Liquid cooling charging module health state prediction system

Abstract

The invention discloses a liquid cooling charging module health state prediction system, and relates to the technical field of electrochemical detection, and the system comprises a multi-parameter collection unit which collects the pressure and flow data of a cooling liquid, and is provided with an independent collection node at each branch of a double-gun charging system, and provides basic data for monitoring; the dynamic calibration unit communicates with the acquisition unit, compensates data drift based on a temperature-vibration interference model of machine learning training, executes baseline calibration by using an idle period to update a reference value, and guarantees data accuracy; the leakage detection unit is used for calculating a data change rate after compensation, outputting an early warning according to a preset condition and positioning a leakage branch; and the health state evaluation unit fuses the early warning signal and the historical operation data to generate a prediction result. The problems that tiny leakage detection is difficult and a sensor is prone to interference can be solved, manual inspection is reduced, the maintenance cost is reduced, and the requirement of a high-power charging scene is met.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, and more specifically to a charging state prediction system. Background Technology

[0002] With the widespread adoption of supercharging technology for electric vehicles, traditional air cooling is no longer sufficient to meet the heat dissipation requirements. Liquid cooling technology uses liquid circulation to control the temperature of the charging module within a safe range, supporting a supercharging capability of 200 kilometers of range in just 5 minutes. A health status prediction system is key to ensuring the reliability of supercharging equipment.

[0003] As infrastructure, charging stations incur high downtime costs and are sensitive to user experience. Predictive maintenance can provide 30% advance warning of potential faults, reducing reliance on manual inspections and lowering maintenance costs. For example, an AI diagnostic engine can identify abnormal noises from transformers using voiceprint recognition, achieving a defect detection rate of 99.7%.

[0004] Liquid cooling systems involve the combination of high-voltage electricity and coolant, requiring solutions to safety issues such as leak detection and insulation failure. The EU has mandated testing of coolant insulation performance in liquid cooling systems, but traditional methods are ineffective in identifying small leaks or single-gun leaks in dual-gun systems.

[0005] Therefore, there is an urgent need for a liquid-cooled charging module health status prediction system. This system faces bottlenecks in sensor accuracy and reliability, including the challenge of detecting small leaks. Traditional pressure monitoring cannot identify leaks as small as one drop per second, and it is difficult to locate leaks in a single gun in a dual-gun system. Environmental interference, such as high temperature and vibration, can easily cause sensor drift. Summary of the Invention

[0006] To address the aforementioned technical problems, a liquid-cooled charging module health status prediction system is provided. This technical solution solves the bottlenecks in sensor accuracy and reliability.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A liquid-cooled charging module health status prediction system, comprising: The multi-parameter acquisition unit acquires at least two physical parameters of the liquid-cooled charging module, including the pressure data and flow data of the coolant, and sets independent acquisition nodes in each branch of the dual-gun charging system. The dynamic calibration unit is communicatively connected to the multi-parameter acquisition unit. Based on a preset temperature-vibration interference model, it performs drift compensation on the pressure and flow data. The temperature-vibration interference model is trained through machine learning, with the input being ambient temperature and vibration frequency values, and the output being sensor drift correction coefficients. During the idle period of the liquid-cooled charging module, baseline calibration is performed to update the baseline values ​​of the pressure and flow data. The leakage detection unit is communicatively connected to the dynamic calibration unit, calculates the rate of change of pressure data and the rate of change of flow data after drift compensation; when the rate of change of pressure data and the rate of change of flow data meet the preset small leakage judgment conditions, it outputs a leakage warning signal; and locates the leakage branch based on the pressure-flow correlation difference of each branch's independent acquisition node. The health status assessment unit is communicatively connected to the leakage detection unit and generates a health status prediction result by combining the leakage early warning signal and the historical operating data of the liquid-cooled charging module.

[0008] Preferably, the multi-parameter acquisition unit specifically includes: The first acquisition component, located in the main circulation pipeline of the liquid-cooled charging module, includes a first pressure sensor, which adopts a differential pressure measurement structure. Its positive and negative pressure interfaces are respectively connected to the inlet and outlet of the main circulation pipeline to acquire the pressure difference of the main coolant. The first flow sensor adopts the ultrasonic Doppler principle and has two built-in probes arranged at an angle on the outer wall of the pipeline to acquire the instantaneous flow rate of the main coolant. The second acquisition component is respectively set at the end interface of each branch of the dual-gun charging system, including the first charging gun branch and the second charging gun branch. The second acquisition component of each branch includes a second pressure sensor, which adopts a miniature piezoresistive structure, integrates a temperature compensation chip, and has a vibration-damping buffer layer in the package shell, for acquiring the real-time pressure value at the end of the corresponding branch; and a second flow sensor, which adopts a thermal mass flow metering method and has a built-in self-cleaning electrode, for acquiring the instantaneous flow value of the corresponding branch.

[0009] Preferably, the multi-parameter acquisition unit further includes: The environmental sensing components include a temperature sensor, which is placed on the surface of the power devices of the liquid-cooled module and inside the coolant tank; and a triaxial vibration sensor, which is fixed to the load-bearing part of the liquid-cooled module housing, for collecting vibration acceleration data during module operation. The outputs of the first acquisition component, the second acquisition component, and the environmental sensing component are all connected to the signal conditioning module via shielded twisted-pair cables. The signal conditioning module is configured to filter, amplify, and perform A / D conversion on the original signal, and transmit it to the dynamic calibration unit via the industrial Ethernet protocol.

[0010] Preferably, the dynamic calibration unit specifically includes: The data interface is configured to receive raw data transmitted by the multi-parameter acquisition unit via the industrial Ethernet protocol, including pressure data, flow data, ambient temperature value, vibration frequency value, and system operating status signal. It performs adaptive Kalman filtering on the raw pressure / flow data to remove high-frequency noise; timestamps the temperature / vibration data; and identifies the operating status of the liquid-cooled charging module as working or idle. The sensor drift correction coefficient is obtained, and a pre-trained temperature-vibration interference model is stored in the built-in storage medium. The model is a random forest regression model or a neural network model. The input layer receives the ambient temperature value, vibration frequency value and sensor type identifier; the output layer outputs the drift correction coefficient of the corresponding sensor.

[0011] Preferably, the dynamic calibration unit further includes: Real-time drift compensation: Obtain correction coefficients under current temperature-vibration conditions from the model storage and retrieval module; perform compensation calculations on preprocessed pressure data and preprocessed flow data; Baseline calibration is automatically triggered when the liquid-cooled charging module is detected to be in an idle state for a duration longer than the predetermined time. Within the calibration window, compensated pressure and flow data are collected, and their average value is calculated as the new baseline value. If the data fluctuation within the calibration window exceeds the preset threshold, the calibration window will be extended to the predetermined duration. If the requirement is still not met, a calibration anomaly will be recorded and a manual inspection prompt will be triggered.

[0012] Preferably, the dynamic calibration unit further includes: The calibration parameters are stored using non-volatile storage media, recording real-time updated pressure / flow reference values; correction factors for each drift compensation and corresponding temperature / vibration parameters; timestamps, durations, and result status of baseline calibration; and support for updating model parameters and calibration strategies via OTA.

[0013] Preferably, the leakage detection unit specifically includes: It receives the compensated pressure data and compensated flow data output by the dynamic calibration unit, and performs timestamp alignment on the multi-dimensional data based on the system's unified clock to ensure the temporal correlation between pressure and flow changes. The rate of change is calculated using a sliding window and least squares fitting to calculate the instantaneous rate of change of pressure and flow data. The rate of change is calculated independently for each branch of the main road and the dual-gun system, and the rate of change of each branch needs to be correlated with the end pressure and flow data of the corresponding branch. The calculation results are smoothed using an exponentially weighted moving average algorithm.

[0014] Preferably, the leakage detection unit further includes: The minor leak detection is based on multi-parameter correlation logic. The first-level detection condition is that the instantaneous change rate of the main road pressure data and the instantaneous change rate of the main road flow data are within a predetermined range. The second-level detection condition is that the ratio of the instantaneous change rate of the pressure data to the instantaneous change rate of the flow data is stable within a predetermined range. The third-level detection condition is that the above first-level and second-level conditions are continuously satisfied for a predetermined duration. When the third-level detection condition is satisfied simultaneously, a minor leak warning signal is output. For branch leakage location, in a dual-gun system, the configuration is to locate the leakage point through branch difference analysis, calculating the difference in pressure change rate and flow change rate between the first and second branches; a location threshold is set, if the pressure change rate difference is greater than the pressure difference threshold and the flow change rate difference is greater than the flow difference threshold, then the branch with the larger absolute value of the change rate is determined to be the leakage branch; if the difference in change rate between the two branches is less than the threshold, then it is determined to be a leakage in the shared main line. The system outputs early warning signals and classifies leak warning signals into levels: minor leaks are classified as Level 1 warnings, and leaks exceeding a predetermined threshold are classified as Level 2 warnings. The warning information includes the leak occurrence timestamp, main / branch identification, real-time monitoring values ​​of the instantaneous change rate of pressure data and the instantaneous change rate of flow data, and the leak probability. The information is transmitted to the health status assessment unit via the industrial bus and triggers a local audible and visual alarm.

[0015] Preferably, the health status assessment unit specifically includes: The system receives early warning signals from the leakage detection unit and simultaneously retrieves historical operating data of the liquid-cooled charging module. This historical data includes pressure / flow baseline calibration records and drift compensation coefficients for the past 12 months; peak pressure fluctuations and flow stability indicators in the past 300 charging cycles; records of handling historical leakage events and performance recovery data after maintenance; and cumulative operating time and temperature cycle count of power devices. The system stores the above data using a time-series database, supporting rapid querying by time dimension. Key features reflecting health status are extracted from the fused data, including leakage-related features such as the frequency of leakage warnings, the trend of single leakage duration, and the drift of the pressure-flow rate change ratio; performance degradation features such as the monthly average decrease in baseline pressure, the attenuation rate of flow response speed, and the increase in coolant circulation power consumption under the same operating conditions; and environmental features such as the leakage sensitivity coefficient under extreme temperatures and the performance recovery capability after high vibration conditions. Feature extraction uses a sliding time window to output a standardized feature vector.

[0016] Preferably, the health status assessment unit further includes: Health status prediction is based on a prediction model using an improved long short-term memory network. The input layer receives the feature vector output by the degradation feature extraction module and the quantization parameters of the current leakage warning signal. The hidden layer strengthens key features through an attention mechanism, introduces a physical constraint layer, and corrects the model output based on the thermal balance equation of the liquid cooling system. The output layer generates a two-dimensional prediction result: a short-term health status score and a long-term remaining life prediction value for key components, along with confidence intervals. The prediction results are output as a visual assessment report, including a health status trend curve, high-risk feature annotations, and remaining life countdown. Tiered decision-making recommendations are generated based on the rating level. Output formats include local touchscreen display, data push to a remote management platform, and printed report generation.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The multi-parameter acquisition unit acquires both pressure and flow parameters, and is equipped with independent acquisition nodes for dual gun paths, which changes the limitations of traditional single-parameter monitoring and lays a data foundation for the detection and location of minute leaks.

[0018] The dynamic calibration unit relies on a temperature-vibration interference model trained by machine learning to compensate for the drift of pressure and flow data in real time, effectively offsetting sensor errors caused by high temperature, vibration and other operating conditions. At the same time, it uses the idle period of the liquid cooling module to perform baseline calibration and update the reference value, further ensuring data accuracy and solving the problem of monitoring inaccuracy caused by environmental interference. The leak detection unit calculates the rate of change of pressure and flow data after compensation and combines it with multi-dimensional preset judgment conditions to accurately identify minute leaks, overcoming the shortcomings of traditional pressure monitoring that cannot detect minute leaks; by using the pressure-flow correlation difference analysis of each branch, it can accurately locate the leaking branch in the dual-gun system and avoid misjudging the location of the leak.

[0019] The health status assessment unit integrates leakage early warning signals with historical operating data of the liquid cooling module, comprehensively considers multi-dimensional health characteristics, and generates comprehensive health status prediction results. It can not only provide timely early warning of potential faults, but also provide a scientific basis for operation and maintenance, reduce reliance on manual inspections, reduce maintenance costs, improve the operational reliability and service life of the liquid cooling charging module, and meet the needs of safe and stable operation of equipment in high-power charging scenarios. Attached Figure Description

[0020] Figure 1 This is an internal framework diagram of a liquid-cooled charging module health status prediction system. Detailed Implementation

[0021] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0022] Reference Figure 1 As shown, a liquid-cooled charging module health status prediction system includes: A liquid-cooled charging module health status prediction system, comprising: The multi-parameter acquisition unit acquires at least two physical parameters of the liquid-cooled charging module, including the pressure data and flow data of the coolant, and sets independent acquisition nodes in each branch of the dual-gun charging system. The dynamic calibration unit is communicatively connected to the multi-parameter acquisition unit. Based on a preset temperature-vibration interference model, it performs drift compensation on the pressure and flow data. The temperature-vibration interference model is trained through machine learning, with the input being ambient temperature and vibration frequency values, and the output being sensor drift correction coefficients. During the idle period of the liquid-cooled charging module, baseline calibration is performed to update the baseline values ​​of the pressure and flow data. The leakage detection unit is communicatively connected to the dynamic calibration unit, calculates the rate of change of pressure data and the rate of change of flow data after drift compensation; when the rate of change of pressure data and the rate of change of flow data meet the preset small leakage judgment conditions, it outputs a leakage warning signal; and locates the leakage branch based on the pressure-flow correlation difference of each branch's independent acquisition node. The health status assessment unit is communicatively connected to the leakage detection unit and generates a health status prediction result by combining the leakage early warning signal and the historical operating data of the liquid-cooled charging module.

[0023] It should be noted that the dynamic calibration unit eliminates the coupling effect of environmental variables on the pressure / flow sensor in real time through a preset temperature-vibration interference model. This model needs to be trained in the laboratory under simulated high and low temperature cycles (-30℃~85℃) and different vibration spectra (0~200Hz) to ensure that the correction coefficients cover the actual working conditions and keep the data drift error within ±1.5%. Idle cycle baseline calibration automatically resets the sensor reference value during charging intervals (such as low-load periods at night), avoiding the accumulation of zero-point drift caused by long-term operation and solving the pain point of traditional static calibration requiring manual intervention.

[0024] Detection of minute leaks: Dual-indicator collaborative judgment, leak warning is based on the joint threshold of pressure change rate (dP / dt) and flow change rate (dQ / dt), dP / dt>0.05MPa / s and dQ / dt<-2% of rated flow, which improves the sensitivity by more than 60% compared with single parameter detection; Branch-level fault isolation: By comparing the correlation differences of the pressure-flow curves of each branch, such as the strong linear relationship in normal branches and the sharp drop in pressure but slow change in flow in leaking branches, the leak source can be located within 5 seconds with the accuracy of branch level.

[0025] The multi-parameter acquisition unit specifically includes: The first acquisition component, located in the main circulation pipeline of the liquid-cooled charging module, includes a first pressure sensor, which adopts a differential pressure measurement structure. Its positive and negative pressure interfaces are respectively connected to the inlet and outlet of the main circulation pipeline to acquire the pressure difference of the main coolant. The first flow sensor adopts the ultrasonic Doppler principle and has two built-in probes arranged at an angle on the outer wall of the pipeline to acquire the instantaneous flow rate of the main coolant. The second acquisition component is respectively set at the end interface of each branch of the dual-gun charging system, including the first charging gun branch and the second charging gun branch. The second acquisition component of each branch includes a second pressure sensor, which adopts a miniature piezoresistive structure, integrates a temperature compensation chip, and has a vibration-damping buffer layer in the package shell, for acquiring the real-time pressure value at the end of the corresponding branch; and a second flow sensor, which adopts a thermal mass flow metering method and has a built-in self-cleaning electrode, for acquiring the instantaneous flow value of the corresponding branch.

[0026] The multi-parameter acquisition unit also includes: The environmental sensing components include a temperature sensor, which is placed on the surface of the power devices of the liquid-cooled module and inside the coolant tank; and a triaxial vibration sensor, which is fixed to the load-bearing part of the liquid-cooled module housing, for collecting vibration acceleration data during module operation. The outputs of the first acquisition component, the second acquisition component, and the environmental sensing component are all connected to the signal conditioning module via shielded twisted-pair cables. The signal conditioning module is configured to filter, amplify, and perform A / D conversion on the original signal, and transmit it to the dynamic calibration unit via the industrial Ethernet protocol.

[0027] It should be noted that the main circulation pipeline uses a differential pressure sensor, which adopts direct differential measurement at the inlet / outlet (not traditional single-point measurement) to eliminate common-mode error caused by coolant temperature expansion; a built-in capillary balance structure between the positive and negative pressure interfaces suppresses pulse noise caused by liquid flow impact, making the pressure difference measurement accuracy ±0.1%FS (full scale), especially suitable for scenarios with large flow fluctuations, such as dual-gun power switching; The ultrasonic Doppler flow meter features dual probes arranged symmetrically at a 45° angle, forming a three-dimensional measurement area through the intersection of sound wave paths, thus solving the signal attenuation problem caused by bubbles / impurities in single probes. The operating frequency is selected at 1MHz (balancing penetration and resolution), and with the adaptive gain control algorithm, the linear error is <±1.5% in the flow rate range of 0.5-50L / min.

[0028] Branch end: Triple protection of piezoresistive pressure sensor, temperature compensation, integrated MEMS temperature chip to correct thermal drift in real time (-40~125℃ compensation range), temperature drift coefficient reduced to 0.01%FS / ℃; mechanical vibration resistance, silicone buffer layer and metal corrugated diaphragm structure to attenuate 10-100Hz mechanical shock during gun wire insertion and removal; media compatibility, 316L stainless steel isolation diaphragm and fluororubber seal, resistant to ethylene glycol-based coolant corrosion; The thermal flow meter has a self-cleaning mechanism that uses pulsed electrode heating (with adjustable duty cycle). During the charging interval, it triggers an instantaneous high temperature of 800℃ to ablate oxide deposits on the electrode surface, ensuring the zero-point stability of long-term measurements (annual drift <1%).

[0029] Environmental sensing: strategic placement of temperature sensors, direct mounting of IGBT / MOSFET heat dissipation substrates on the surface of power devices, monitoring chip junction temperature with an accuracy of ±0.5℃; coolant tank with dual-probe redundancy design (top gas phase and bottom liquid phase) to capture coolant stratification.

[0030] Vibration sensor modal analysis guides the use of a triaxial accelerometer (±50g range) fixed at the point of maximum mechanical impedance of the housing (determined by finite element analysis) to effectively capture unbalanced vibration of the water pump impeller (fundamental frequency 30-100Hz), shock of solenoid valve action (transient >500Hz), and pipeline resonance (characteristic frequency band 150-300Hz).

[0031] For interference-resistant transmission, the shielded twisted pair (STP) cable uses a double shielding layer of aluminum foil and braided copper mesh, and the cable capacitance is controlled below 60pF / m to suppress EMI interference of 10kV / μs level in the charging pile; the connector is reinforced with a metal shell M12 interface (IP67 protection level) and gold-plated contact (>3μm) to ensure contact resistance <5mΩ in humid environments. Signal conditioning module: Adaptive filtering strategy, pressure / flow signals, switching 4th order Butterworth filter (cutoff frequency 1Hz~1kHz programmable), dynamically filtering out pump pulsation noise; vibration signals, parallel processing path, low-pass filter (0-500Hz) for fault diagnosis, high-pass filter (>5kHz) to capture early bearing damage. The 24-bit Σ-Δ ADC conversion, with an oversampling rate of 1024x, combined with digital noise reduction technology, improves the effective resolution to 20 bits, making minute changes in flow rate (<0.1L / min) identifiable.

[0032] The dynamic calibration unit specifically includes: The data interface is configured to receive raw data transmitted by the multi-parameter acquisition unit via the industrial Ethernet protocol, including pressure data, flow data, ambient temperature value, vibration frequency value, and system operating status signal. It performs adaptive Kalman filtering on the raw pressure / flow data to remove high-frequency noise; timestamps the temperature / vibration data; and identifies the operating status of the liquid-cooled charging module as working or idle. The sensor drift correction coefficient is obtained, and a pre-trained temperature-vibration interference model is stored in the built-in storage medium. The model is a random forest regression model or a neural network model. The input layer receives the ambient temperature value, vibration frequency value and sensor type identifier; the output layer outputs the drift correction coefficient of the corresponding sensor.

[0033] The dynamic calibration unit also includes: Real-time drift compensation: Obtain correction coefficients under current temperature-vibration conditions from the model storage and retrieval module; perform compensation calculations on preprocessed pressure data and preprocessed flow data; Baseline calibration is automatically triggered when the liquid-cooled charging module is detected to be in an idle state for a duration longer than the predetermined time. Within the calibration window, compensated pressure and flow data are collected, and their average value is calculated as the new baseline value. If the data fluctuation within the calibration window exceeds the preset threshold, the calibration window will be extended to the predetermined duration. If the requirement is still not met, a calibration anomaly will be recorded and a manual inspection prompt will be triggered.

[0034] The dynamic calibration unit also includes: The calibration parameters are stored using non-volatile storage media, recording real-time updated pressure / flow reference values; correction factors for each drift compensation and corresponding temperature / vibration parameters; timestamps, durations, and result status of baseline calibration; and support for updating model parameters and calibration strategies via OTA.

[0035] It should be noted that the temperature-vibration disturbance model is a multi-physics coupled model, which adopts a dual-model fusion architecture of gradient boosting decision tree (GBDT) and 1D-CNN. The GBDT model processes structured data such as temperature / vibration, and through feature importance analysis, it determines that the 100-150Hz component in the vibration frequency has a weight of up to 45% on the drift of the pressure sensor. The 1D-CNN model extracts the time-frequency features of the vibration signal (after wavelet transform preprocessing) to capture the nonlinear drift caused by transient impact.

[0036] Sensor type identification coding converts pressure / flow sensor model numbers into One-Hot vectors, enabling the model to automatically adapt to different sensitivity characteristics.

[0037] Real-time compensation: The correction coefficient is dynamically loaded, and a three-dimensional lookup table (LUT) for temperature-vibration-correction coefficient is established. The pre-calculated values ​​are stored with 10℃ / 5Hz as the grid. The table is looked up first during operation (response time ≤1ms). Data between grids is filled by bilinear interpolation. Pressure value compensation:

[0038] In the formula, The pressure value after compensation; and This is the compensation coefficient for temperature and vibration; This is the original pressure value; This is the pressure offset. Flow value compensation is used to perform frequency domain phase correction on the ultrasonic Doppler signal to eliminate the deviation of the sound wave refraction angle caused by vibration.

[0039] Baseline calibration: State machine control, operating state, charging power > 5kW; The working state changes to an idle state, and the power consumption remains zero for 30 seconds. The idle state is changed to calibration ready, which lasts for 300s and the temperature gradient is <0.1℃ / min; The calibration ready state is changed to calibration execution state, with vibration RMS < 0.5g; The fluctuation threshold is adaptively adjusted. The preset threshold changes dynamically with historical stability. If the standard deviation of calibration is less than 0.1% for 10 consecutive calibrations, the threshold is tightened to ±0.3% to improve calibration sensitivity.

[0040] Fail-safe strategy, graded response mechanism: Level 1 anomaly, fluctuation exceeds limit → extend calibration window to 900s; Level 2 anomaly, still exceeding limits → Freeze baseline value update, triggering sensor health assessment (e.g., comparing data from adjacent branches). Level 3 anomaly, assessment fails → The location of the faulty sensor is highlighted on the operation interface to guide priority replacement.

[0041] Key parameters of non-volatile memory: Storage Class Record content Application value Baseline Log Pressure / flow baseline + timestamp + ambient temperature Tracing long-term drift trends to predict sensor lifespan Correction coefficient set Temperature / vibration parameters + correction factor + model version Verify the effectiveness of the interference model and support OTA rollback. Calibration event Triggering cause / duration / result status Statistical calibration success rate and optimization of triggering strategies OTA upgrade, dual-zone storage architecture, dividing the space into an Active zone (running model) and a Backup zone (old version). The new model will be overwritten in the Active zone only after a 72-hour trial run verification. Differential upgrades transmit only the incremental model parameters (average 50KB / time), reducing data transmission volume by 90%, making them suitable for narrowband IoT environments.

[0042] The leakage detection unit specifically includes: It receives the compensated pressure data and compensated flow data output by the dynamic calibration unit, and performs timestamp alignment on the multi-dimensional data based on the system's unified clock to ensure the temporal correlation between pressure and flow changes. The rate of change is calculated using a sliding window and least squares fitting to calculate the instantaneous rate of change of pressure and flow data. The rate of change is calculated independently for each branch of the main road and the dual-gun system, and the rate of change of each branch needs to be correlated with the end pressure and flow data of the corresponding branch. The calculation results are smoothed using an exponentially weighted moving average algorithm.

[0043] The leakage detection unit also includes: The minor leak detection is based on multi-parameter correlation logic. The first-level detection condition is that the instantaneous change rate of the main road pressure data and the instantaneous change rate of the main road flow data are within a predetermined range. The second-level detection condition is that the ratio of the instantaneous change rate of the pressure data to the instantaneous change rate of the flow data is stable within a predetermined range. The third-level detection condition is that the above first-level and second-level conditions are continuously satisfied for a predetermined duration. When the third-level detection condition is satisfied simultaneously, a minor leak warning signal is output. For branch leakage location, in a dual-gun system, the configuration is to locate the leakage point through branch difference analysis, calculating the difference in pressure change rate and flow change rate between the first and second branches; a location threshold is set, if the pressure change rate difference is greater than the pressure difference threshold and the flow change rate difference is greater than the flow difference threshold, then the branch with the larger absolute value of the change rate is determined to be the leakage branch; if the difference in change rate between the two branches is less than the threshold, then it is determined to be a leakage in the shared main line. The system outputs early warning signals and classifies leak warning signals into levels: minor leaks are classified as Level 1 warnings, and leaks exceeding a predetermined threshold are classified as Level 2 warnings. The warning information includes the leak occurrence timestamp, main / branch identification, real-time monitoring values ​​of the instantaneous change rate of pressure data and the instantaneous change rate of flow data, and the leak probability. The information is transmitted to the health status assessment unit via the industrial bus and triggers a local audible and visual alarm.

[0044] It should be noted that the parameter spatiotemporal correlation detection uses a high-precision time alignment mechanism, employing the IEEE 1588 Precision Time Protocol (PTP) to achieve microsecond-level clock synchronization, eliminating time offset between main and branch data. When the time difference between pressure drop and flow decay is ≤50ms, a correlation flag is triggered. For example, the typical physical response delay for pipeline leakage is 30-80ms. Adaptive sliding window, the window width dynamically adjusts with traffic: When the flow rate Q is greater than 30 L / min (high flow rate condition), the window width W is set to 5 seconds; When the flow rate Q is between 10 L / min and 30 L / min, the window width W is set to 10 seconds. When the flow rate Q is less than 10 L / min (low flow condition), the window width W is set to 20 seconds.

[0045] The least squares linear fitting method is superior to the traditional difference method in calculating dP / dt and dQ / dt, and the tolerance to noise is improved by 3 times (stable output is possible when the signal-to-noise ratio is ≥10dB).

[0046] The rate of change is smoothed, and an exponentially weighted moving average (EWMA) is applied. The weighting factor α is adaptively adjusted according to the vibration intensity. In the formula, The effective value of the vibration; Decrease the α value during high vibration (enhanced smoothness), and increase the α value during low vibration (preserved transient characteristics), thereby compressing the fluctuation amplitude of the flow rate change curve by 60%.

[0047] Environmental coupling interference is eliminated. When the vibration frequency is greater than 200Hz or the temperature gradient is greater than 2℃ / min, the judgment threshold is automatically raised to prevent misjudgment caused by sudden environmental changes. Historical data comparison involves performing DTW dynamic time warping matching between the current dP / dt-dQ / dt curve and typical leakage patterns in the database. An alert is only triggered if the similarity is greater than 85%.

[0048] Branch leakage location: Differential analysis method is used for location. Calculate Δ(dP / dt) and Δ(dQ / dt) of branch 1 / 2. Check if Δ(dP / dt) > 0.05MPa / s. If so, check if Δ(dQ / dt) > 1.2% / s. If so, determine that the branch with the greater rate of change is leaking. Otherwise, determine that the main branch is leaking. If Δ(dP / dt) > 0.05MPa / s, trigger branch sensor cross-verification; Additional verification of main road leakage: If main road leakage is determined, it is necessary to simultaneously detect whether the main road flow attenuation rate is greater than the sum of the branch road flow attenuation rates. In asymmetrical operating conditions, load balancing compensation is implemented. When the power difference between the two guns exceeds 30%, such as gun 1 = 200kW and gun 2 = 50kW, the threshold is automatically scaled according to the power ratio.

[0049] In the formula, To indicate the power of the two guns; to avoid misjudging the normal voltage drop of the high-load branch as a leak.

[0050] Tiered early warning and output, early warning grading standards:

[0051] Early warning information, leakage probability model, based on Bayesian network calculation:

[0052] In the formula, The probability of a leak occurring; This represents the probability that the sensor will detect a leak in the event of a leak. This represents the total probability that the sensor detects a leak. The prior probability includes historical leakage frequency, sensor health, and environmental severity level; When the output probability is greater than 90%, the spare parts scheduling instruction will be automatically pushed to the cloud platform. Human-computer interaction for audible and visual alarms: Yellow slow flashing light (1Hz) and 800Hz buzzer → Level 1 warning; Red flashing lights (5Hz) and a 2kHz beeping sound → Level 2 warning.

[0053] The health status assessment unit specifically includes: The system receives early warning signals from the leakage detection unit and simultaneously retrieves historical operating data of the liquid-cooled charging module. This historical data includes pressure / flow baseline calibration records and drift compensation coefficients for the past 12 months; peak pressure fluctuations and flow stability indicators in the past 300 charging cycles; records of handling historical leakage events and performance recovery data after maintenance; and cumulative operating time and temperature cycle count of power devices. The system stores the above data using a time-series database, supporting rapid querying by time dimension. Key features reflecting health status are extracted from the fused data, including leakage-related features such as the frequency of leakage warnings, the trend of single leakage duration, and the drift of the pressure-flow rate change ratio; performance degradation features such as the monthly average decrease in baseline pressure, the attenuation rate of flow response speed, and the increase in coolant circulation power consumption under the same operating conditions; and environmental features such as the leakage sensitivity coefficient under extreme temperatures and the performance recovery capability after high vibration conditions. Feature extraction uses a sliding time window to output a standardized feature vector.

[0054] The health status assessment unit also includes: Health status prediction is based on a prediction model using an improved long short-term memory network. The input layer receives the feature vector output by the degradation feature extraction module and the quantization parameters of the current leakage warning signal. The hidden layer strengthens key features through an attention mechanism, introduces a physical constraint layer, and corrects the model output based on the thermal balance equation of the liquid cooling system. The output layer generates a two-dimensional prediction result: a short-term health status score and a long-term remaining life prediction value for key components, along with confidence intervals. The prediction results are output as a visual assessment report, including a health status trend curve, high-risk feature annotations, and remaining life countdown. Tiered decision-making recommendations are generated based on the rating level. Output formats include local touchscreen display, data push to a remote management platform, and printed report generation.

[0055] It should be noted that the degradation trajectory analysis of the baseline calibration record, through the time series change of the pressure / flow baseline value over 12 months (such as a monthly average decrease of 0.08% → 0.15%), identifies the fatigue trend of the sensor diaphragm or the fouling of the pipeline, and provides an early warning 6 months earlier than the traditional threshold alarm. Environmental adaptability assessment of drift compensation coefficient, statistical analysis of correction coefficient increase under extreme working conditions (>65℃ or vibration>15g), calculation of cumulative environmental damage (e.g., cumulative duration of coefficient>1.2), and quantification of equipment disturbance rejection degradation.

[0056] Leakage event correlation analysis establishes a mapping relationship between maintenance records and performance recovery (e.g., flow stability improved by 32% after seal replacement), forming a maintenance knowledge graph to guide subsequent decision-making.

[0057] The construction logic of the three-dimensional feature space:

[0058] A sliding time window, an adaptive window mechanism, has a 30-day window width during the stable period to capture slow degradation; and a 3-day window width during the failure period to track accelerated failures. Feature standardization innovation, using load condition weighted Z-score:

[0059] In the formula, x is the original data, μ is the mean, σ is the standard deviation, Prated is the rated power, and Pactual is the actual power; Two-dimensional prediction, Short-term Health Index (HHI):

[0060] In the formula, LeakScore represents the leakage score, PerfDecline represents the performance degradation, and EnvRisk represents the environmental risk; Risk level classification: ≥80: Healthy (Green); 60-79: Attenuation (yellow); <60: High risk (red); Remaining life prediction (RUL) was performed using Monte Carlo Dropout to quantify uncertainty: RUL for water pump = 850 ± 120 cycles (95% confidence interval). Visualized reports, health trend curves overlaid with key events, high-risk feature annotations, attribution analysis to pinpoint root causes of problems, and intelligent generation of tiered decision-making:

[0061] Multi-channel output collaboration, local touchscreen, interactive 3D device health model; remote platform, RUL countdown interface for connecting to maintenance system; printed reports, diagnostic documents compliant with ISO13379 standard.

[0062] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A health status prediction system for a liquid-cooled charging module, characterized in that, include: The multi-parameter acquisition unit acquires at least two physical parameters of the liquid-cooled charging module, including the pressure data and flow data of the coolant, and sets independent acquisition nodes in each branch of the dual-gun charging system. The dynamic calibration unit is communicatively connected to the multi-parameter acquisition unit. Based on a preset temperature-vibration interference model, it performs drift compensation on the pressure and flow data. The temperature-vibration interference model is trained through machine learning, with the input being ambient temperature and vibration frequency values, and the output being sensor drift correction coefficients. During the idle period of the liquid-cooled charging module, baseline calibration is performed to update the baseline values ​​of the pressure and flow data. The leakage detection unit is communicatively connected to the dynamic calibration unit, calculates the rate of change of pressure data and the rate of change of flow data after drift compensation; when the rate of change of pressure data and the rate of change of flow data meet the preset small leakage judgment conditions, it outputs a leakage warning signal; and locates the leakage branch based on the pressure-flow correlation difference of each branch's independent acquisition node. The health status assessment unit is communicatively connected to the leakage detection unit and generates a health status prediction result by combining the leakage early warning signal and the historical operating data of the liquid-cooled charging module.

2. The liquid-cooled charging module health status prediction system according to claim 1, characterized in that, The multi-parameter acquisition unit specifically includes: The first acquisition component, located in the main circulation pipeline of the liquid-cooled charging module, includes a first pressure sensor, which adopts a differential pressure measurement structure. Its positive and negative pressure interfaces are respectively connected to the inlet and outlet of the main circulation pipeline to acquire the pressure difference of the main coolant. The first flow sensor adopts the ultrasonic Doppler principle and has two built-in probes arranged at an angle on the outer wall of the pipeline to acquire the instantaneous flow rate of the main coolant. The second acquisition component is respectively set at the end interface of each branch of the dual-gun charging system, including the first charging gun branch and the second charging gun branch. The second acquisition component of each branch includes a second pressure sensor, which adopts a miniature piezoresistive structure, integrates a temperature compensation chip, and has a vibration-damping buffer layer in the package shell, for acquiring the real-time pressure value at the end of the corresponding branch; and a second flow sensor, which adopts a thermal mass flow metering method and has a built-in self-cleaning electrode, for acquiring the instantaneous flow value of the corresponding branch.

3. The liquid-cooled charging module health status prediction system according to claim 2, characterized in that, The multi-parameter acquisition unit also includes: The environmental sensing components include a temperature sensor, which is placed on the surface of the power devices of the liquid-cooled module and inside the coolant tank; and a triaxial vibration sensor, which is fixed to the load-bearing part of the liquid-cooled module housing, for collecting vibration acceleration data during module operation. The outputs of the first acquisition component, the second acquisition component, and the environmental sensing component are all connected to the signal conditioning module via shielded twisted-pair cables. The signal conditioning module is configured to filter, amplify, and perform A / D conversion on the original signal, and transmit it to the dynamic calibration unit via the industrial Ethernet protocol.

4. The liquid-cooled charging module health status prediction system according to claim 1, characterized in that, The dynamic calibration unit specifically includes: The data interface is configured to receive raw data transmitted by the multi-parameter acquisition unit via the industrial Ethernet protocol, including pressure data, flow data, ambient temperature value, vibration frequency value, and system operating status signal. It performs adaptive Kalman filtering on the raw pressure / flow data to remove high-frequency noise; timestamps the temperature / vibration data; and identifies the operating status of the liquid-cooled charging module as working or idle. The sensor drift correction coefficient is obtained, and a pre-trained temperature-vibration interference model is stored in the built-in storage medium. The model is a random forest regression model or a neural network model. The input layer receives the ambient temperature value, vibration frequency value and sensor type identifier; the output layer outputs the drift correction coefficient of the corresponding sensor.

5. The liquid-cooled charging module health status prediction system according to claim 4, characterized in that, The dynamic calibration unit also includes: Real-time drift compensation: Obtain correction coefficients under current temperature-vibration conditions from the model storage and retrieval module; perform compensation calculations on preprocessed pressure data and preprocessed flow data; Baseline calibration is automatically triggered when the liquid-cooled charging module is detected to be in an idle state for a duration longer than the predetermined time. Within the calibration window, compensated pressure and flow data are collected, and their average value is calculated as the new baseline value. If the data fluctuation within the calibration window exceeds the preset threshold, the calibration window will be extended to the predetermined duration. If the requirement is still not met, a calibration anomaly will be recorded and a manual inspection prompt will be triggered.

6. The liquid-cooled charging module health status prediction system according to claim 1, characterized in that, The dynamic calibration unit also includes: The calibration parameters are stored using non-volatile storage media, recording real-time updated pressure / flow reference values; correction factors for each drift compensation and corresponding temperature / vibration parameters; timestamps, durations, and result status of baseline calibration; and support for updating model parameters and calibration strategies via OTA.

7. The liquid-cooled charging module health status prediction system according to claim 6, characterized in that, The leakage detection unit specifically includes: It receives the compensated pressure data and compensated flow data output by the dynamic calibration unit, and performs timestamp alignment on the multi-dimensional data based on the system's unified clock to ensure the temporal correlation between pressure and flow changes. The rate of change is calculated using a sliding window and least squares fitting to calculate the instantaneous rate of change of pressure and flow data. The rate of change is calculated independently for each branch of the main road and the dual-gun system, and the rate of change of each branch needs to be correlated with the end pressure and flow data of the corresponding branch. The calculation results are smoothed using an exponentially weighted moving average algorithm.

8. The liquid-cooled charging module health status prediction system according to claim 7, characterized in that, The leakage detection unit also includes: The minor leak detection is based on multi-parameter correlation logic. The first-level detection condition is that the instantaneous change rate of the main road pressure data and the instantaneous change rate of the main road flow data are within a predetermined range. The second-level detection condition is that the ratio of the instantaneous change rate of the pressure data to the instantaneous change rate of the flow data is stable within a predetermined range. The third-level detection condition is that the above first-level and second-level conditions are continuously satisfied for a predetermined duration. When the third-level detection condition is satisfied simultaneously, a minor leak warning signal is output. For branch leakage location, in a dual-gun system, the configuration is to locate the leakage point through branch difference analysis, calculating the difference in pressure change rate and flow change rate between the first and second branches; a location threshold is set, if the pressure change rate difference is greater than the pressure difference threshold and the flow change rate difference is greater than the flow difference threshold, then the branch with the larger absolute value of the change rate is determined to be the leakage branch; if the difference in change rate between the two branches is less than the threshold, then it is determined to be a leakage in the shared main line. The system outputs early warning signals and classifies leak warning signals into levels: minor leaks are classified as Level 1 warnings, and leaks exceeding a predetermined threshold are classified as Level 2 warnings. The warning information includes the leak occurrence timestamp, main / branch identification, real-time monitoring values ​​of the instantaneous change rate of pressure data and the instantaneous change rate of flow data, and the leak probability. The information is transmitted to the health status assessment unit via the industrial bus and triggers a local audible and visual alarm.

9. The liquid-cooled charging module health status prediction system according to claim 1, characterized in that, The health status assessment unit specifically includes: The system receives early warning signals from the leakage detection unit and simultaneously retrieves historical operating data of the liquid-cooled charging module. This historical data includes pressure / flow baseline calibration records and drift compensation coefficients for the past 12 months; peak pressure fluctuations and flow stability indicators in the past 300 charging cycles; records of handling historical leakage events and performance recovery data after maintenance; and cumulative operating time and temperature cycle count of power devices. The system stores the above data using a time-series database, supporting rapid querying by time dimension. Key features reflecting health status are extracted from the fused data, including leakage-related features such as the frequency of leakage warnings, the trend of single leakage duration, and the drift of the pressure-flow rate change ratio; performance degradation features such as the monthly average decrease in baseline pressure, the attenuation rate of flow response speed, and the increase in coolant circulation power consumption under the same operating conditions; and environmental features such as the leakage sensitivity coefficient under extreme temperatures and the performance recovery capability after high vibration conditions. Feature extraction uses a sliding time window to output a standardized feature vector.

10. The liquid-cooled charging module health status prediction system according to claim 9, characterized in that, The health status assessment unit also includes: Health status prediction is based on a prediction model using an improved long short-term memory network. The input layer receives the feature vector output by the degradation feature extraction module and the quantization parameters of the current leakage warning signal. The hidden layer strengthens key features through an attention mechanism, introduces a physical constraint layer, and corrects the model output based on the thermal balance equation of the liquid cooling system. The output layer generates a two-dimensional prediction result: a short-term health status score and a long-term remaining life prediction value for key components, along with confidence intervals. The prediction results are output as a visual assessment report, including a health status trend curve, high-risk feature annotations, and remaining life countdown. Tiered decision-making recommendations are generated based on the rating level. Output formats include local touchscreen display, data push to a remote management platform, and printed report generation.

Images