outer cylindrical surface of a transformer winding and / or to detect the object sequence within an ROI); but not limited to this.

[0221] One or several of the classifiers can be implemented using trained ML models. More than one ML model can be used. For illustration, image processing system 50 can include a first ML model configured to determine at which image acquisition times the most relevant images are captured for further analysis. Alternatively or additionally, image processing system 50 can include a second ML model configured to detect ROL. Alternatively or additionally, image processing system 50 can include a third ML model configured to perform object classification.

[0222] Several trained ML models or other classifiers can be deployed and used to perform any of these functions. For illustration, several ROI classifiers can be applied to the same image data to achieve more accurate detection of possible errors during the winding process.

[0223] Figure 5 Image processing system 50 including at least one ML model 101 is shown. ML model 100 can include an input layer 101, an output layer 102, and a hidden layer 103. The specific configuration of ML model 100 can respectively depend on the function it will perform. Input layer 101 and output layer 102 similarly depend on the function that ML model 100 has in the computer vision system.

[0224] A computer vision system may use a first ML model to determine at least one image acquisition time of an image to be processed to perform object classification. In this case, the first ML model may have a first input layer 101 that receives the image pixels of an image stream. The first ML model may have a first output layer 102 that outputs a value indicating whether further analysis is to be performed on any of the images. Alternatively or additionally, the first input layer may receive the winding device position, such as the rotational and / or translational position of a winding bracket on which a transformer winding is formed.

[0225] Identifying the ROI may include applying at least one ROI classifier to at least one image. Several ROI classifiers may be applied to at least one image. Each of one or more ROI classifiers may indicate for each pixel whether the corresponding pixel is classified as being within the ROI or outside the ROI.

[0226] One or several ROI classifiers may include at least one second ML model. The at least one second ML model may have a second input layer 101 that receives the pixels of at least one image and / or receives the (multiple) winding device positions. The at least one second ML model may have a second output layer 102 that outputs for each pixel a value indicating whether the corresponding pixel is within the ROI. Alternatively or additionally, the second input layer may receive the winding device position, such as the rotational and / or translational position of a winding bracket on which a transformer winding is formed.

[0227] The at least one second ML model 100 may include a deep learning ML model, such as a U-Net ML model.

[0228] At least one third ML model may be used to perform object classification. The at least one third ML model may have a third input layer 101 that receives the pixels located within the ROI. The at least one third ML model may have a third output layer 102 that outputs an object classification result indicating to which of several object categories the corresponding pixel in the ROI is assigned. The third ML model may include a random forest (RF) model, but is not limited thereto.

[0229] Figure 6 An image 120 captured by an image acquisition device 41 while a transformer winding 20 is being formed on a winding device 30 is shown.

[0230] As mentioned above, optional preprocessing may be performed on the raw image data. The optional preprocessing may be used to take into account the perspective effect caused by the arrangement of the optical axis relative to the topmost layer 21. The optional preprocessing may also enhance certain features, such as edges and / or contrast and / or periodicity.

[0231] The ROI detection identifies the ROI 125 in the image 120. The ROI 125 can be defined by the edge 126 of the shield 121. The ROI 125 can also be defined in the radial direction by the inner and outer circumferential boundaries of the innermost and outermost conductors as seen in the image 120.

[0232] Perform object classification to classify objects such as the conductor 122 and the filler 123. The sequence of these object types can be determined in the radial direction (i.e., in the radially inward or outward direction). Then, check the consistency of the object category sequence with the design data. The target sequence of the object categories expected when conforming to the design data can be determined. The object category sequence determined from the image 120 can be compared with the target sequence.

[0233] Figure 7 Another image with the ROI 125 is shown (in this case, determined by using a trained U-Net ML model as a classifier). The ROI detection works reliably. In particular, the top layer 21 can be reliably distinguished from the outer cylindrical surface of the already wound transformer winding. The edge of the shield can be reliably determined.

[0234] It is not necessary to perform object classification for every image obtained or obtainable by the image acquisition device 41. The image acquisition time of the images for detailed analysis (especially by performing object classification) can be determined by the computer vision system 40 (e.g., by its image processing system 50).

[0235] Figure 8 A functional block diagram 130 is shown, which illustrates the operation of the computer vision system 40 to determine the image acquisition time of the images they capture around further analysis.

[0236] (Multiple) winding device positions 131 and / or the image stream 132 can be used as inputs for determining 133 the relevant image acquisition time. Optionally, the design data can be taken into account to determine for which (which) winding device positions 131 a minor occlusion of the relevant feature can be detected, which is then translated into the corresponding image acquisition time during the winding process.

[0237] The determined image acquisition time of the images to be analyzed in detail can be used in various ways.

[0238] Camera control operations 134 can be performed to cause the image acquisition device 41 to perform image acquisition at the desired time. Alternatively or additionally, at block 135, images can be selected from the image stream (which can be continuously acquired) according to the (multiple) determined image acquisition times.

[0239] As an alternative to determining a specific image acquisition time, image processing can be performed on an ongoing basis. However, it is not generally necessary (but possible) to analyze every image frame.

[0240] Figure 9 is a flowchart of process 140. Process 140 can be executed by computer vision system 40 or a separate computing system coupled to computer vision system 40. Process 140 transforms design data into a format that can be easily compared with the results of object classification.

[0241] At process block 141, design data is obtained. The design data can be retrieved from a database. The design data may be specific to the particular specifications of a transformer in which the transformer winding is to be used. The design data can define the three-dimensional configuration of the transformer winding. The design data can include cross-sectional diagrams, but can have alternative formats, such as instruction codes executed by a winding device or a winding device controller.

[0242] At process block 142, the design data is processed to infer a target sequence of object classes for corresponding winding device positions based on the design data and the state of the winding process, which determines the winding device position.

[0243] The target sequence of object classes (also referred to herein as the target sequence of the object) can be compared with the results of object classification performed on at least one image to detect possible differences.

[0244] The computer vision system can include one or several ML models, but other processing techniques can also be used. One challenge encountered when using one or several ML models is providing a suitable training set, because fortunately, images showing errors during the winding process are rare compared to images showing the perfect operation of the winding process. The methods and systems disclosed herein can synthetically generate additional images showing winding errors, thereby providing a more balanced training data set. This improves the accuracy during inference.

[0245] ML model training can be performed by the computer vision system. ML model training can be repeated during on-site use of the computer vision system. ML model training can also be performed by a computing system separate from the computer vision system.

[0246] Figure 10 A flowchart of method 150 is shown. Method 150 can be executed to train one or several ML models that can be used in a method or system for monitoring a winding process.

[0247] At process block 151, labeled training images are obtained. The labeled training images are possibly but not necessarily captured on the same winding device as the ML model used in the inference phase.

[0248] At process block 152, training images marked as showing winding errors are identified. Additional training images are synthetically generated. Generating additional synthetic images with markings indicating winding errors may include transforming images marked as showing winding errors, such as rotation, translation, distortion, and / or reflection.

[0249] At process block 153, ML model training is performed using the images, especially the synthetically generated ones.

[0250] At process block 154, the trained ML model can be deployed to a computer vision system for inference during on-site use of the computer vision system.

[0251] The ML model can be an ROI detector or an object classifier, but is not limited thereto.

[0252] According to an embodiment, a method of forming a transformer winding on a winding device includes monitoring the winding process while it is being performed using the computer vision techniques described in detail herein.

[0253] The obtained windings can be assembled to form a transformer core or a transformer, such as a power transformer for a high-voltage or medium-voltage power transmission network, or a distribution transformer, a shunt reactor, and / or a variable reactor, for example.

[0254] Figure 11 A transformer 160 is schematically shown, which includes at least one transformer winding 163 formed using the method according to an embodiment. The at least one transformer winding 163 is assembled with a yoke 162 to form a transformer core. The transformer 160 may include additional components, such as a transformer tank 161, insulating oil inside the transformer tank 161, bushings (not shown), a transformer breather (not shown), etc.

[0255] Figure 12 is a block diagram of a manufacturing system 170 for a transformer core configured to manufacture a transformer. The manufacturing system 170 includes a winding device 30 and a yoke manufacturing system 171. The manufacturing system 170 includes a computer vision system 40 according to an embodiment.

[0256] The transformer winding formed on the winding device 30 is used in association with other transformer components, such as a yoke, under the monitoring of the computer vision system 40 to assemble a transformer core or a transformer. The assembly can be performed by an assembly system 172.

[0257] Figure 13 is a flowchart of a method 180. The method 180 can be executed by the manufacturing system 170.

[0258] At process block 181, visual quality control is performed during the winding process in which a transformer winding is formed on a winding device.

[0259] At process block 182, it is determined whether any differences have been detected during the winding process that would indicate that the object seen by the computer vision system does not match or imperfectly matches the object expected according to the design data.

[0260] At process block 183, if no differences are detected, the transformer winding can be used to assemble the transformer core or transformer.

[0261] At process block 184, if differences are detected, the transformer winding can be deprecated.

[0262] In response to the detection of differences, if differences are detected but are subsequently rectified by appropriate corrective actions, the transformer winding can still potentially be used at process block 183.

[0263] Although embodiments have been described with reference to the drawings, modifications and changes can be implemented in other embodiments.

[0264] For illustration, although embodiments have been described in which the transformer winding can include several disks and shields between the disks, the techniques disclosed herein are not limited to monitoring the winding process of such transformer windings.

[0265] For further illustration, although embodiments have been described in which images captured by one image acquisition device are analyzed, these techniques are also applicable when images are captured by more than one image acquisition device. Using several image acquisition devices can be beneficial for reading potential adverse effects using specific occlusions.

[0266] For further illustration, although embodiments have been described in which a trained ML model can be used to perform certain processing operations, other processing techniques can be used.

[0267] For further illustration, although embodiments have been described in which a classifier is used to perform ROI detection, if ROI detection is performed, other techniques that do not involve a classifier can be used to perform ROI detection.

[0268] For still further illustration, although embodiments have been described in connection with monitoring the winding process of a transformer winding, these techniques can also be used to monitor the manufacture of windings used in reactive or inductive systems such as shunt reactors or variable reactors.

[0269] Various effects and advantages are achieved by embodiments of the present invention. For illustration, embodiments allow errors to be detected while the winding process is ongoing. This allows corrective and / or mitigation actions to be taken during the winding process. The risk of transformer failure caused by winding errors is reduced.

[0270] The methods, systems, and apparatuses are particularly suitable in the context of being associated with power transformers of a power transmission system, such as a high-voltage, medium-voltage power transmission system, or a distribution system. These methods, systems, and apparatuses are not limited to these applications. By way of illustration, the techniques disclosed herein can also be used with railway transformers.

[0271] This description and the drawings illustrating aspects and embodiments of the invention should not be regarded as restrictive—the claims define the protected invention. In other words, although the invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. Various mechanical, compositional, structural, electrical, and operational changes may be made without departing from the spirit and scope of this description and the claims. In some instances, well-known circuits, structures, and techniques have not been shown in detail so as not to obscure the invention. Accordingly, it will be understood that modifications and changes may be made by those of ordinary skill within the scope and spirit of the appended claims. In particular, the invention encompasses further embodiments having any combination of features from different embodiments described above and below.

[0272] This disclosure also encompasses all further features shown separately in the drawings, even though they may not be described in the preceding or following description. Also, a single alternative of an embodiment described in the drawings and description, and a single alternative of its features, may be excluded from the subject matter of the invention or from the disclosed subject matter. This disclosure includes the subject matter consisting of the features defined in the claims or exemplary embodiments and the subject matter including such features.

[0273] The word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality. A single unit or step may perform the functions of several features recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Components described as being coupled or connected may be directly coupled or connected electrically or mechanically, or they may be indirectly coupled or connected via one or more intermediate components. Any reference signs in the claims should not be construed as limiting the scope.

[0274] Machine-readable instruction code may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via a wide area network or other wired or wireless telecommunication system. Additionally, the machine-readable instruction code may also be a data structure product or a signal for embodying a particular method, such as a method according to an embodiment.

Claims

1. A method for monitoring a winding process of forming a transformer winding (20) on a winding device (30), the method comprising: Processing, by a computer vision system (40), at least one image (42) captured during the winding process and showing at least a part of the transformer winding (20), wherein processing the at least one image (42) includes performing object classification; Detecting, by the computer vision system (40), a difference between the result of the object classification and the design data (49) of the transformer winding (20); Implementing, by the computer vision system (40), at least one action in response to the detected difference; and Determining, based on at least one position of the winding device (30) in combination with the design data (49), at least one image acquisition time of the at least one image (42) during the winding process.

2. The method according to claim 1, Among them, The at least one action includes: causing a human-machine interface HMI (37) to output information depending on the detected difference by the computer vision system (40).

3. The method according to claim 2, Among them, The information includes one, several or all of the following: Alarms and / or warnings; Information indicating the location of the detected difference; Information indicating the time when the detected difference occurs; Information about the root cause of the detected difference; Instructions for correcting the detected difference.

4. The method according to any one of the preceding claims, Among them, The at least one action includes one or both of the following: Mitigation actions; Correction actions.

5. The method according to any one of the preceding claims, Among them, The action includes stopping the rotation and / or translational displacement of the winding device (30).

6. The method according to any one of the preceding claims, further comprising: Performing image acquisition device control and / or image selection based on the determined at least one image acquisition time.

7. The method according to any one of the preceding claims, Among them, Processing the at least one image (42) includes identifying a region of interest ROI (125) in which the object classification is to be performed. Optionally, the boundary of the ROI (125) includes the edge (126) of a shield of the transformer winding (20).

8. The method according to any one of the preceding claims, Among them, Processing the at least one image (42) includes: Using one or several position measurements of the winding device (30); and / or Executing at least one machine learning ML model (100), the at least one machine learning ML model having: an input layer (101) that receives at least a part of the at least one image (42); and an output layer (102), wherein the output layer provides a region of interest ROI (125) in which the object classification is to be performed, and / or an object classifier.

9. The method according to any one of the preceding claims, Among them, Detecting the difference includes: determining a target sequence of an object in a radial direction along the transformer winding (20) based on the design data (49); and comparing the target sequence with the result of classifying the object.

10. The method according to any one of the preceding claims, Among them, wherein the object classification classifies: at least conductors, fillers, and shields in at least one of the at least one image (42), and / or crossings, and / or other components of the transformer winding (20), and / or other structural features of the transformer winding (20).

11. A method of manufacturing a transformer winding (20), comprising: controlling a winding device (30) on which the transformer winding (20) is formed by at least one control device (35); and performing the method according to any one of the preceding claims while forming the transformer winding (20).

12. A method of manufacturing a transformer core (162, 163) or a transformer (160), comprising: manufacturing at least one transformer winding (20) by the method according to claim 11; and assembling a transformer core (162, 163) or a transformer (160) including the at least one transformer winding (20).

13. A system configured to monitor a winding process of forming a transformer winding (20) on a winding device (30), the system comprising: a computer vision system (40) configured to: process at least one image (42) captured during the winding process and showing at least a part of the transformer winding (20) to perform object classification; detect a difference between the result of the object classification and the design data (49) of the transformer winding (20); perform an action in response to the detected difference; and determine at least one image acquisition time of the at least one image (42) during the winding process based on at least one position of the winding device (30) in combination with the design data (49).

14. The system according to claim 13, further comprising: a human-machine interface HMI (37), wherein the computer vision system (40) is configured to control the HMI in response to the detected difference.

15. A manufacturing system for manufacturing a transformer winding (20), a transformer core (162, 163) or a transformer (160), the system comprising: a winding device (30); and the system according to claim 13 or claim 14.

16. A machine-readable instruction code that, when executed by at least one programmable circuit, causes the at least one programmable circuit to execute the method according to any one of claims 1 to 11.

Images