Liquid leakage monitoring and predicting method and system for liquid cooling system of electronic device

Abstract

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

Description

Technical Field

[0001] This invention relates to a method and system for monitoring and predicting leakage in liquid cooling systems of electronic devices. Background Technology

[0002] With the advent of the high-computing era of Artificial Intelligence (AI), the demand for high-performance computing and high-frequency, high-speed transmission is increasing daily, leading to a continuous rise in server power consumption and driving upgrades in heat dissipation technology. Currently, air cooling is gradually becoming insufficient, making liquid cooling technology a necessity. However, liquid cooling technology still carries the potential risk of leakage, which can lead to electronic component malfunctions.

[0003] Current technologies for monitoring leaks in liquid cooling systems mostly employ leak detection cables. These cables are typically deployed under the raised floor of data centers or inside servers. However, due to their large size and high cost, leak detection cables can only be deployed in key areas. Furthermore, they cannot detect minute leaks; they are usually only detected when the leak has reached a considerable extent, by which time damage to electronic components or circuits has often already occurred. Summary of the Invention

[0004] In view of the above, the present invention provides a method for monitoring and predicting leakage in a liquid cooling system of an electronic device. The method includes: receiving an instantaneous audio signal of fluid flow in a pipeline; converting the instantaneous audio signal into a time spectrum; and performing image recognition on the time spectrum to determine whether there is leakage in the pipeline.

[0005] In some embodiments, the step of converting the audio signal into a time-spectrum includes: performing a Fast Fourier Transform on the real-time audio signal to generate the time-spectrum; the time-spectrum is a three-dimensional spectrum that includes time information, frequency information, and sound intensity information.

[0006] In some embodiments, the step of determining whether the pipeline is leaking based on the time-spectrum graph includes: inputting the time-spectrum graph into a classification model to generate a leak determination result for the pipeline. In some embodiments, the classification model is a neural network model.

[0007] In some embodiments, the classification model includes a Bidirectional Long Short-Term Memory (BLSTM) model, which includes an input layer, a forward LLSTM layer, a reverse LLSTM layer, a fully connected layer, and an output layer. The input layer is used to input the time-spectrum image; the forward LLSTM layer processes the forward sequence data in the time-spectrum image; the reverse LLSTM layer processes the reverse sequence data in the time-spectrum image; the fully connected layer integrates the outputs of the forward LLSTM layer and the reverse LLSTM layer; and the output layer outputs the corresponding leakage judgment result.

[0008] In some embodiments, after obtaining the leakage judgment result, the method further includes: determining whether the leakage judgment result matches an actual situation result; and if the leakage judgment result does not match the actual situation result, inputting the time spectrum into a training dataset to retrain the classification model using the training dataset.

[0009] In some embodiments, the training dataset includes the following steps: initializing the weights and biases 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 the gap between the prediction result and the actual result; calculating the gradients of the weights and biases using a backpropagation algorithm and using an optimization algorithm to update the weights and biases to minimize the loss function; and repeating the aforementioned steps until the model converges or reaches a set number of training iterations.

[0010] In some embodiments, after training the classification model, the steps include: using test data to make predictions and evaluating the model's performance; and adjusting the model architecture, hyperparameters, or retraining the model based on the evaluation results.

[0011] In some embodiments, the classification model includes a Support Vector Machine (SVM) classification model.

[0012] In some embodiments, establishing the classification model includes the following steps: providing multiple time-spectrum images with and without leakage, and sequentially performing image pyramid processing and feature extraction processing on these time-spectrum images before outputting them to the support vector machine classification model for classification.

[0013] In some embodiments, inputting the time-spectrum image into the classification model to generate a leakage judgment result for the pipeline includes the following steps: performing image pyramid processing on the time-spectrum images at different scales to generate multiple audio region images at different scales; segmenting each of the different scale audio region images to generate multiple sub-blocks; calculating at least one feature value for each of the sub-blocks, and calculating the texture correlation between each sub-block and a reference model associated with no leakage; and using the support vector machine classification model, generating a leakage judgment result for the pipeline based on the feature value and the texture correlation of each sub-block.

[0014] In some embodiments, establishing the reference model includes the following steps: averaging the pixel values ​​of the sub-time spectrum obtained by performing the image pyramid processing on multiple time spectrum maps without leakage to obtain the reference model.

[0015] In some embodiments, at least one feature value of each of the sub-blocks includes at least one of the standard deviation of a plurality of pixel values ​​of the sub-block and histogram skewness.

[0016] In some embodiments, the texture correlation between each sub-block and the reference model is the correlation coefficient of the local binary pattern (LBP) between the sub-block and the reference model.

[0017] In view of the foregoing, the present invention provides a leakage monitoring and prediction system for a liquid cooling system of an electronic device, comprising: a receiving module for receiving a real-time audio signal of fluid flow in a pipeline; and a processor coupled to the receiving module and configured to convert the real-time audio signal into a time spectrum; and to perform image recognition on the time spectrum to determine whether there is a leakage in the pipeline.

[0018] In some embodiments, the receiving module includes an automotive audio bus (A2B) and multiple miniature radio units configured in a daisy chain and electrically connected to the automotive audio bus.

[0019] In some embodiments, the receiving module is configured on one side of a fluid cooling line of an electronic device and is adapted to continuously receive the real-time audio signal generated by the flow of cooling fluid within the line.

[0020] In some embodiments, the classification model includes a bidirectional long short-term memory model, which includes an input layer, a forward long short-term memory layer, a reverse long short-term memory layer, a fully connected layer, and an output layer; wherein, the input layer is used to input the time spectrum, the forward long short-term memory layer is used to process the forward sequence data in the time spectrum, the reverse long short-term memory layer is used to process the reverse sequence data in the time spectrum, the fully connected layer is used to integrate the outputs of the forward long short-term memory layer and the reverse long short-term memory layer, and the output layer is used to output the corresponding leakage judgment result.

[0021] In some embodiments, the classification model includes a support vector machine classification model; the processor is configured to: perform image pyramid processing on the time-spectrum maps at different scales to generate multiple audio region images at different scales; segment each of the different scale audio region images to generate multiple sub-blocks; calculate at least one feature value for each of the sub-blocks, and calculate the texture correlation between each sub-block and a reference model associated with no leakage; and use the support vector machine classification model to generate a leakage judgment result for the pipeline based on the feature value and the texture correlation of each sub-block.

[0022] In view of the foregoing, the present invention provides a leakage monitoring and prediction system for a liquid cooling system of an electronic device, comprising: a receiving module for receiving a real-time audio signal of fluid flow in a pipeline; and a processor coupled to the receiving module and configured to perform a fast Fourier transform on the real-time audio signal to generate a time spectrum; and inputting the time spectrum into a classification model to perform image recognition, thereby determining whether there is a leakage in the pipeline. Attached Figure Description

[0023] Figure 1 This is a system block diagram of a leakage monitoring and prediction system according to some embodiments of the present invention;

[0024] Figure 2 These are schematic diagrams of leakage monitoring and prediction systems according to some embodiments of the present invention;

[0025] Figure 3A This is a schematic diagram of the time spectrum when no leakage occurs;

[0026] Figure 3B This is a time-frequency spectrum diagram showing the occurrence of continuous leakage;

[0027] Figure 4 This is a main flowchart of the leakage monitoring and prediction method according to some embodiments of the present invention;

[0028] Figure 5 This is a schematic diagram of the classification model for establishing leakage monitoring and prediction methods in some embodiments of the present invention;

[0029] Figure 6This is a flowchart of the leakage judgment process of the leakage monitoring and prediction method according to some embodiments of the present invention;

[0030] Figure 7 This is a flowchart of the training classification model of the leakage monitoring and prediction method in some embodiments of the present invention;

[0031] Figure 8 This is an architecture diagram of the bidirectional long short-term memory model in the leakage monitoring and prediction method of some embodiments of the present invention.

[0032] Symbol Explanation

[0033] 1: Leakage Monitoring and Prediction System

[0034] 2: Receiver module

[0035] 3: Processor

[0036] 4: Storage Module

[0037] 21: Vehicle audio bus

[0038] 22: Miniature radio unit

[0039] 41: Database

[0040] 42: Classification Model

[0041] 421: Input Layer

[0042] 422: Forward Long Short-Term Memory Layer

[0043] 423: Reverse Long Short-Term Memory Layer

[0044] 424: Fully Connected Layer

[0045] 425: Output Layer

[0046] CDU: Coolant Monitoring Unit

[0047] D: Electronic devices

[0048] D1: Water-cooled plate

[0049] N: Real-time audio signal

[0050] P: Piping

[0051] SVM: Support Vector Machine classification model

[0052] S405, S410, S415, Sa, Sb, Sc, S605, S610, S615, S705, S710, S715, S720, S725, S730: Steps Detailed Implementation

[0053] Various embodiments are described in detail below. These embodiments are merely illustrative and do not limit the scope of protection intended by the present invention. Furthermore, some elements are omitted in the accompanying drawings of the embodiments to clearly show the technical features of the present invention. Moreover, the same reference numerals will be used to denote the same or similar elements in all the drawings, and the drawings of the present invention are for illustrative purposes only, are not necessarily drawn to scale, and not all details may be shown in the drawings.

[0054] Please also refer to Figure 1 and Figure 2 , Figure 1 This is a system block diagram of the leakage monitoring and prediction system 1 according to some embodiments of the present invention. Figure 2 This is a schematic diagram of a leakage monitoring and prediction system 1 according to some embodiments of the present invention. As shown in the figure, the leakage monitoring and prediction system 1 includes a receiving module 2, a processor 3, and a storage module 4. The processor 3 is electrically connected to the receiving module 2 and the storage module 4. The receiving module 2 is used to receive a real-time audio signal N of fluid flow in a pipe P, and the processor 3 is configured to convert the real-time audio signal N into a time-frequency spectrum; and to perform image recognition on the time-frequency spectrum to determine whether there is a leakage in the pipe P.

[0055] In one embodiment, the leakage monitoring and prediction system 1 can be housed within an electronic device D, which may be a web server, AI server, data center, switch, high-performance computing (HPC) machine, or other electronic devices that generate high heat. The electronic device D is equipped with multiple pipes P and multiple water-cooled plates D1, wherein the water-cooled plates D1 are mounted on the heat-generating chips or electronic components, and the pipes P connect the water-cooled plates D1 and the coolant monitoring host CDU. The coolant monitoring host CDU is primarily responsible for supplying low-temperature coolant to the water-cooled plates D1. The coolant within the water-cooled plates D1 absorbs the heat generated by the chips or electronic components and then flows back to the coolant monitoring host CDU for heat exchange to lower the coolant temperature, thus continuously circulating.

[0056] In some embodiments, the receiving module 2 is configured on one side of the fluid cooling pipe P of the electronic device D and continuously receives the real-time audio signal N generated by the flow of cooling fluid within the pipe P. Furthermore, the receiving module 2 may include an in-vehicle audio bus 21 and a plurality of miniature microphone units 22. These miniature microphone units 22 may be microphones of a microelectromechanical system (MEMS) and are daisy-chained and electrically connected to the in-vehicle audio bus 21. Accordingly, by daisy-chaining the miniature microphone units 22, each daisy-chain unit includes four miniature microphone units 22, thereby improving the accuracy of interpretation and making it easier to identify leakage points. In addition, the use of the in-vehicle audio bus 21 improves scalability, allowing for the continuous extension and connection of miniature microphone units 22 as needed, and also possesses the characteristic of fixed delay for each audio node, reducing errors caused by inconsistent signal delays.

[0057] In some embodiments, the processor 3 is, for example, a central processing unit (CPU), a graphics processing unit (GPU), or other programmable microprocessor, digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), programmable logic device (PLD), or other similar device.

[0058] In some embodiments, the storage module 4 is, for example, any type of fixed or removable random access memory, read-only memory, flash memory, hard disk, or other similar device or combination thereof. The storage module 4 includes a database 41 and a classification model 42. The database 41 stores a training dataset. The training dataset includes multiple known (as a result of actual conditions) audio data with and without leakage, which are collected beforehand and can be used to build and train the classification model 42. Here, the classification model 42 is, for example, a neural network (NN) model with multiple layers, trained using deep learning. The concept of deep learning is to use a large amount of known data to inform the neural network model of the relationship between input and output, thereby adjusting parameters such as weights and biases in the neural network model.

[0059] Please refer to the following: Figure 3A , Figure 3B and Figure 4 , Figure 3AThis is a schematic diagram of the time spectrum when no leakage occurs. Figure 3B This is a time-frequency spectrum diagram showing the occurrence of continuous leakage. Figure 4 This is a main flowchart of the leakage monitoring and prediction method according to some embodiments of the present invention. According to some embodiments, after the receiving module 2 receives the real-time audio signal N of the fluid flow in the pipeline P, i.e., step S405; the processor 3 converts the real-time audio signal N into a time spectrum, i.e., step S410; and performs image recognition on the time spectrum, i.e., step S415, to determine whether there is a leakage in the pipeline P.

[0060] like Figure 3A and Figure 3B The time-frequency spectra shown represent the states with and without leakage, respectively; among them, the... Figure 3A It is evident that there are only multiple independent surges, representing ambient audio, such as noise emitted by parts due to impact or vibration. Additionally, in Figure 3B In addition to multiple independent surges, the leak also includes multiple continuous audio signals, which are special audio signals generated by continuous leakage in pipe P. Generally speaking, leaks in metal water pipes transmit sound at higher frequencies, with an audio range between 500Hz and 1500Hz; while leaks in plastic or PVC water pipes transmit sound at lower frequencies, with an audio range typically between 70Hz and 850Hz.

[0061] In step S410, the processor 3 performs a Fast Fourier Transform (FFT) on the instantaneous audio signal N to generate the time-frequency spectrum. However, in some embodiments, the time-frequency spectrum can be a three-dimensional spectrum, which includes time information, frequency information, and sound intensity information. In other embodiments, the time-frequency spectrum can be a two-dimensional spectrum. The purpose of converting the instantaneous audio signal N into a time-frequency spectrum through FFT is that, after converting the time-domain signal into a time-frequency spectrum, the audio of a leak will exhibit temporal continuity and energy clustering in the time-frequency spectrum, and subtle features can be preserved. The subtle features of the sound will not be lost during the conversion process, which is beneficial for subsequent computer vision technology to determine whether a leak has occurred.

[0062] In step S415, the processor 3 inputs the converted time-frequency spectrum to the classification model 42 to generate a leakage judgment result for the pipeline P. In some embodiments, the classification model 42 may be a support vector machine (SVM) classification model. The classification model 42 may be trained by the processor 3 itself or obtained from other processing devices; this invention is not limited thereto.

[0063] Please refer to the following: Figure 5This is a schematic diagram of the classification model 42 established according to some embodiments of the leakage monitoring and prediction method of the present invention. In some embodiments, establishing the classification model 42 includes the following steps: First, multiple time-spectrum maps with and without leakage are provided respectively. These time-spectrum maps are time-spectrum maps with known leakage results, including states with clearly defined features and states with unclear features. Next, the processor 3 sequentially performs image pyramid processing on these time-spectrum maps to generate images of different scales, i.e. Figure 5 Step Sa in the process. Furthermore, processor 3 performs feature extraction processing on the image generated in the previous step, that is... Figure 5 Step Sb in the process. Finally, processor 3 outputs the image to the support vector machine classification model (SVM) for classification, that is... Figure 5 The steps in Sc are followed, and a classification model 42 is established accordingly.

[0064] More specifically, in some embodiments, an artificial intelligence (AI) classification model 42 is used to analyze the data. This mainly involves collecting all sounds and building a training database, which contains a large amount of normal, leak-free audio data, as well as filtered audio data of various abnormal states, including modalities containing various occasional environmental noises and manually confirmed leaks. Once the classification model 42 has learned from sufficient samples, it can make distinctions from complex noise.

[0065] Furthermore, in some embodiments, the sound data is analyzed and recorded using a three-dimensional spectrogram (including time, frequency, and volume). Since leakage may be a very quiet but continuous audio mode, which may also contain a lot of noise from the venue equipment, it is necessary to first manually select and classify such modes as training samples for abnormal states, and then supplement them with a large number of normal samples and non-leakage abnormal samples. After machine learning, the classification model 42 can distinguish leakage.

[0066] Please refer to the following: Figure 6 This is a flowchart illustrating the leakage judgment process of the leakage monitoring and prediction method according to some embodiments of the present invention. The following describes the specific steps in one embodiment of inputting a time-spectrum image into the classification model 42 to generate a leakage judgment result for the pipeline P. First, the processor 3 performs image pyramid processing at different scales on the time-spectrum images, thereby generating multiple audio region images at different scales; then, the processor 3 segments each of the different scale audio region images to generate multiple sub-blocks, which are of the same size, i.e. Figure 6 Step S605 in the process.

[0067] Next, in step S610, the processor 3 calculates at least one feature value for each of the sub-blocks and calculates the texture correlation between each sub-block and the reference model associated with the leak-free area. In some embodiments, establishing the reference model includes the following steps: averaging the pixel values ​​of the sub-time spectrum maps obtained after performing the image pyramid processing on multiple time spectrum maps of the leak-free area to obtain the reference model; wherein the size of each sub-time spectrum map is the same.

[0068] In some embodiments, the processor 3 may, for example, calculate the standard deviation or histogram skewness of the pixel values ​​of each sub-block as a feature value for each sub-block. Of course, in other embodiments, both the standard deviation and histogram skewness can be used together to calculate the feature value of each sub-block, thus obtaining more accurate results. Furthermore, in some embodiments, the texture correlation between each sub-block and the reference model may be the correlation coefficient of the local binary patterns between the sub-block and the reference model.

[0069] Finally, in step S615, the Support Vector Machine (SVM) classification model is used to generate a leakage judgment result for the pipeline P based on the feature value and texture correlation of each sub-block. That is, the processor 3 then inputs the feature value and texture correlation corresponding to each sub-block into the classification model 42 (SVM classification model) to generate an output result, which will indicate whether leakage has occurred.

[0070] In some embodiments, such as during the initial setup of the system or the tuning of modules, it is necessary to further verify the leakage judgment results of the system. The following method for verifying the leakage judgment results includes the following steps: after obtaining the leakage judgment result, manually determining whether the leakage judgment result matches an actual current situation; and if the leakage judgment result does not match the actual current situation, the processor 3 inputs the time-frequency spectrum into a training dataset to retrain the classification model 42 using the training dataset. The actual current situation result can be the result of whether or not there is a leakage, obtained through human observation, manual detection, or other leakage detection systems. In some embodiments, the training dataset is stored in the database 41 of the storage module 4.

[0071] Please see Figure 7This is a flowchart of the training classification model 42 of the leakage monitoring and prediction method according to some embodiments of the present invention. The following provides a method for refining and tuning the classification model 42, applicable when the leakage judgment result is incorrect, allowing for retraining of the classification model 42. First, in step S705, the weights and biases of the classification model 42 are initialized. Next, in step S710, training data is input into the classification model 42, and a prediction result is obtained through forward propagation. Then, in step S715, the loss function is calculated and the gap between the prediction result and the actual result is evaluated. Next, in step S720, the gradients of the weights and biases are calculated using the backpropagation algorithm, and an optimization algorithm is used to update the weights and biases to minimize the loss function. Accordingly, steps S710 to S720 are repeated until the model converges or reaches the set number of training iterations (step S725), thus completing the tuning and training process of the classification model 42, i.e., step S730.

[0072] In addition, in some embodiments, after training the classification model 42, the following steps can be performed for minor adjustments: first, the classification model 42 is used to make predictions and evaluate its performance using test data; then, the model architecture, hyperparameters, or the model is retrained based on the evaluation results. These steps are all aimed at continuously refining the classification model 42 to improve the accuracy and efficiency of judgments.

[0073] Please see Figure 8 This is an architectural diagram of the bidirectional long short-term memory model in the leakage monitoring and prediction method according to some embodiments of the present invention. In other embodiments, the classification model 42 may also adopt a bidirectional long short-term memory model, which, in addition to determining whether leakage has occurred, can further analyze the state, scale, and location of leakage in other embodiments.

[0074] In some embodiments, the bidirectional long short-term memory model includes an input layer 421, a forward long short-term memory layer 422, a reverse long short-term memory layer 423, a fully connected layer 424, and an output layer 425. The input layer 421 is used to input the time-spectrum graph; the forward long short-term memory layer 422 is used to process the forward sequence data in the time-spectrum graph; the reverse long short-term memory layer 423 is used to process the reverse sequence data in the time-spectrum graph; the fully connected layer 424 is used to integrate the outputs of the forward long short-term memory layer 422 and the reverse long short-term memory layer 423; and the output layer 425 is used to output the corresponding leakage judgment result.

[0075] To further explain, the time-spectrum graph input to the input layer 421 can be divided into multiple sub-time-spectrum graphs according to the time sequence of the time-spectrum graph (from low frequency to high frequency). Next, the forward long short-term memory layer 422 performs forward sequence data calculation on the sub-time-spectrum graph to obtain feature data. Furthermore, the reverse long short-term memory layer 423 performs reverse sequence data calculation on the sub-time-spectrum graph to obtain another set of feature data. Then, the fully connected layer 424 integrates the feature data output by the forward long short-term memory layer 422 and the reverse long short-term memory layer 423 respectively, and obtains a probability after weight calculation. Finally, the output layer 425 outputs the corresponding leakage judgment result based on the probability. In other embodiments, the fully connected layer 424 further transmits information such as the leakage location and leakage scale to the output layer 425, and the output layer 425, in addition to outputting the leakage judgment result, also simultaneously outputs information such as the leakage location and leakage scale.

[0076] In summary, while most sensitive sound recording systems capture all ambient noise, including air conditioning, occasional events, cooling fans, pumps, and rack resonance, the noise from a server coolant pipe leak is very low and generally difficult to detect. Therefore, in some embodiments, an AI classification model 42 based on machine deep learning is used to identify the leak noise from the mixed noise.

[0077] Furthermore, in some embodiments, the leakage monitoring and prediction system 1 collects sound signals in real time and inputs them into the classification model 42 for analysis. When an abnormal sound (such as coolant leakage) is detected, the system immediately issues an alarm and records the location, scale, and time of the leak, facilitating timely handling by maintenance personnel. In addition, through an integrated data management module, the system can store and analyze historical sound data and generate reports to help managers understand the equipment's operating status and perform preventative maintenance.

[0078] Furthermore, in some embodiments, the training dataset may include pre-collected audio data from multiple known (as a result of actual conditions) early minor leaks or impending leaks in pipe P or its joints, used to build and train the classification model 42. Moreover, in some embodiments, since these miniature microphone units 22 are microelectromechanical systems (MEMS) microphones with actual dimensions of approximately 2mm to 3mm, they can sense extremely small vibrations (acceleration) and acoustic fluctuations, making them suitable for sensing early minor leaks or impending leaks in pipe P or its joints. Thus, in addition to real-time monitoring and detection of leaks, leak prediction and prevention can also be achieved. That is, when early signs of leaks occur, such as changes in the fluid flow path or flow pattern within pipe P or its joints, the system will immediately issue an alarm, enabling leak prediction and prevention.

[0079] Although the present invention has been disclosed above by way of embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the appended claims.

Claims

1. A method for monitoring and predicting leakage in a liquid cooling system of an electronic device, the method comprising: Receive real-time audio signals of fluid flow within the pipeline; Convert the real-time audio signal into a time-spectrum graph; as well as Image recognition is performed on the time spectrum to determine if there is a leak in the pipeline.

2. The leakage monitoring and prediction method as described in claim 1, wherein the step of converting the audio signal into the time-spectrum graph includes: A fast Fourier transform is performed on the real-time audio signal to generate the time-frequency spectrum; the time-frequency spectrum is a three-dimensional spectrum, which includes time information, frequency information and sound intensity information.

3. The leakage monitoring and prediction method as described in claim 1, wherein the step of determining whether there is a leakage in the pipeline based on the time-frequency spectrum includes: The time spectrum is input into the classification model to generate a leak detection result for the pipeline.

4. The leakage monitoring and prediction method as described in claim 3, wherein the classification model is a neural network model.

5. The leakage monitoring and prediction method as described in claim 4, wherein the classification model includes a bidirectional long short-term memory model, which includes an input layer, a forward long short-term memory layer, a reverse long short-term memory layer, a fully connected layer, and an output layer; wherein, The input layer is used to input the time spectrum, the forward long short-term memory layer is used to process the forward sequence data in the time spectrum, the reverse long short-term memory layer is used to process the reverse sequence data in the time spectrum, the fully connected layer is used to integrate the outputs of the forward long short-term memory layer and the reverse long short-term memory layer, and the output layer is used to output the corresponding leakage judgment result.

6. The leakage monitoring and prediction method as described in claim 5, wherein after obtaining the leakage judgment result, it further includes: Determine whether the leak assessment result matches the actual situation. If the leakage judgment result does not match the actual situation, the time spectrum is input into the training dataset to retrain the classification model.

7. The leakage monitoring and prediction method as described in claim 6, wherein the training dataset includes the following steps: Initialize the weights and biases of the classification model; The training data is input into the classification model, and the prediction results are obtained through forward propagation; Calculate the loss function and evaluate the gap between the prediction and the actual result; The gradients of the weights and biases are computed using the backpropagation algorithm, and an optimization algorithm is used to update the weights and biases to minimize the loss function; and Repeat the aforementioned steps until the model converges or the set number of training iterations is reached.

8. The leakage monitoring and prediction method as described in claim 7, wherein after training the classification model, the following steps are included: Use test data to make predictions and evaluate the model's performance; as well as Adjust the model architecture, hyperparameters, or retrain the model based on the evaluation results.

9. The leakage monitoring and prediction method as described in claim 3, wherein the classification model includes a support vector machine classification model.

10. The leakage monitoring and prediction method as described in claim 9, wherein establishing the classification model includes the following steps: Multiple time-spectrum images with and without leakage are provided respectively. These time-spectrum images are then processed sequentially using image pyramid processing and feature extraction before being output to the support vector machine classification model for classification.

11. The leakage monitoring and prediction method as described in claim 9, wherein inputting the time-frequency spectrum into the classification model to generate the leakage judgment result of the pipeline includes the following steps: Different scales of image pyramid processing are performed on these time-spectrum images to generate multiple audio region images of different scales; and each of these audio region images of different scales is segmented to generate multiple sub-blocks. Calculate at least one feature value for each of the sub-blocks, and calculate the texture correlation between each of the sub-blocks and the reference model associated with the leak-free liquid; as well as Using the support vector machine classification model, based on the feature value of each sub-block and the texture correlation, the leakage judgment result of the pipeline is generated.

12. The leakage monitoring and prediction method as described in claim 11, wherein establishing the reference model includes the following steps: The reference model is obtained by averaging the pixel values ​​of the sub-time spectrum obtained after performing image pyramid processing on the time spectrum maps of those without leakage.

13. The leakage monitoring and prediction method of claim 11, wherein the at least one feature value of each of the sub-blocks includes at least one of the standard deviation of a plurality of pixel values ​​of the sub-block and histogram skewness.

14. The leakage monitoring and prediction method as described in claim 11, wherein the texture correlation between each sub-block and the reference model is the correlation coefficient of the local binary pattern between the sub-block and the reference model.

15. A leakage monitoring and prediction system for a liquid cooling system in an electronic device, comprising: The receiving module is used to receive real-time audio signals of fluid flow within the pipeline. as well as The processor, coupled to the receiving module, is configured to convert the real-time audio signal into a time-spectrum graph; and to perform image recognition on the time-spectrum graph to determine whether there is a leak in the pipeline.

16. The leakage monitoring and prediction system of claim 15, wherein the receiving module includes an in-vehicle audio bus and a plurality of miniature radio units, the miniature radio units being configured in a daisy-chain manner and electrically connected to the in-vehicle audio bus.

17. The leakage monitoring and prediction system of claim 15, wherein the receiving module is configured on one side of the fluid cooling pipeline of the electronic device and is adapted to continuously receive the real-time audio signal generated by the flow of cooling fluid in the fluid cooling pipeline.

18. A leakage monitoring and prediction system for a liquid cooling system in an electronic device, comprising: The receiving module is used to receive real-time audio signals of fluid flow within the pipeline. as well as The processor is coupled to the receiving module and configured to perform a fast Fourier transform on the real-time audio signal to generate a time-spectrum; and to input the time-spectrum into a classification model to perform image recognition, thereby determining whether there is a leak in the pipeline.

19. The leakage monitoring and prediction system as described in claim 18, wherein, The classification model includes a bidirectional long short-term memory model, which comprises an input layer, a forward long short-term memory layer, a reverse long short-term memory layer, a fully connected layer, and an output layer. The input layer is used to input the time-spectrum graph, the forward long short-term memory layer is used to process the forward sequence data in the time-spectrum graph, the reverse long short-term memory layer is used to process the reverse sequence data in the time-spectrum graph, the fully connected layer is used to integrate the outputs of the forward long short-term memory layer and the reverse long short-term memory layer, and the output layer is used to output the corresponding leakage judgment result.

20. The leakage monitoring and prediction system of claim 18, wherein the classification model includes a support vector machine classification model; the processor is configured to: perform image pyramid processing on the time-spectrum images at different scales to generate multiple audio region images at different scales; segment each of the audio region images at different scales to generate multiple sub-blocks; calculate at least one feature value for each of the sub-blocks, and calculate the texture correlation between each sub-block and a reference model associated with no leakage; and use the support vector machine classification model to generate a leakage judgment result for the pipeline based on the feature value and the texture correlation of each sub-block.

Images