Liquid leakage detection method, model training method, device and equipment for liquid cooling server

Abstract

The invention relates to the technical field of equipment detection, and particularly discloses a liquid leakage detection method of a liquid cooling server, a model training method, a device and equipment, and the liquid cooling server comprises a liquid cooling device. The method comprises the following steps: acquiring a target image of a target liquid cooling device acquired by an image acquisition device; inputting the target image into a trained reconstruction model, extracting potential characterization of the target image by the reconstruction model, and reconstructing based on the potential characterization to obtain a reconstructed image of the target liquid cooling device, the reconstruction model being obtained by training based on a first sample image of the first liquid cooling device having no liquid leakage risk; and determining the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image. The probability that the liquid cooling server is damaged can be reduced.

Description

Technical Field

[0001] The present application relates to the technical field of equipment detection, and in particular to a method for detecting liquid leakage of a liquid-cooled server, a model training method, a device and equipment. Background Art

[0002] In a liquid-cooled server, in addition to cooling devices such as fans, there is also a liquid cooling device. By transporting a flowable liquid (such as water) in the liquid cooling device, the heat in the liquid-cooled server can be taken out. Since both fans and liquid cooling devices are included, the heat dissipation efficiency of a liquid-cooled server is usually good and can meet the heat dissipation requirements of high-density computing. However, there is also a risk of leakage in the liquid cooling device, and the leaked liquid may cause damage to the circuit boards and the like in the liquid-cooled server. Therefore, it is necessary to inspect the liquid cooling device and promptly detect liquid cooling devices with leakage problems.

[0003] Some current technologies can only detect liquid cooling devices that have already leaked. This means that when liquid cooling devices with leaks are detected, the leaked liquid from these liquid cooling devices has already begun to damage the liquid cooling server, or has already damaged the liquid cooling server for some time. Therefore, it is necessary to improve these technologies, preferably to promptly detect liquid cooling devices that are about to leak but have not yet leaked, and to promptly replace or repair these liquid cooling devices, thereby reducing the probability of liquid cooling servers being damaged. Summary of the invention

[0004] In view of this, the implementation method of the present application provides a liquid leakage detection method, a model training method, a liquid leakage detection device, a model training device, an electronic device and a computer-readable storage medium for a liquid-cooled server, which can detect a liquid cooling device with a risk of leakage, thereby reducing the probability of damage to the liquid-cooled server.

[0005] On one hand, the present application provides a method for detecting liquid leakage of a liquid-cooled server, wherein the liquid-cooled server includes a liquid cooling device; the method includes:

[0006] Acquire a target image of the target liquid cooling device acquired by an image acquisition device;

[0007] Inputting the target image into a trained reconstruction model, extracting a potential representation of the target image by the reconstruction model, and reconstructing a reconstructed image of the target liquid cooling device based on the potential representation, wherein the reconstruction model is trained based on a first sample image of a first liquid cooling device without a risk of liquid leakage;

[0008] Based on a first reconstruction error between the target image and the reconstructed image, a liquid leakage risk of the target liquid cooling device is determined.

[0009] On the other hand, the present application also provides a model training method, the method comprising:

[0010] Acquire a first sample image of a first liquid cooling device having no risk of liquid leakage;

[0011] Inputting each of the first sample images into a reconstruction model to be trained respectively, extracting potential representations of the first sample images by the reconstruction model, and reconstructing a first sample reconstructed image of the first liquid cooling device based on the potential representations;

[0012] Determine a first sample reconstruction error between the first sample image and the first sample reconstructed image, and adjust model parameters of the reconstruction model based on the deviation of the first sample reconstruction error relative to the target reconstruction error, so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than a deviation threshold.

[0013] On the other hand, the present application further provides a liquid leakage detection device for a liquid cooling server, wherein the liquid cooling server comprises a liquid cooling device; the device comprises:

[0014] An image acquisition module, used to acquire a target image of the target liquid cooling device acquired by an image acquisition device;

[0015] a reconstruction module, configured to input the target image into a trained reconstruction model, extract a potential representation of the target image by the reconstruction model, and reconstruct a reconstructed image of the target liquid cooling device based on the potential representation, wherein the reconstruction model is trained based on a first sample image of a first liquid cooling device without a risk of liquid leakage;

[0016] The liquid leakage detection module is used to determine the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image.

[0017] On the other hand, the present application also provides a model training device, the device comprising:

[0018] A sample image acquisition module, used to acquire a first sample image of a first liquid cooling device without a risk of liquid leakage;

[0019] a sample image reconstruction module, used for inputting each of the first sample images into a reconstruction model to be trained, extracting a potential representation of the first sample image by the reconstruction model, and reconstructing a first sample reconstructed image of the first liquid cooling device based on the potential representation;

[0020] A model training module is used to determine a first sample reconstruction error between the first sample image and the first sample reconstructed image, and based on the deviation of the first sample reconstruction error relative to the target reconstruction error, adjust the model parameters of the reconstruction model so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than a deviation threshold.

[0021] On the other hand, the present application further provides an electronic device, which includes a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method described above is implemented.

[0022] On the other hand, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0023] In the technical solutions of some embodiments of the present application, since the reconstruction model is mainly trained based on the first sample image of the first liquid cooling device without the risk of leakage, the extraction method of the potential representation and the reconstruction method of the image learned by the reconstruction model are mainly for the image of the liquid cooling device without the risk of leakage. Furthermore, when the target image input to the reconstruction model is an image of a liquid cooling device with the risk of leakage or leakage, the first reconstruction error between the target image and the reconstructed image can be large, and when the target image is an image of a liquid cooling device without the risk of leakage, the first reconstruction error between the target image and the reconstructed image can be small. In this way, based on the first reconstruction error, it can be determined whether the target liquid cooling device has the risk of leakage, so that when the target liquid cooling device has the risk of leakage but has not actually caused the leakage problem, the target liquid cooling device can be changed or repaired in advance, thereby greatly reducing the probability of the liquid cooling server where the target liquid cooling device is located being damaged by leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The features and advantages of the present application will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present application in any way. In the accompanying drawings:

[0025] Figure 1 A schematic diagram of a flow chart of a model training method provided by an embodiment of the present application is shown;

[0026] Figure 2 A schematic diagram of a flow chart of a liquid leakage detection method provided by an embodiment of the present application is shown;

[0027] Figure 3 A schematic diagram showing the relationship between a target reconstruction error and various error thresholds provided by an embodiment of the present application is shown;

[0028] Figure 4 A schematic diagram showing a normal distribution of a second reconstruction error provided by an embodiment of the present application is shown;

[0029] Figure 5 A schematic diagram of a module of a reconstruction model provided by an embodiment of the present application is shown;

[0030] Figure 6 A schematic diagram of a module of a liquid leakage detection device provided by an embodiment of the present application is shown;

[0031] Figure 7 A schematic diagram of a module of a model training device provided by an embodiment of the present application is shown;

[0032] Figure 8 A schematic diagram of an electronic device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the implementation method of this application clearer, the technical scheme in the implementation method of this application will be clearly and completely described below in conjunction with the drawings in the implementation method of this application. Obviously, the implementation method described is a part of the implementation method of this application, not all of the implementation methods. Based on the implementation method in this application, all other implementation methods obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0034] In some technologies, an image acquisition device is usually installed in a liquid cooling server, and the image of the liquid cooling device is regularly acquired. Then, based on the image difference between the liquid cooling device when there is a leakage problem and when there is no leakage problem, the acquired image is identified by manual or intelligent means to determine whether there is a leakage in the liquid cooling device. For example, when there is a leakage problem in the liquid cooling device, liquid drops or rust will appear at the leakage point of the liquid cooling device, while when there is no leakage problem in the liquid cooling device, there will be no liquid drops and rust on the surface of the liquid cooling device. Therefore, by regularly acquiring images of the liquid cooling device and identifying whether there are liquid drops or rust in the image, it can be determined whether there is a leakage problem in the liquid cooling device. For another example, there may be differences in the infrared imaging of a liquid cooling device with a leakage problem and a liquid cooling device without a leakage problem. Therefore, by regularly acquiring infrared imaging of the liquid cooling device and identifying the acquired infrared imaging, it can be determined whether there is a leakage problem in the liquid cooling device.

[0035] The liquid cooling devices identified by these technologies are liquid cooling devices that have already had leakage problems, that is, the leaked liquid from these liquid cooling devices has begun to damage the liquid cooling server, or has damaged the liquid cooling server for some time. Maintenance personnel can only perform some post-inspection and processing on the liquid cooling server. The probability of the liquid cooling server being damaged by the leaked liquid from the liquid cooling device is still high.

[0036] In view of this, the present application provides a model training method and a liquid leakage detection method for a liquid cooling server, which can detect liquid cooling devices that have a risk of leakage but have not actually leaked, so that maintenance personnel can replace or repair these liquid cooling devices in time, reducing the probability of damage to the liquid cooling server. Among them, the model training method is used to train a reconstruction model, and the leakage detection method can be executed based on the trained reconstruction model.

[0037] Combined with reference Figure 1 , is a flow chart of a model training method provided by an embodiment of the present application. The model training method can be applied to electronic devices. The electronic devices include but are not limited to tablet computers, laptop computers, desktop computers, etc. Figure 1 In , the model training method includes the following steps:

[0038] Step S101 : acquiring a first sample image of a first liquid cooling device having no risk of liquid leakage.

[0039] Specifically, there may be multiple first liquid cooling devices, so that a large number of first sample images for model training can be obtained. For the first sample image of any first liquid cooling device, the first sample image may include an overall image of the corresponding first liquid cooling device and images at different positions. The so-called different positions of the first liquid cooling device refers to dividing the first liquid cooling device into multiple parts, and the image of each part represents the image at one position of the first liquid cooling device.

[0040] Step S102: input each first sample image into a reconstruction model to be trained, extract the potential representation of the first sample image by the reconstruction model, and reconstruct a first sample reconstructed image of the first liquid cooling device based on the potential representation.

[0041] Specifically, the latent representation refers to the key features of the first sample image extracted by the reconstruction model, and these features are not necessarily the features directly observed from the first sample image, but a simplified or abstract representation of the first sample image. Based on the extracted latent representation, the reconstruction model can reconstruct the image of the first liquid cooling device, that is, the above-mentioned first sample reconstructed image.

[0042] Step S103, determining a first sample reconstruction error between the first sample image and the first sample reconstructed image, and adjusting the model parameters of the reconstruction model based on the deviation of the first sample reconstruction error relative to the target reconstruction error, so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than a deviation threshold.

[0043] The first sample reconstruction error refers to the difference between the first sample image and the first sample reconstructed image. Specifically, the first pixel value distribution of the first sample image and the second pixel value distribution of the first sample reconstructed image can be obtained, and then the difference between the first pixel value distribution and the second pixel value distribution is used as the first sample reconstruction error. Of course, the first sample reconstruction error can also be determined by other means. For example, the first sample reconstruction error can also be determined based on the mean square error of the image pixel values, etc., and this application does not limit this.

[0044] The deviation of the first sample reconstruction error relative to the target reconstruction error may be the absolute value of the difference between the first sample reconstruction error and the target reconstruction error. The target reconstruction error may be a smaller value, so that when the deviation of the first sample reconstruction error relative to the target reconstruction error is less than the deviation threshold, it can be considered that the similarity between the first sample image and the first sample reconstructed image has met the requirement. Training the reconstruction model is to determine whether the deviation of the first sample reconstruction error relative to the target reconstruction error is less than the deviation threshold. If so, the training of the reconstruction model can be terminated. If not, it is necessary to continue training the reconstruction model (i.e., continue to adjust the model parameters of the reconstruction model).

[0045] Generally, in the training process of the reconstruction model, the model parameters of the reconstruction model are adjusted, that is, the extraction method of the potential representation of the first sample image and the reconstruction method of the first sample reconstructed image are adjusted, so that the similarity between the constructed first sample reconstructed image and the first sample image meets the requirements. This training process can also be called the learning process of the reconstruction model, that is, learning the extraction method of the potential representation of the first sample image and the reconstruction method of the first sample reconstructed image.

[0046] Furthermore, since the first sample image is an image of the first liquid cooling device without the risk of liquid leakage, the extraction method of the potential representation and the image reconstruction method learned by the trained reconstruction model can be for the image of the liquid cooling device without the risk of liquid leakage. In other words, after the reconstruction model is trained using the first sample image, the reconstruction model can reconstruct the image of the liquid cooling device without the risk of liquid leakage more accurately, so that the similarity between the reconstructed image and the original image meets the requirements.

[0047] In addition, there are certain differences between the images of the liquid cooling device with or without leakage risk and the images of the liquid cooling device without leakage risk. For example, in the images of the liquid cooling device with or without leakage risk, the pipeline will be deformed, have abnormal bubbles, and have abnormal infrared temperature distribution. In the images of the liquid cooling device with or without leakage risk, there will be liquid droplets and rust. However, in the images of the liquid cooling device without leakage risk, there will be no deformation, abnormal bubbles, and other related conditions. Therefore, after the reconstruction model is trained, after the images of the liquid cooling device with or without leakage risk are input into the reconstruction model, if the reconstruction model reconstructs the corresponding liquid cooling device images according to the extraction method of the latent representation and the reconstruction method of the image that have been learned, there may be a large difference between the reconstructed image and the original image, and the deviation between the error size of the reconstructed image relative to the original image and the target reconstruction error may be greater than the deviation threshold. In this way, after the reconstruction model is trained, by inputting the liquid cooling device image into the reconstruction model, and based on the error size of the reconstructed image relative to the original image, it is possible to determine whether the liquid cooling device has a risk of leakage or has a leakage problem, that is, the liquid cooling device with a risk of leakage can be detected in advance, so that the liquid cooling device can be replaced or repaired in time before the leakage problem occurs, thereby reducing the probability of damage to the liquid cooling server. Of course, it is also possible to detect the liquid cooling server that has already had a leakage problem, thereby reducing the degree of damage to the liquid cooling server.

[0048] In some embodiments, in order to avoid the overfitting problem of the reconstruction model, in the process of acquiring the first sample image and training the reconstruction model based on the first sample image, the method of the present application may further include:

[0049] acquiring a second sample image of a second liquid cooling device;

[0050] Inputting each second sample image into the reconstruction model to be trained respectively, extracting the potential representation of the second sample image by the reconstruction model, and reconstructing a second sample reconstructed image of the second liquid cooling device based on the potential representation;

[0051] Determine a second sample reconstruction error between the second sample image and the second sample reconstructed image, and adjust the model parameters of the reconstruction model based on the deviation of the second sample reconstruction error relative to the target reconstruction error, so that the deviation of the second sample reconstruction error relative to the target reconstruction error is less than a deviation threshold;

[0052] Wherein, the second liquid cooling device includes at least one of the following liquid cooling devices:

[0053] Liquid cooling devices that are at risk of leaking but have not yet leaked;

[0054] Liquid cooling device that has already leaked;

[0055] Furthermore, the number of images of the first sample images is greater than the number of images of the second sample images.

[0056] Specifically, the deviation of the second sample reconstruction error relative to the target reconstruction error may be the absolute value of the difference between the second sample reconstruction error and the target reconstruction error.

[0057] Among all the sample images used to train the reconstruction model, the proportion of the first sample images can be much larger than the proportion of the second sample images. For example, the proportion of the first sample images can be 90%, and the proportion of the second sample images can be 10%. Since the proportion of the first sample images is much larger than the proportion of the second sample images, the influence of the second sample images on the model training may not be significant. Therefore, the extraction method of the potential representation and the image reconstruction method mainly learned by the reconstruction model can still be for the image of the liquid cooling device without the risk of leakage. However, because a small amount of second sample images are added to the sample images, the overfitting problem of the reconstruction model is avoided, so that the reconstruction model obtained by training can have better generalization ability.

[0058] In some embodiments, after the reconstruction model is trained based on the above training method, the obtained reconstruction model can have basic image reconstruction capabilities, so that the error size of the reconstructed image relative to the original image can be within a threshold range. However, in order to improve the model accuracy, the reconstruction model obtained by the above training can be trained again to make subtle adjustments to the model parameters of the reconstruction model, so that the reconstruction model can learn the difference between the first sample image and the second sample image more accurately on the basis of having basic image reconstruction capabilities, so that the first sample reconstruction error is different from the second sample reconstruction error.

[0059] Specifically, when re-training the reconstructed image, the deviation threshold may include a first deviation threshold, a second deviation threshold and a third deviation threshold, wherein the first deviation threshold is less than the second deviation threshold, and the second deviation threshold is less than the third deviation threshold. The first sample reconstruction error between the first sample image and the first sample reconstructed image may be annotated, and the second sample reconstruction error between the second sample image and the second sample reconstructed image may be annotated, so that the deviation magnitudes of the first sample reconstruction error and the second sample reconstruction error relative to the target reconstruction error may both be less than the third deviation threshold, but the deviation magnitude of the first sample reconstruction error relative to the target reconstruction error may be between the first deviation threshold and the second deviation threshold, and the deviation magnitude of the second sample reconstruction error relative to the target reconstruction error may be between the second deviation threshold and the third deviation threshold. After re-training the reconstruction model according to such annotations, the reconstruction model can better distinguish between images of liquid cooling devices that are at risk of leakage or have already leaked, and images of liquid cooling devices that are not at risk of leakage, with higher accuracy.

[0060] Based on the trained reconstruction model, the leakage detection method of the present application can be executed. The leakage detection method can be applied to electronic devices. The electronic devices include but are not limited to tablet computers, laptop computers, desktop computers, etc. Figure 2 , which is a flow chart of a liquid leakage detection method provided in one embodiment of the present application. Figure 2 In the present invention, the leakage detection method comprises the following steps:

[0061] Step S201, acquiring a target image of a target liquid cooling device acquired by an image acquisition device.

[0062] Specifically, the image acquisition device may include but is not limited to an infrared camera, a camera, etc. The target liquid cooling device is a liquid cooling device that needs to be detected for leakage. In the liquid cooling server where the target liquid cooling device is located, multiple image acquisition devices can be deployed, and these image acquisition devices can be used to regularly acquire the target image of the target liquid cooling device. For example, the target image of the target liquid cooling device is acquired every 1 hour or every 30 minutes. For each target image acquired, step S202 and step S203 can be executed respectively to determine whether the target liquid cooling device has a leakage problem.

[0063] Step S202 , inputting the target image into a trained reconstruction model, extracting the potential representation of the target image by the reconstruction model, and reconstructing a reconstructed image of the target liquid cooling device based on the potential representation.

[0064] As described in the above model training method, the reconstructed model can be obtained by training based on the first sample image of the first liquid cooling device without leakage risk, or based on the first sample image and the second sample image of the second liquid cooling device.

[0065] The principle of step S202 is similar to that of step S102 and will not be described in detail here.

[0066] Step S203: determining the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image.

[0067] Specifically, through the above training process, it can be learned that if the deviation between the first reconstruction error and the target reconstruction error is smaller than the deviation threshold, it can be considered that there is no leakage risk in the target liquid cooling device; if the deviation between the first reconstruction error and the target reconstruction error is larger than the deviation threshold, it can be considered that there is a leakage risk in the target liquid cooling device or a leakage problem has already occurred.

[0068] In summary, in the technical solutions of some embodiments of the present application, since the reconstruction model is mainly trained based on the first sample image of the first liquid cooling device without leakage risk, the extraction method of the potential representation and the reconstruction method of the image learned by the reconstruction model are mainly for the image of the liquid cooling device without leakage risk. Furthermore, when the target image input to the reconstruction model is an image of a liquid cooling device with leakage risk or leakage, the first reconstruction error between the target image and the reconstructed image can be large, and when the target image is an image of a liquid cooling device without leakage risk, the first reconstruction error between the target image and the reconstructed image can be small. In this way, based on the first reconstruction error, it can be determined whether the target liquid cooling device has leakage risk, so that when the target liquid cooling device has leakage risk but has not actually leakage problem, the target liquid cooling device can be changed or repaired in advance, so the probability of the liquid cooling server where the target liquid cooling device is located being damaged by leakage can be greatly reduced.

[0069] In some embodiments, determining the liquid leakage risk of the target liquid cooling device based on the first reconstruction error between the target image and the reconstructed image in step S203 may include:

[0070] If the first reconstruction error is between the first error threshold and the second error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is greater than the second error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the second error threshold is greater than the first error threshold; and / or

[0071] If the first reconstruction error is between the third error threshold and the fourth error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is less than the fourth error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the fourth error threshold is less than the third error threshold.

[0072] Specifically, the target reconstruction error may be a statistical reconstruction error of most liquid cooling device images without leakage risk. If the deviation of the first reconstruction error relative to the target reconstruction error is larger, it means that the target liquid cooling device has a greater leakage risk, or even has leakage problem.

[0073] The first error threshold, the second error threshold, the third error threshold and the fourth error threshold may be located on both sides of the target reconstruction error. Figure 3 , is a schematic diagram of the relationship between the target reconstruction error and various error thresholds provided in an embodiment of the present application. Figure 3In the example, the deviation magnitude of the first error threshold and the third error threshold relative to the target reconstruction error may be equal to the deviation threshold, and the deviation magnitude of the second error threshold and the fourth error threshold relative to the target reconstruction error may be greater than the deviation threshold. This means that if the first reconstruction error between the target image and the reconstructed image is between the target reconstruction error and the first error threshold, or between the target reconstruction error and the third error threshold, it can be considered that the target liquid cooling device has no leakage risk; if the first reconstruction error is between the first error threshold and the second error threshold, or between the third error threshold and the fourth error threshold, it can be considered that the target liquid cooling device has a leakage risk but has not yet experienced a leakage problem; if the first reconstruction error is greater than the second error threshold or less than the fourth error threshold, it can be considered that the target liquid cooling device has already experienced a leakage problem.

[0074] In the above embodiment, multiple threshold intervals are divided to distinguish between a situation where there is a risk of leakage but the leakage problem has not yet occurred and a situation where the leakage problem has occurred, so that the problem detection of the target liquid cooling device is more accurate.

[0075] In some embodiments, the first error threshold, the second error threshold, the third error threshold, and the fourth error threshold are obtained based on the following method:

[0076] 11) After the reconstruction model is trained, each sample image used to train the reconstruction model is input into the trained reconstruction model again, and the reconstruction model extracts the potential representation of the sample image, and reconstructs the sample reconstructed image corresponding to each sample image based on the extracted potential representation;

[0077] 12) respectively determining the second reconstruction error between each sample image and the corresponding sample reconstructed image, and obtaining a full set of second reconstruction errors;

[0078] 13) In the full set, second reconstruction errors with the same value are grouped together to obtain a plurality of subsets of second reconstruction errors;

[0079] 14) determining the occurrence probability of the second reconstruction error in each subset according to the number of the second reconstruction errors in the full set and the number of the second reconstruction errors in each subset;

[0080] 15) Taking the second reconstruction error as the abscissa and the occurrence probability as the ordinate, determine the normal distribution of the second reconstruction error in the full set;

[0081] 16) In the normal distribution, the second reconstruction error with the highest probability of occurrence is used as the reference reconstruction error, and the first error threshold, the second error threshold, the third error threshold and the fourth error threshold are determined among the second reconstruction errors on both sides of the reference reconstruction error.

[0082] For easier understanding, please refer to Figure 4 , is a schematic diagram of the normal distribution of the second reconstruction error provided by an embodiment of the present application. Figure 4 In the above, the probability at T0 is the largest, which is represented by the reconstruction error of most liquid cooling device images without leakage risk. Therefore, T0 can be determined as the target reconstruction error. On both sides of T0, T1, T2, T3 and T4 for representing the first error threshold, the second error threshold, the third error threshold and the fourth error threshold can be determined according to actual needs.

[0083] In the above embodiment, the target reconstruction error, the first error threshold, the second error threshold, the third error threshold and the fourth error threshold are determined by normal distribution, and the obtained thresholds can be relatively objective and accurate.

[0084] In some embodiments, the target image is associated with a server identifier of a liquid cooling server where the target liquid cooling device is located; after determining the first reconstruction error, the method of the present application may further include:

[0085] If it is determined based on the first reconstruction error that the target liquid cooling device has a risk of leakage but has not yet leaked, sending a first alarm; and if it is determined based on the first reconstruction error that the target liquid cooling device has already leaked, sending a second alarm;

[0086] The urgency of the first alarm is lower than that of the second alarm, and the first alarm and the second alarm include a server identifier associated with the target image.

[0087] Specifically, the urgency of the first alarm and the second alarm can be distinguished by the alarm color. For example, the first alarm can be blue, and the second alarm can be red. In this way, it is convenient for maintenance personnel to distinguish alarms of different urgency by alarm color, and promptly handle the liquid cooling device that has leaked. At the same time, replace or repair the liquid cooling device that has the risk of leaking but has not leaked in advance.

[0088] In the above embodiment, by associating the target image with the server identifier of the liquid cooling server where the target liquid cooling device is located, it is possible to locate the liquid cooling server where the target liquid cooling device is located when the target liquid cooling device has a leakage problem or a risk of leakage, and then replace or maintain the target liquid cooling device.

[0089] Furthermore, in some embodiments, the liquid cooling server where the target liquid cooling device is located includes multiple image acquisition devices, and different image acquisition devices are used to acquire target images at different positions of the target liquid cooling device. The target image is also associated with a device identifier of the image acquisition device that acquires the target image; the first alarm and the second alarm also include the device identifier associated with the target image.

[0090] In this way, the fault position of the target liquid cooling device can be located.

[0091] Combined with reference Figure 5 , which is a module diagram of a reconstruction model provided by an embodiment of the present application. Figure 5 In the embodiment, the reconstruction model includes an encoder and a decoder, wherein the encoder is used to extract the potential representation of the target image, and the decoder is used to reconstruct the reconstructed image of the target liquid cooling device based on the potential representation; wherein the encoder and the decoder each include a preset number of convolutional layers. Specifically, the encoder and the decoder may each include 3 convolutional layers.

[0092] In the above embodiment, by setting the convolution layer in the encoder and the decoder, it is convenient to process the target image.

[0093] In some embodiments, before inputting the target image into the reconstruction model, the target image may be preprocessed in advance, including but not limited to normalization processing, grayscale processing, image size adjustment, etc. of the target image.

[0094] Corresponding to the liquid leakage detection method, the present application also provides a liquid leakage detection device for a liquid cooling server. Figure 6 , which is a module schematic diagram of a liquid leakage detection device provided in one embodiment of the present application. Figure 6 In the invention, the leakage detection device comprises:

[0095] An image acquisition module 601 is used to acquire a target image of a target liquid cooling device acquired by an image acquisition device;

[0096] A reconstruction module 602 is used to input a target image into a trained reconstruction model, extract a potential representation of the target image by the reconstruction model, and reconstruct a reconstructed image of the target liquid cooling device based on the potential representation, wherein the reconstruction model is trained based on a first sample image of a first liquid cooling device that does not have a risk of liquid leakage;

[0097] The liquid leakage detection module 603 is used to determine the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image.

[0098] In some embodiments, the reconstruction model is also trained based on at least one of the following second sample images of the second liquid cooling device:

[0099] Liquid cooling devices that are at risk of leaking but have not yet leaked;

[0100] Liquid cooling device that has already leaked;

[0101] The number of images of the first sample image is greater than the number of images of the second sample image.

[0102] In some embodiments, the liquid leakage detection module 603 is specifically used to:

[0103] If the first reconstruction error is between the first error threshold and the second error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is greater than the second error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the second error threshold is greater than the first error threshold; and / or

[0104] If the first reconstruction error is between the third error threshold and the fourth error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is less than the fourth error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the fourth error threshold is less than the third error threshold.

[0105] In some embodiments, the first error threshold, the second error threshold, the third error threshold and the second error threshold are obtained based on the following method:

[0106] After the reconstruction model is trained, each sample image used to train the reconstruction model is input into the trained reconstruction model again, and the reconstruction model extracts the potential representation of the sample image, and reconstructs the sample reconstructed image corresponding to each sample image based on the extracted potential representation;

[0107] Determine the second reconstruction error between each sample image and the corresponding sample reconstructed image respectively, and obtain a full set of second reconstruction errors;

[0108] In the full set, second reconstruction errors with the same value are taken as a group to obtain a plurality of subsets of second reconstruction errors;

[0109] Determining respectively the occurrence probability of the second reconstruction error in each subset according to the number of the second reconstruction errors in the full set and the number of the second reconstruction errors in each subset;

[0110] Taking the second reconstruction error as the abscissa and the occurrence probability as the ordinate, determining the normal distribution of the second reconstruction error in the entire set;

[0111] In the normal distribution, the second reconstruction error with the highest probability is taken as the reference reconstruction error, and the first error threshold, the second error threshold, the third error threshold and the second error threshold are determined among the second reconstruction errors on both sides of the reference reconstruction error.

[0112] In some embodiments, the target image is associated with a server identifier of a liquid cooling server where the target liquid cooling device is located; after determining the first reconstruction error, the liquid leakage detection module 603 is further used to:

[0113] If it is determined based on the first reconstruction error that the target liquid cooling device has a risk of leakage but has not yet leaked, sending a first alarm; and if it is determined based on the first reconstruction error that the target liquid cooling device has already leaked, sending a second alarm;

[0114] The urgency of the first alarm is lower than that of the second alarm, and the first alarm and the second alarm include a server identifier associated with the target image.

[0115] In some embodiments, the liquid cooling server where the target liquid cooling device is located includes multiple image acquisition devices, and different image acquisition devices are used to acquire target images at different positions of the target liquid cooling device, and the target image is also associated with a device identifier of the image acquisition device that acquires the target image;

[0116] The first alarm and the second alarm also include a device identifier associated with the target image.

[0117] In some embodiments, the reconstruction model includes an encoder and a decoder, wherein the encoder is used to extract a potential representation of the target image, and the decoder is used to reconstruct a reconstructed image of the target liquid cooling device based on the potential representation;

[0118] The encoder and the decoder each include a preset number of convolutional layers.

[0119] Corresponding to the model training method, the present application also provides a model training device. Figure 7 , which is a module diagram of a model training device provided in one embodiment of the present application. Figure 7 In the model training device, the model training device includes:

[0120] A sample image acquisition module 701 is used to acquire a first sample image of a first liquid cooling device without a risk of liquid leakage;

[0121] The sample image reconstruction module 702 is used to input each first sample image into the reconstruction model to be trained, extract the potential representation of the first sample image by the reconstruction model, and reconstruct the first sample reconstructed image of the first liquid cooling device based on the potential representation;

[0122] The model training module 703 is used to determine the first sample reconstruction error between the first sample image and the first sample reconstructed image, and adjust the model parameters of the reconstruction model based on the deviation of the first sample reconstruction error relative to the target reconstruction error, so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than the deviation threshold.

[0123] In some embodiments, in the process of acquiring the first sample image and training the reconstruction model based on the first sample image, the model training module 703 is further used to:

[0124] acquiring a second sample image of the second liquid cooling device;

[0125] Inputting each second sample image into the reconstruction model to be trained respectively, extracting the potential representation of the second sample image by the reconstruction model, and reconstructing a second sample reconstructed image of the second liquid cooling device based on the potential representation;

[0126] Determine a second sample reconstruction error between the second sample image and the second sample reconstructed image, and adjust the model parameters of the reconstruction model based on the deviation of the second sample reconstruction error relative to the target reconstruction error, so that the deviation of the second sample reconstruction error relative to the target reconstruction error is less than a deviation threshold;

[0127] Wherein, the second liquid cooling device includes at least one of the following liquid cooling devices:

[0128] Liquid cooling devices that are at risk of leaking but have not yet leaked;

[0129] Liquid cooling device that has already leaked;

[0130] Furthermore, the number of images of the first sample images is greater than the number of images of the second sample images.

[0131] Combined with reference Figure 8 , is a schematic diagram of an electronic device provided by an embodiment of the present application. The electronic device includes a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the above method is implemented.

[0132] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0133] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, that is, implementing the method in the above method embodiment.

[0134] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0135] An embodiment of the present disclosure further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0136] Although the implementation method of the present application is described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for detecting liquid leakage in a liquid-cooled server, characterized in that: The liquid cooling server comprises a liquid cooling device; the method comprises: Acquire a target image of the target liquid cooling device acquired by an image acquisition device; Inputting the target image into a trained reconstruction model, extracting a potential representation of the target image by the reconstruction model, and reconstructing a reconstructed image of the target liquid cooling device based on the potential representation, wherein the reconstruction model is trained based on a first sample image of a first liquid cooling device without a risk of liquid leakage; Based on a first reconstruction error between the target image and the reconstructed image, a liquid leakage risk of the target liquid cooling device is determined.

2. The method according to claim 1, characterized in that The reconstruction model is also obtained by training based on at least one of the following second sample images of the second liquid cooling device: Liquid cooling devices that are at risk of leaking but have not yet leaked; Liquid cooling device that has leaked liquid; The number of images of the first sample images is greater than the number of images of the second sample images.

3. The method according to claim 1 or 2, characterized in that The step of determining the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image includes: If the first reconstruction error is between a first error threshold and a second error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is greater than the second error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the second error threshold is greater than the first error threshold; and / or If the first reconstruction error is between a third error threshold and a fourth error threshold, it is determined that the target liquid cooling device has a risk of leakage but has not yet leaked, and if the first reconstruction error is less than the fourth error threshold, it is determined that the target liquid cooling device has a leakage problem, wherein the fourth error threshold is less than the third error threshold.

4. The method according to claim 3, characterized in that The first error threshold, the second error threshold, the third error threshold, and the fourth error threshold are obtained based on the following method: After the reconstruction model is trained, each sample image used to train the reconstruction model is input into the trained reconstruction model again, the reconstruction model extracts the potential representation of the sample image, and based on the extracted potential representation, reconstructs a sample reconstructed image corresponding to each sample image; Determine the second reconstruction error between each of the sample images and the corresponding sample reconstructed image respectively, and obtain a full set of second reconstruction errors; In the full set, second reconstruction errors with the same value are taken as a group to obtain a plurality of subsets of second reconstruction errors; Determining respectively the occurrence probability of the second reconstruction error in each subset according to the number of second reconstruction errors in the full set and the number of second reconstruction errors in each subset; Taking the second reconstruction error as the abscissa and the occurrence probability as the ordinate, determining a normal distribution of the second reconstruction error in the full set; In the normal distribution, the second reconstruction error with the highest probability of occurrence is used as the reference reconstruction error, and the first error threshold, the second error threshold, the third error threshold and the fourth error threshold are determined among the second reconstruction errors on both sides of the reference reconstruction error.

5. The method according to claim 3, characterized in that The target image is associated with a server identifier of a liquid cooling server where the target liquid cooling device is located; After determining the first reconstruction error, the method further comprises: If it is determined based on the first reconstruction error that the target liquid cooling device has a risk of leakage but has not yet leaked, sending a first alarm; and if it is determined based on the first reconstruction error that the target liquid cooling device has already leaked, sending a second alarm; The urgency of the first alarm is lower than that of the second alarm, and the first alarm and the second alarm include a server identifier associated with the target image.

6. The method according to claim 5, characterized in that The liquid cooling server where the target liquid cooling device is located includes a plurality of the image acquisition devices, and different image acquisition devices are used to acquire target images at different positions of the target liquid cooling device, and the target image is also associated with a device identifier of the image acquisition device that acquires the target image; The first alarm and the second alarm also include a device identifier associated with the target image.

7. The method according to claim 1, characterized in that The reconstruction model includes an encoder and a decoder, wherein the encoder is used to extract a potential representation of the target image, and the decoder is used to reconstruct the reconstructed image of the target liquid cooling device based on the potential representation; The encoder and the decoder each include a preset number of convolutional layers.

8. A model training method, characterized in that: The method comprises: Acquire a first sample image of a first liquid cooling device having no risk of liquid leakage; Inputting each of the first sample images into a reconstruction model to be trained respectively, extracting potential representations of the first sample images by the reconstruction model, and reconstructing a first sample reconstructed image of the first liquid cooling device based on the potential representations; Determine a first sample reconstruction error between the first sample image and the first sample reconstructed image, and adjust model parameters of the reconstruction model based on the deviation of the first sample reconstruction error relative to the target reconstruction error, so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than a deviation threshold.

9. The method according to claim 8, characterized in that In the process of acquiring the first sample image and training the reconstruction model based on the first sample image, the method further includes: acquiring a second sample image of the second liquid cooling device; Inputting each of the second sample images into a reconstruction model to be trained respectively, extracting a potential representation of the second sample image by the reconstruction model, and reconstructing a second sample reconstructed image of the second liquid cooling device based on the potential representation; determining a second sample reconstruction error between the second sample image and the second sample reconstructed image, and adjusting a model parameter of the reconstruction model based on a deviation of the second sample reconstruction error relative to the target reconstruction error, so that a deviation of the second sample reconstruction error relative to the target reconstruction error is less than a deviation threshold; Wherein, the second liquid cooling device includes at least one of the following liquid cooling devices: Liquid cooling devices that are at risk of leaking but have not yet leaked; Liquid cooling device that has leaked liquid; Furthermore, the number of images of the first sample images is greater than the number of images of the second sample images.

10. A liquid leakage detection device for a liquid cooling server, characterized in that: The liquid cooling server comprises a liquid cooling device; the device comprises: An image acquisition module, used to acquire a target image of the target liquid cooling device acquired by an image acquisition device; a reconstruction module, configured to input the target image into a trained reconstruction model, extract a potential representation of the target image by the reconstruction model, and reconstruct a reconstructed image of the target liquid cooling device based on the potential representation, wherein the reconstruction model is trained based on a first sample image of a first liquid cooling device without a risk of liquid leakage; The liquid leakage detection module is used to determine the liquid leakage risk of the target liquid cooling device based on a first reconstruction error between the target image and the reconstructed image.

11. A model training device, characterized in that: The device comprises: A sample image acquisition module, used to acquire a first sample image of a first liquid cooling device without a risk of liquid leakage; a sample image reconstruction module, used for inputting each of the first sample images into a reconstruction model to be trained, extracting a potential representation of the first sample image by the reconstruction model, and reconstructing a first sample reconstructed image of the first liquid cooling device based on the potential representation; A model training module is used to determine a first sample reconstruction error between the first sample image and the first sample reconstructed image, and based on the deviation of the first sample reconstruction error relative to the target reconstruction error, adjust the model parameters of the reconstruction model so that the deviation of the first sample reconstruction error relative to the target reconstruction error is less than a deviation threshold.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, it implements the method according to any one of claims 1 to 7, or implements the method according to any one of claims 8 to 9.

13. An electronic device, characterized in that: The electronic device includes a processor and a memory, the memory is used to store a computer program, and when the computer program is executed by the processor, it implements the method according to any one of claims 1 to 7, or implements the method according to any one of claims 8 to 9.

Images