Asset management system

Abstract

This application provides an asset management system belonging to the field of data center infrastructure management technology. It involves setting up multiple sensor nodes distributed along the target deployment direction using a sensing array. The system activates a radio frequency (RF) field for data exchange only after a magnetic smart tag is detected approaching and generates a location signal at the sensor node. The RF field is immediately deactivated after the data exchange is complete. This achieves precise U-position-level positioning and identification in the complex electromagnetic environment of a metal server rack, effectively suppressing signal crosstalk between adjacent U-positions and avoiding the impact of continuous electromagnetic radiation on the server room environment. This results in high-precision, low-interference automatic asset location and identification. Furthermore, the digital twin platform can construct digital twin models of the assets, enabling visualized monitoring of these assets.

Description

Technical Field

[0001] This application relates to the field of data center infrastructure management technology, specifically to an asset management system. Background Technology

[0002] With the rapid development of cloud computing, artificial intelligence, and big data technologies, data centers are continuously expanding in scale, and the number of IT devices in server racks is increasing dramatically. Traditional data center asset management mainly relies on manual inventory and spreadsheet records, which has the following prominent problems: Asset change information updates are severely delayed—according to IBM research data, 67% of asset change information is updated after more than 24 hours; rack space utilization is severely fragmented, with the average rack density only reaching 55%-65% of the design capacity; traditional RFID solutions suffer severe signal attenuation in metal rack environments, with positioning errors reaching ±5U (1U=44.45mm). Furthermore, cooling systems are often in a state of "oversupply," lacking real-time linkage with IT loads, resulting in serious energy waste.

[0003] To address the aforementioned issues, the industry has successively experimented with EIC (1-Wire bus) contact electronic tag technology and ordinary RFID tag solutions. The EIC contact solution relies on the metal contacts on the tag to make contact with the contacts on the U-position asset module to achieve data reading. After long-term use, the contacts are prone to oxidation, wear, electrostatic breakdown, and poor contact due to vibration, resulting in a high failure rate and making it unsuitable for large-scale deployment. While ordinary RFID solutions offer some improvement in non-contact reading, they face severe signal multipath reflection and attenuation problems in all-metal cabinet environments, making it impossible to reliably achieve precise U-position level positioning. Summary of the Invention

[0004] In view of this, the embodiments of this application aim to provide an asset management system to solve the problem that asset management in the prior art cannot reliably achieve U-level precise positioning.

[0005] This application provides an asset management system, including:

[0006] Magnetic smart tags are installed on the corresponding asset equipment;

[0007] The sensing module includes a sensing array and a radio frequency data reading unit. The sensing array has multiple sensing nodes distributed along the target deployment direction. Each sensing node can sense the proximity of the magnetic smart tag in a non-contact manner and generate a position signal characterizing the position of the sensing node. In response to the generation of the position signal, the radio frequency data reading unit activates a radio frequency field at the position of the sensing node to exchange data with the magnetic smart tag, and turns off the radio frequency field after the data exchange is completed.

[0008] The digital twin platform communicates with the sensing module, receives data acquired by the sensing module, and constructs a digital twin model of the asset equipment based on the data acquired by the sensing module.

[0009] In some embodiments, the magnetic smart tag is an RFID tag with an integrated permanent magnet that is magnetically attached to the surface of the asset device; the sensing node includes a Hall sensor, which is used to sense the proximity of the permanent magnet in a non-contact manner and generate the position signal; the radio frequency data reading unit includes an RF antenna and a microcontroller, and the RF antenna is controlled by the microcontroller to emit electromagnetic wave energy to activate the radio frequency field energy.

[0010] In some embodiments, the digital twin platform also acquires temperature data of the environment in which the asset device is located and power consumption data of the asset device, and uses a spatial interpolation algorithm to generate an energy consumption-heat dissipation thermal map of the asset device.

[0011] In some embodiments, when the temperature of any region in the energy consumption-heat dissipation thermal map exceeds a preset threshold, the digital twin platform causes the corresponding region to flash brightly as an alarm.

[0012] In some embodiments, the digital twin platform also automatically adjusts the operating parameters of the cooling equipment based on the energy consumption-heat dissipation thermal map.

[0013] In some embodiments, the digital twin platform performs a graded response based on the temperature rise obtained from the energy consumption-heat dissipation thermogram: when the temperature reaches the first interval, the return air temperature setpoint or supply air speed of the precision air conditioner is adjusted to increase the supply of cold air; when the temperature reaches the second interval, the cooling output power of the row air conditioner or the coolant circulation flow rate of the liquid cooling distribution unit is increased; when the temperature exceeds the third interval, the computing power scheduling system is linked to migrate the computing tasks from the overheated area to the low temperature area.

[0014] In some embodiments, when the linked computing power scheduling system migrates computing tasks from the overheated region to the low-temperature region, the digital twin platform also reduces the cooling output power of the row-level air conditioner or the coolant circulation flow of the liquid cooling distribution unit corresponding to the overheated region, while increasing the cooling output power of the row-level air conditioner or the coolant circulation flow of the liquid cooling distribution unit in the low-temperature region.

[0015] In some embodiments, the digital twin platform further includes a CFD simulation engine, which uses the real-time power consumption of the asset equipment as a heat source boundary condition to predict the temperature field distribution and calculate the supplementary cooling capacity required to eliminate hot spots. The cooling equipment adjusts its operating parameters according to the supplementary cooling capacity.

[0016] In some embodiments, the digital twin platform also records the entire lifecycle operation trajectory of the asset device from listing to delisting, and stores the operation records on the blockchain using a hash algorithm.

[0017] In some embodiments, the digital twin platform also uses a time series forecasting model to predict the supply and demand gap of U-positions based on historical U-position utilization data and business growth trends, and generates expansion proposals.

[0018] This application provides an asset management system that uses a sensing array with multiple sensing nodes distributed along the target deployment direction. The system activates a radio frequency (RF) field for data exchange only after a magnetic smart tag is detected and a location signal is generated at the sensing node. The RF field is immediately deactivated after the data exchange is complete. This achieves precise U-position-level positioning and identification in the complex electromagnetic environment of a metal cabinet, effectively suppressing signal crosstalk between adjacent U-positions and avoiding the impact of continuous electromagnetic radiation on the data center environment. This results in high-precision, low-interference automatic asset location and identification. Furthermore, the digital twin platform can construct digital twin models of assets and equipment, enabling visualized monitoring of these assets. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the architecture of an asset management system provided in an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of the deployment structure of the sensing module and the magnetic smart tag provided in one embodiment of this application.

[0021] Figure 3 A flowchart illustrating the working principle of a radio frequency data reading unit and a magnetic smart tag exchanging data, as provided in an embodiment of this application.

[0022] Figure 4 An example diagram showing the three-dimensional representation of the U-level energy consumption-heat dissipation thermal map in a digital twin platform provided in an embodiment of this application.

[0023] Figure 5 This is a closed-loop control flowchart of the U-position thermal map and cooling optimization linkage logic provided in an embodiment of this application.

[0024] Figure 6 This is a schematic diagram of a scheduling process provided in an embodiment of this application. Detailed Implementation

[0025] In data center rack asset management environments, to achieve the location and identification of asset equipment, relevant technologies typically employ either ordinary RFID tag solutions or EIC contact electronic tag solutions. Ordinary RFID solutions use a reader to emit radio frequency signals, exchanging data without contact with the RFID tag attached to the equipment to obtain equipment identification information. EIC contact solutions, on the other hand, rely on the metal contacts on the tag to make contact with the contacts on the U-position asset module installed on the rack rails to achieve data reading. The former eliminates physical wear and tear, while the latter avoids the attenuation of radio frequency signals in a metallic environment, and both are used in different scenarios.

[0026] However, when these solutions are applied to the precise management of densely packed U-positions within all-metal cabinets, they all struggle to simultaneously achieve both precise U-position positioning and reliable identification. Specifically, ordinary RFID solutions face severe signal multipath reflection and attenuation inside metal cabinets, resulting in positioning errors of up to ±5U, making it difficult to distinguish devices in adjacent U-positions. While contact-based solutions can achieve U-position-level positioning, long-term use can lead to oxidation, wear, electrostatic breakdown, and poor contact due to vibration, resulting in identification failures. Therefore, there is a contradiction in these technologies: precise U-position positioning and long-life contactless identification cannot be simultaneously achieved.

[0027] In-depth analysis revealed that the root cause of the aforementioned contradiction lies in the fact that the attenuation and multipath reflection of radio frequency signals by metal cabinets are inherent constraints of the physical environment. Ordinary RFID solutions continuously or omnidirectionally transmit radio frequency energy without spatially limiting the energy radiation range, causing crosstalk between adjacent U-position signals. Contact-based solutions, on the other hand, rely on mechanical contact conduction, which cannot avoid the aging and oxidation of metal contacts. Both solutions are physically limited by bottlenecks in electromagnetic propagation characteristics and material durability, respectively.

[0028] To overcome the aforementioned contradictions, this application proposes an asset management system. Its core concept lies in: arranging multiple sensor nodes in an array along the target deployment direction. Each sensor node employs a collaborative working mechanism of first magnetic positioning, then initiating radio frequency (RF) reading at specific locations, and immediately shutting off after reading. RF energy is activated only at specific sensor node locations where a permanent magnet is detected approaching for data exchange, and the RF energy is immediately deactivated after reading is complete. Thus, while maintaining the reliability of contactless identification, the effective range of the RF energy is limited to the target U-position, avoiding crosstalk between adjacent U-position signals and reducing continuous electromagnetic radiation, achieving a balance between precise U-position-level positioning and identification.

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] Figure 1 This is a schematic diagram of the architecture of an asset management system provided in one embodiment of this application. The system is deployed in a data center environment and consists of a perception layer, a network layer, a platform / digital twin layer, and an application layer.

[0031] The sensing layer is based on MC-RFID (Magnetic Control-Radio Frequency Identification) U-position level asset sensing technology. Its field equipment mainly includes sensing modules installed on the rack side and magnetic smart tags attached to the asset equipment side. The sensing modules are embedded in the vertical guide rails (also called U-position strips or rack columns) of the rack with a snap-in structure. Multiple sensing nodes and corresponding RF antennas are arranged in an array along the guide rail direction. Each U-position corresponds to at least one independent sensing node and RF antenna.

[0032] The sensing layer may also include environmental sensors and power metering devices distributed throughout the server room. Environmental sensors, such as high-precision infrared temperature sensors or digital temperature probes deployed at the air inlets, outlets, and key unit locations of the server racks, are used to collect data on the surface temperature of the equipment and the temperature of incoming and outgoing air. Power metering devices, such as smart meters or smart power distribution units, are used to collect real-time power consumption data for each unit's equipment.

[0033] The aforementioned field devices are connected to the network layer via wired and / or wireless networks. Location signals and identity data from the sensing module, temperature data from the environmental sensing device, and power consumption data from the power metering device are transmitted via the network to the data acquisition gateway or directly to the back-end server to complete data aggregation, cleaning, and timestamp alignment.

[0034] The digital twin platform runs on top of the platform layer / digital twin layer, continuously acquiring and processing data through communication interfaces (such as WebSocket or HTTP protocols). This digital twin platform includes a 3D visualization engine (digital twin 3D engine), data storage and analysis modules, an AI analysis engine, and a control command issuance interface. The 3D visualization engine constructs a digital twin model of the data center based on the building information model of the data center or WebGL technology, visually representing the data center space, rack arrays, and asset equipment. The control command issuance interface is used to communicate with external systems such as cooling equipment controllers and computing power scheduling platforms to transmit control commands.

[0035] Figure 2 This is a schematic diagram illustrating the deployment structure of a sensing module and a magnetic smart tag according to an embodiment of this application. Figure 2 As shown, magnetic smart tags are attached to the corresponding asset equipment. In a typical implementation, the magnetic smart tag is an RFID tag with an integrated permanent magnet (embedded with a permanent magnet and a radio frequency identification chip). It is magnetically attached to the surface of the casing of asset equipment such as servers and switches, allowing deployment without disassembling the asset equipment. The term "asset equipment" broadly refers to various IT infrastructure devices that can be installed in standard rack rails and occupy one or more U-spaces, such as rack servers (including general-purpose servers, GPU servers, storage servers, etc.), network switches (including core switches, access switches, etc.), routers, hardware firewalls, load balancers, disk arrays / network storage devices (NAS / SAN), blade server chassis and their blade nodes, UPS power distribution units, and various IT infrastructure devices that can be installed in standard 19-inch racks. Any physical device that can be installed in a standard rack rail and occupies one or more U-spaces can be identified using magnetic smart tags and included in the management scope of this system.

[0036] The sensing module includes a sensing array and a radio frequency data reading unit. The sensing array has multiple sensing nodes distributed along the target deployment direction. In this embodiment, the target deployment direction corresponds to the vertical guide rail direction of the cabinet, i.e., the arrangement direction of the U-positions. Each sensing node can sense the approach of the magnetic smart tag in a non-contact manner and generate a position signal characterizing the position of the sensing node. Specifically, each sensing node may include a Hall sensor. When the device is installed in a certain U-position of the cabinet, the permanent magnet inside the magnetic smart tag approaches and aligns with the Hall sensor corresponding to that U-position. After sensing the change in magnetic field, the Hall sensor outputs a position signal, thereby completing the accurate identification of the physical position. This position signal clearly indicates the specific U-position coordinates of the asset device in the cabinet, and the positioning accuracy can reach within ±0.5mm.

[0037] In response to the generation of the aforementioned position signal, the radio frequency data reading unit activates the radio frequency field energy at the location of the sensing node to exchange data with the magnetic smart tag, and then turns off the radio frequency field energy after the data exchange is completed. Figure 3 A flowchart illustrating the working principle of a radio frequency data reading unit exchanging data with a magnetic smart tag according to an embodiment of this application is shown below. Figure 3 As shown, the radio frequency (RF) data reading unit may include an RF antenna and a microcontroller. Upon receiving a position signal, the microcontroller controls the RF antenna corresponding to the sensing node to emit electromagnetic wave energy, thereby activating the RF field energy and turning on the RF identification chip within the magnetic smart tag. This allows for contactless reading of device identification information, such as device serial number, model, and configuration parameters. After data reading is complete, the microcontroller immediately shuts off the electromagnetic wave emission of the RF antenna, thus terminating the RF field energy at the sensing node's location. By integrating a permanent magnet and an RFID chip into a single magnetically attachable tag, and utilizing a Hall sensor as a contactless position sensing element and a microcontroller-driven RF antenna as the RF field energy generating unit, tool-free installation and removal of asset tags and precise coordination between magnetic field-triggered positioning and RF data reading are achieved. This design allows for deployment by simply attaching the tag to a suitable location on the asset's casing, without requiring any structural modifications to the asset, significantly reducing deployment complexity. Furthermore, since each U-position is equipped with an independent Hall sensor and RF antenna, electrical and logical independence between U-positions is ensured, further guaranteeing the consistency of U-position-level positioning and identification response and anti-crosstalk capability. Furthermore, this collaborative working mechanism of "first magnetic induction positioning, then RF reading initiated by position, and then shutdown after reading" ensures that the system only performs brief RF operations when it detects a magnetic component approaching a specific U position. This fundamentally avoids interference from continuous electromagnetic radiation to other electronic devices in the computer room and effectively suppresses signal crosstalk between adjacent U positions, because at any given moment, only the U position triggered by magnetic induction generates RF field energy.

[0038] It is understandable that the components for achieving non-contact magnetic sensing are not limited to Hall sensors; other magnetic sensing devices such as magnetoresistive sensors and magnetic switches can also be used. Magnetic fixing methods can also be combined with mechanical clips to adapt to different device housing materials.

[0039] Furthermore, the digital twin platform communicates with the sensing module via wired or wireless networks, receiving data acquired by the sensing module, including location signals, identity information, and status change events of each asset device, and constructing digital twin models of the assets and devices based on this data. Specifically, the digital twin platform can utilize Building Information Modeling (BIM) or the Three.js / WebGL web-based 3D engine to create a high-fidelity 3D model of the data center, geometrically recreating the data center space, each rack, each U-space, cable trays, and the asset devices installed within it at a 1:1 scale. The platform uses modern front-end frameworks such as Vue 3 / React combined with the Three.js engine to achieve pure web-based rendering, supporting interactive operations such as device rotation, scaling, panning, and roaming, and supporting multi-level drill-down (campus → building → data center → rack → device). The 3D model and the location information collected by the sensing module, as well as the temperature and power consumption data collected by environmental sensors, achieve second-level data synchronization via WebSocket or HTTP polling. Subsequently, the location and identity data of the assets uploaded in real time by the sensing module are bound to the corresponding 3D objects, thereby forming a dynamic digital mapping of the physical assets in virtual space. It should be noted that the bit-by-bit activation and instant shutdown mechanism of the aforementioned RF data reading unit provides a stable, crosstalk-free source of precise asset location and identity information for the digital twin model. This enables subsequent advanced applications such as thermal field monitoring, cooling optimization, and capacity planning to be built on a reliable data foundation, collectively solving the challenge of U-level precise asset sensing in environments with strong attenuation and multipath reflection in metal cabinets.

[0040] Building upon the digital twin mapping of asset location and identity at the unit level described above, this application further expands the functionality of the digital twin platform to facilitate intuitive understanding of the internal thermal environment of the server rack for maintenance personnel. In some embodiments, the digital twin platform also acquires temperature data of the environment where the asset equipment is located and power consumption data of the asset equipment, and uses spatial interpolation algorithms to generate an energy consumption-heat dissipation thermal map of the asset equipment. Specifically, temperature data can be collected in real time through high-precision infrared temperature sensor arrays or digital temperature probes deployed at the air inlets, outlets, and key unit locations of the server rack, while power consumption data can be obtained from intelligent power metering modules (such as smart meters or intelligent power distribution units, PDUs). Centered on each unit location, the digital twin platform uses spatial interpolation algorithms (such as Kriging interpolation or Inverse Distance Weighted Interpolation (IDW)) to spatially continuousize the discrete temperature and power consumption data, generating a temperature field estimate with unit-level accuracy, and finally superimposing it onto the surface of the digital twin model in the form of a three-dimensional thermal map. Figure 4 An example diagram illustrating the three-dimensional representation of a U-level energy consumption-heat dissipation thermal map in a digital twin platform provided in an embodiment of this application, as shown below. Figure 4 As shown, on the front and back views of the cabinet, each unit is covered with a three-dimensional color gradient band. The dark blue area represents the low-temperature zone (e.g., 22-25℃), the green area represents the normal temperature zone (e.g., 25-30℃), the yellow to orange area represents the heating zone (e.g., 30-35℃), and the red to dark red area represents the hot spot zone (above 35℃). This allows maintenance personnel to directly identify local hot spots and uneven cooling distribution from the continuous color bands when dragging and rotating the digital twin model in the digital twin interface, providing a refined visual basis for optimizing cooling strategies.

[0041] To further reduce the latency from hotspot generation to maintenance personnel response, based on the aforementioned energy consumption-heat dissipation heat map, when the temperature of any area in the energy consumption-heat dissipation heat map exceeds a preset threshold, the digital twin platform can trigger an alarm by highlighting the corresponding area in a flashing manner, or by using color coding to indicate the alarm status (e.g., green for normal, yellow for warning, and red for alarm). For example, the preset threshold can be set to 35℃. Once the estimated or measured temperature of a certain U-position area exceeds 35℃, the corresponding heat map color of that area will automatically switch to a flashing bright red. In this way, alarm information is precisely bound to the three-dimensional spatial location of the cabinet. Maintenance personnel can immediately capture the precise U-position of the hotspot area from any perspective of the digital twin interface, thereby enabling rapid intervention measures and reducing the risk of equipment frequency reduction or downtime due to undetected local overheating.

[0042] In data center cooling systems, cold aisles and hot aisles are designed for the entire server room layout, rather than each device having its own dedicated cold aisle. Specifically, data centers typically employ a "closed cold and hot aisle" layout—multiple rows of server racks are arranged face-to-face, with the air intake surfaces of the front rows forming a closed cold aisle area. Cool air is evenly distributed to the air intakes of each rack through underfloor air supply or horizontal air supply from row-level air conditioning. Based on this, after revealing the internal thermal state of the racks using a thermal map, to change the passive operation mode of the cooling system, in some embodiments of this application, the digital twin platform can also automatically adjust the operating parameters of the cooling equipment based on the energy consumption-heat dissipation thermal map. Figure 5 This is a closed-loop control flowchart of the U-position thermal map and cooling optimization linkage logic provided in one embodiment of this application, as follows: Figure 5As shown, the digital twin platform, through a closed loop of "sensing-analysis-decision-execution," takes the U-level temperature distribution reflected in the heat map as input. It then generates control commands for the cooling equipment via preset control logic or optimization algorithms. These commands are transmitted to cooling equipment such as precision air conditioners (CRAC / CRAH), in-row cooling systems, and cooling distribution units (CDUs) through a control command distribution interface, dynamically adjusting their operating parameters such as supply air temperature, air velocity, and coolant flow rate. This closed-loop linkage mechanism enables the cooling supply to follow the changes in heat consumption of the assets in real time, achieving a shift from "blindly over-supplying" to "precisely supplying on demand." This eliminates localized hotspots and reduces the overall energy consumption of the cooling system, thereby improving the power usage efficiency (PUE) of the data center.

[0043] To avoid overreacting to slight temperature increases and wasting energy, or underreacting to severe overheating and damaging equipment, the aforementioned strategy of automatically adjusting the operating parameters of cooling equipment based on heat maps can be further refined into a tiered response. Specifically, the digital twin platform executes a tiered response based on the temperature rise obtained from the energy consumption-heat dissipation heat map: when the temperature reaches the first range, the return air temperature setpoint or supply air velocity of the precision air conditioner (the precision air conditioner corresponding to the cold aisle where the asset equipment is located) is adjusted to increase the supply of cold air; when the temperature reaches the second range, the cooling output power of the row-level air conditioner is increased, and for high-density cabinets with liquid-cooled backplanes or cold plate liquid cooling, the coolant circulation flow rate of the liquid-cooled distribution unit (CDU) can be increased simultaneously; when the temperature exceeds the third range, the load migration plan is automatically triggered, and the computing power scheduling system is linked to migrate high-load computing tasks from the overheated area (servers) to the low-temperature area (servers). In a specific implementation, the first range can be set to 30-35℃, the second range to 35-40℃, and the third range to above 40℃. Level 1 response involves mild adjustment, suppressing the temperature rise simply by fine-tuning the precision air conditioning settings. Level 2 response activates localized enhanced cooling measures, specifically increasing the cooling capacity of high-density areas. Level 3 response addresses the root cause, using computational scheduling to transfer the heat load to low-temperature areas, achieving "cooling follows computation." This tiered, progressive control allows for intervention with methods commensurate with the level of thermal risk, preventing equipment overheating while maximizing energy savings in cooling systems.

[0044] In some embodiments, when migrating computing tasks, an intelligent scheduling engine can be built based on reinforcement learning algorithms (such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO)). The engine uses comprehensive computing power efficiency (the amount of computing tasks completed per unit of energy consumption), thermal risk index (the degree to which the temperature of each node deviates from the safe threshold), and sufficiency of cooling supply as core evaluation indicators to generate the optimal task allocation strategy in real time.

[0045] During the migration of computational tasks from an overheated region to a cryogenic region, if only the tasks are migrated without adjusting the operating parameters of the refrigeration equipment accordingly, it may result in redundant waste of refrigeration resources in the original overheated region, while the target cryogenic region may experience insufficient cooling capacity due to a sudden increase in heat load. Therefore, in a preferred embodiment, when the linked computing power scheduling system migrates computational tasks from an overheated region to a cryogenic region, the digital twin platform can reduce the cooling output power of the row-level air conditioners or the coolant circulation flow rate of the liquid-cooled distribution units (CDUs) in the overheated region, while simultaneously increasing the cooling output power of the row-level air conditioners or the coolant circulation flow rate of the liquid-cooled CDUs in the cryogenic region. Figure 6 A schematic diagram of the scheduling process provided in an embodiment of this application is shown below. Figure 6 As shown, task scheduling instructions are synchronously pushed to the cooling control system, achieving coordinated linkage between "computing power moving and cooling capacity following." The cooling supply to the original overheated area is immediately reduced after the computing tasks are moved out, avoiding ineffective cooling; the target low-temperature area increases cooling capacity reserves in advance to cope with the instantaneous increase in heat consumption caused by the incoming load. Through this coordinated adjustment of cooling operating parameters, the system achieves dynamic redistribution of cooling resources as the computing load shifts, achieving precise coordination between "computing power moving and cooling capacity following," further optimizing the overall energy efficiency of the data center.

[0046] Of course, in some embodiments, the air-liquid co-source cooling architecture (i.e., a system architecture in which the same cabinet has both air cooling and liquid cooling methods) can be linked to intelligently switch between air cooling and liquid cooling modes according to load changes—air cooling is used first under low load to reduce system complexity, and liquid cooling is automatically switched to improve heat dissipation efficiency under high load, thereby matching the load requirements of different intensities.

[0047] To avoid lag between cooling regulation and changes in the thermal field, some embodiments of the digital twin platform also incorporate predictive capabilities. Specifically, the digital twin platform includes a Computational Fluid Dynamics (CFD) simulation engine. This CFD simulation engine uses the real-time power consumption of the asset equipment as the heat source boundary condition to predict the temperature field distribution and calculate the supplementary cooling required to eliminate hotspot areas. The cooling equipment adjusts its operating parameters based on this supplementary cooling. Figure 5As shown in the thermal field prediction and cooling capacity matching section, the CFD simulation engine performs numerical simulation of the airflow organization in the computer room. The simulation process is as follows: First, a three-dimensional geometric model of the computer room is established (including parameters such as rack layout, raised floor height, air conditioning supply / return vent locations, and equipment heating power). Then, the real-time power consumption of each unit's equipment is used as the heat source boundary condition input into the model, and initial conditions such as the current air conditioning supply temperature, wind speed, and floor opening ratio are set. The CFD solver (such as ANSYS Fluent or OpenFOAM based on the finite volume method) is used to iteratively solve the airflow velocity field and temperature field distribution in the computer room space, predicting the temperature field change trend at various locations in the computer room under different load scenarios, especially the location and diffusion range of hotspot areas. Based on the thermal map trend and simulation prediction results, the system calculates the supplementary cooling capacity Q required to eliminate these hotspots based on the thermodynamic formula Q=c·m·ΔT, where c is the specific heat capacity of air, m is the required air mass flow rate, and ΔT is the expected temperature reduction. Based on this, the refrigeration equipment adjusts parameters such as supply air temperature, air velocity, valve opening, and the number of operating cold towers in advance. The refrigeration control is upgraded from "passive response" to "active prevention," effectively avoiding the expansion of hot spots caused by response delays and the energy waste caused by overcompensation, and realizing "on-demand refrigeration."

[0048] In actual operation, continuous comparison of energy consumption data before and after scheduling is conducted for post-evaluation and effect verification. By continuously monitoring and calculating the changing trend of PUE (Power Usage Effectiveness, PUE = Total Energy Consumption of Data Center / Energy Consumption of Asset Equipment), the impact of each cooling optimization strategy on overall energy consumption is quantitatively evaluated. By comparing and analyzing the PUE values ​​before and after optimization (e.g., daily, weekly, monthly dimensions), the system automatically evaluates the effectiveness of the strategy and continuously iterates and optimizes parameters. After applying the above "cooling-following-calculation" scheduling process, the total energy consumption of the cooling system was reduced by 18.7%, the interruption rate of tasks due to overheating decreased by 90%, the annual average PUE decreased from 1.48 to 1.32, and the annual electricity saving exceeded 3 million kWh. This mechanism not only effectively extends the service life of hardware equipment (avoiding chip aging and increased failure rate caused by long-term high-temperature operation), but also supports the continuous and stable operation of AI training clusters in ultra-high power density scenarios of more than 50kW per rack.

[0049] In terms of the full lifecycle management of assets and equipment, to ensure that operation records are authentic, complete, and tamper-proof to meet compliance audit requirements, in some embodiments of this application, the digital twin platform also records the entire lifecycle operation trajectory of the asset equipment from racking to de-racking, and stores the operation records on the blockchain using a hash algorithm. The operation trajectory includes all key events such as equipment racking and installation, configuration and allocation, U-type displacement, maintenance, and de-racking and scrapping. Each operation record includes information such as operation time, operator, operation type, and equipment status before and after the change. These operation records are uploaded to the blockchain for storage after generating unique digital fingerprints using a hash algorithm. Due to the collision-resistant characteristics of the hash algorithm and the distributed and tamper-proof characteristics of the blockchain, any modification to historical operation records will result in a change in the hash value and will be detected, thereby ensuring the high credibility and traceability of asset audit data and providing a solid compliance foundation for data center asset management. Relying on the batch concurrent reading capability of MC-RFID (processing more than 2,000 tag data per second), the system can complete a full automatic inventory audit of thousands of asset devices within 10 minutes.

[0050] To shift from passive, reactive capacity expansion to proactive capacity planning, in some embodiments of this application, the digital twin platform also uses time-series forecasting models (such as LSTM or Prophet) to predict the supply-demand gap for server racks based on historical rack utilization data and business growth trends, and generates expansion recommendations. For example, the digital twin platform can aggregate long-term rack occupancy records for each rack, combine them with expansion plans or historical growth curves provided by business departments, and use time-series forecasting models such as LSTM or Prophet to predict changes in rack demand for each rack over a future period. The system can predict which racks will experience rack shortages up to 14 days in advance and automatically generate expansion recommendations including expansion timing, expansion quantity, and suggested rack locations. This allows the operations team to rationally plan expansion based on data-driven forecasts, avoiding idle space and excessive construction of power supply and cooling facilities caused by blind expansion, thus improving the overall efficiency of data center resource utilization.

[0051] The following is a specific application scenario of the asset management system:

[0052] The asset management system described in this application was deployed in the data center of a large internet company to manage over two thousand servers and network devices located in standard 42U racks. The operations and maintenance team attached magnetic smart tags (requiring no disassembly and installation, compatible with mainstream 19-inch standard rack devices) to the surface of each device, while sensing modules of sensing nodes arranged in an array along the U-position direction were inserted into the vertical guide rails on both sides of the rack, so that each U-position corresponds to an independent Hall sensor and RF antenna.

[0053] One morning, a new batch of GPU servers needed to be deployed. The maintenance personnel pushed one of the GPU servers into position 22U-23U of rack 3. The instant the server was in place, the magnetic field generated by the permanent magnet inside the magnetic smart tag attached to its front panel was detected by Hall effect sensors at positions 22U and 23U. Each sensor node then generated a position signal carrying its own U-position coordinates. Upon receiving this signal, the microcontroller in the sensing module immediately controlled the corresponding U-position's RF antenna to emit electromagnetic energy, activating the RFID chip in the magnetic smart tag and reading the device's serial number, model, and factory configuration. The entire process was completed within tens of milliseconds, and the RF field was immediately shut off once the data exchange was finished. Almost simultaneously, the device's identity and precise U-position coordinates were uploaded to the digital twin platform via an encrypted link. The platform automatically created a digital twin model of the GPU server for positions 22U and 23U in the data center's 3D model and displayed the device model, serial number, rated power, and other information as an asset card on the interface, with a green status light. This mechanism of "first magnetic positioning, then starting RF reading by position, and turning off immediately after reading" makes the entire racking process require no manual scanning or contact operation, and there is no RF crosstalk between adjacent U positions. It still achieves U-position level accuracy identification even in the strong attenuation environment of metal cabinets.

[0054] As GPU servers begin to handle large model training tasks, their power consumption continues to rise. Digital temperature probes and intelligent power distribution units deployed at the air inlets, outlets, and key unit locations of the rack collect real-time temperature data for each unit and power consumption data for each device, pushing this data to the digital twin platform via a WebSocket channel. The digital twin platform uses the Kriging space interpolation algorithm to generate local temperature field estimates centered on each unit, and plots energy consumption-heat dissipation thermogram chromatogram bands on the surface of the 3D model. When maintenance personnel rotate the digital twin model in the interface, they can clearly see that the area from 22U to 25U gradually transitions from green to orange, indicating that the temperature in this area has entered the 30-35℃ range.

[0055] When the estimated temperature of U23 reached 33°C, the digital twin platform automatically triggered a Level 1 response: via the control command interface, the return air temperature setpoint of the nearest precision air conditioner was lowered from 24°C to 22°C, increasing the supply of cold air, and the temperature rise in that area was initially contained. However, as the training workload further increased, the temperature in areas 22U to 25U exceeded 35°C after twenty minutes. The corresponding location on the heat map automatically switched to a bright red flashing alarm, and the alarm information in the three-dimensional space was precisely bound to the U location. Maintenance personnel immediately received a system push notification and quickly located the hotspot area. At the same time, the platform entered a Level 2 response, increasing the cooling output power of the row-level air conditioner where the cabinet is located and increasing the coolant circulation flow rate of the liquid cooling distribution unit to enhance local heat exchange capacity.

[0056] An hour later, the peak workload caused the temperature in area 22U to exceed 40°C, triggering a Level 3 response on the platform: Through the coordinated computing power scheduling system, some high-load computing tasks were migrated from the overheated server area to servers in a lower-temperature zone within the same data center, achieving "cooling follows computing." The task migration command was simultaneously pushed to the cooling control system. The cooling power and coolant circulation flow of the row-level air conditioners in the original overheated area automatically decreased to avoid ineffective cooling; the cooling output power and coolant circulation flow of the row-level air conditioners in the original low-temperature area increased accordingly, pre-stocking cooling capacity to cope with the increased transient heat consumption. The entire process was aided by the platform's built-in Computational Fluid Dynamics (CFD) simulation engine. This engine used the real-time power consumption of each U-bit as the heat source boundary condition to predict the temperature field evolution trend and calculate the additional cooling capacity required to eliminate the hotspot. This ensured precise matching of control parameters. After the migration was completed, the temperature in the hotspot area dropped below 32°C within five minutes, during which time no service interruption occurred. In this scenario, this application enables the operations and maintenance team to intuitively grasp thermal field changes and obtain automatic graded response support on the large screen of the digital twin platform without having to manually troubleshoot in a noisy data center. The fault detection latency is reduced from tens of minutes that relied on manual inspection in related technologies to seconds. The risk of local overheating causing equipment frequency reduction or downtime is greatly reduced. The energy consumption of the cooling system is also significantly reduced due to the shift from "blindly over-supplying" to "precise supply on demand". The energy efficiency of data center power usage continues to improve.

[0057] Throughout the GPU server's subsequent lifecycle, the digital twin platform automatically records a complete trajectory, including the operation time, personnel, type, and pre- and post-operation states, for each operation involving equipment relocation, configuration changes, maintenance, and eventual decommissioning. Each operation record is hashed to generate a unique digital fingerprint and then uploaded to the blockchain for evidence storage. Auditors can use the platform to trace the historical operation chain of any asset; this historical operation chain is tamper-proof and meets compliance requirements. Simultaneously, based on long-term accumulated U-slot utilization data and business growth trends across racks, the digital twin platform utilizes a Long Short-Term Memory (LSTM) network time series prediction model to predict several days in advance that rack 3 will experience a U-slot supply-demand gap in the following month. It automatically generates expansion suggestions to assist the operations and maintenance team in planning rack space and corresponding cooling and power resources in advance, avoiding idle waste caused by blind expansion.

[0058] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0059] It should also be noted that although the preferred embodiments have been disclosed above, these embodiments are not intended to limit the scope of this application. For any person skilled in the art, many possible variations and modifications can be made to the technical solutions of this application based on the disclosed technical content, or equivalent embodiments can be modified accordingly, without departing from the scope of the technical solutions of this application. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of this application, without departing from the content of the technical solutions of this application, shall still fall within the scope of protection of the technical solutions of this application.

[0060] It should also be understood that, unless otherwise specified or indicated, the terms “first,” “second,” “third,” etc., in the specification are used only to distinguish the various components, elements, steps, etc. in the specification, and not to indicate the logical or sequential relationships between the various components, elements, steps, etc.

[0061] Furthermore, it should be recognized that the terminology described herein is used only to describe particular embodiments and is not intended to limit the scope of this application. It must be noted that the singular forms “an” and “a” as used herein include plural bases unless the context clearly indicates the opposite. For example, a reference to “a step” or “a device” means a reference to one or more steps or devices, and may include secondary steps and secondary devices. All conjunctions used should be understood in the broadest sense. The word “or” should be understood as having the definition of logical “or”, not logical “exclusive OR”, unless the context clearly indicates the opposite. Furthermore, implementations of the methods and / or devices in the embodiments of this application may include performing selected tasks manually, automatically, or in combination.

Claims

1. An asset management system, characterized in that, include: Magnetic smart tags are installed on the corresponding asset equipment; The sensing module includes a sensing array and a radio frequency data reading unit. The sensing array has multiple sensing nodes distributed along the target deployment direction. Each sensing node can sense the proximity of the magnetic smart tag in a non-contact manner and generate a position signal characterizing the position of the magnetic smart tag. In response to the generation of the position signal, the radio frequency data reading unit activates radio frequency energy at the position of the sensing node to exchange data with the magnetic smart tag, and turns off the radio frequency energy after the data exchange is completed. The digital twin platform communicates with the sensing module, receives data acquired by the sensing module, and constructs a digital twin model of the asset equipment based on the data acquired by the sensing module.

2. The asset management system according to claim 1, characterized in that, The magnetic smart tag is an RFID tag with an integrated permanent magnet that is magnetically attached to the surface of the asset equipment. The sensing node includes a Hall sensor, which is used to sense the proximity of the permanent magnet in a non-contact manner and generate the position signal. The radio frequency data reading unit includes an RF antenna and a microcontroller. The RF antenna is controlled by the microcontroller to emit electromagnetic wave energy to activate the radio frequency field.

3. The asset management system according to claim 1, characterized in that, The digital twin platform also acquires temperature data of the environment in which the asset is located and power consumption data of the asset, and uses a spatial interpolation algorithm to generate an energy consumption-heat dissipation thermal map of the asset.

4. The asset management system according to claim 3, characterized in that, When the temperature of any region in the energy consumption-heat dissipation thermal map exceeds a preset threshold, the digital twin platform will issue an alarm by highlighting and flashing the corresponding region.

5. The asset management system according to claim 3, characterized in that, The digital twin platform also automatically adjusts the operating parameters of the refrigeration equipment based on the energy consumption-heat dissipation thermal map.

6. The asset management system according to claim 5, characterized in that, The digital twin platform performs a graded response based on the temperature rise obtained from the energy consumption-heat dissipation thermogram: when the temperature reaches the first range, the return air temperature setpoint or supply air speed of the precision air conditioner is adjusted to increase the supply of cold air; when the temperature reaches the second range, the cooling output power of the row air conditioner or the coolant circulation flow rate of the liquid cooling distribution unit is increased. When the temperature exceeds the third range, the linked computing power scheduling system will migrate the computing tasks from the overheated area to the low-temperature area.

7. The asset management system according to claim 6, characterized in that, When the linked computing power scheduling system migrates computing tasks from the overheated area to the low-temperature area, the digital twin platform also reduces the cooling output power of the row-level air conditioner or the coolant circulation flow of the liquid cooling distribution unit corresponding to the overheated area, while increasing the cooling output power of the row-level air conditioner or the coolant circulation flow of the liquid cooling distribution unit in the low-temperature area.

8. The asset management system according to claim 5, characterized in that, The digital twin platform also includes a CFD simulation engine, which uses the real-time power consumption of the asset equipment as the heat source boundary condition to predict the temperature field distribution and calculate the supplementary cooling capacity required to eliminate hot spots. The refrigeration equipment adjusts its operating parameters according to the supplementary cooling capacity.

9. The asset management system according to claim 1, characterized in that, The digital twin platform also records the entire lifecycle operation trajectory of the asset equipment from listing to delisting, and stores the operation records on the blockchain using a hash algorithm.

10. The asset management system according to claim 1, characterized in that, The digital twin platform also uses a time series forecasting model to predict the supply and demand gap of U-positions based on historical U-position utilization data and business growth trends, and generates expansion suggestions.

Images