On-line cleanliness closed-loop control process and control system for liquid cooling pipeline

Abstract

The invention relates to the technical field of liquid cooling system operation and maintenance, and provides a liquid cooling pipeline online cleanliness closed-loop control method and system. The method comprises the following steps: firstly, carrying out topological modeling and segmentation on a liquid cooling pipeline, arranging cleanliness detection nodes such as granularity, pressure difference, flow and temperature at an inlet and / or an outlet of each segment, collecting multi-source operation data in real time, constructing a multi-parameter fusion cleanliness index after preprocessing, and realizing threshold value self-adaption by combining with the temperature. The method comprises the following steps: judging pollution sections and pollution types through cleanliness index gradient and pressure difference distribution, controlling execution nodes to switch valve positions, constructing a local cleaning loop between the pollution sections and a bypass pipeline, and selecting pulse flushing, flow direction reversal, bypass fine filtration and a combined cleaning mode according to pollution grades. In the cleaning process, the flow, the pulse parameters and the reversal period are adjusted in a self-adaptive mode according to the cleanliness index change rate, and after a termination criterion is met, cleaning is automatically quitted, and operation of a main loop is recovered. The system can write back whole-process data for threshold and strategy self-learning optimization, realizes refined online maintenance of the liquid cooling pipeline, and reduces shutdown and operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling system operation and maintenance technology, and in particular to a closed-loop control process and control system for online cleanliness of liquid cooling pipelines. Background Technology

[0002] Liquid cooling technology has become an important heat dissipation method in data centers, high-power servers, power electronic devices, and electronic control units for new energy vehicles in recent years, offering significant advantages in heat transfer efficiency, energy consumption, and system compactness. However, the piping of liquid cooling systems operates in a closed loop, inevitably carrying with it particles or soft contaminants such as welding slag, metal shavings, sealant fragments, rust products, and microbial slime. These contaminants accumulate in complex piping and cold plate microchannels, leading to increased local pressure drop, decreased heat exchange efficiency, valve jamming, and even leaks, severely impacting the long-term reliability of the liquid cooling system. Current engineering practices primarily control cleanliness through pre-flushing, offline filtration, and periodic coolant replacement, but lack a complete technical system for "online fine-grained management and closed-loop control of liquid cooling piping cleanliness under service conditions."

[0003] In the field of hydraulic and cooling pipeline cleaning, various cleaning systems targeting "high cleanliness" have been developed. Chinese patent CN113323945A discloses a high-cleanliness hydraulic pipeline cleaning system. This system achieves closed-loop cleaning of hydraulic pipelines through a combination of an oil tank, a pulse cleaning circuit, a static pressure testing circuit, and a circulating filtration circuit. A particle size detection device is integrated into the cleaning circuit for real-time monitoring of oil cleanliness. Cleaning ends when both the cleaning time and a specified cleanliness level are met. This system employs pulse cleaning, a series closed-loop circuit, and an external filtration unit, effectively removing adhering impurities from the pipes and improving pipeline cleanliness. However, this type of solution primarily targets initial cleaning or post-maintenance cleaning of hydraulic pipelines, typically performed under equipment shutdown conditions. Cleanliness is ensured through the overall operation of the entire cleaning circuit, without differentiating between different pipe sections with varying degrees of contamination during normal equipment operation. It also lacks the refined control capability to automatically adjust cleaning strategies and parameters based on online monitoring results.

[0004] CN105964632B discloses an online circulating cleaning process for hydraulic pipeline systems. This process targets complex, long, and large hydraulic pipeline systems. It involves disconnecting the hydraulic pump station, actuators, and control components, short-connecting the oil ports using temporary jumpers, dividing the pipeline into several loops according to the system schematic, and using specially molded valve blocks for online circulating cleaning to improve cleaning quality and efficiency. While this process is targeted in terms of loop division and cleaning process design, the cleaning process often requires manual configuration of the cleaning loops and relying on experience to set parameters such as cleaning time, flow rate, and pressure while the system is shut down. Regarding changes in cleanliness during the cleaning process, this literature mainly relies on sampling and detection before and after cleaning, making it difficult to achieve adaptive adjustment based on real-time cleanliness status. Furthermore, it does not address the issues of local contamination identification and segmented online maintenance during long-term system operation.

[0005] In the field of online monitoring of oil cleanliness, Chinese patent CN1877323A discloses an online oil contamination monitoring device. This device obtains the oil contamination level by rotating a differential pressure probe to sample and query a database. It can monitor the oil contamination level online during hydraulic system operation and generate alarm signals. Additionally, CN107560984A proposes an online oil contamination monitoring device and method for hydraulic systems. This method uses an oil particle counter combined with parameters such as pressure and temperature to correct the cleanliness results output by the sensor, improving monitoring accuracy. These technical solutions are relatively mature in terms of "online monitoring" and "cleanliness level assessment," providing data support for proactive maintenance. However, they typically only reach the level of alarms and manual decision-making, lacking deep coupling with pipeline structure models and cleaning actuators. They cannot form a closed-loop control link from "online detection" to "online cleaning" and "automatic recovery," and they do not address differentiated diagnosis and localized treatment for different pipeline sections.

[0006] In addition, there are some utility model patents, such as CN209550148U, which discloses a hydraulic pipeline cleaning device that integrates hydraulic oil line cleaning, filtration, and pressure testing onto a mobile device. It achieves cleaning and pressure testing of hydraulic pipelines through modules such as a high-pressure oil pump, pulse generator, filter, and bypass detection. While such devices improve the convenience and integration of cleaning operations, they are primarily geared towards centralized cleaning tasks during maintenance and construction phases, lacking the ability to continuously perceive and automatically intervene in the cleanliness status of the system during operation.

[0007] As data centers and high-power electronic devices evolve towards higher heat flux densities and greater integration, the microchannel structures, flexible hoses, and multi-branch loops in liquid cooling pipelines are significantly increasing, making the deposition and redistribution of contaminants in different sections more complex. Against this backdrop, there is an urgent need for an online cleanliness closed-loop control process and system that can integrate liquid cooling pipeline topology for multi-parameter online monitoring and fusion evaluation of cleanliness in each section, automatically locate contaminated sections and construct local cleaning loops, adaptively adjust cleaning parameters based on changes in the cleanliness index, and use historical operating data for threshold and strategy optimization. This would reduce downtime and overall liquid replacement frequency, improving the reliability and economy of liquid cooling system operation. Summary of the Invention

[0008] The technical objective of this invention is to provide an online cleanliness closed-loop control process and system for liquid-cooled pipelines. By combining pipeline topology segmentation modeling, multi-point and multi-parameter online detection, and cleanliness fusion evaluation, the invention achieves real-time monitoring and precise identification of the contamination status of each segment of the liquid-cooled pipeline. Based on this, it automatically constructs local cleaning loops and adaptively adjusts cleaning modes and parameters. Thus, without interrupting or minimizing the interruption of normal system operation, it promptly eliminates local contamination, maintains a long-term stable cleanliness level of the liquid-cooled pipeline, improves the heat exchange efficiency and operational reliability of the liquid-cooled system, and reduces maintenance costs.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A closed-loop online cleanliness control process for liquid-cooled pipelines is applied to the operation of circulating liquid cooling media in liquid-cooled pipelines. The process includes the following steps: S1, establishing a liquid-cooled pipeline topology model, dividing the liquid-cooled pipeline into at least two independent control segments, and arranging cleanliness detection nodes and execution control nodes at the inlet and / or outlet of each pipeline segment. The cleanliness detection nodes include at least one of particle size sensors and differential pressure sensors; S2, during system operation, real-time acquisition of flow rate, temperature, particle count data, and differential pressure data output by the cleanliness detection nodes for each pipeline segment, and performing noise reduction, outlier removal, and temperature compensation processing on the acquired data; S3, for each pipeline segment, constructing a multi-parameter fusion cleanliness index CI based on the particle count data and differential pressure data. i The cleanliness index CI i The particle concentration, pressure difference increment, and flow rate normalization value are weighted by preset weights, and the baseline threshold is updated online adaptively according to the type of liquid cooling medium and temperature; S4, the cleanliness index CI of each pipeline section is calculated. iCompared with the corresponding target cleanliness range, when the cleanliness index of one or more pipeline sections exceeds the upper limit threshold of the target cleanliness, the contaminated section and its degree of contamination are automatically determined based on the cleanliness index gradient and pressure difference distribution of adjacent sections; S5, for the determined contaminated section, the corresponding execution control node is controlled to isolate the contaminated section from the normal section liquid cooling circuit, construct a local cleaning circuit, and automatically select or combine at least one cleaning mode among pulse flushing, flow reversal flushing, and bypass filtration circuit according to the degree of contamination; S6, during the local cleaning process, the contaminated section is collected in real time. The system tracks the cleanliness index changes at the inlet and outlet, and automatically adjusts cleaning parameters based on the real-time value and rate of change of the cleanliness index. These parameters include pulse frequency, flushing flow rate, flow direction switching cycle, and online backwashing time of the filter unit. Cleaning is considered complete when the cleanliness index falls back to the target cleanliness range and the rate of change is below the threshold within a preset time window. S7: After cleaning, the system restores the connection between the contaminated section and the main liquid cooling circuit, and writes the cleanliness index data and corresponding cleaning parameters of each pipeline section as historical operating conditions into the closed-loop control model for subsequent self-learning optimization of the target cleanliness threshold and cleaning strategy under different operating conditions.

[0011] Preferably, the multi-parameter fusion cleanliness index CI in step S3 i The calculation includes: S3.1, classifying and statistically analyzing particle counts according to particle size range to form a particle level vector; S3.2, normalizing the pressure difference data to remove pressure difference disturbances caused by flow fluctuations; S3.3, linearly or nonlinearly fusing the particle level vector and the pressure difference normalization result according to preset weights to obtain the target cleanliness index; S3.4, the preset weights are based on historical operating data of the liquid cooling system and are trained offline or updated online using machine learning algorithms.

[0012] Preferably, the automatic determination of contamination segment in step S4 includes: S4.1, calculating the difference in cleanliness index and the difference in pressure between adjacent pipeline segments; S4.2, comparing the difference in cleanliness index and the difference in pressure with corresponding set thresholds respectively, and when both exceed the threshold, determining the corresponding pipeline segment as a priority contamination segment; S4.3, when multiple adjacent segments meet the contamination determination conditions, determining multi-level contamination segments in descending order of cleanliness index difference, and assigning different cleaning intensities and cleaning durations to each segment.

[0013] Preferably, the construction of the local cleaning circuit in step S5 includes: S5.1, separating the contaminated section from the main circuit by means of an electric three-way valve or an electric ball valve installed at both ends of each pipeline section, and forming a closed cleaning branch with an independent circulation pump and filter unit through a bypass pipeline; S5.2, the bypass pipeline is provided with at least one set of fine filter cartridges or magnetic trapping units for simultaneously removing solid particles and ferromagnetic impurities during the local cleaning process; S5.3, for pipeline sections with cold plates or microchannel structures, alternating forward and reverse flushing is achieved by switching valves.

[0014] Preferably, the automatic adjustment of cleaning parameters in step S6 includes: S6.1, when the rate of decrease of the real-time cleanliness index is less than the first threshold, increasing the pulse frequency and flushing flow rate; S6.2, when the real-time cleanliness index rises or fluctuates beyond the second threshold within a preset time, shortening the flow direction switching cycle and triggering online backwashing of the filter unit; S6.3, when the real-time cleanliness index remains within the target cleanliness range for multiple consecutive sampling cycles and the rate of change is less than the third threshold, gradually reducing the flushing flow rate and extending the pulse cycle to achieve energy-saving final cleaning.

[0015] Preferably, the closed-loop control model is a digital twin model based on the liquid-cooled pipeline topology. The digital twin model is used to: couple the geometric parameters, material, and flow resistance characteristics of each pipeline segment with the measured flow rate and cleanliness index for calculation; predict the changing trend of the cleanliness index under different cleaning modes and combinations of cleaning parameters; and automatically recommend the optimal cleaning mode and initial cleaning parameters for subsequent operating conditions.

[0016] Furthermore, the present invention also provides an online cleanliness closed-loop control system for liquid-cooled pipelines to realize the online cleanliness closed-loop control process of the liquid-cooled pipelines. The system includes: a liquid-cooled pipeline and a pipeline topology that divides the liquid-cooled pipeline into multiple pipeline segments; cleanliness detection nodes disposed at the inlet and / or outlet of each pipeline segment, the cleanliness detection nodes including at least two of particle size sensors, differential pressure sensors, and flow and temperature sensors; execution control nodes disposed at both ends of each pipeline segment and on bypass pipelines, the execution control nodes including electric valves and circulating pumps; a filter unit and its online backwashing assembly; and a controller electrically connected to the cleanliness detection nodes and execution control nodes; wherein the controller is configured to: a) divide the liquid-cooled pipeline into multiple pipeline segments according to the liquid-cooled pipeline topology information, and establish corresponding cleanliness index calculation models and target cleanliness intervals; b) receive the detection data output by each cleanliness detection node in real time, calculate the cleanliness index of each pipeline segment, and compare it with the corresponding target cleanliness interval; c) When the cleanliness index of one or more pipeline sections is detected to exceed the target cleanliness range, the control execution node constructs the corresponding local cleaning loop and outputs the cleaning mode and cleaning parameter control instructions; d) During the local cleaning process, the cleaning parameters are adjusted in real time according to the changes in the cleanliness index and the normal operation of the liquid cooling pipeline is restored after the cleaning is completed.

[0017] Preferably, the controller includes: a data acquisition module for acquiring and preprocessing multi-source data from particle size sensors, differential pressure sensors, flow sensors, and temperature sensors; a cleanliness assessment module for performing multi-parameter fusion cleanliness index calculation and automatic contamination segmentation determination; a cleaning strategy decision module for outputting local cleaning modes and initial cleaning parameters based on contamination levels and a digital twin model; a closed-loop control execution module for issuing control commands to each execution control node and adjusting cleaning parameters in real time during the cleaning process; a working condition recording and self-learning module for storing parameters and cleanliness change curves for each cleaning process and updating cleanliness thresholds and cleaning strategies based on historical data; and / or, the particle size sensor is an online liquid particle counter, and its measurement results are converted to ISO 4406. The NAS level may be used for cleanliness index calculation; the differential pressure sensors are respectively arranged at the front and rear ends of each pipeline segment to monitor the increase in local resistance caused by contamination; and / or, the liquid-cooled pipeline is used for cold plate or immersion liquid cooling systems for data centers or power electronic devices. The system also includes: an interface module that communicates with the liquid cooling system management platform to report the cleanliness index and cleaning status of each pipeline segment in real time, and to receive operating mode switching instructions and cleanliness strategy configuration parameters from the host computer; and a human-machine interaction terminal to display the liquid cooling pipeline topology, cleanliness distribution, and current and historical cleaning conditions, and to support manual triggering of local cleaning or modification of the target cleanliness range.

[0018] Furthermore, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0019] Furthermore, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the method.

[0020] This invention achieves quantitative and visual online assessment of the contamination status of each segment of the liquid-cooled pipeline by topologically segmenting the pipeline and arranging multi-source sensing nodes for particle size, differential pressure, flow rate, and temperature at the inlet / outlet of each segment. It also constructs a multi-parameter fusion cleanliness index based on particle concentration, differential pressure increment, and flow rate normalization. Through the linkage analysis of cleanliness index gradient and differential pressure distribution, this invention can accurately locate contaminated segments and their contamination levels in complex multi-branch pipelines, avoiding large-scale indiscriminate flushing. Furthermore, by automatically switching valve states through execution control nodes, it constructs local cleaning loops only for contaminated segments and adaptively combines various cleaning modes such as pulse flushing, flow reversal, and bypass fine filtration. Simultaneously, it dynamically adjusts the flushing flow rate, pulse frequency, flow reversal cycle, and filter unit backflushing time based on the real-time value and rate of change of the cleanliness index, ensuring that the cleaning process achieves the target cleanliness while optimizing time efficiency and energy consumption. By writing back the cleanliness curves and control parameters of each cleaning process to the closed-loop control model, this invention can automatically optimize the cleanliness thresholds and cleaning strategies under different media, temperatures and operating conditions over time, thereby achieving long-term online self-maintenance of the liquid cooling pipeline: on the one hand, it significantly reduces the operation and maintenance costs and system risks caused by overall shutdown for cleaning or frequent replacement of coolant; on the other hand, it effectively suppresses local blockage and heat exchange deterioration, maintaining high heat exchange efficiency and high reliability of the liquid cooling system throughout its entire life cycle. Attached Figure Description

[0021] Figure 1 Schematic diagram of the overall structure of the online cleanliness closed-loop control system for liquid-cooled pipelines.

[0022] Figure 2 Schematic diagram of pipeline segmentation and inspection / execution node layout.

[0023] Figure 3 Schematic diagram of the internal functional modules of the cleanliness closed-loop controller.

[0024] Figure 4 Schematic diagram of the closed-loop control method for online cleanliness of liquid cooling pipelines.

[0025] Figure 5 : Schematic diagram of a local cleaning circuit structure for a single contamination segment. Detailed Implementation

[0026] The preferred embodiments of the present invention will be further described clearly and completely below with reference to the accompanying drawings. It should be noted that the present invention is not limited to the specific embodiments described below, and any equivalent substitutions or modifications made by those skilled in the art without departing from the concept of the present invention should fall within the protection scope of the present invention.

[0027] I. Overview of the Implementation Examples

[0028] This embodiment provides a closed-loop control method and system for online cleanliness of liquid-cooled pipelines, applicable to liquid cooling systems for data center servers, liquid cooling systems for cold plates in power electronic devices, or other industrial scenarios employing liquid circulation cooling. The system performs topology modeling and segmentation of the liquid-cooled pipeline, arranging cleanliness detection nodes and execution control nodes at the inlet and / or outlet of each segment. It collects multi-source data such as particle count, differential pressure, flow rate, and temperature in real time, constructing a multi-parameter fused cleanliness index to finely assess the contamination status of each segment. When the cleanliness of a segment exceeds the limit, the system automatically determines the contaminated segment and contamination level, constructs a local cleaning loop, and selects an appropriate cleaning mode and parameters for online cleaning. During the cleaning process, the parameters are adaptively adjusted based on changes in the cleanliness index. After cleaning, the cleaning process data is written back to the control model for subsequent self-learning optimization of thresholds and control strategies, thereby achieving online self-maintenance of the liquid-cooled pipeline.

[0029] II. Terminology Explanation

[0030] To facilitate understanding, some of the terms used in this manual will be explained first:

[0031] Liquid cooling piping: refers to a closed piping system used to transport liquid cooling media (such as ethylene glycol aqueous solution, deionized water or other coolants), including main pipes, branch pipes, hoses, internal channels of cold plates, etc.

[0032] Pipeline segmentation: Based on topology modeling, continuous liquid cooling pipelines are divided into several sub-segments according to functional units or structural characteristics. Each segment has a relatively independent inlet and outlet, which facilitates independent assessment of cleanliness and performance of local cleaning.

[0033] Cleanliness detection node: refers to a sensor combination installed at the inlet and / or outlet of a pipeline section to detect parameters such as particulate contamination level, differential pressure, flow rate and temperature in the liquid cooling medium, including at least an online particulate size sensor and a differential pressure sensor.

[0034] Execution control nodes: These refer to actuators such as electric valves and circulating pumps installed on liquid cooling pipelines and bypass pipelines, which are used to switch liquid flow paths, construct local cleaning loops, and adjust cleaning flow rates according to control commands.

[0035] Cleanliness Index: A comprehensive quantitative indicator used to characterize the degree of contamination in a specific pipeline section. It is obtained by weighted fusion of multiple parameters such as particle count, pressure differential increment, and flow rate, and is denoted as [missing information]. ,in For segment numbering.

[0036] Local cleaning loop: When a segment is identified as a contaminated segment, the execution control node is switched to form a closed loop path that connects the segment, bypass pipeline and filter unit, and only the segment and its adjacent area are cleaned online.

[0037] Cleaning mode: refers to the operating mode of the local cleaning circuit, including pulse flushing, reverse flow flushing, bypass fine filtration or a combination thereof.

[0038] Self-learning optimization: refers to the process by which a control system iteratively updates cleanliness threshold parameters, weighting coefficients, and cleaning strategies based on historical cleanliness data and cleaning process data during long-term operation, thereby gradually optimizing control performance.

[0039] III. System Structure (Combined with) Figures 1-3 )

[0040] 3.1 Overall Structure Figure 1 )

[0041] like Figure 1 As shown, the liquid-cooled pipeline online cleanliness closed-loop control system of this embodiment includes: a main liquid-cooled loop 10; multiple pipeline segment units 20; cleanliness detection nodes 30 arranged at the inlet / outlet of each segment; execution control nodes 40 set on the pipeline and bypass; a filter unit and online backwashing assembly 50; a local cleaning bypass pipeline 60; a cleanliness closed-loop controller 70; and a human-machine interface terminal 80 connected to a host computer or liquid-cooled management platform. The main liquid-cooled loop 10 can be a conventional liquid-cooled system, including a liquid-cooled pump, main heat exchanger, expansion tank, main and branch pipes, cold plate assembly, etc. The controller 70 communicates with the cleanliness detection nodes 30 and execution control nodes 40 through an industrial network to realize data acquisition and control command issuance.

[0042] 3.2 Pipeline segmentation and node layout structure ( Figure 2 )

[0043] like Figure 2As shown, the piping in the main liquid cooling circuit 10 is divided into several piping segment units 20, for example: Segment 1: from the pump outlet to the main liquid supply trunk line; Segment 2: from the main liquid supply trunk line to the inlet of rack A; Segment 3: the liquid supply branch of the internal cold plate of rack A; Segment 4: the liquid return branch of the internal cold plate of rack A; Segment 5: the liquid return from rack A to the main liquid return trunk line; and so on to other racks and the main circuit section. Cleanliness detection nodes 30a and 30b are respectively set at the inlet and outlet of each segment 20. Node 30 includes an online particle size sensor 31 and a differential pressure sensor 32, and optionally also includes a flow sensor 33 and a temperature sensor 34. Execution control nodes 40 are set at the beginning and end of each segment and at the bypass connection points, including an electric ball valve 41, an electric three-way valve 42, a bypass circulation pump 43, etc. With the above arrangement, when a certain segment is determined to be a contaminated segment, it can be isolated from the main circuit by control valves 41 and 42, and a local cleaning circuit can be constructed through bypass pipeline 60 and circulation pump 43.

[0044] 3.3 Controller Structure ( Figure 3 )

[0045] like Figure 3 As shown, the cleanliness closed-loop controller 70 includes:

[0046] Data acquisition module 701: responsible for periodically acquiring data such as particle count, differential pressure, flow rate, and temperature from cleanliness detection node 30; Data preprocessing module 702: responsible for synchronization, filtering, outlier removal, temperature compensation, and differential pressure baseline processing; Cleanliness assessment module 703: calculates the cleanliness index of each segment based on a multi-parameter fusion model. ;Pollution segmentation determination module 704: Based on The system compares the contamination level with thresholds and uses gradient analysis to determine the contamination segment and level. The cleaning strategy decision module 705 selects the local cleaning mode and initial cleaning parameters based on the location and degree of contamination. The closed-loop control execution module 706 issues instructions such as valve action and pump speed setting to the execution control node 40 and dynamically adjusts the cleaning parameters during the cleaning process. The operating condition recording and self-learning module 707 records cleaning operating condition data and performs parameter statistical analysis and strategy optimization. The controller 70 connects to the host computer / liquid cooling management platform 80 via a communication interface to display operating status, configure parameters, provide alarm prompts, and perform remote maintenance.

[0047] IV. Method Implementation Path – Steps S1 to S7

[0048] The following is combined with Figure 4 The flowchart shown illustrates the specific implementation of steps S1 to S7 of the method of the present invention.

[0049] S1: Liquid Cooling Piping Topology Modeling and Segmentation, Node Arrangement

[0050] During the system deployment phase, engineers first perform topology modeling of the liquid cooling piping based on the liquid cooling system design drawings and the on-site piping layout. This includes the following steps:

[0051] 1. Topology node identification

[0052] By reading CAD piping diagrams, BIM models, or P&ID drawings, identify key locations as topology nodes, such as: liquid-cooled pump outlets, main heat exchanger inlets / outlets; distribution / collection manifolds of the main supply and return manifolds; inlets, outlets, and internal cold plate manifolds for each rack; key valves, bypass connection points, etc. Define these locations as nodes in the diagram structure and assign them unique identifiers.

[0053] 2. Topological edge creation and attribute assignment

[0054] A continuous pipe segment between two adjacent nodes is defined as a topological edge, and its geometric parameters (pipe diameter, pipe length), material, design flow rate, and theoretical resistance are recorded. The controller 70 stores this graph structure as a pipeline topology model for subsequent segmentation and positioning calculations.

[0055] 3. Definition of pipeline sections

[0056] Based on the topology model, and considering maintenance needs and cleaning controllability, the pipeline is divided into sections. The principles include: each section should ideally cover similar equipment (e.g., all cold plate loops on a rack) or the same functional unit (e.g., main trunk and branch sections); cleanliness monitoring nodes and control valves should be installed at both ends of each section if possible; easily contaminated cold plate microchannels and return liquid sections with extremely low flow rates should be prioritized for independent sectioning. After division, the sections are numbered. Label each segment and record its upstream and downstream adjacent relationships.

[0057] 4. Cleanliness testing node layout

[0058] For each segment, a cleanliness detection node 30 is installed at its inlet and / or outlet. The configuration of the node 30 can be selected as follows:

[0059] Online particle size sensor: Detects particle size distribution and particle count in liquids using optical, laser scattering, or impedance methods;

[0060] Differential pressure sensor: measures the pressure difference across the two ends of a segment, reflecting changes in local resistance;

[0061] Flow sensor: monitors the actual flow rate of this segment and is used for flow rate normalization of the cleanliness index;

[0062] Temperature sensor: Used for temperature compensation and threshold adaptation of sensor output.

[0063] In practical engineering, for cost-sensitive sections, a single integrated node that includes particle size and differential pressure measurements can be installed on the outlet side.

[0064] 5. Execution control node deployment

[0065] An electric ball valve 41-1 is installed at the inlet of each section, and an electric ball valve 41-2 is installed at the outlet, for connecting or isolating the section from the main circuit. Electric three-way valves 42-1 and 42-2 are installed at the connections between each section and the bypass pipeline 60, for switching between "main circuit mode" and "bypass local cleaning mode". A bypass circulation pump 43, a regulating valve, and necessary pressure testing ports are installed on the bypass pipeline 60.

[0066] Through the above steps, the system completes the topology modeling and segmentation of the liquid cooling pipeline, and maps the cleanliness detection nodes and execution control nodes into the topology model, laying the foundation for subsequent refined control.

[0067] S2: Real-time acquisition and preprocessing of multi-source data

[0068] During system operation, the data acquisition module 701 collects raw data from each cleanliness detection node 30 according to a preset sampling period (e.g., 1s to 60s), specifically including particle count, pressure difference, flow rate, and temperature. To improve the reliability of subsequent evaluations, preprocessing is required, as follows:

[0069] 1. Time synchronization and data caching

[0070] The controller 70 periodically sends a sampling synchronization signal, or synchronizes the clocks of each node via a network time protocol. During each sampling, all sensor data is added with a unified timestamp and stored in a buffer, and then categorized according to segment number.

[0071] 2. Noise filtering

[0072] For particle count signals, moving average or median filtering methods are used to eliminate instantaneous fluctuations. For example, the most recent... Calculate the average of the sampled values:

[0073] ,

[0074] in, For segmentation Particle size range At any moment Smooth particle count, For the original count, This is the length of the sliding window. A first-order low-pass filter or median filter can be used for the differential pressure signal to filter out sharp transients caused by flow fluctuations.

[0075] 3. Outlier Identification and Removal

[0076] If a sampled data point exceeds the reasonable physical range (e.g., particle count exceeds the sensor's upper limit, pressure difference is negative, or the change is abnormally large), the sampled value is marked as abnormal, and the value is replaced by interpolation between two normal values ​​to avoid distortion in the cleanliness index calculation.

[0077] 4. Temperature compensation and flow correction

[0078] Since particle size sensor readings are affected by temperature and flow rate, the preprocessing module performs temperature compensation on the particle count based on the temperature sensor reading and flow rate correction on the particle count based on the flow sensor reading. Those skilled in the art can implement this based on calibration curves or empirical formulas provided by the sensor manufacturer.

[0079] 5. Establishment and updating of differential pressure baseline

[0080] After the system is initially put into operation or after a thorough cleaning, the differential pressure values ​​of each section under clean conditions are recorded as the baseline differential pressure. Real-time pressure difference during operation The difference from the baseline value is taken as the pressure differential increment: ;in, Reflecting segmentation Additional resistance caused by contaminant deposits.

[0081] Through the above processing, the effective data set of each segment in each sampling period is obtained. This information is used for the next step of cleanliness assessment. All of the above steps are based on conventional signal processing methods and are easily implemented by those skilled in the art.

[0082] S3: Construction of Multi-parameter Fusion Cleanliness Index and Threshold Adaptation

[0083] 3.1 Input Data and Variable Definitions

[0084] After completing the S2 data preprocessing, for each sampling period, each pipeline segment is numbered as follows: The following valid data can be obtained (at time...) ):

[0085] Particle size range Average particle count: ,

[0086] Segmented pressure differential increment: (Baseline has been subtracted)

[0087] Segmented volumetric flow rate: ,

[0088] Segmented medium temperature: ,

[0089] Meanwhile, the parameters preset or maintained by the system include:

[0090] Differential pressure reference value: ,

[0091] Traffic reference value: ,

[0092] Weights for different particle size ranges: ,

[0093] The fusion weights of the three factors—particle size, pressure difference, and flow rate—are as follows: (satisfy ),

[0094] Initial cleanliness target threshold: ,

[0095] Temperature reference value and temperature sensitivity coefficient: .

[0096] S3.2 Particulate Pollutants Calculation

[0097] 1) Constructing granular vectors

[0098] For segmentation At any moment Constructing granular-level vectors:

[0099] ;

[0100] in, Segmentation At any moment The particle count vector; Segmentation In particle size range Average number of particles per unit volume.

[0101] 2) Weighted nonlinear mapping by particle size interval

[0102] Considering that large particles are more sensitive to clogging and wear risks, a logarithmic weighted summation is used to obtain the particulate contamination factor. :

[0103]

[0104] in, Segmentation At any moment Scalar indicators contributed by particulate pollution; Particle size range The weighting coefficient is such that the larger the value, the greater the impact of the particle size on the risk of clogging. Compress the number of particles to avoid a single interval's maximum value dominating the overall result. Engineering implementation suggestions: Three to five particle size ranges can be selected, for example , , wait; It can be set empirically to an increasing sequence that increases with particle size, such as... It can also be fine-tuned later through S7 self-learning.

[0105] S3.3 Pressure Difference Factor Calculation

[0106] 1) Pressure difference normalization

[0107] Normalize the differential pressure increment to the reference value:

[0108] ;

[0109] in, Segmentation The increment after subtracting the baseline pressure difference from the real-time differential pressure; : The reference differential pressure selected in the project (such as the design value or the allowable differential pressure increment); : Normalized pressure difference factor.

[0110] 2) Non-negative cutoff and magnification factor

[0111] To avoid the negative impact of negative values ​​(such as short-term pressure drop caused by flow rate changes) on pollution indicators, negative values ​​are truncated and multiplied by an amplification factor. :

[0112] ;

[0113] in, Segmentation Pressure difference contribution factor; : The coefficient for adjusting the differential pressure as a "sensitivity" factor in the overall cleanliness index. Engineering implementation suggestion: If the system is more concerned with microchannel clogging, this coefficient can be appropriately increased. If the flow rate fluctuates significantly and is prone to noise, it can be reduced. And it works in conjunction with a stronger filter in S2.

[0114] S3.4 Flow Factor Calculation

[0115] Particle deposition is more likely to occur when the flow rate is too low, so an additional penalty is imposed when the flow rate is below the reference value.

[0116] 1) Flow normalization

[0117] ;

[0118] in, Segmentation Real-time volumetric flow rate; Reference flow rate (e.g., design rated flow rate); Normalized flow (usually in) scope).

[0119] 2) Deposition risk penalty function

[0120] when When, a penalty term is introduced; when At the same time, without increasing risk:

[0121] ;

[0122] in, Segmentation Traffic contribution factor; : Flow risk weight, which can be set to a larger value when the system is more sensitive to "low flow rate deposition".

[0123] S3.5 Multi-parameter Fusion Cleanliness Index Calculation

[0124] The particulate factor, pressure difference factor, and flow rate factor are combined into a single cleanliness index:

[0125] ;

[0126] in, Segmentation At any moment Multi-parameter fusion cleanliness index; : Granularity factor weight, usually taken as ; Pressure differential factor weight, usually taken as ; : Flow factor weight, usually taken as ; Project implementation path: In the controller parameter configuration interface, preset... Parameters; calculated from the data output by S2 in each sampling period. ; calculated using the above linear combination formula ;Will It is stored in a circular buffer and can be used by S4 to calculate the gradient and S6 to calculate the rate of change.

[0127] The above process uses only basic operations such as addition, subtraction, multiplication, division, logarithms, and maximum values, which are easy to implement on PLCs or industrial PCs. All parameters can be configured through the host computer interface and can be slowly adjusted in S7 based on historical data, thereby completing the construction of multi-parameter fusion cleanliness index and threshold adaptive control without increasing the complexity of implementation.

[0128] S4: Contamination segmentation determination based on cleanliness index gradient and pressure difference distribution

[0129] This method is based on the assumption that the cleanliness index of each segment has already been calculated in S3. and pressure differential increment Based on this, spatial positioning is achieved through simple difference and threshold comparison.

[0130] S4.1 Input Data and Symbol Conventions

[0131] In each sampling period The controller has received the following data from S3: each segment Cleanliness index: Each segment Temperature adaptive upper limit threshold: Each segment Temperature adaptive lower threshold: Each segment Normalized value of differential pressure increment: Additionally, from the topology model, we can obtain: each segment The upstream adjacent segment set: Each segment The downstream adjacent segment set: .

[0132] To simplify the description, the following example uses a typical series system with "single input and single output", i.e., segmented. There is only one upstream segment and a downstream segment When there are multiple branches in an engineering project, the same judgment logic can be applied independently to each branch. The following judgment parameters are preset: Cleanliness gradient threshold: Pressure gradient threshold: Pollution level threshold: (Boundary between mild / moderate / severe).

[0133] S4.2 Preliminary Over-Limit Segmentation Screening

[0134] In each sampling cycle, a "one-dimensional over-limit judgment" is first performed on all segments to screen out candidate contaminated segments.

[0135] For each segment Calculate the excess amount: ,in, For segmentation At any moment The cleanliness level exceeds the limit; if Then it is considered to be segmented. The current limit has not been exceeded.

[0136] Forming a candidate set: if Then it will be segmented. Add to the "candidate contamination segment set" ;like Then segment This indicates that the current cleanliness level meets the standard. This step is based on a single point only. Compared to its threshold, its complexity is extremely low, providing a set of objects for subsequent spatial positioning.

[0137] S4.3 Cleanliness gradient calculation of adjacent segments

[0138] For each candidate segment The difference in cleanliness index between the segment and its adjacent segments is calculated to determine the spatial distribution trend of pollution.

[0139] In the case of single-input and single-output, the upstream segment is denoted as... The downstream segment is (If one end does not exist, only the side that exists will be counted):

[0140] upstream cleanliness gradient: ;in, To segment from upstream Up to the current segment The cleanliness level "leap"; when And if it is relatively large, it indicates that it has entered a segmentation phase. Post-contamination pollution increased significantly.

[0141] Downstream cleanliness gradient: ,in, To start from the current segment Downstream segment The change in cleanliness; when And when the absolute value is large, it indicates segmentation. It is relatively dirtier than the downstream. If the system has multiple upstream or downstream branches (such as a confluence or branch structure), the calculation can be performed separately for each adjacent segment, and the maximum absolute value or a flow-weighted average can be taken. This is a standard engineering process and will not be elaborated on here.

[0142] S4.4 Calculation of pressure gradient between adjacent segments

[0143] To avoid relying solely on Gradient misjudgment requires auxiliary pressure difference information. For each candidate segment... Calculate the pressure difference increment with the adjacent segment:

[0144] Upstream pressure gradient: ,in, To segment from upstream To segment The normalized increment of the pressure difference; when it is positive and large, it indicates that the segmentation has begun. The local resistance increased significantly afterward.

[0145] Downstream pressure gradient: ,in, To segment Downstream segment The normalized increment of the pressure difference; when it is negative and has a large absolute value, it indicates segmentation. The resistance is higher than that downstream.

[0146] Pressure gradient can help distinguish between pressure gradients caused by changes in flow rate. "Small differences" and "real pollution concentration caused by sediment blockage".

[0147] S4.5 Joint Judgment Rule – Determining the Pollution Type (Uniform / Concentrated)

[0148] Yes After applying the gradient and pressure gradient, each candidate segment can be... Execute joint judgment:

[0149] 1. Uniform contamination segmentation

[0150] If the following conditions are met: (Exceeded limits); and ; and Then determine the segmentation. The area and its surrounding region have a high overall pollution level but a relatively uniform spatial distribution, which is called a "uniform pollution segment". This type of segment is suitable for a milder cleaning mode (such as bypass fine filtration + medium intensity pulse).

[0151] 2. Forward concentrated pollution segmentation (pollution increases sharply from upstream to this segment)

[0152] If the following conditions are met: ; (from arrive (Cleanliness deteriorated significantly) (from arrive If the resistance increases significantly, then segmentation is determined. The term "forward-concentrated pollution segmentation" indicates that pollution deposition mainly occurs in these segments. Near the inlet or inside the segment, it is suitable to enhance the local flushing intensity and frequency of that segment.

[0153] 3. Downstream concentrated pollution segmentation (this segment → downstream pollution mitigation)

[0154] If the following conditions are met: ; (from arrive (Cleanliness significantly improved) (from arrive If the resistance is significantly reduced, then segmentation is determined. The term "reverse concentrated pollution segmentation" describes the segmentation process. The microchannels or local structures have the greatest resistance and the heaviest contamination, which may require a combination of flow reversal and strong pulse.

[0155] 4. Serialization determination of multiple candidate segments

[0156] In actual engineering projects, it is possible for multiple consecutive sections to exceed the limits (e.g.) All in (In the middle). At this point, the following strategy can be adopted: First, perform the above joint determination on each candidate segment; classify the continuous region as the main pollution segment according to the "maximum gradient point", and the rest as "secondary pollution segments"; in S5, prioritize the construction of local cleaning loops for the main pollution segments, and extend to adjacent secondary pollution segments for cleaning if necessary.

[0157] S4.6 Pollution Level Classification and Output Results

[0158] After determining the type of pollution, it is also necessary to determine the extent of exceeding the limit. The severity of contamination is classified and used for selecting the S5 cleaning mode and initial setting of S6 parameters.

[0159] Set two level thresholds: ,For example: The dividing line between mild and moderate; : The dividing line between moderate and severe.

[0160] For each candidate segment Classified according to the following formula: If ,but It is classified as "lightly polluted"; if ,but It is classified as "moderately polluted"; if ,but It is classified as "severely polluted".

[0161] The final output to S5 includes: pollution segment number: Pollution type: uniform / forward-concentrated / backward-concentrated; Pollution level: light / moderate / heavy; Adjacent segment numbers and their gradient data (optional). This information will directly drive the "local cleaning loop construction" and "preliminary selection of cleaning mode" in S5.

[0162] Through the above specific implementation methods, step S4 can not only accurately locate the pollution segments and their concentration directions in space, but also distinguish the degree of severity, providing clear input conditions for S5 and S6.

[0163] S5: Construction of Segmented and Localized Cleaning Circuits and Selection of Cleaning Modes

[0164] This step takes the judgment result of S4 as input. Based on the known pollution segment number, pollution type and pollution level, it constructs a local cleaning loop on the physical pipeline by controlling the execution control node, and completes the selection of cleaning mode and initial parameter setting according to the rule base.

[0165] S5.1 Step Input and Basic Conditions

[0166] In each control cycle, the output from S4 includes at least the following information:

[0167] Pollution segment number: ;

[0168] Contamination types: uniform contamination, up-concentrated contamination, down-concentrated contamination;

[0169] Pollution levels: light, medium, heavy;

[0170] Meanwhile, the topology model and system architecture provide:

[0171] Segmentation Inlet valve number: e.g., inlet electric ball valve ;

[0172] Segmentation Outlet valve number: such as outlet electric ball valve ;

[0173] With segmentation Bypass three-way valve connected at both ends: , ;

[0174] The positional relationship between the circulation pump 43 and the filter unit 50 on the bypass pipeline 60;

[0175] Based on this, the controller performs local cleaning loop construction and cleaning mode selection.

[0176] S5.2 Construction of Localized Cleaning Circuit for Segmented Contamination

[0177] S5.2.1 Main circuit isolation and bypass access

[0178] By segmentation For example, the closed-loop control execution module switches valve positions in the following order (this can be achieved through a preset "action sequence"):

[0179] 1) Reduce the flow rate of this branch (optional pre-operation)

[0180] For branches carrying critical loads, the flow rate can be gradually reduced by adjusting the upstream regulating valve (if applicable) or decreasing the opening of the branch control valve. Reduce the main circuit flow and minimize the impact of instantaneous switching on system voltage drop and flow balance.

[0181] 2) Close the segmented inlet / outlet ball valves

[0182] The controller issues a command to shut down and To make segments Disconnect from the main circuit series path. At this time, the main liquid cooling circuit will no longer supply liquid to this section, and the bypass pipeline will bear the subsequent cleaning flow.

[0183] 3) Open the bypass three-way valve and connect the bypass pipeline.

[0184] The three-way valve at the segment inlet Switch to the "Bypass-Segment" channel and turn on the three-way valve at the segment outlet. Switch to the "Segment-Bypass" channel to allow the bypass pipe 60 to enter the segment from one end. It flows back from the other end to the bypass pipeline, forming a closed circuit.

[0185] S5.2.2 Bypass circulation and filter unit connection

[0186] After completing the valve position switching, the controller connects the filter unit to the local circuit according to the preset cleaning mode:

[0187] 1) Start the bypass circulation pump 43

[0188] Set initial speed Or flow rate setting This allows coolant to enter the section via bypass pipe 60. The entrance is divided into sections. It then returns from the outlet to the bypass pipeline 60, forming a local circulation loop.

[0189] 2) Filter units connected in series

[0190] For applications requiring rapid particle removal, the coarse and fine filters are connected in series in the local loop. For example: Bypass 60 → Coarse Filter → Fine Filter → Segmentation. →Bypass 60, the flow path is connected by controlling the corresponding inlet and outlet valves.

[0191] 3) Magnetic trapping unit access (optional)

[0192] If there are a large number of ferromagnetic particles in the system, a magnetic trap can be set before or after the fine filter to specifically capture abrasive contaminants.

[0193] Through the above operations, without significantly affecting the normal operation of other segments, the segmentation is completed. Establish a relatively independent and controllable local cleaning circuit.

[0194] S5.3 Cleaning Mode Library and Selection Logic

[0195] This invention defines three basic cleaning modes and one combined mode. The controller maintains a "mode selection rule table" to select modes based on the type and level of contamination.

[0196] S5.3.1 Basic Cleaning Mode Definition

[0197] Mode M1: Bypass Fine Filtering Mode

[0198] Main action: Maintain a small, stable flow rate and circulate the filter over a long period to gradually remove fine particles; there is no need to drastically change the flow direction and flow rate.

[0199] Applicable scenarios: Branches with light or uniform pollution, where minimal disturbance to the system is required.

[0200] Mode M2: Pulse flushing mode

[0201] Main action: Periodically change the speed of the circulating pump or the opening of the throttle valve in the local loop to make the flow rate form periodic high and low changes (flow velocity pulses), generate impact shear force, and peel off the deposits in the pipe wall and cold plate channel.

[0202] By setting the pulse period High flow and low flow Control the rinsing intensity.

[0203] Mode M3: Reverse flow flushing mode

[0204] Main action: Periodically switch the flow direction of segmented inlet / outlet. By controlling a set of switching valves, the fluid alternates between "normal flow direction" and "reverse flow direction" to flush out dead corner deposits that are difficult to remove under conventional unidirectional flow.

[0205] It is particularly suitable for segmentation with complex structures such as cold plate microchannels and sharp bends.

[0206] Mode M4: Comprehensive Enhancement Mode (Combined Mode)

[0207] Using M2+M3 in combination with M1 simultaneously or in stages: This method combines pulsed flow with reversed flow direction, and the fluid passes through a fine filter in each cycle, achieving a combined cleaning of "strong flushing + continuous filtration". It is suitable for heavily polluted sections with a high risk of concentrated blockage.

[0208] Example of mode selection rules in S5.3.2

[0209] A set of rules can be pre-configured in tabular form in the controller, for example (this is just an example and can be optimized according to actual needs):

[0210] Pollution type / level Mild Medium Heavy Uniform M1 M2+M1 M4 Up-concentrated M2+M1 (weak) M2+M1 (Middle) M4 (High Strength) Down-concentrated M2+M3 (weak) M2+M3+M1 M4 (High Intensity + Reverse Rotation)

[0211] After receiving the "contamination type" and "contamination level" output by S4, the controller directly queries the rule table to obtain the corresponding cleaning mode combination and intensity level, which serves as the basis for subsequent parameter settings.

[0212] S5.4 Initial Cleaning Parameter Settings

[0213] After the mode is determined, initial control parameters need to be assigned to each mode. These parameters can be determined by engineering experience or historical data, or they can be gradually optimized during S7 self-learning.

[0214] S5.4.1 Bypass Fine Filtering Mode Parameters

[0215] For M1, the main parameter is the local circulation flow rate setpoint. A simple empirical formula can be used: ,in, This represents the initial circulation flow rate of the local cleaning circuit. This is the design flow rate for this segment under normal operating conditions; This is a proportionality coefficient; it can be taken as [value] for light pollution. During moderate to heavy pollution, it can be taken To avoid insufficient flow rate leading to low cleaning efficiency.

[0216] S5.4.2 Pulse Flushing Mode Parameters

[0217] For M2, define the pulse period and upper / lower limits of the flow rate:

[0218] Pulse period: (e.g., 10-30 seconds);

[0219] High flow: Low flow: .

[0220] A simple setting is as follows: ,in, ,For example Used to generate high-velocity scouring; ,For example This allows particles time to be carried into the filter unit during the low flow rate phase; larger units can be selected for higher contamination levels. and smaller and shorten appropriately Pulse control can be achieved by periodically changing the speed of the bypass circulation pump 43, or it can be combined with the action of the regulating valve.

[0221] S5.4.3 Flow Reversal Mode Parameters

[0222] For M3, the main parameter is the flow direction switching cycle. And forward / reverse flow settings. General settings: 30~120s, set according to the volume of the cold plate and the expected flushing cycle; the forward and reverse flow rates can be the same or slightly higher than the normal operating value to improve the flushing effect.

[0223] The following simple rules can be adopted:

[0224] Moderate pollution: Take the median value, such as 60s;

[0225] Severe pollution: Use a shorter time, such as 30 seconds, to increase the frequency of backwashing.

[0226] In practice, by controlling the valve positions in groups, the segmented inlet and outlet can be switched between "main bypass direction A→B" and "reverse B→A".

[0227] S5.4.4 Integrated Enhancement Mode Parameters

[0228] The parameters for mode M4 can be directly selected from those for M2 and M3, choosing the "high intensity" level, for example: Pick Double the design flow rate; Take 10-15 seconds; Take about 30 seconds; The flow rate should be close to or slightly higher than the design flow rate. These initial parameters are only a starting point; S6 will automatically adjust them based on changes in the cleanliness index.

[0229] S5.5 multi-segment joint cleaning (optional implementation)

[0230] When S4 determines that multiple consecutive adjacent segments are contaminated segments and the gradient change is not significant, it can choose to perform joint cleaning on multiple segments. Two simple strategies can be used in this embodiment:

[0231] 1) Series-combined cleaning

[0232] Segmentation and Treat it as an extended segment, uniformly close the inlet / outlet valves at both ends of the extended segment, and connect the bypass loop across the entire extended segment; in the topology, only the extended "equivalent segment" is treated as a new object, and the above mode and parameter rules are applied.

[0233] 2) Segmented sequential cleaning

[0234] Select segments for local cleaning in descending order of the degree of exceeding limits; after the current segment is cleaned and the main circuit is restored, switch to the next segment.

[0235] The specific strategy to be selected can be configured by the host computer or maintenance personnel based on the importance of the system and the acceptable level of disturbance; this invention does not impose any restrictions.

[0236] Through the above implementation, S5 completes the key steps in the control logic from "passively identifying contamination" to "actively constructing a local cleaning loop and selecting a cleaning mode", providing complete initial conditions for the dynamic adjustment of S6.

[0237] S6: Closed-loop regulation and termination criteria for the cleaning process

[0238] This step involved segmenting the pollution in the previous stage, S5. With the establishment of a local cleaning loop, selection of cleaning mode and initial parameters as prerequisites, the rinsing intensity is adaptively adjusted by online monitoring of the cleanliness index and calculation of its changing trend, and the cleaning process is automatically terminated when the termination conditions are met.

[0239] S6.1 Input Conditions and Running Status Definitions

[0240] Initiating a pollution segment During a localized cleaning task, the controller already possesses the following information: Segment number: Current cleaning mode combination: such as M1 / M2 / M3 / M4 (see S5), initial cleaning parameters: such as local circulation flow rate Pulse period Flow reversal cycle Cleanliness index and threshold: , , (Obtained from S3); meanwhile, S2 / S3 can be used in each sampling period. Provide the latest: Current cleanliness index Normalized differential pressure increment Normalized flow, :temperature.

[0241] The controller maintains a "cleaning task state machine" during the cleaning process. Typical states include: RUNNING_ENHANCE (enhanced cleaning phase), RUNNING_STABLE (stable cleaning phase), WAITING_TERMINATE (termination conditions met, entering the final stage), and FINISHED (cleaning completed).

[0242] S6.2 Calculation of the rate of change of cleanliness index

[0243] To determine the cleaning effectiveness and adjust the intensity accordingly, it is necessary to estimate the rate of change of the cleanliness index over time. In each sampling period, the percentage of segments... calculate:

[0244] ,

[0245] in, Current sampling period time The cleanliness index; Cleanliness index of the previous sampling period; Sampling period (e.g.) ); At any moment The discrete-time rate of change estimate of the cleanliness index. The controller can use the data from the most recent sampling points. Stored in a circular buffer for smoothing trend judgment (can also be used for...) Perform another moving average.

[0246] S6.3 Threshold and Logic Parameter Settings

[0247] To achieve closed-loop regulation, several threshold parameters need to be preset:

[0248] First rate of change threshold: (Judging from "the descent is very slow or not at all")

[0249] Second rate of change threshold: (Judging from the statement that the situation has "stable out"),

[0250] Third rate of change threshold: (Used as a termination criterion, with stricter requirements) ),

[0251] Termination Criterion Time Window: (like );

[0252] These parameters can be set by engineers during the debugging phase, or they can be gradually optimized in S7 using historical data.

[0253] S6.4 Closed-Loop Regulation Strategy – Enhancement / Stability / Abnormal Phases

[0254] S6.4.1 Enhanced Cleaning Phase (RUNNING_ENHANCE)

[0255] When local cleaning is initiated, it typically enters enhanced cleaning mode first. The typical decision logic is: if... and If the current cleanliness level is still above the standard and the rate of decrease is too slow, then it is considered that "the cleaning intensity needs to be increased".

[0256] Examples of enhancement strategies:

[0257] 1) Increase local circulation flow

[0258] The target flow rate or speed of the bypass circulation pump is changed from Upgraded to: ,in For example, the gain coefficient. .

[0259] 2) Increase pulse intensity (if using M2)

[0260] Increase pulse high flow : .

[0261] Reduce pulse period : in Coefficients less than 1 (e.g.) ), indicating a gradual increase in intensity.

[0262] 3) Shorten the flow reversal cycle (if using M3 / M4)

[0263] Will Reduce, for example: ,in For the shortest acceptable reversal period (e.g.) ), It's a reduction in proportion.

[0264] When repeated enhancement of regulation still does not show significant improvement (e.g., persistent) After one cycle If the value is still far above the threshold, an alarm can be triggered to prompt maintenance personnel to check for any abnormalities (such as severe blockage of the cold plate or saturation of the filter).

[0265] S6.4.2 Stable Cleaning Phase (RUNNING_STABLE)

[0266] As the cleanliness gradually approaches the target range, the logical judgment can be: If and If so, it is considered that the cleaning process has entered a "stable phase".

[0267] At this stage, the strategy mainly focuses on reducing energy consumption and minimizing disturbances to other parts of the system: appropriately reducing the circulation flow. For example, decreasing by a certain percentage; increasing the pulse period. Alternatively, reduce the high flow rate amplitude to weaken the impact; appropriately extend the flow reversal period. This reduces valve switching frequency and extends valve life. Simultaneously, records are kept of the valves during this stable phase. Fluctuations provide a basis for termination criteria.

[0268] S6.4.3 Abnormal Fluctuation Stage (Abnormal Handling)

[0269] If any of the following situations occur during the cleaning process: A short-term, obvious reversal and rise Continue to increase; or Frequent and significant fluctuations, with the differential pressure showing a continuous upward trend, may indicate: filter unit blockage, leading to a sharp increase in pressure drop; or the re-accumulation of contaminant clumps removed during cleaning in the loop. In this case, closed-loop control can perform the following: pause the current pulse / reverse operation, maintaining a moderate and stable flow rate; trigger the filter unit's backwashing process: switch the filter inlet / outlet valve positions, allowing the cleaning fluid to backwash the filter element and discharge contaminants to the drain line; after backwashing is complete, resume the local cleaning mode and re-enter the enhanced or stabilization phase.

[0270] S6.5 Cleaning Termination Criterion – Two-Condition Method

[0271] To avoid "insufficient cleaning" or "over-cleaning," this invention employs a dual-condition termination criterion:

[0272] 1) Conditions for long-term cleanliness compliance

[0273] At some point Before the inspection Cleanliness index sequence within a time window (e.g., the past 30 minutes) If the following conditions are always met within this window: ;in, , : Real-time threshold under temperature adaptation; the above formula means "cleanliness is always maintained within the target range throughout the entire time window".

[0274] 2) Steady-state condition of rate of change

[0275] Within the same time window, the rate of change satisfies: ,in, The third rate of change threshold is usually set to be higher than the third rate of change threshold. The smaller value is used to confirm that "cleanliness has essentially stopped fluctuating." Only when both conditions are met simultaneously does the controller consider the cleaning of that segment to be sufficiently complete and allow it to proceed to termination. This design ensures both compliance and stability, preventing accidental termination due to short-term fluctuations and avoiding unnecessary extension of cleaning time.

[0276] S6.6 Cleaning End and Final Process

[0277] Once the termination criterion is met, the closed-loop control execution module changes the cleaning task status from RUNNING_STABLE to WAITING_TERMINATE and performs the following cleanup operations:

[0278] 1) Gradual flow rate reduction and stop pulse / reverse

[0279] Gradually reduce the local circulation flow rate in several small steps. Set the flow rate to a smaller value to avoid pressure drop fluctuations caused by sudden stops; at the same time, stop pulse control and flow reversal control, and switch the circulation pump to stable low flow operation for a short period of time (such as 1 to 5 minutes) to ensure that residual particles are carried out and intercepted by the filter unit.

[0280] 2) Bypass circuit deactivation and main circuit restoration

[0281] Close the three-way valve connecting the bypass and the sectional section. , Switch back to "main circuit straight-through" mode; enable segmentation. Inlet / outlet ball valve , Reconnect the segment to the main liquid cooling circuit; gradually increase the target flow rate of this branch of the main circuit to the design value or the flow rate required for the current operating conditions.

[0282] 3) Stop the local circulation pump

[0283] When the segmentation is confirmed After the main circuit has been fully restored to normal flow, stop the bypass circulation pump 43 and close the auxiliary valves related to local cleaning.

[0284] 4) Status recording and task completion

[0285] The time, mode, parameters, and... of this cleaning... The curve is written into the historical database for S7 to perform statistical analysis and self-learning; the task status is set to FINISHED to notify the host computer or liquid cooling management platform that the segment has been cleaned.

[0286] Through the above specific implementation method, step S6 connects the "cleanliness index - cleaning intensity - termination criterion" in a closed loop, so that the present invention can not only identify pollution and construct a local cleaning loop, but also automatically adjust the intensity and reasonably end the task during the cleaning process, thereby taking into account energy consumption and operation and maintenance costs while ensuring that the cleanliness meets the standards and is stable.

[0287] S7: Operating Data Writeback and Self-Learning Optimization

[0288] The present invention further utilizes historical data through working condition recording and self-learning module 707 to gradually optimize the cleanliness index model and cleaning strategy over time.

[0289] 1. Data Recording

[0290] In each localized cleaning task, the system records at least the following data: contaminated segment number, segment type (main trunk, branch, cold plate, etc.), topological location information; cleanliness index and differential pressure data before cleaning (baseline); and data for each sampling period during cleaning. , , and traffic Cleaning mode combination and initial control parameters; parameter adjustment records during the cleaning process (such as pulse cycle, pump speed changes, backwashing times, etc.); result evaluation after cleaning, including total cleaning time, energy consumption estimation, and cleanliness stability over a certain period of time thereafter.

[0291] 2. Statistical Analysis and Parameter Correction

[0292] The working condition recording and self-learning module periodically performs statistical analysis on historical data, evaluating the rationality of current cleanliness index parameters and threshold settings for different segment types and working conditions. For example, if long-term observation reveals that a certain type of segment... If significant blockage or cooling performance degradation occurs nearby, the cooling capacity of this type of section can be appropriately reduced. Or increase particle weight If some segments become contaminated again quickly after cleaning, the cleaning mode can be adjusted or the cleaning time extended. Adjustments can be made through simple manual adjustments based on experience, or by employing simple self-learning algorithms such as multiple linear regression or rule updates. Since this invention does not rely on complex deep learning models, those skilled in the art can easily implement it using existing statistical software.

[0293] 3. Strategy optimization and knowledge reuse

[0294] As the operating time increases, the system forms a set of "experience parameter library" for different segment types and operating conditions. When a segment contamination with similar topology and operating conditions is detected again, the optimal or better cleaning mode and initial parameter values ​​can be directly called, reducing debugging time and improving control effect.

[0295] Through step S7, this invention enables the online cleanliness control system for liquid-cooled pipelines to have the characteristic of "getting better with use," achieving intelligent optimization of thresholds and control parameters while ensuring that the implementation difficulty is not high.

[0296] In summary, the embodiments of this invention clearly demonstrate the implementation method and system for online cleanliness closed-loop control of liquid cooling pipelines through terminology explanation, system structure, and specific implementation paths for steps S1 to S7. The sensor types, valve structures, control logic, and mathematical models used in this embodiment are all based on existing mature devices and technologies. Those skilled in the art, after reading this specification, can engineer liquid cooling systems of different scales and media without creative effort, and obtain the beneficial technical effects of this invention, such as refined contamination identification, localized online cleaning, and self-learning optimization, in practical applications.

[0297] V. Application Example: Application in a Data Center Rack Liquid Cooling System

[0298] 1. Application Scenarios and System Configuration

[0299] A data center has constructed a new single-story server room using liquid cooling, deploying 48 liquid-cooled racks. Each rack houses 32 CPU / GPU cooling plates and is connected to the main liquid cooling circuit of the floor via rack-level supply and return manifolds. The cooling medium is a 30% ethylene glycol aqueous solution, with a designed supply temperature of 20°C and a return temperature of approximately 28°C. The design flow rate per rack is 1.5 m³ / h. In traditional solutions, this data center only undergoes a complete shutdown for flushing and coolant replacement during initial system commissioning and annual scheduled maintenance. Interim measures rely solely on simple online filtration, which is insufficient for precise identification and treatment of localized contamination.

[0300] This embodiment adds an online cleanliness closed-loop control system for the liquid cooling pipeline to the existing liquid cooling system. The pipeline is divided into five typical sections: the main supply / return main pipe on each floor, the supply section of each rack, the return section of each rack, and the cold plate manifold. Cleanliness detection nodes (particle size + differential pressure + flow rate + temperature) are arranged at the inlet of the supply section and the outlet of the return section of each rack. Electric ball valves and three-way valves are installed at both ends of the supply and return sections of each rack, and connected to the filter unit via bypass pipelines. A variable frequency circulating pump is configured on the bypass pipeline. The cleanliness closed-loop controller is connected to the computer room integrated monitoring system to realize parameter setting, data acquisition, and control strategy distribution.

[0301] 2. Baseline establishment during initial operation

[0302] Before system commissioning, according to the requirements of S1-S2 of this invention: topology modeling and segment numbering of the main circuit and each rack branch circuit are performed; after the initial flushing and filtration, the pressure difference of each segment under "clean state" is recorded to establish a baseline pressure difference; at the same time, the particle sensor output under different flow conditions and temperatures is calibrated as the basis for subsequent temperature compensation and flow correction; the initial cleanliness target threshold range is set in the controller, for example... , (Dimensionless), and set the temperature reference value to 25℃. After a period of stable operation, the controller monitors each segment... Statistical analysis confirmed that, under baseline conditions, the cleanliness index of the vast majority of rack return liquid sections was concentrated between 7 and 10.

[0303] 3. Automatic detection and segmentation of localized contamination

[0304] About eight months after the computer room started operating, the controller detected the cleanliness index of the return liquid section of rack A-17 one day. It continued to rise and exceeded the upper limit threshold of temperature adaptation. The data is as follows (after normalization):

[0305] Normal period: , ;

[0306] Abnormal period: Gradually rising to 16-18, It rose to around 0.45;

[0307] The return liquid sections of adjacent racks A-16 and A-18 The pressure difference remains within the normal range.

[0308] After the controller calculates the multi-parameter fused cleanliness index according to step S3, it executes the contamination segmentation judgment process in step S4:

[0309] The A-17 return fluid segment exceeded the limit. Add it to the candidate set;

[0310] Compared with the upstream section (rack A-17 liquid supply section), the following was found:

[0311] This is significantly greater than the cleanliness gradient threshold.

[0312] The value is also clearly positive, indicating that the pressure loss is mainly concentrated in the return liquid section;

[0313] Compared with the liquid return sections of adjacent racks A-16 and A-18, the cleanliness index and pressure difference of A-17 are significantly higher.

[0314] Based on the above gradient information, the controller determines that "Rack A-17 return liquid segment" is a backward-concentrated medium-contamination segment and transmits the information "Segment number = A-17 return liquid, type = down_concentrated, level = medium" to the S5 module.

[0315] 4. Local cleaning loop construction and automatic cleaning mode selection

[0316] According to the S5 rule table, for "backward concentration + moderate contamination", the system automatically selects the combination mode of M2 (pulse flushing) + M3 (flow reversal) + M1 (bypass fine filtration), with an intensity level of medium.

[0317] The controller automatically performs the following actions (without human intervention):

[0318] 1) Gradually reduce the main flow rate of the A-17 return branch to avoid excessive fluctuations during switching;

[0319] 2) Close the inlet and outlet electric ball valves of the A-17 return liquid section to isolate it from the main circuit;

[0320] 3) Open the bypass three-way valves connected to both ends of the A-17 return liquid section to make the bypass pipeline 60 form a closed local loop with the section;

[0321] 4) Start the bypass circulation pump, set the initial local circulation flow rate to approximately 0.8 times the design value, and connect the coarse filter + fine filter unit in series to the local loop;

[0322] 5) Set initial pulse parameters: pulse period High traffic low flow ;

[0323] 6) Set the flow reversal cycle It automatically switches between "forward flushing" and "reverse flushing" every 60 seconds.

[0324] At this time, the IT load of rack A-17 is slightly reduced through the whole data center scheduling platform, but the entire data center does not need to be shut down, and the liquid cooling operation of the other 47 racks is not affected.

[0325] 5. Closed-loop regulation and key data changes during the cleaning process

[0326] After the cleaning begins, the controller presses S6. and Perform real-time calculations and monitoring:

[0327] 1) 10 minutes before cleaning (enhanced phase)

[0328] initial It slowly descends under the action of pulse + reverse cleaning, but in the first 5 minutes The absolute value is still relatively small; the controller determines that the decrease is "slow" and automatically increases the pulse high flow rate to [a higher value]. The pulse period was shortened to 15 seconds, and the flow reversal period was shortened to 45 seconds; then, within the next 5 minutes, The decline was more pronounced, dropping from approximately 17 to around 13.

[0329] 2) 10-30 minutes after cleaning (stabilization phase)

[0330] when It dropped to 11-12 and is close to the temperature correction value. At that time, the controller detected Upon determining that the system has entered a stable cleaning phase, the system automatically and slightly reduces the circulation flow rate and pulse intensity, extends the pulse cycle to 25 seconds, and restores the flow reversal cycle to 60 seconds to reduce energy consumption and valve impact. During this period, It has basically stabilized in the range of 9 to 11, and the differential pressure increment has gradually decreased from the original 0.45 to about 0.18, which is close to the level of other normal racks.

[0331] 3) 30-60 minutes after cleaning (termination criterion monitoring phase)

[0332] The controller detected the following within a 30-minute time window: Always in the personalized range Inside, and Always below the set value Trigger the termination criterion and enter the final stage: gradually reduce the local circulation flow, stop the pulse and flow reversal operations, and restore the local flow to a stable state; close the bypass valve, reopen the inlet and outlet ball valves of the A-17 return liquid section, and reconnect it to the main circuit; finally, stop the bypass circulation pump, and the local cleaning task ends.

[0333] The entire cleaning process lasted about 60 minutes. During this time, the A-17 rack only needed to reduce some of its IT load and was not completely taken offline; the remaining racks and the entire liquid cooling system continued to operate stably.

[0334] 6. Comparison of effects and proof of technical effectiveness

[0335] 6.1 Improvement in cleanliness and differential pressure

[0336] By comparing data before and after cleaning, the key indicators of the A-17 return fluid segment changed as follows:

[0337] Cleanliness index: dropped from 17.0 to around 9.5, returning to the normal rack average level (approximately 9.0 to 10.0).

[0338] Normalized differential pressure increment: decreased from 0.45 to approximately 0.18, which is basically consistent with the return liquid segment of other racks (0.15~0.20);

[0339] The temperature difference between the cold plate inlet and outlet returned to normal, and the highest temperature of the CPU / GPU chip inside the rack decreased by about 3-5°C compared to before cleaning, eliminating the risk of local overheating.

[0340] 6.2 Comparison with traditional maintenance methods

[0341] If the traditional method of "periodic complete shutdown flushing + experience-based coolant replacement" is adopted:

[0342] If a rack becomes locally contaminated, it usually needs to be shut down for treatment during the next centralized maintenance. During this period, the rack may continue to experience local overheating or performance degradation.

[0343] If a special shutdown for flushing is arranged in advance for safety reasons, it is often necessary to shut down the main circuit or the entire rack at the same time, which has a significant impact on business operations.

[0344] If the rinsing strategy is not precise, problems such as "heavy washing for light dirt" or "severe local contamination not being fully cleaned" are likely to occur, while also consuming a lot of manpower and coolant.

[0345] The system of the present invention can:

[0346] 1) Accurately identify the location and extent of pollution.

[0347] Automatically detects abnormalities in the A-17 return liquid segment and locks the contaminated segment through cleanliness index gradient and differential pressure gradient, avoiding large-scale blind inspection and cleaning.

[0348] 2) No need for system shutdown, local online cleaning

[0349] A local cleaning loop was built only for the A-17 return liquid section, while the other 47 racks and main loop continued to operate normally, significantly reducing the impact of maintenance on business.

[0350] 3) Automatically adjusts cleaning intensity to avoid over- or under-cleaning.

[0351] During the cleaning process, according to The changing trend automatically adjusts the pulse intensity, flow rate, and flow direction reversal cycle to ensure that contaminants are removed without causing unnecessary energy consumption and valve fatigue.

[0352] 4) Reduce the frequency of coolant changes and maintenance costs.

[0353] Through multiple similar local cleaning tasks, the data center did not require a complete coolant replacement within a year. Statistics show that the overall maintenance cost was reduced by about 30% compared to traditional solutions, and the number of unplanned downtimes was significantly reduced.

[0354] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

Claims

1. A closed-loop control process for online cleanliness of liquid-cooled pipelines, characterized in that, The process, applied to the operation of circulating liquid cooling media in liquid cooling pipelines, includes the following steps: S1. Establish a liquid cooling pipeline topology model, divide the liquid cooling pipeline into at least two independent control segments, and arrange cleanliness detection nodes and execution control nodes at the inlet and / or outlet of each pipeline segment. The cleanliness detection nodes include at least one of particle size sensor and differential pressure sensor. S2, during system operation, real-time collection of flow rate, temperature, particle count data and differential pressure data output by the cleanliness detection node of each pipeline segment, and noise reduction, outlier removal and temperature compensation processing of the collected data. S3, for each pipeline segment, construct a multi-parameter fusion cleanliness index CI based on the particle count data and pressure difference data. i The cleanliness index CI i The particle concentration, pressure difference increment, and flow rate normalization value are weighted by preset weights, and the baseline threshold is updated online adaptively according to the type of liquid cooling medium and temperature. S4, the cleanliness index CI of each pipeline section. i Compared with the corresponding target cleanliness range, when the cleanliness index of one or more pipeline segments exceeds the upper limit threshold of the target cleanliness, the contaminated segment and its degree of contamination are automatically determined based on the cleanliness index gradient and pressure difference distribution of adjacent segments. S5, for the determined contaminated segment, control the action of the execution control node corresponding to the contaminated segment, isolate the contaminated segment from the normal segment liquid cooling circuit, construct a local cleaning circuit, and automatically select or combine at least one cleaning mode among pulse flushing, flow reversal flushing and bypass filtration circuit according to the degree of contamination. S6, during the local cleaning process, the cleanliness index change curves at the inlet and outlet of the contaminated section are collected in real time, and the cleaning parameters, including pulse frequency, flushing flow rate, flow direction switching cycle and online backwashing time of filter unit, are automatically adjusted according to the real-time value of the cleanliness index and its rate of change. The cleaning is determined to be completed when the cleanliness index falls back to the target cleanliness range and the rate of change is lower than the threshold within the preset time window. S7. After cleaning is completed, the connection between the contaminated section and the main liquid cooling circuit is restored, and the cleanliness index data of each pipeline section and the corresponding cleaning parameters are written into the closed-loop control model as historical operating conditions for subsequent self-learning optimization of target cleanliness thresholds and cleaning strategies under different operating conditions.

2. The closed-loop control process for online cleanliness of liquid-cooled pipelines according to claim 1, characterized in that, In step S3, the multi-parameter fusion cleanliness index CI i The calculations include: S3.1, Perform graded statistical analysis on particle counts according to particle size range to form a particle grade vector; S3.2, Perform operating condition normalization processing on the differential pressure data to eliminate differential pressure disturbances caused by flow fluctuations; S3.3, perform linear or nonlinear fusion of the particle size vector and the pressure difference normalization result according to the preset weights to obtain the target cleanliness index; S3.4 The preset weights are based on historical operating data of the liquid cooling system and are trained offline or updated online using machine learning algorithms.

3. The closed-loop control process for online cleanliness of liquid-cooled pipelines according to claim 1, characterized in that, The automatic determination of pollution segments in step S4 includes: S4.1 Calculate the difference in cleanliness index and pressure difference between adjacent pipeline sections; S4.2, compare the cleanliness index difference and pressure difference with the corresponding set thresholds respectively. When both exceed the threshold, the corresponding pipeline segment is determined as the priority contamination segment. S4.3 When multiple adjacent segments meet the contamination judgment conditions, multi-level contamination segments are determined in descending order of cleanliness index difference, and different cleaning intensities and cleaning durations are assigned to each segment.

4. The closed-loop control process for online cleanliness of liquid-cooled pipelines according to claim 1, characterized in that, The construction of the local cleaning circuit in step S5 includes: S5.1, by using electric three-way valves or electric ball valves installed at both ends of each pipeline section, the contaminated section is separated from the main circuit, and a closed cleaning branch is formed by bypass pipelines and independent circulation pumps and filter units. S5.2, the bypass pipeline is provided with at least one set of fine filter cartridges or magnetic collection units, which are used to simultaneously remove solid particles and ferromagnetic impurities during local cleaning. S5.3 For pipeline sections with cold plates or microchannel structures, alternating forward and reverse flushing is achieved by switching valves.

5. The closed-loop control process for online cleanliness of liquid-cooled pipelines according to claim 1, characterized in that, The automatic adjustment of cleaning parameters in step S6 includes: S6.1 When the rate of decrease of the real-time cleanliness index is less than the first threshold, increase the pulse frequency and flushing flow rate; S6.2 When the real-time cleanliness index rises or fluctuates beyond the second threshold within a preset time, the flow direction switching cycle is shortened and the online backwashing of the filter unit is triggered. S6.3 When the real-time cleanliness index remains within the target cleanliness range for multiple consecutive sampling cycles and the rate of change is lower than the third threshold, the flushing flow rate is gradually reduced and the pulse cycle is extended to achieve energy-saving final cleaning.

6. The closed-loop control process for online cleanliness of liquid-cooled pipelines according to claim 1, characterized in that, The closed-loop control model is a digital twin model based on the liquid cooling pipeline topology, and the digital twin model is used for: The geometric parameters, material, and flow resistance characteristics of each pipeline segment are coupled with the measured flow rate and cleanliness index for calculation. Predict the changing trend of the cleanliness index under different cleaning modes and combinations of cleaning parameters; It automatically recommends the optimal cleaning mode and initial cleaning parameters for subsequent working conditions.

7. A closed-loop control system for online cleanliness of liquid-cooled pipelines for implementing the closed-loop control process for online cleanliness of liquid-cooled pipelines as described in any one of claims 1 to 6, characterized in that, The system includes: Liquid cooling piping and a piping topology that divides the liquid cooling piping into multiple piping segments; Cleanliness detection nodes are installed at the inlet and / or outlet of each pipeline section. The cleanliness detection nodes include at least two of the following: particle size sensor, differential pressure sensor, and flow and temperature sensor. The execution control nodes are set at both ends of each pipeline section and on the bypass pipeline, and the execution control nodes include electric valves and circulating pumps; Filter unit and its online backwashing assembly; The controller is electrically connected to the cleanliness detection node and the execution control node; The controller is configured as follows: a) Divide the liquid cooling pipeline into multiple pipeline segments based on the liquid cooling pipeline topology information, and establish corresponding cleanliness index calculation models and target cleanliness ranges; b) Receive the detection data output by each cleanliness detection node in real time, calculate the cleanliness index of each pipeline segment, and compare it with the corresponding target cleanliness range; c) When the cleanliness index of one or more pipeline sections is detected to exceed the target cleanliness range, the control execution node constructs the corresponding local cleaning loop and outputs the cleaning mode and cleaning parameter control instructions. d) During the local cleaning process, the cleaning parameters are adjusted in real time according to the changes in the cleanliness index, and the normal operation of the liquid cooling pipeline is restored after the cleaning is completed.

8. The online cleanliness closed-loop control system for liquid-cooled pipelines according to claim 7, characterized in that, The controller includes: The data acquisition module is used to acquire and preprocess multi-source data from particle size sensors, differential pressure sensors, flow sensors, and temperature sensors; The cleanliness assessment module is used to perform multi-parameter fusion cleanliness index calculation and automatic determination of contamination segments; The cleaning strategy decision module is used to output local cleaning modes and initial cleaning parameters based on the level of contamination and the digital twin model. The closed-loop control execution module is used to issue control commands to each execution control node and adjust the cleaning parameters in real time during the cleaning process; The working condition recording and self-learning module is used to store the parameters and cleanliness change curves of each cleaning process, and update the cleanliness threshold and cleaning strategy based on historical data. And / or, the particle size sensor is an online liquid particle counter, and its measurement results are converted into ISO 4406 or NAS levels for cleanliness index calculation; the differential pressure sensors are respectively arranged at the front and rear ends of each pipeline section to monitor the increase in local resistance caused by contamination; And / or, the liquid cooling piping is used in a cold plate or immersion liquid cooling system for data centers or power electronic devices, the system further comprising: The interface module that communicates with the liquid cooling system management platform is used to report the cleanliness index and cleaning status of each pipeline segment in real time, and to receive the operation mode switching command and cleanliness strategy configuration parameters from the host computer. The human-machine interface terminal is used to display the topology of the liquid cooling pipeline, the cleanliness distribution, and the current and historical cleaning conditions, and supports manually triggering local cleaning or modifying the target cleanliness range.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method of any one of claims 1–6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the method of any one of claims 1–6.

Images