Disassembly Method and System for Distribution Transformer Cover Plate

Abstract

The present application discloses a method and system for disassembling a cover plate of a distribution transformer. The method includes: after the manipulator starts to perform the disassembly operation on the cover plate of the distribution transformer, obtaining an optical cover plate image and an infrared cover plate image captured of the cover plate of the distribution transformer; based on the optical cover plate image and the infrared cover plate image, using a disassembly model to generate a disassembly plan and send a control instruction to the manipulator; the training process of the disassembly model includes: obtaining a sample optical image and a sample infrared image corresponding to a sample distribution transformer; using a model to be trained to encode the sample optical image and the sample infrared image, and obtaining optical features and infrared features generated during the encoding process; based on the optical features, the infrared features and the information of the sample distribution transformer, using a large model to generate a first disassembly plan; training the model to be trained based on the optical features, the infrared features and the first disassembly plan to obtain a disassembly model, thereby improving the disassembly efficiency and disassembly safety of the cover plate of the distribution transformer.

Description

Technical Field

[0001] This application relates to the field of automatic control technology, and particularly relates to a method and system for disassembling the cover plate of a distribution transformer. Background Art

[0002] The cover plate of a distribution transformer is one of the important components of the distribution transformer. It is mainly used to protect the electrical components inside the transformer from the influence of the external environment, and at the same time provide safety protection to prevent personnel from contacting the live parts. Before overhauling the internal components of the distribution transformer, it is usually necessary to disassemble the cover plate of the distribution transformer by manually operating the manipulator throughout the process.

[0003] However, manually operating the manipulator throughout the process for disassembly not only has low efficiency, but also highly depends on the experience of the operator. It is difficult to ensure the rationality of parameter settings such as the disassembly speed, angle, and force of the manipulator during the disassembly process, which is likely to cause damage to the distribution transformer, thereby threatening the safety of equipment and personnel. Summary of the Invention

[0004] To solve the above technical problems, the embodiments of this application propose a method and system for disassembling the cover plate of a distribution transformer, which can improve the disassembly efficiency and safety of the cover plate of the distribution transformer.

[0005] In a first aspect, the embodiments of this application provide a method for disassembling the cover plate of a distribution transformer, including:

[0006] After detecting that the manipulator starts to perform the disassembly operation on the cover plate of the distribution transformer, use an image acquisition device to capture an optical cover plate image and an infrared cover plate image of the cover plate of the distribution transformer;

[0007] Based on the optical cover plate image and the infrared cover plate image, use a disassembly model to generate a disassembly plan, and send a control command to the manipulator according to the disassembly plan, where the control command is used to regulate the disassembly operation of the manipulator on the cover plate;

[0008] Among them, the training process of the disassembly model includes:

[0009] Obtain a sample optical image and a sample infrared image corresponding to a sample distribution transformer;

[0010] Use a model to be trained to encode the sample optical image and the sample infrared image respectively, and obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process respectively;

[0011] Based on the optical features, the infrared features, and the information of the sample distribution transformer, use a large model to generate a first disassembly plan;

[0012] Train the model to be trained based on the optical features, the infrared features, and the first disassembly plan to obtain the disassembly model.

[0013] Optionally, the step of generating a first disassembly plan based on the optical features, the infrared features, and the information of the sample distribution transformer by using a large model includes:

[0014] Perform an inner product operation on the optical features and the infrared features to obtain inner product features.

[0015] Input the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model.

[0016] Optionally, the information of the sample distribution transformer includes the standard screw profile template diagram and the standard temperature distribution diagram of the sample distribution transformer. The step of inputting the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model includes:

[0017] Input the inner product features into the large model so that the large model generates a simulated screw profile template diagram and a simulated temperature distribution diagram according to the inner product features.

[0018] Input the standard screw profile template diagram and the standard temperature distribution diagram into the large model so that the large model generates and outputs the first disassembly plan based on the difference between the simulated screw profile template diagram and the standard screw profile template diagram, and the difference between the simulated temperature distribution diagram and the standard temperature distribution diagram.

[0019] Optionally, the step of training the model to be trained based on the optical features, the infrared features, and the first disassembly plan to obtain the disassembly model includes:

[0020] Perform an inner product operation on the optical features and the infrared features to obtain inner product features.

[0021] Input the inner product features into the model to be trained to generate a second disassembly plan.

[0022] Extract the first text features of the first disassembly plan and extract the second text features of the second disassembly plan.

[0023] Use a preset loss function to determine the text feature difference between the first text features and the second text features.

[0024] Train the model to be trained based on the text feature difference to obtain the disassembly model.

[0025] Optionally, the sample optical image is cropped into multiple optical local images, and the sample infrared image is cropped into multiple infrared local images. N optical local images among the multiple optical local images correspond one-to-one with N infrared local images among the multiple infrared local images, where N is a positive integer. The local regions indicated by the corresponding optical local image and infrared local image are the same. The optical feature at least includes the optical local features corresponding to the N optical local images respectively, and the infrared feature at least includes the infrared local features corresponding to the N infrared local images respectively;

[0026] Using the model to be trained, encoding the sample optical image and the sample infrared image respectively, and obtaining the optical feature corresponding to the sample optical image and the infrared feature corresponding to the sample infrared image generated during the encoding process respectively, includes:

[0027] Using the model to be trained, encoding the sample optical image, and obtaining the optical local features corresponding to the N optical local images respectively generated during the encoding process,

[0028] Using the model to be trained, encoding the sample infrared image, and obtaining the infrared local features corresponding to the N infrared local images respectively generated during the encoding process.

[0029] Optionally, each optical local image among the N optical local images and its corresponding infrared local image form a local image pair. Performing an inner product process on the optical feature and the infrared feature to obtain an inner product feature, includes:

[0030] For each local image pair, performing an inner product process on the optical local feature and the infrared local feature corresponding to the local image pair to obtain the inner product local feature corresponding to the local image pair;

[0031] Based on the inner product local features corresponding to all the local image pairs respectively, determining the inner product feature.

[0032] Optionally, the training process further includes:

[0033] According to the local region indicated by each local image pair, determining the weight corresponding to each local image pair;

[0034] The step of determining the inner product feature based on the inner product local features corresponding to all the local image pairs respectively, includes:

[0035] Based on the weight corresponding to each local image pair, performing a weighting process on the inner product local feature corresponding to each local image pair;

[0036] Concatenate the weighted inner product local features corresponding to all the local images into the inner product feature.

[0037] Optionally, before generating a disassembly plan using a disassembly model based on the optical cover plate image and the infrared cover plate image, the method further includes:

[0038] Obtain information about the distribution transformer;

[0039] The generating a disassembly plan using a disassembly model based on the optical cover plate image and the infrared cover plate image includes:

[0040] Generate a disassembly plan using a disassembly model based on the information about the distribution transformer, the optical cover plate image, and the infrared cover plate image.

[0041] Optionally, before generating a disassembly plan using a disassembly model based on the optical cover plate image and the infrared cover plate image, the method further includes:

[0042] Preprocess the optical cover plate image and the infrared cover plate image;

[0043] The generating a disassembly plan using a disassembly model based on the information about the distribution transformer, the optical cover plate image, and the infrared cover plate image includes:

[0044] Input the information about the distribution transformer, the preprocessed optical cover plate image, and the preprocessed infrared cover plate image into the disassembly model to generate a disassembly plan.

[0045] In a second aspect, an embodiment of the present application provides a disassembly system for a distribution transformer cover plate, including:

[0046] A manipulator;

[0047] An image acquisition device; and,

[0048] A control device, communicatively connected to the manipulator and the image acquisition device, and the control device is configured to:

[0049] After detecting that the manipulator starts to perform a cover plate disassembly operation on the distribution transformer, capture an optical cover plate image and an infrared cover plate image of the cover plate of the distribution transformer through the image acquisition device;

[0050] Generate a disassembly plan using a disassembly model based on the optical cover plate image and the infrared cover plate image, and send a control instruction to the manipulator according to the disassembly plan, where the control instruction is used to regulate the cover plate disassembly operation of the manipulator;

[0051] Wherein, the training process of the disassembly model includes:

[0052] Obtain the sample optical image and the sample infrared image corresponding to the sample distribution transformer;

[0053] Use the model to be trained to encode the sample optical image and the sample infrared image respectively, and obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process respectively;

[0054] Based on the optical features, the infrared features and the information of the sample distribution transformer, use a large model to generate a first disassembly plan;

[0055] Based on the optical features, the infrared features and the first disassembly plan, train the model to be trained to obtain the disassembly model.

[0056] In summary, the embodiments of the present application have at least the following beneficial effects:

[0057] By adopting the embodiments of the present application, after detecting that the manipulator starts to execute the cover disassembly operation for the distribution transformer, an optical cover image and an infrared cover image of the cover of the distribution transformer are obtained by an image acquisition device; based on the optical cover image and the infrared cover image, a disassembly plan is generated using a disassembly model, and a control instruction is sent to the manipulator according to the disassembly plan, where the control instruction is used to regulate the cover disassembly operation of the manipulator; wherein, the training process of the disassembly model includes: obtaining the sample optical image and the sample infrared image corresponding to the sample distribution transformer; using the model to be trained to encode the sample optical image and the sample infrared image respectively, and obtaining the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process respectively; based on the optical features, the infrared features and the information of the sample distribution transformer, using a large model to generate a first disassembly plan; based on the optical features, the infrared features and the first disassembly plan, training the model to be trained to obtain the disassembly model, thereby being able to improve the disassembly efficiency and disassembly safety of the distribution transformer cover. Description of the Drawings

[0058] Figure 1 is a schematic flowchart of the method for disassembling the cover of the distribution transformer provided by the embodiment of the present application;

[0059] Figure 2 is a schematic flowchart of the training process of the disassembly model provided by the embodiment of the present application;

[0060] Figure 3 is a schematic structural diagram of the disassembly system of the distribution transformer cover provided by the embodiment of the present application. Detailed Embodiments

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0062] In the description of the present application, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", "third", etc. may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise stated, the meaning of "a plurality" is two or more. In the description of the present application, the term "comprising" and its variations are open-ended, that is, "including but not limited to". The term "based on" is "at least partially based on". The term "according to" is "at least partially according to". The term "an embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments".

[0063] In the description of the present application, it should be noted that, unless otherwise clearly defined and limited, the terms "installed", "connected", "connected to" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0064] In the description of the present application, it should be noted that, unless otherwise defined, all the technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the technical field to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0065] In a first aspect, referring to Figure 1 , a schematic flowchart of a disassembly method for a distribution transformer cover plate provided by an embodiment of the present application is shown. The method includes steps S101-S102, which are specifically as follows:

[0066] S101. After detecting that the manipulator starts to perform the operation of disassembling the cover plate of the distribution transformer, use an image acquisition device to capture an optical cover plate image and an infrared cover plate image of the cover plate of the distribution transformer.

[0067] It should be noted that in this embodiment, the optical cover plate image and the infrared cover plate image can be captured at the same moment to facilitate characterizing the situation of the cover plate of the distribution transformer at the same time point. Here, this same moment can be the moment after a preset duration since the manipulator starts to perform the operation of disassembling the cover plate of the distribution transformer. However, it should be understood that the shooting times of the optical cover plate image and the infrared cover plate image can also be different, as long as the difference between the shooting times of the optical cover plate image and the infrared cover plate image is within a reasonable error range.

[0068] It can be understood that in this embodiment, the optical cover plate image can be used to indicate the appearance of the cover plate of the distribution transformer, so as to further characterize the disassembly speed, disassembly angle, etc. of the current cover plate disassembly operation of the manipulator.

[0069] It can be understood that in this embodiment, the infrared cover plate image can be used to indicate the temperature distribution of the cover plate of the distribution transformer, so as to further characterize whether the materials at various parts of the cover plate of the distribution transformer are overheated due to improper disassembly.

[0070] In one example, the image acquisition device can include an optical imaging device and an infrared imaging device. Among them, the optical imaging device can include the camera of a general mobile device (such as a terminal device, a digital camera, etc.), an industrial camera dedicated to the disassembly of the cover plate of the distribution transformer, etc. The infrared imaging device can include an infrared imager, an infrared sensor, etc. In addition, all or part of the devices in the image acquisition device in this embodiment can also be installed on the manipulator.

[0071] S102. Based on the optical cover plate image and the infrared cover plate image, use a disassembly model to generate a disassembly plan, and send a control instruction to the manipulator according to the disassembly plan, where the control instruction is used to regulate the cover plate disassembly operation of the manipulator.

[0072] In one example, the disassembly plan is used to indicate at least one of the following control parameters of the manipulator: disassembly force, disassembly speed, disassembly angle. In this embodiment, after the manipulator receives this control instruction, it can adjust the parameters of the manipulator according to the various control parameters indicated by the disassembly plan carried in the control instruction, so that the manipulator can operate according to the disassembly plan in the subsequent process, thereby realizing real-time adjustment of the working parameters of the manipulator, facilitating adaptation to the real-time disassembly situation, and avoiding damage to the distribution transformer and its cover plate due to improper disassembly.

[0073] Among them, seeFigure 2 , which shows a schematic flowchart of the training process of the disassembly model provided by the embodiments of the present application. The training process includes steps S201 - S204, specifically as follows:

[0074] S201, Obtain the sample optical image and the sample infrared image corresponding to the sample distribution transformer.

[0075] It should be noted that both the sample optical image and the sample infrared image in this embodiment are taken when the cover of the sample distribution transformer is disassembled, and both can be used to indicate the situation when the cover of the sample distribution transformer is disassembled.

[0076] S202, Use the model to be trained to encode the sample optical image and the sample infrared image respectively, and obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process respectively.

[0077] In one example, the model to be trained may include a general encoding model, which can be used to encode optical images and infrared images. At this time, the corresponding optical features and infrared features can be used to characterize the situation of the sample distribution transformer captured by the corresponding images.

[0078] S203, Based on the optical features, the infrared features, and the information of the sample distribution transformer, use a large model to generate a first disassembly plan; wherein, the information of the sample distribution transformer may include detailed information such as the model, size, structure, and material of the transformer.

[0079] In one example, based on the optical features, the infrared features, and the information of the sample distribution transformer, using a large model to generate a first disassembly plan may include: inputting the optical features, the infrared features, and the information of the sample distribution transformer into the large model, so that the large model generates an initial disassembly plan according to the information of the sample distribution transformer and the optical features, and uses the infrared features to adjust the initial disassembly plan to generate a first disassembly plan. In this embodiment, the large model may be a complex deep learning model, which may be a convolutional neural network, a recurrent neural network, a Transformer, or other types of neural networks. It is pre-trained with a large amount of relevant data, so as to have the ability to generate an initial disassembly plan according to the information of the sample distribution transformer and the optical features, and can use the infrared features to adjust the generated plan.

[0080] S204, Based on the optical features, the infrared features, and the first disassembly plan, train the model to be trained to obtain the disassembly model.

[0081] In one example, training the model to be trained based on the optical features, the infrared features, and the first disassembly plan to obtain the disassembly model may include: inputting the optical features and the infrared features into the model to be trained, so that the model to be trained generates a fourth disassembly plan according to the optical features and the infrared features; extracting the first text feature of the first disassembly plan and the fourth text feature of the fourth disassembly plan; using a preset loss function to determine the text feature difference between the first text feature and the fourth text feature; and training the model to be trained based on the text feature difference to obtain the disassembly model. In this embodiment, the model to be trained may include a convolutional neural network, a recurrent neural network, a Transformer, or other types of neural networks, and has the ability to output the desired content according to specific feature inputs, so that corresponding disassembly plans can be generated.

[0082] It should be noted that in each embodiment related to model training in this application, corresponding model training can be performed through backpropagation and gradient descent or other common model training methods, which will not be elaborated hereinafter.

[0083] In an alternative embodiment, the generating of the first disassembly plan by using a large model based on the optical features, the infrared features, and the information of the sample distribution transformer includes:

[0084] Performing an inner product operation on the optical features and the infrared features to obtain inner product features.

[0085] Inputting the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model.

[0086] Generally speaking, large models are usually deployed in the cloud. That is, information transmission is usually involved when calling a large model. In this embodiment, the inner product can be used to calculate the projection of one vector on another vector, so that the inner product features can simultaneously have the information of both the optical features and the infrared features. In this way, the amount of information to be transmitted when interacting with the large model can be reduced.

[0087] In an alternative embodiment, the information of the sample distribution transformer includes the standard screw profile template diagram and the standard temperature distribution diagram of the sample distribution transformer. The inputting of the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model includes:

[0088] Inputting the inner product features into the large model, so that the large model generates a simulated screw profile template diagram and a simulated temperature distribution diagram according to the inner product features.

[0089] Input the standard screw profile template diagram and the standard temperature distribution diagram into the large model, so that the large model generates and outputs the first disassembly plan based on the differences between the simulated screw profile template diagram and the standard screw profile template diagram, and between the simulated temperature distribution diagram and the standard temperature distribution diagram.

[0090] In one example, the differences between the simulated screw profile template diagram and the standard screw profile template diagram, and between the simulated temperature distribution diagram and the standard temperature distribution diagram can be obtained through the following steps: Extract the image features of the simulated screw profile template diagram, the image features of the standard screw profile template diagram, the image features of the simulated temperature distribution diagram, and the image features of the standard temperature distribution diagram respectively. Calculate the cosine similarity between the image features of the simulated screw profile template diagram and the image features of the standard screw profile template diagram to characterize the difference between the simulated screw profile template diagram and the standard screw profile template diagram; Calculate the cosine similarity between the image features of the simulated temperature distribution diagram and the image features of the standard temperature distribution diagram to characterize the difference between the simulated temperature distribution diagram and the standard temperature distribution diagram.

[0091] In an alternative implementation, training the model to be trained based on the optical features, the infrared features, and the first disassembly plan to obtain the disassembly model includes:

[0092] Perform an inner product operation on the optical features and the infrared features to obtain inner product features.

[0093] Input the inner product features into the model to be trained to generate a second disassembly plan.

[0094] Extract the first text features of the first disassembly plan and extract the second text features of the second disassembly plan.

[0095] Use a preset loss function to determine the text feature difference between the first text features and the second text features.

[0096] Train the model to be trained based on the text feature difference to obtain the disassembly model.

[0097] In this embodiment, it is equivalent to borrowing the powerful computing power of the large model to simulate an optimal first disassembly plan, and then adjusting the model parameters of the model to be trained according to the difference between the plan generated by the model to be trained and the first disassembly plan, so that the output of the model to be trained can be close to the output of the large model.

[0098] In one example, the loss function may include mean square error, root mean square error, mean absolute error, cross-entropy loss, cosine similarity loss, L1 regularization, L2 regularization, etc.

[0099] In one example, after obtaining the inner product feature, training the model to be trained to obtain the disassembly model may further include:

[0100] Determine a corresponding solution database according to the information of the sample distribution transformer.

[0101] Select a third disassembly solution matching the inner product feature from the solution database.

[0102] Extract the third text feature of the third disassembly solution.

[0103] Using a preset loss function, determine the text feature difference between the first text feature and the third text feature.

[0104] Based on the text feature difference between the first text feature and the third text feature, train the model to be trained to obtain the disassembly model.

[0105] In an alternative embodiment, the sample optical image is cropped into multiple optical local images, the sample infrared image is cropped into multiple infrared local images, N optical local images among the multiple optical local images correspond one-to-one with N infrared local images among the multiple infrared local images, N is a positive integer, the local regions indicated by the corresponding optical local image and infrared local image are the same, the optical feature at least includes the optical local features corresponding to the N optical local images respectively, and the infrared feature at least includes the infrared local features corresponding to the N infrared local images respectively.

[0106] Using the model to be trained to encode the sample optical image and the sample infrared image respectively, and respectively obtain the optical feature corresponding to the sample optical image and the infrared feature corresponding to the sample infrared image generated during the encoding process, includes:

[0107] Using the model to be trained to encode the sample optical image, and obtain the optical local features corresponding to the N optical local images respectively generated during the encoding process,

[0108] Using the model to be trained to encode the sample infrared image, and obtain the infrared local features corresponding to the N infrared local images respectively generated during the encoding process.

[0109] It can be understood that in this embodiment, the sample infrared image can be cropped into multiple infrared local images according to a certain cropping method. Correspondingly, the sample infrared image can also be cropped into multiple infrared local images according to the corresponding cropping method. At this time, it can be ensured that at least N of the cropped infrared local images correspond to each other, that is, the indicated local regions are the same.

[0110] In an alternative embodiment, each of the N optical local images and its corresponding infrared local image form a local image pair. The inner product processing of the optical feature and the infrared feature to obtain an inner product feature includes:

[0111] For each of the local image pairs, perform inner product processing on the optical local feature and the infrared local feature corresponding to the local image pair to obtain the inner product local feature corresponding to the local image pair.

[0112] Based on the inner product local features corresponding to all the local image pairs respectively, determine the inner product feature.

[0113] In an example, based on the inner product local features corresponding to all the local image pairs respectively, determining the inner product feature may include: concatenating the inner product local features corresponding to all the local image pairs respectively to form the inner product feature.

[0114] In an alternative embodiment, the training process further includes:

[0115] According to the local region indicated by each of the local image pairs, determine the weight corresponding to each of the local image pairs.

[0116] The determining the inner product feature based on the inner product local features corresponding to all the local image pairs respectively includes:

[0117] Based on the weight corresponding to each of the local image pairs, perform weighted processing on the inner product local feature corresponding to each of the local image pairs.

[0118] Concatenate the weighted inner product local features corresponding to all the local image pairs respectively to form the inner product feature.

[0119] In an alternative embodiment, before generating a disassembly plan using the disassembly model based on the optical cover image and the infrared cover image, the method further includes:

[0120] Obtain information about the distribution transformer.

[0121] The generating a disassembly plan using the disassembly model based on the optical cover image and the infrared cover image includes:

[0122] Generate a disassembly plan using a disassembly model based on the information of the distribution transformer, the optical cover image, and the infrared cover image.

[0123] In an alternative embodiment, before generating a disassembly plan using a disassembly model based on the optical cover image and the infrared cover image, the method further includes:

[0124] Preprocess the optical cover image and the infrared cover image.

[0125] The generating a disassembly plan using a disassembly model based on the information of the distribution transformer, the optical cover image, and the infrared cover image includes:

[0126] Input the information of the distribution transformer, the preprocessed optical cover image, and the preprocessed infrared cover image into the disassembly model to generate a disassembly plan.

[0127] In one example, the preprocessing may include normalization processing, cropping processing, scaling processing, rotation processing, translation processing, denoising processing, etc., which are selected according to needs.

[0128] In a second aspect, correspondingly, an embodiment of the present application further provides a disassembly system for a distribution transformer cover, which can implement all processes of the disassembly method for the distribution transformer cover provided in the above embodiment.

[0129] See Figure 3 , which shows a schematic structural diagram of the disassembly system for a distribution transformer cover provided in an embodiment of the present application. The disassembly system 300 for the distribution transformer cover includes:

[0130] A manipulator 301.

[0131] An image acquisition device 302; and,

[0132] A control device 303, which is communicatively connected to both the manipulator 301 and the image acquisition device 302, and the control device is configured to:

[0133] After detecting that the manipulator 301 starts to perform the cover disassembly operation on the distribution transformer, obtain an optical cover image and an infrared cover image of the cover of the distribution transformer through the image acquisition device 302.

[0134] Generate a disassembly plan based on the optical cover image and the infrared cover image, and send a control instruction to the manipulator 301 according to the disassembly plan, where the control instruction is used to regulate the cover disassembly operation of the manipulator 301.

[0135] Among them, the training process of the disassembly model includes:

[0136] Obtain the sample optical image and sample infrared image corresponding to the sample distribution transformer.

[0137] Use the model to be trained to encode the sample optical image and the sample infrared image respectively, and obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process respectively.

[0138] Based on the optical features, the infrared features and the information of the sample distribution transformer, use a large model to generate a first disassembly plan.

[0139] Based on the optical features, the infrared features and the first disassembly plan, train the model to be trained to obtain the disassembly model.

[0140] In an optional implementation manner, the step of using a large model to generate a first disassembly plan based on the optical features, the infrared features and the information of the sample distribution transformer includes:

[0141] Perform an inner product operation on the optical features and the infrared features to obtain inner product features.

[0142] Input the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model.

[0143] In an optional implementation manner, the information of the sample distribution transformer includes the standard screw contour template diagram and the standard temperature distribution diagram of the sample distribution transformer. The step of inputting the inner product features and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model includes:

[0144] Input the inner product features into the large model, so that the large model generates a simulated screw contour template diagram and a simulated temperature distribution diagram according to the inner product features.

[0145] Input the standard screw contour template diagram and the standard temperature distribution diagram into the large model, so that the large model generates and outputs the first disassembly plan based on the difference between the simulated screw contour template diagram and the standard screw contour template diagram, and the difference between the simulated temperature distribution diagram and the standard temperature distribution diagram.

[0146] In an optional implementation manner, the step of training the model to be trained based on the optical features, the infrared features and the first disassembly plan to obtain the disassembly model includes:

[0147] Perform an inner product operation on the optical feature and the infrared feature to obtain an inner product feature.

[0148] Input the inner product feature into the model to be trained to generate a second disassembly plan.

[0149] Extract the first text feature of the first disassembly plan and extract the second text feature of the second disassembly plan.

[0150] Use a preset loss function to determine the text feature difference between the first text feature and the second text feature.

[0151] Train the model to be trained based on the text feature difference to obtain the disassembly model.

[0152] In an alternative embodiment, the sample optical image is cropped into multiple optical local images, the sample infrared image is cropped into multiple infrared local images, N optical local images in the multiple optical local images correspond one-to-one with N infrared local images in the multiple infrared local images, N is a positive integer, the local regions indicated by the corresponding optical local image and infrared local image are the same, the optical feature at least includes the optical local features corresponding to the N optical local images respectively, and the infrared feature at least includes the infrared local features corresponding to the N infrared local images respectively.

[0153] Using the model to be trained to encode the sample optical image and the sample infrared image respectively, and obtaining the optical feature corresponding to the sample optical image and the infrared feature corresponding to the sample infrared image generated during the encoding process respectively, includes:

[0154] Use the model to be trained to encode the sample optical image and obtain the optical local features corresponding to the N optical local images respectively generated during the encoding process.

[0155] Use the model to be trained to encode the sample infrared image and obtain the infrared local features corresponding to the N infrared local images respectively generated during the encoding process.

[0156] In an alternative embodiment, each optical local image in the N optical local images and its corresponding infrared local image form a local image pair. The performing an inner product operation on the optical feature and the infrared feature to obtain an inner product feature includes:

[0157] For each local image pair, perform an inner product operation on the optical local feature and the infrared local feature corresponding to the local image pair to obtain the inner product local feature corresponding to the local image pair.

[0158] Determine the inner product feature based on the inner product local features corresponding to all the local images respectively.

[0159] In an alternative embodiment, the training process further includes:

[0160] Determine the weight corresponding to each local image pair according to the local region indicated by each local image pair.

[0161] The determining the inner product feature based on the inner product local features corresponding to all the local images respectively includes:

[0162] Perform weighted processing on the inner product local features corresponding to each local image pair based on the weight corresponding to each local image pair.

[0163] Concatenate the weighted inner product local features corresponding to all the local image pairs respectively into the inner product feature.

[0164] In an alternative embodiment, before generating a disassembly plan using the disassembly model based on the optical cover plate image and the infrared cover plate image, the control device is further configured to:

[0165] Obtain the information of the distribution transformer.

[0166] The generating a disassembly plan using the disassembly model based on the optical cover plate image and the infrared cover plate image includes:

[0167] Generate a disassembly plan using the disassembly model based on the information of the distribution transformer, the optical cover plate image and the infrared cover plate image.

[0168] In an alternative embodiment, before generating a disassembly plan using the disassembly model based on the optical cover plate image and the infrared cover plate image, the control device is further configured to:

[0169] Preprocess the optical cover plate image and the infrared cover plate image.

[0170] The generating a disassembly plan using the disassembly model based on the information of the distribution transformer, the optical cover plate image and the infrared cover plate image includes:

[0171] Input the information of the distribution transformer, the preprocessed optical cover plate image and the preprocessed infrared cover plate image into the disassembly model to generate a disassembly plan.

[0172] In summary, the embodiments of the present application have at least the following beneficial effects:

[0173] By adopting the embodiments of the present application, after detecting that the manipulator starts to perform the operation of disassembling the cover plate of the distribution transformer, an optical cover plate image and an infrared cover plate image of the distribution transformer are obtained by an image acquisition device; based on the optical cover plate image and the infrared cover plate image, a disassembly plan is generated using a disassembly model, and a control instruction is sent to the manipulator according to the disassembly plan, where the control instruction is used to regulate the cover plate disassembly operation of the manipulator; wherein, the training process of the disassembly model includes: obtaining a sample optical image and a sample infrared image corresponding to a sample distribution transformer; using a model to be trained, encoding the sample optical image and the sample infrared image respectively, and obtaining an optical feature corresponding to the sample optical image and an infrared feature corresponding to the sample infrared image generated during the encoding process respectively; based on the optical feature, the infrared feature and the information of the sample distribution transformer, generating a first disassembly plan using a large model; training the model to be trained based on the optical feature, the infrared feature and the first disassembly plan to obtain the disassembly model, so as to improve the disassembly efficiency and disassembly safety of the cover plate of the distribution transformer.

[0174] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary hardware platform, and of course, it can also be implemented entirely by hardware. Based on such an understanding, all or part of the technical solutions of the present application that contribute to the background art can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0175] The above is the preferred embodiment of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present application.

Claims

1. A disassembly method for the cover plate of a distribution transformer, characterized in that, Including: After detecting that the manipulator starts to perform the operation of disassembling the cover plate of the distribution transformer, an optical cover plate image and an infrared cover plate image of the cover plate of the distribution transformer are obtained by an image acquisition device. Among them, the optical cover plate image is used to characterize the disassembly speed and disassembly angle of the current cover plate disassembly operation of the manipulator. Among them, the infrared cover plate image is used to indicate the temperature distribution of the cover plate of the distribution transformer, so as to characterize whether the materials at various parts of the cover plate of the distribution transformer are overheated due to improper disassembly; Based on the optical cover plate image and the infrared cover plate image, a disassembly plan is generated using a disassembly model, and a control instruction is sent to the manipulator according to the disassembly plan. Among them, the control instruction is used to regulate the cover plate disassembly operation of the manipulator, and the disassembly plan is used to indicate at least one of the following control parameters of the manipulator: disassembly force, disassembly speed, disassembly angle; Among them, the training process of the disassembly model includes: Obtain a sample optical image and a sample infrared image corresponding to a sample distribution transformer; Using the model to be trained, encode the sample optical image and the sample infrared image respectively, and respectively obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process; Based on the optical features, the infrared features and the information of the sample distribution transformer, use a large model to generate a first disassembly plan. Among them, the information of the sample distribution transformer includes at least one of the following: size, structure, material; Based on the optical features, the infrared features and the first disassembly plan, train the model to be trained to obtain the disassembly model; Among them, the using the large model to generate a first disassembly plan based on the optical features, the infrared features and the information of the sample distribution transformer includes: Perform an inner product process on the optical features and the infrared features to obtain an inner product feature; Input the inner product feature and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model; Among them, the sample optical image is cropped into multiple optical local images, the sample infrared image is cropped into multiple infrared local images, N optical local images among the multiple optical local images correspond one by one to N infrared local images among the multiple infrared local images, N is a positive integer, the local regions indicated by the corresponding optical local image and infrared local image are the same, the optical features at least include the optical local features corresponding to the N optical local images respectively, and the infrared features at least include the infrared local features corresponding to the N infrared local images respectively; The using the model to be trained to encode the sample optical image and the sample infrared image respectively, and respectively obtain the optical features corresponding to the sample optical image and the infrared features corresponding to the sample infrared image generated during the encoding process includes: Using the model to be trained, encode the sample optical image, and obtain the optical local features corresponding to the N optical local images respectively generated during the encoding process, Using the to-be-trained model, encode the sample infrared image, and obtain the infrared local features corresponding to each of the N infrared local images generated during the encoding process; Wherein, each of the N optical local images and its corresponding infrared local image form a local image pair, and the inner product processing of the optical feature and the infrared feature to obtain the inner product feature includes: For each of the local image pairs, perform inner product processing on the optical local feature and the infrared local feature corresponding to the local image pair to obtain the inner product local feature corresponding to the local image pair; Based on the inner product local features corresponding to all the local image pairs, determine the inner product feature; Wherein, the training process further includes: According to the local area indicated by each of the local image pairs, determine the weight corresponding to each of the local image pairs; The determining the inner product feature based on the inner product local features corresponding to all the local image pairs includes: Based on the weight corresponding to each of the local image pairs, perform weighted processing on the inner product local feature corresponding to each of the local image pairs; Concatenate the weighted inner product local features corresponding to all the local image pairs into the inner product feature.

2. The disassembly method of the distribution transformer cover plate according to claim 1, characterized in that, The information of the sample distribution transformer includes the standard screw contour template diagram and the standard temperature distribution diagram of the sample distribution transformer. The inputting the inner product feature and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model includes: Input the inner product feature into the large model, so that the large model generates a simulated screw contour template diagram and a simulated temperature distribution diagram according to the inner product feature; Input the standard screw contour template diagram and the standard temperature distribution diagram into the large model, so that the large model generates and outputs the first disassembly plan based on the difference between the simulated screw contour template diagram and the standard screw contour template diagram, and the difference between the simulated temperature distribution diagram and the standard temperature distribution diagram.

3. The disassembly method of the distribution transformer cover plate according to claim 1, characterized in that, The training the to-be-trained model based on the optical feature, the infrared feature and the first disassembly plan to obtain the disassembly model includes: Perform inner product processing on the optical feature and the infrared feature to obtain the inner product feature; Input the inner product feature into the to-be-trained model to generate a second disassembly plan; Extract the first text feature of the first disassembly plan and extract the second text feature of the second disassembly plan; Use a preset loss function to determine the text feature difference between the first text feature and the second text feature; Based on the text feature difference, train the to-be-trained model to obtain the disassembly model.

4. The disassembly method of the distribution transformer cover plate according to any one of claims 1-3, characterized in that, Before using the disassembly model to generate a disassembly plan based on the optical cover image and the infrared cover image, the method further includes: Obtain the information of the distribution transformer; The using the disassembly model to generate a disassembly plan based on the optical cover image and the infrared cover image includes: Generate a disassembly plan using a disassembly model based on the information of the distribution transformer, the optical cover image, and the infrared cover image.

5. The disassembly method of the distribution transformer cover plate according to claim 4, characterized in that, Before generating a disassembly plan using the disassembly model based on the optical cover image and the infrared cover image, the method further includes: Preprocess the optical cover image and the infrared cover image; Generating a disassembly plan using a disassembly model based on the information of the distribution transformer, the optical cover image, and the infrared cover image includes: Input the information of the distribution transformer, the preprocessed optical cover image, and the preprocessed infrared cover image into the disassembly model to generate a disassembly plan.

6. A disassembly system for a distribution transformer cover plate, characterized in that, Including: A manipulator; An image acquisition device; And, A control device, communicatively connected to both the manipulator and the image acquisition device, the control device being configured to: After detecting that the manipulator starts to perform a cover disassembly operation on the distribution transformer, obtain an optical cover image and an infrared cover image of the cover of the distribution transformer through the image acquisition device, wherein the optical cover image is used to characterize the disassembly speed and disassembly angle of the current cover disassembly operation of the manipulator, and wherein the infrared cover image is used to indicate the temperature distribution of the cover of the distribution transformer, thereby characterizing whether the materials at various parts of the cover of the distribution transformer are overheated due to improper disassembly; Generate a disassembly plan using a disassembly model based on the optical cover image and the infrared cover image, and send a control instruction to the manipulator according to the disassembly plan, wherein the control instruction is used to regulate the cover disassembly operation of the manipulator, and the disassembly plan is used to indicate at least one of the following control parameters of the manipulator: disassembly force, disassembly speed, disassembly angle; Wherein, the training process of the disassembly model includes: Obtain a sample optical image and a sample infrared image corresponding to a sample distribution transformer; Use a model to be trained to encode the sample optical image and the sample infrared image respectively, and obtain an optical feature corresponding to the sample optical image and an infrared feature corresponding to the sample infrared image generated during the encoding process respectively; Generate a first disassembly plan using a large model based on the optical feature, the infrared feature, and the information of the sample distribution transformer, wherein the information of the sample distribution transformer includes at least one of the following: size, structure, material; Train the model to be trained based on the optical feature, the infrared feature, and the first disassembly plan to obtain the disassembly model; Wherein, generating a first disassembly plan using a large model based on the optical feature, the infrared feature, and the information of the sample distribution transformer includes: Perform an inner product process on the optical feature and the infrared feature to obtain an inner product feature; Input the inner product feature and the information of the sample distribution transformer into the large model to obtain the first disassembly plan output by the large model; Among them, the sample optical image is cropped into multiple optical local images, the sample infrared image is cropped into multiple infrared local images, N optical local images among the multiple optical local images correspond one-to-one with N infrared local images among the multiple infrared local images, N is a positive integer, the local regions indicated by the corresponding optical local image and infrared local image are the same, the optical feature at least includes the optical local features corresponding to the N optical local images respectively, and the infrared feature at least includes the infrared local features corresponding to the N infrared local images respectively; Using the model to be trained, encoding the sample optical image and the sample infrared image respectively, and obtaining the optical feature corresponding to the sample optical image and the infrared feature corresponding to the sample infrared image generated during the encoding process respectively, includes: Using the model to be trained, encoding the sample optical image, and obtaining the optical local features corresponding to the N optical local images respectively generated during the encoding process; Using the model to be trained, encoding the sample infrared image, and obtaining the infrared local features corresponding to the N infrared local images respectively generated during the encoding process; Among them, each optical local image among the N optical local images and its corresponding infrared local image form a local image pair, and performing an inner product process on the optical feature and the infrared feature, obtaining an inner product feature, includes: For each local image pair, performing an inner product process on the optical local feature and the infrared local feature corresponding to this local image pair, obtaining the inner product local feature corresponding to this local image pair; Based on the inner product local features corresponding to all the local image pairs respectively, determining the inner product feature; Among them, the training process further includes: According to the local region indicated by each local image pair, determining the weight corresponding to each local image pair; The determining the inner product feature based on the inner product local features corresponding to all the local image pairs respectively, includes: Based on the weight corresponding to each local image pair, performing a weighting process on the inner product local feature corresponding to each local image pair; Concatenating the weighted inner product local features corresponding to all the local image pairs respectively into the inner product feature.

Images