CN-121994426-A - Liquid leakage monitoring and predicting method and system for liquid cooling system of electronic device

Abstract

The invention discloses a leakage monitoring and predicting method and a leakage monitoring and predicting system for a liquid cooling system of an electronic device, and discloses a related leakage monitoring technology. In some embodiments, an instant audio signal of fluid flow in a pipeline is received in real time, the instant audio signal is converted into a time-frequency spectrogram, and image recognition is performed on the time-frequency spectrogram to judge whether the pipeline has liquid leakage or not.

Inventors

• TANG TONGYANG
• PAN LIPING
• GAO MENGCHAO
• CAI ZHUJIA
• LIU WENHUA

Assignees

• 纬创资通股份有限公司

Dates

Publication Date

20260508

Application Date

20241126

Priority Date

20241101

Claims (20)

1. 1. A leakage monitoring and predicting method for a liquid cooling system of an electronic device comprises the following steps: receiving an instant audio signal of fluid flow in the pipeline; converting the instant audio signal into a time-frequency spectrogram, and And executing image recognition on the time-frequency spectrogram to judge whether the pipeline has liquid leakage.
2. 2. The leakage monitoring and prediction method according to claim 1, wherein the step of converting the audio signal into the time-frequency spectrogram comprises: And performing fast Fourier transform on the instant audio signal to generate the time-frequency spectrogram, wherein the time-frequency spectrogram is a three-dimensional spectrogram, and the three-dimensional spectrogram comprises time information, frequency information and sound intensity information.
3. 3. The method for monitoring and predicting leakage according to claim 1, wherein the step of determining whether the pipeline has leakage according to the time-frequency spectrum chart comprises: And inputting the time-frequency spectrogram into a classification model to generate a liquid leakage judging result of the pipeline.
4. 4. The fluid leakage monitoring and predicting method according to claim 3, wherein the classification model is a neural network model.
5. 5. The method of claim 4, wherein the classification model comprises a two-way long-short-term memory model including an input layer, a forward long-short-term memory layer, a reverse long-short-term memory layer, a full-connection layer and an output layer, wherein the input layer is used for inputting the time-frequency spectrogram, the forward long-short-term memory layer is used for processing forward sequence data in the time-frequency spectrogram, the reverse long-short-term memory layer is used for processing reverse sequence data in the time-frequency spectrogram, the full-connection layer is used for integrating the outputs of the forward long-short-term memory layer and the reverse long-short-term memory layer, and the output layer is used for outputting the corresponding judgment result of the leakage.
6. 6. The method of claim 5, further comprising determining whether the leakage determination result is consistent with an actual current result after the leakage determination result is obtained, and inputting the time-frequency spectrogram into a training dataset to retrain the classification model through the training dataset if the leakage determination result is not consistent with the actual current result.
7. 7. The leakage monitoring and prediction method according to claim 6, wherein the training data set comprises the steps of: Initializing the weight and deviation of the classification model; inputting training data into the classification model, and obtaining a prediction result through forward propagation; Calculating a loss function and evaluating a gap between the predicted result and the actual; Calculating a gradient of weights and biases using a back propagation algorithm and updating the weights and biases using an optimization algorithm to minimize a loss function, and Repeating the steps until the model converges or the set training times are reached.
8. 8. The leakage monitoring and prediction method according to claim 7, wherein the training of the classification model is completed by: Predicting the classification model using the test data and evaluating the performance of the model, and And adjusting the model architecture, super parameters or retraining the model according to the evaluation result.
9. 9. The leakage monitoring and prediction method of claim 3, wherein the classification model comprises a support vector machine classification model.
10. 10. The leakage monitoring and prediction method according to claim 9, wherein the step of establishing the classification model comprises the steps of: and providing a plurality of time-frequency spectrograms with and without liquid leakage respectively, sequentially carrying out image pyramid processing and feature extraction processing on the time-frequency spectrograms, and outputting the time-frequency spectrograms to the support vector machine classification model for classification.
11. 11. The method of claim 9, wherein inputting the time-frequency spectrum into the classification model to generate the leakage determination result of the pipeline comprises the steps of: image pyramid processing with different scales is carried out on the time-frequency spectrograms to respectively generate a plurality of audio region images with different scales; Calculating at least one characteristic value of each sub-block, and calculating texture correlation between each sub-block and a reference model associated with no leakage of liquid, and And generating a liquid leakage judging result of the pipeline according to the characteristic value and the texture correlation of each sub-block by using the support vector machine classification model.
12. 12. The leakage monitoring and prediction method according to claim 11, wherein establishing the reference model comprises the steps of: And averaging pixel values of the sub-spectrum spectrograms obtained after the image pyramid processing is carried out on the time-spectrum spectrograms without liquid leakage so as to obtain the reference model.
13. 13. The method of claim 11, wherein the at least one characteristic value of each of the sub-blocks comprises at least one of a standard deviation and a histogram bias of a plurality of pixel values of the sub-block.
14. 14. The method of claim 11, wherein the texture correlation between each sub-block and the reference model is a correlation coefficient of a local binary pattern between the sub-block and the reference model.
15. 15. A leakage monitoring and prediction system for an electronic device liquid cooling system, comprising: a receiving module for receiving the instant audio signal of the fluid flow in the pipeline, and The processor is coupled to the receiving module and configured to convert the real-time audio signal into a time-frequency spectrogram, and perform image recognition on the time-frequency spectrogram to determine whether the pipeline has liquid leakage.
16. 16. The system of claim 15, wherein the receiving module comprises a car audio bus and a plurality of micro-radio units configured in a daisy chain manner and electrically connected to the car audio bus.
17. 17. The leakage monitoring and prediction system according to claim 15, wherein the receiving module is disposed at one side of a fluid cooling circuit of the electronic device and adapted to continuously receive the real-time audio signal generated by the flow of the cooling fluid in the fluid cooling circuit.
18. 18. A leakage monitoring and prediction system for an electronic device liquid cooling system, comprising: a receiving module for receiving the instant audio signal of the fluid flow in the pipeline, and The processor is coupled to the receiving module and configured to perform fast Fourier transform on the real-time audio signal to generate a time-frequency spectrogram, and input the time-frequency spectrogram into the classification model to perform image recognition, thereby judging whether the pipeline has liquid leakage.
19. 19. The fluid leakage monitoring and predicting system according to claim 18, wherein the classification model comprises a two-way long-short-term memory model comprising an input layer, a forward long-short-term memory layer, a reverse long-short-term memory layer, a full connection layer and an output layer, wherein the input layer is used for inputting the time-frequency spectrogram, the forward long-short-term memory layer is used for processing forward sequence data in the time-frequency spectrogram, the reverse long-short-term memory layer is used for processing reverse sequence data in the time-frequency spectrogram, the full connection layer is used for integrating the outputs of the forward long-short-term memory layer and the reverse long-short-term memory layer, and the output layer is used for outputting a corresponding fluid leakage judging result.
20. 20. The fluid leakage monitoring and predicting system of claim 18, wherein the classification model comprises a support vector machine classification model, wherein the processor is configured to perform different-scale image pyramid processing on the time-frequency spectrograms to generate a plurality of different-scale audio region images, respectively, divide each of the different-scale audio region images to generate a plurality of sub-blocks, calculate at least one characteristic value of each of the sub-blocks, calculate a texture correlation between each of the sub-blocks and a reference model associated with no fluid leakage, and generate a fluid leakage judgment result of the pipeline according to the characteristic value and the texture correlation of each of the sub-blocks by using the support vector machine classification model.

Description

Liquid leakage monitoring and predicting method and system for liquid cooling system of electronic device Technical Field The invention relates to a liquid leakage monitoring and predicting method and a liquid leakage monitoring and predicting system for a liquid cooling system of an electronic device. Background With the advent of the high-power era of artificial intelligence (ARTIFICIAL INTELLIGENCE, AI), the demands for high-performance operation and high-frequency high-speed transmission are increasing, so that the power consumption of a server is continuously improved, and the heat dissipation technology is also driven to be upgraded. The current air cooling heat dissipation capacity is not used, so that the liquid cooling technology is used for jumping the table top. However, liquid cooling techniques still present a potential risk of leakage that can lead to electronic component failure problems. In the prior art, a liquid leakage detection cable (Leak Detection Cable) is mostly adopted for liquid leakage monitoring of a liquid cooling system to detect liquid, and the liquid leakage detection cable is generally deployed under an overhead floor of a data center or in a server. However, the leakage detection cable is large in size and high in cost, so that the leakage detection cable can be only deployed in important areas. Moreover, a trace of leakage cannot be detected, which is usually detected when the leakage has reached a considerable level, but the electronic device or circuit is often damaged. Disclosure of Invention In view of the above, the present invention provides a method for monitoring and predicting leakage of a liquid cooling system of an electronic device, which includes receiving an instant audio signal of fluid flowing in a pipeline, converting the instant audio signal into a time-frequency spectrogram, and performing image recognition on the time-frequency spectrogram to determine whether the pipeline has leakage. In some embodiments, the step of converting the audio signal into a time-frequency spectrogram comprises performing a fast Fourier transform (Fast Fourier Transform) on the real-time audio signal to generate the time-frequency spectrogram, wherein the time-frequency spectrogram is a three-dimensional spectrogram, and the three-dimensional spectrogram comprises time information, frequency information and sound intensity information. In some embodiments, determining whether the pipeline has a leakage according to the time-frequency spectrum chart includes inputting the time-frequency spectrum chart into a classification model to generate a leakage determination result of the pipeline. In some embodiments, the classification model is a neural network model. In some embodiments, the classification model includes a two-way long and short Memory (BLSTM) model including an input layer, a forward long and short Memory layer, a reverse long and short Memory layer, a full connection layer, and an output layer, wherein the input layer is used for inputting the time-frequency spectrogram, the forward long and short Memory layer is used for processing forward sequence data in the time-frequency spectrogram, the reverse long and short Memory layer is used for processing reverse sequence data in the time-frequency spectrogram, the full connection layer is used for integrating outputs of the forward long and short Memory layer and the reverse long and short Memory layer, and the output layer is used for outputting a corresponding leakage judgment result. In some embodiments, after the leakage judging result is obtained, the method further comprises judging whether the leakage judging result is consistent with an actual current situation result or not, and if the leakage judging result is not consistent with the actual current situation result, inputting the time-frequency spectrogram into a training data set so as to retrain the classification model through the training data set. In some embodiments, the training data set includes the steps of initializing weights and biases for the classification model, inputting a training data into the classification model and obtaining a predicted result by forward propagation, calculating a loss function and evaluating a gap between the predicted result and the actual, calculating gradients of weights and biases using a back propagation algorithm and updating the weights and biases using an optimization algorithm to minimize the loss function, and repeating the foregoing steps until the model converges or reaches a set number of exercises. In some embodiments, training the classification model includes predicting the classification model using a test data and evaluating the performance of the model, and adjusting the model architecture, hyper-parameters, or retraining the model based on the evaluation. In some embodiments, the classification model includes a support vector machine (Support Vector Machine) classification model. In some embodiments, t