A method for diagnosing operating conditions of an SVG water-cooling heat dissipation system

Abstract

This invention belongs to the field of reactive power compensation in power systems and relates to a method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system. The method includes: collecting multi-source operating parameters of the SVG water-cooled heat dissipation system; preprocessing the collected multi-source operating parameters to obtain preprocessed data; performing feature processing on the preprocessed data to extract operating condition feature vectors reflecting the operating conditions; the operating condition feature vectors include turbulence state features and pressure fluctuation statistical features; inputting the operating condition feature vectors into a fault diagnosis model to obtain diagnostic results; the fault diagnosis model is a two-layer architecture model integrating support vector machines and Bayesian networks; the diagnostic results are used to indicate the operating condition category and / or fault type of the water-cooled heat dissipation system. Through multi-source data fusion and intelligent algorithms, early abnormal characteristics of the water-cooled system can be accurately captured, improving diagnostic accuracy and robustness, extending equipment maintenance cycles, and reducing operation and maintenance costs. This method can be widely applied to the health management of SVG equipment in power systems.

Description

Technical Field

[0001] This invention relates to the field of reactive power compensation in power systems, and specifically discloses a method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system. Background Technology

[0002] Static var generators (SVGs), as core reactive power compensation devices in modern power systems, are crucial for the power quality and reliability of the grid. The power modules (such as IGBTs) in SVGs generate a significant amount of heat during operation, requiring efficient water-cooling systems to ensure the junction temperatures of the power devices remain within safe ranges. However, water-cooling systems operate under complex conditions and are susceptible to fluctuations in ambient temperature, load changes, and equipment aging. This can lead to faults such as abnormal coolant temperature, flow fluctuations, and pressure exceeding limits, resulting in increased IGBT junction temperatures, increased output harmonic distortion, and increased power losses. In severe cases, this can cause system shutdowns or even equipment damage.

[0003] Currently, monitoring the operating conditions of water-cooled heat dissipation systems mostly relies on single-sensor threshold judgment methods, such as triggering alarms by setting upper and lower limits for temperature, pressure, and flow. These methods struggle to capture the dynamic correlations and potential coupling effects between multiple parameters, and are insensitive to early anomalies under complex operating conditions, easily leading to missed or false alarms. Furthermore, some studies have attempted to introduce intelligent algorithms such as neural networks for fault diagnosis, but these often depend on a large number of labeled samples, and the models lack generalization ability, making it difficult to adapt to changes in actual operating conditions.

[0004] Therefore, there is an urgent need for a method for SVG water-cooled heat dissipation system operation condition diagnosis that can integrate multi-source sensor data, mine dynamic characteristics of operating conditions, and achieve high-precision real-time diagnosis, so as to improve equipment operation reliability and maintenance efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system, addressing the problems of low diagnostic accuracy and difficulty in capturing abnormal features under complex operating conditions in existing technologies; the specific solution is as follows:

[0006] A method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system includes: collecting multi-source operating parameters of the SVG water-cooled heat dissipation system; the multi-source operating parameters include at least coolant temperature, coolant flow rate, and loop pressure; preprocessing the collected multi-source operating parameters to obtain preprocessed data; performing feature processing on the preprocessed data to extract operating condition feature vectors reflecting the operating conditions; the operating condition feature vectors include turbulence state features and pressure fluctuation statistical features; inputting the operating condition feature vectors into a fault diagnosis model to obtain a diagnosis result; the fault diagnosis model is a two-layer architecture model integrating support vector machines and Bayesian networks; the diagnosis result is used to indicate the operating condition category and / or fault type of the water-cooled heat dissipation system.

[0007] Furthermore, the preprocessing includes time synchronization and filtering preprocessing; the time synchronization adopts an FPGA-based synchronization triggering mechanism; the filtering preprocessing adopts sliding window mean filtering.

[0008] Furthermore, feature processing is performed on the preprocessed data to extract operating condition feature vectors that reflect the operating conditions. This includes: performing wavelet packet decomposition on the preprocessed coolant flow rate to extract the energy proportion and energy entropy of each frequency band as turbulent state features; calculating the peak-to-peak value and standard deviation of the preprocessed loop pressure within a sliding time window to construct pressure fluctuation statistical features; and concatenating the turbulent state features and pressure fluctuation statistical features to obtain the operating condition feature vector.

[0009] Furthermore, the wavelet packet decomposition has 8 layers, the decomposition frequency band is 0-500Hz, and the energy entropy value of the 20-100Hz frequency band is extracted to characterize the blockage risk of the heat dissipation substrate flow channel.

[0010] Furthermore, the length of the sliding time window is 5 minutes. When the peak-to-peak value of the loop pressure is higher than 0.25 MPa for 5 consecutive minutes and the standard deviation increases by more than 30%, the probability of bearing wear of the circulating pump group is determined to rise to 82%.

[0011] Furthermore, the operating condition feature vector is input into the fault diagnosis model to obtain the diagnosis results, including: the first layer uses a support vector machine to perform preliminary classification of the operating condition feature vector and outputs the probability of each type of fault; the second layer uses the output of the support vector machine as the input of a Bayesian network, and calculates the joint probability based on the prior probability of historical fault samples; it determines whether the joint probability exceeds a set threshold, and triggers a fault warning and outputs the diagnosis results when the joint probability exceeds the set threshold; when the joint probability is less than the set threshold, no action is taken.

[0012] Furthermore, the kernel function parameters of the support vector machine are optimized through grid search.

[0013] Furthermore, the Bayesian network constructs a conditional probability table based on the prior probabilities of historical fault samples, and calculates the joint probability based on the conditional probability table.

[0014] Furthermore, this also includes: utilizing online learning mechanisms to periodically update the fault diagnosis model using newly added operating condition data.

[0015] Furthermore, the diagnostic results include pump jamming, heat sink leakage, heat exchanger scaling, and / or bearing wear.

[0016] The present invention has the following advantages and beneficial effects:

[0017] This invention integrates multi-source data, including temperature, flow rate, and pressure, to uncover the dynamic correlation of operating conditions, overcoming the limitations of single threshold judgment.

[0018] This invention uses wavelet packet decomposition to extract the frequency domain energy entropy of the flow signal, which can sensitively capture changes in turbulent state and provide early warning of flow channel blockage.

[0019] This invention combines a two-layer diagnostic architecture of SVM and Bayesian network, which leverages the advantages of SVM in small sample classification and utilizes Bayesian network to handle uncertainty and prior knowledge, thereby improving diagnostic accuracy and robustness.

[0020] The model of this invention has online learning capabilities, can adapt to changes in working conditions, and continuously improve detection sensitivity. Attached Figure Description

[0021] Figure 1 An exemplary flowchart of a method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] like Figure 1 As shown, the present invention provides a method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system, comprising the following steps:

[0024] Step 1: Collect multi-source operating parameters of the SVG water cooling system.

[0025] Temperature sensors (accuracy ±0.5℃), pressure sensors (accuracy ±0.2%FS), and flow meters (accuracy ±1.5% of reading) are installed at the outlet of the circulating pump, the inlet of the heat sink, and the outlet of the plate heat exchanger in the SVG water-cooled heat dissipation system, respectively. The temperature sensor sampling rate is 50Hz, the pressure sensor sampling rate is 100Hz, and the flow meter sampling rate is 20Hz. All sensor signals are connected to an FPGA-based data acquisition card, employing a synchronous triggering mechanism to ensure that the timestamp deviation of multi-channel signals is less than 10ms, avoiding distortion of the flow-temperature correlation due to timing misalignment.

[0026] Step 2: Preprocess the collected multi-source operating parameters to obtain preprocessed data.

[0027] The collected raw data is first subjected to sliding window mean filtering with a window length of 500 samples to remove high-frequency noise while retaining useful information such as changes in flow rate. The filtered data is then downsampled to 200Hz to meet the computational requirements of subsequent processing. The effective utilization rate of the preprocessed data exceeds 98%.

[0028] Step 3: Perform feature processing on the preprocessed data to extract the operating condition feature vectors that reflect the operating conditions.

[0029] The preprocessed coolant flow signal was decomposed into eight layers of wavelet packets using a time-frequency analysis algorithm. The energy percentage of each of the eight sub-bands in the 0-500Hz frequency band was extracted as the feature vector of the turbulent state, and the energy entropy in the 20-100Hz frequency band was also extracted. When the energy entropy value exceeds 0.7, it indicates that there is a risk of local blockage in the heat sink substrate flow channel.

[0030] For the preprocessed pressure signal, the peak-to-peak value and standard deviation of the pressure fluctuation within a 5-minute sliding time window are calculated. When the peak-to-peak value is above 0.25 MPa for 5 consecutive minutes and the standard deviation increases by more than 30%, the probability of bearing wear in the circulating pump unit is determined to have increased to 82%.

[0031] The original 12-dimensional features, including flow energy entropy, pressure peak-to-peak value, pressure standard deviation, and the mean and variance of temperature signals, were reduced to 3-dimensional principal components by principal component analysis (PCA). The cumulative variance contribution rate exceeded 86%, which was used as the final feature vector.

[0032] Step 4: Input the operating condition feature vector into the fault diagnosis model to obtain the diagnosis results.

[0033] A two-layer diagnostic model combining Support Vector Machine (SVM) and Bayesian network was constructed. The first layer uses an SVM classifier with a radial basis function (RBF) kernel. The kernel parameters were optimized through grid search, achieving a classification accuracy of 92.4% in cross-validation. The SVM performs preliminary classification on the feature vectors, outputting the probability of belonging to five states: pump jamming, substrate leakage, heat exchanger fouling, sensor drift, and normal operation.

[0034] The second layer is a Bayesian network, with nodes including various probabilities output by the SVM and measured values ​​of temperature, flow rate, and pressure. The network's conditional probability table is constructed based on 300 historical fault samples, and the parameters are learned using maximum likelihood estimation. When a feature vector is input, the Bayesian network infers the joint probability of each fault. If the joint probability exceeds a set threshold (e.g., 0.65), a corresponding fault warning is triggered. For pump jamming, the joint probability threshold is set to 0.65, reducing the false alarm rate by 19% compared to the traditional threshold method.

[0035] The model has online learning capabilities. Every 24 hours, it uses the latest 5,000 operating condition data within the sliding time window to incrementally update the training set, retrain the SVM and fine-tune the Bayesian network parameters, which improves the detection sensitivity of heat exchanger scaling and blockage from 78% to 89% and shortens the response time to less than 3 seconds, meeting the real-time diagnostic requirements of SVG for continuous operation for 72 hours at an ambient temperature of 40℃.

[0036] In some embodiments, the diagnostic model can be deployed on an edge computing unit to output fault type and confidence level in real time and display it through a visual interface. Validated on a hardware-in-the-loop simulation platform (based on LabVIEW), using 120 hours of continuous operation data at 45°C as a benchmark, the model's response time to junction temperature exceeding the limit is less than 3 seconds, and the Kappa coefficient of consistency with manual inspection results is 0.82. After field application, the fault location error was reduced from the initial ±15% to ±5%, the maintenance cycle was extended from 42 days to 58 days, and coolant consumption decreased by 37% year-on-year.

[0037] Example 1

[0038] The data acquisition module consists of a temperature sensor (accuracy ±0.5℃), a pressure sensor (accuracy ±0.2%FS), a flow meter (accuracy ±1.5% reading), and an FPGA data acquisition card, enabling multi-channel synchronous acquisition.

[0039] Preprocessing module: performs time alignment and moving average filtering on the acquired signal, and downsamples it to 200Hz.

[0040] Feature extraction module: Built-in wavelet packet decomposition algorithm and statistical feature calculation unit to compress 12-dimensional original features into 3-dimensional principal components.

[0041] Fault diagnosis module: A two-layer diagnostic model integrating SVM and Bayesian network, which receives feature vectors and outputs fault type and confidence level.

[0042] Output and Display Module: Generates diagnostic reports, displays real-time operating conditions, historical trends and early warning information through the HMI interface, and supports remote operation and maintenance.

[0043] Example 2

[0044] A field test platform was built on an SVG water-cooled heat dissipation device in an industrial park. The ambient temperature ranged from -10℃ to 50℃, the coolant flow rate was 10-20 L / min, the pressure fluctuation was ±0.15 MPa, and the data sampling frequency was 100 Hz. After the diagnostic model was imported into the edge computing gateway, it identified two types of faults—pump bearing wear and heat exchanger scaling—from 180 hours of accumulated operating data. The accuracy rates were 94.1% and 89.6%, respectively, with the false alarm rate remaining stable below 3.8%. When the low temperature (-10℃) caused the coolant viscosity to increase to 5.2 mPa·s, the model dynamically adjusted the feature weights, reducing the substrate temperature difference prediction deviation from ±2.3℃ to ±0.9℃. Based on the diagnostic report, the maintenance team extended the preventive maintenance interval from 35 days to 52 days, reducing the average annual energy consumption by 12.4%.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system, characterized in that, include: Collect multi-source operating parameters of the SVG water-cooling system; The multi-source operating parameters include at least coolant temperature, coolant flow rate, and circuit pressure; The collected multi-source operating parameters are preprocessed to obtain preprocessed data; Feature processing is performed on the preprocessed data to extract operating condition feature vectors that reflect the operating conditions. The operating condition feature vector includes turbulent state characteristics and pressure fluctuation statistical characteristics; The operating condition feature vector is input into the fault diagnosis model to obtain the diagnosis result; the fault diagnosis model is a two-layer architecture model that integrates support vector machine and Bayesian network; the diagnosis result is used to indicate the operating condition category and / or fault type of the water cooling heat dissipation system.

2. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 1, characterized in that, Preprocessing includes time synchronization and filtering preprocessing; The time synchronization adopts an FPGA-based synchronization triggering mechanism; The filtering preprocessing uses sliding window mean filtering.

3. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 1, characterized in that, The preprocessed data undergoes feature processing to extract operating condition feature vectors that reflect the operating conditions, including: Wavelet packet decomposition was performed on the pre-processed coolant flow rate to extract the energy proportion and energy entropy of each frequency band as turbulent state features. The peak-to-peak value and standard deviation of the pre-treated loop pressure were calculated within the sliding time window to construct statistical characteristics of pressure fluctuations. The turbulent state characteristics and pressure fluctuation statistical characteristics are concatenated to obtain the operating condition feature vector.

4. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 3, characterized in that, The wavelet packet decomposition has 8 layers and a decomposition frequency band of 0-500Hz. The energy entropy value of the 20-100Hz frequency band is extracted to characterize the blockage risk of the heat dissipation substrate flow channel.

5. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 1, characterized in that, The sliding time window is 5 minutes long. When the peak value of the loop pressure is higher than 0.25 MPa for 5 consecutive minutes and the standard deviation increases by more than 30%, the probability of bearing wear of the circulating pump group is determined to be 82%.

6. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 1, characterized in that, Inputting the operating condition feature vector into the fault diagnosis model yields the diagnosis results, including: The first layer uses a support vector machine to perform preliminary classification of the operating condition feature vectors and outputs the probability of each type of fault. The second layer uses the output of the support vector machine as the input of the Bayesian network and calculates the joint probability based on the prior probability of historical fault samples. Determine whether the joint probability exceeds a set threshold. If the joint probability exceeds the set threshold, trigger a fault warning and output the diagnostic result; if the joint probability is less than the set threshold, do not take any action.

7. The method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system according to claim 6, characterized in that, The kernel function parameters of the support vector machine are optimized through grid search.

8. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 6, characterized in that, The Bayesian network constructs a conditional probability table based on the prior probabilities of historical fault samples, and calculates the joint probability based on the conditional probability table.

9. The method for diagnosing the operating conditions of the SVG water-cooled heat dissipation system according to claim 1, characterized in that, Also includes: By utilizing an online learning mechanism, the fault diagnosis model is incrementally updated periodically using newly added operating condition data.

10. The method for diagnosing the operating conditions of an SVG water-cooled heat dissipation system according to claim 1, characterized in that, The diagnostic results include pump jamming, heat sink leakage, heat exchanger scaling and / or bearing wear.

Images