Integrated rack electric cooling conveying equipment

Abstract

The invention relates to an integrated rack electric cooling conveying device. A power distribution system may include: a power distribution subsystem including power distribution circuitry for providing power from an input source to a load through an outlet; a liquid cooling subsystem for providing active cooling to at least some of the loads wherein the liquid cooling subsystem includes one or more manifolds providing one or more supply nozzles and one or more return nozzles for directing fluid for active cooling; and one or more housings configured to surround the power distribution subsystem and the liquid cooling subsystem, where the one or more housings further provide a thermal coupling between the power distribution subsystem and the liquid cooling subsystem for at least partially cooling the power distribution subsystem. The system may also capture load diagnostic data associated with the loads and classify the loads into two or more categories based on the load diagnostic data.

Description

[0001] Cross - Reference to Related Applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 604,461, filed on November 30, 2023, entitled "INTEGRATED RACK POWER - COOLING DELIVERY DEVICE", with inventors Philip R. Aldag and Kevin R. Ferguson, under 35 U.S.C.§119(e), which is hereby incorporated by reference in its entirety. Technical Field

[0003] The present disclosure generally relates to power distribution systems, and more particularly to power distribution systems that incorporate liquid cooling and profiling monitoring. Background Art

[0004] High - density (HD) servers that require liquid cooling pose practical challenges when installed in server racks. For example, even with liquid cooling, HD servers may have air - side heat losses of 15% to 20% or more of the thermal load. The zero - U area on the back of the server rack must also accommodate various components, including a power distribution system, cables (e.g., power cables, network cables, etc.), and a liquid cooling manifold with associated pipes. As a result, the back of the server rack can become highly congested and / or difficult to access. In such server racks, server performance and / or reliability may be compromised due to restricted exhaust air flow, thermally stressed power distribution circuitry, and / or obstructed physical infrastructure maintenance. Therefore, there is a need to develop systems and methods that address the above - mentioned deficiencies. Summary of the Invention

[0005] In an embodiment, the technology described herein relates to a power distribution system, comprising: a power distribution subsystem including a power distribution circuitry configured to provide power from an input source to a plurality of loads through a plurality of outlets, wherein each of the plurality of outlets is configured to provide an electrical connection to any connected load among the plurality of loads; a liquid cooling subsystem configured to provide active cooling to at least some of the plurality of loads, wherein the liquid cooling subsystem includes one or more manifolds that provide one or more supply nozzles and one or more return nozzles for directing fluid for active cooling; and one or more enclosures configured to at least partially surround the power distribution subsystem and the liquid cooling subsystem, wherein the one or more enclosures also provide a thermal coupling between the power distribution subsystem and the liquid cooling subsystem for at least partially cooling at least a portion of the power distribution subsystem.

[0006] In an embodiment, the techniques described herein relate to a power distribution system, wherein one or more housings include a single housing that at least partially encloses a power distribution subsystem and a liquid cooling subsystem.

[0007] In an embodiment, the techniques described herein relate to a power distribution system, wherein one or more housings include: a first housing that at least partially encloses a liquid cooling subsystem; and a second housing that at least partially encloses a power distribution subsystem, wherein the first housing and the second housing are thermally coupled.

[0008] In an embodiment, the techniques described herein relate to a power distribution system, further comprising: one or more hinges to provide access to a thermal interface material that provides thermal coupling by rotation of at least one of the first housing or the second housing.

[0009] In an embodiment, the techniques described herein relate to a power distribution system, further comprising: one or more controllers, wherein each of a plurality of outlets is coupled to at least one of the one or more controllers, and wherein a respective one of the one or more controllers includes one or more processors configured to execute program instructions stored on a memory device, and wherein the program instructions are configured to cause the one or more processors to: receive load diagnostic data for any load connected to any of the plurality of outlets; and classify a plurality of loads connected to at least one of the plurality of outlets into two or more categories based on the load diagnostic data.

[0010] In an embodiment, the techniques described herein relate to a power distribution system, wherein load diagnostic data for a respective one of the plurality of loads includes at least one of the following: a temperature of the respective load; a temperature of a fluid of a respective supply nozzle exiting one or more supply nozzles; a temperature of a fluid exiting any of one or more manifolds; a temperature of a fluid entering a respective return nozzle of one or more return nozzles; a temperature of a fluid entering any of one or more manifolds; an ambient temperature of the power distribution system; an ambient temperature of the respective load; a die temperature associated with a processor of the respective load; utilization data of at least one of a central processing unit or a graphics processing unit of the respective load; a current drawn by the respective load; or a voltage drawn by the respective load.

[0011] In an embodiment, the techniques described herein relate to a power distribution system, wherein the two or more categories include: one or more normal categories associated with one or more acceptable operating conditions, and wherein the two or more categories further include one or more atypical categories associated with one or more atypical operating conditions.

[0012] In an implementation, the techniques described herein relate to a power distribution system, where two or more categories include: a binary category set, where one or more normal categories include a single normal category, and one or more atypical categories include a single atypical category.

[0013] In an implementation, the techniques described herein relate to a power distribution system, where two or more categories include: two or more category sets, where each of the two or more category sets corresponds to a different load type, and each of the two or more category sets includes at least one normal category among one or more normal categories and at least one atypical category among one or more atypical categories.

[0014] In an implementation, the techniques described herein relate to a power distribution system, where one or more controllers are further configured to execute a subset of program instructions that cause the one or more controllers to generate one or more alert signals when at least one of a plurality of loads is classified as an atypical category among one or more atypical categories.

[0015] In an implementation, the techniques described herein relate to a power distribution system, where one or more controllers are further configured to execute a subset of program instructions that cause the one or more controllers to disconnect power to the at least one of the plurality of loads when at least one of the plurality of loads is classified as an atypical category among one or more atypical categories.

[0016] In an implementation, the techniques described herein relate to a power distribution system, where one or more controllers include: one or more first controllers configured to be communicatively coupled to a plurality of outlets, where the one or more first controllers are configured to execute a subset of program instructions that cause the one or more first controllers to classify a plurality of loads into two or more categories using a machine learning model; and one or more second controllers configured to be communicatively coupled to the one or more first controllers, where the one or more second controllers are configured to execute a subset of program instructions that cause the one or more second controllers to train the machine learning model using training data including labeled load diagnostic data associated with the two or more categories.

[0017] In an implementation, the techniques described herein relate to a power distribution system, where one or more first controllers and one or more second controllers are located in one or more enclosures.

[0018] In an embodiment, the techniques described herein relate to a power distribution system, where at least one of one or more first controllers or one or more second controllers is located outside one or more enclosures.

[0019] In an embodiment, the techniques described herein relate to a power distribution system, where one or more first controllers include one or more power output modules (POMs), and where one or more second controllers include one or more interchangeable monitoring devices (IMDs).

[0020] In an embodiment, the techniques described herein relate to a power distribution system, where one or more first controllers utilize embedded memory, and where one or more second controllers utilize external memory.

[0021] In an embodiment, the techniques described herein relate to a power distribution system, where at least some of the labeled load diagnostic data is associated with historical load diagnostic data.

[0022] In an embodiment, the techniques described herein relate to a power distribution system, where the historical load diagnostic data is provided by at least one of: the power distribution system, one or more other power distribution systems, known historical failures of multiple loads, or a data lake.

[0023] In an embodiment, the techniques described herein relate to a power distribution system, where one or more controllers are further configured to execute a subset of program instructions that cause the one or more controllers to display information associated with at least one of multiple loads on a display device based on an associated classification according to the load diagnostic data.

[0024] In an embodiment, the techniques described herein relate to a power distribution method, including: using a power distribution system to capture load diagnostic data for a plurality of loads connected to a plurality of outlets to receive power from an input power source connected to the plurality of outlets, wherein the power distribution system includes a liquid cooling subsystem configured to provide active cooling to at least some of the plurality of loads, wherein the liquid cooling subsystem includes one or more manifolds that provide one or more supply nozzles and one or more return nozzles for guiding fluid for active cooling; classifying the plurality of loads connected to at least one of the plurality of outlets into two or more categories based on the load diagnostic data, wherein the two or more categories include one or more normal categories associated with one or more acceptable operating conditions, and wherein the two or more categories further include one or more atypical categories associated with one or more atypical operating conditions; and when a particular load among the plurality of loads is classified into an atypical category among the one or more atypical categories, performing at least one of the following: generating one or more alert signals; or disconnecting the connection of power to the particular load.

[0025] It should be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the claimed invention. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the general description, serve to explain the principles of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Those skilled in the art can better understand many advantages of the present disclosure by referring to the accompanying drawings.

[0027] Figure 1A is a block diagram showing a power distribution system according to one or more embodiments of the present disclosure.

[0028] Figure 1B is a simplified schematic diagram of a power distribution system with a liquid cooling (LC) subsystem thermally coupled to a power distribution subsystem according to one or more embodiments of the present disclosure.

[0029] Figure 1C is a block diagram view of a power input module (PIM) according to one or more embodiments of the present disclosure.

[0030] Figure 1D is a block diagram view of a power output module (POM) according to one or more embodiments of the present disclosure.

[0031] Figure 1E is a block diagram view of an interchangeable monitoring device (IMD) according to one or more embodiments of the present disclosure.

[0032] Figure 2 is a flowchart showing steps performed in a method for classifying a load according to one or more embodiments of the present disclosure.

[0033] Figure 3 is a flowchart depicting classification using a machine learning model according to one or more embodiments of the present disclosure. Detailed Description

[0034] Reference will now be made in detail to the disclosed subject matter, which is illustrated in the accompanying drawings. The present disclosure has been particularly shown and described with respect to certain embodiments and specific features thereof. The embodiments set forth herein are to be considered illustrative and not restrictive. It should be readily apparent to those of ordinary skill in the art that various changes and modifications can be made in form and detail without departing from the spirit and scope of the present disclosure.

[0035] Embodiments of the present disclosure relate to systems and methods for providing high-capacity power management with integrated liquid cooling for connected loads. Additional embodiments of the present disclosure relate to systems and methods for providing monitoring and classification of the operating conditions of a load and / or an associated liquid cooling system based at least in part on ambient data associated with the load, the liquid cooling system, or environmental conditions. For example, a machine learning model that receives load diagnostic data as input can be used to perform load operating condition monitoring and classification, where the load diagnostic data includes at least temperature data directly associated with the load or temperature data associated with a liquid cooling subsystem.

[0036] In an embodiment, a power distribution system can include a power distribution circuitry configured to distribute power to loads via a set of outlets, and can also include a liquid cooling (LC) system that provides dedicated liquid cooling conduits for loads connected to at least some of the outlets. The power distribution circuitry and the LC system can be disposed in a common housing or in separate housings that are thermally coupled to facilitate heat transfer. Additionally, the liquid cooling conduits can be physically located near the associated outlets and associated power distribution components (e.g., circuit breakers, etc.), which can provide distinct electrical and thermal connections for each load, as well as provide cooling of the power distribution components to facilitate efficient operation of both the load and the power distribution system.

[0037] The power distribution system may also include an environmental monitoring circuitry for generating load diagnostic data and / or a port for receiving load diagnostic data, where the load diagnostic data may include information including, but not limited to, temperature, humidity, or moisture of a load or a server rack. The power distribution system may then include one or more controllers configured to use machine learning techniques to predict and / or classify the operating condition (e.g., status) of the connected load based on the load diagnostic data. For example, the power distribution system may classify the operating condition of the load and / or an associated liquid cooling system into one or more normal categories associated with acceptable operating conditions and one or more atypical categories associated with a failure or pre-failure condition. In this way, loads in a failed condition, pre-failure condition, or impaired operating condition can be quickly identified and repaired.

[0038] The systems and methods disclosed herein may provide any number of classifications or sub-classifications of a load (and an associated liquid cooling system) based on load diagnostic data. In some embodiments, a binary class classification system including a single normal category and a single atypical category is used to classify the load. Such a configuration may be well-suited for, but not limited to, identifying desired atypical behavior. In some embodiments, a multi-class classification system including a single normal category and multiple atypical categories is used to classify the load signal. Such a configuration may be well-suited for, but not limited to, distinguishing different atypical categories. For example, different atypical categories may be associated with different failure mechanisms and / or different pre-failure conditions.

[0039] Additional embodiments of the present disclosure also relate to generating one or more alert signals (e.g., light, sound, interruption, maintenance request, etc.) when a load signal in an atypical category is identified. It is contemplated that the systems and methods disclosed herein may provide predictive failure analysis for the connected load. For example, the systems and methods disclosed herein may identify when the connected load is operating in an atypical manner based on the power signal to the connected load. Associated alerts may signal various actions, including but not limited to user intervention, disconnection of the load (manually by the user or automatically based on the classification), replacement of the load, repair of the load, or maintenance of the load.

[0040] The systems and methods disclosed herein may provide many benefits for the operation and monitoring of liquid-cooled loads. Proximity to the liquid cooling manifold and the power distribution system provides an opportunity to improve server performance and / or health monitoring. For example, integrating the power distribution circuitry with the liquid cooling manifold can effectively utilize space, facilitate the organization of power components and cooling components, and enable cooling of the power distribution circuitry. As another example, integrating the power distribution circuitry with the liquid cooling manifold conveniently allows for collection of load-specific environmental data for monitoring and classifying the operating condition.

[0041] Now referring to Figures 1A to 3 , according to one or more embodiments of the present disclosure, systems and methods for power distribution with integrated liquid cooling and temperature-based profiling are described in more detail. Specifically, Figures 1A to 1E depicts a power distribution system 100 adapted to distribute power to various loads 102, while Figures 2 to 3 depicts processing steps for classifying the loads 102 based at least on environmental data.

[0042] Figure 1A is a block diagram showing the power distribution system 100 according to one or more embodiments of the present disclosure. The power distribution system 100 may, but is not required to be, characterized as a power distribution unit (PDU).

[0043] In an embodiment, the power distribution system 100 includes a power distribution subsystem 104 and an LC subsystem 106. The power distribution subsystem 104 may include various electrical components to selectively distribute input power 108 to any number of loads 102 connected to any number of outlets 110, such as, but not limited to, transistors, relays, amplifiers, voltage converters, rectifiers, alternating current (AC) to direct current (DC) converters, DC to AC converters, DC to DC converters, etc. The power distribution subsystem 104 may also include one or more controllers to direct and / or control such components (e.g., via control signals) to selectively distribute the input power 108 to the respective outlets 110 and thus to the connected loads 102. The outlets 110 may include plugs or other sockets configured to provide electrical connections to various loads 102. The power distribution subsystem 104 may also include circuit breakers 112 or other electrical components associated with the outlets 110.

[0044] The various components of the power distribution subsystem 104 may, but are not required to be, divided into various modules that physically or functionally provide discrete functions. For example, as Figure 1A shown, the power distribution subsystem 104 may include a power input module (PIM) 114 configured to receive the input power 108 from one or more input sources (not shown) and at least one power output module (POM) 116 coupled to the outlets 110. A particular POM 116 may be connected to any number of outlets 110 and may control the distribution of power to any number of outlets 110. Additionally, the power distribution system 100 may include any number of POMs 116 and associated outlets 110. In other words, the outlets 110 may be distributed among any number of POMs 116. Thus, the power distribution system 100 may selectively distribute the received input power 108 to any load 102 connected to the corresponding outlet 110 via the combined operation of the PIM 114 and the corresponding POM 116.

[0045] In an embodiment, the power distribution system 100 further includes at least one interchangeable monitoring device (IMD) 118, which can be communicatively coupled to the PIM 114 and one or more POMs 116. The IMD 118 may or may not be interchangeable or replaceable. In this way, the acronym IMD is thus merely illustrative and should not be construed as limiting the scope of the present disclosure. The IMD 118 can receive data from the PIM 114 and the respective POMs 116 and / or direct the PIM 114 and the respective POMs 116 (e.g., via control signals). For convenience, Figure 1A FIG. 100 depicts a power distribution system 100 having a single PIM 114, a single POM 116 connected to multiple outlets 110, and a single IMD 118. However, it should be understood that this is also merely illustrative and should not be construed as limiting the scope of the present disclosure. The power distribution system 100 can generally have any number of PIMs 114, POMs 116, or IMDs 118.

[0046] The power distribution system 100 can also include one or more components to provide visual and / or audio signals to a user. Such components can be adapted to warn the user of the status of one or more connected loads 102 (or associated liquid cooling components), such as but not limited to an indication of a situation where the load 102 is classified into an atypical category or any other alarm condition.

[0047] For example, the power distribution system 100 can include one or more visual display devices 120. In some embodiments, the one or more visual display devices 120 include a display screen, which can provide any combination of text or graphic information. Additionally, the backlight on the display screen or the displayed background can flash or display a selected color to provide additional information or an alarm.

[0048] In some embodiments, the one or more visual display devices 120 include one or more light emitting diodes (LEDs), which can provide information based on color, brightness, blinking, etc. In some cases, the power distribution system 100 includes one or more LEDs associated with each outlet 110 (or at least one outlet 110) to provide a separate visual indication for the associated load 102. As another example, the power distribution system 100 can include one or more speakers 122 to provide audio signals, such as but not limited to sounds or spoken text.

[0049] The LC subsystem 106 may include any component or combination of components suitable for providing active liquid cooling to one or more loads 102 connected to an outlet 110. In an embodiment, the LC subsystem 106 includes one or more manifolds 124 to hold a fluid (e.g., coolant) and direct the fluid through various paths for cooling the various loads 102. For example, the one or more manifolds 124 may provide various supply nozzles 126 and return nozzles 128. In this way, the fluid may be directed from the supply nozzles 126 through pipes to one or more loads 102 to provide active cooling and then returned through the return nozzles 128 to form one or more cooling loops. The LC subsystem 106 may also include a heat exchanger 130 to control the temperature of the fluid (e.g., maintain the temperature of the fluid when the fluid is heated by one or more loads 102 via the cooling loop). The LC subsystem 106 may also include one or more pumps 132 and / or pipes 134 to circulate the fluid.

[0050] The LC subsystem 106 may also include one or more environmental sensors 136 configured to generate load diagnostic data and / or one or more data ports 138 for receiving load diagnostic data associated with any of the loads 102. For example, the load diagnostic data may include data collected by the environmental sensors 136 within the enclosure of the power distribution system 100, such as but not limited to the temperature associated with the supply nozzle 126 or the supply assembly of the manifold 124 (e.g., supply temperature), the temperature associated with the return nozzle 128 or the return assembly of the manifold 124 (e.g., return temperature), or the environmental conditions near the power distribution system 100 (e.g., temperature, humidity, moisture, etc.). As another example, the load diagnostic data may include data received by the data ports 138 from an external source, such as but not limited to the temperature of the load 102 (e.g., graphics processing unit (GPU) temperature, central processing unit (CPU) temperature, chassis temperature, die temperature, etc.), the environmental conditions near the load 102 (e.g., temperature, humidity, moisture, etc.), or utilization data (e.g., utilization data from the CPU, GPU, etc. of the load 102).

[0051] The LC subsystem 106 and the power distribution subsystem 104 may be physically connected or housed in any suitable arrangement. In some embodiments, the LC subsystem 106 and the power distribution subsystem 104 are at least partially surrounded by a single enclosure. In some embodiments, the LC subsystem 106 and the power distribution subsystem 104 are at least partially surrounded by separate enclosures but may be physically attached and in some cases thermally bonded to provide efficient heat transfer between the two. More generally, the power distribution system 100 may include one or more enclosures to at least partially surround any portion of the power distribution subsystem 104 and / or the LC subsystem 106.

[0052] Figure 1B is a simplified schematic diagram of a power distribution system 100 having an LC subsystem 106 thermally coupled to a power distribution subsystem 104, in accordance with one or more embodiments of the present disclosure.

[0053] The LC subsystem 106 can provide cooling for any number of loads 102 using any arrangement of components. For example, the LC subsystem 106 can include dedicated supply nozzles 126 and / or dedicated return nozzles 128 for one or more of the outlets 110.

[0054] Figure 1B Depicts a particular non-limiting configuration in which the electronic components and cooling components for each outlet 110 are physically grouped. For example, the LC subsystem 106 in this configuration includes a supply nozzle 126 and a return nozzle 128 dedicated to each outlet 110. In addition, electronic components such as, but not limited to, circuit breakers 112 and data ports 138 are also physically grouped by the corresponding outlet 110.

[0055] It is contemplated herein that physically grouping the electronic components and LC components for each outlet 110 can provide numerous benefits for a high-power liquid cooling system. For example, physically grouping the dedicated supply nozzles 126 and return nozzles 128 with the associated outlet 110 can provide an intuitive and unambiguous arrangement of the various components associated with each outlet 110 (e.g., each connected load 102). Such an arrangement can provide ease of use during setup and / or maintenance by reducing potential confusion. Such a configuration can provide an efficient use of rack space by avoiding long coolant lines and / or preventing entanglement of power lines and coolant lines. As another example, physically grouping the dedicated supply nozzles 126 and return nozzles 128 with the associated outlet 110 can enable the generation of differentiated load diagnostic data for the loads 102 connected to each outlet 110, which can be used for load operating condition classification as described herein.

[0056] In an embodiment, the LC subsystem 106 additionally provides cooling for the power distribution subsystem 104 (or a portion thereof), which can mitigate overheating of the power distribution subsystem 104 and thus facilitate reliable operation. For example, a portion of the LC subsystem 106 can be thermally coupled to a portion of the power distribution subsystem 104.

[0057] By way of illustration, Figure 1BDepicts placing the circuit breaker 112 near (or connected to) the supply nozzle 126 (e.g., cold side) for active thermal cooling. However, this is merely illustrative and should not be construed as limiting the scope of the present disclosure. In some cases, the circuit breaker 112 or other components of the power distribution subsystem 104 may be placed near (or connected to) the return nozzle 128 (e.g., warm side). Such a configuration may be applicable to, but not limited to, applications where the temperature difference (ΔT) between the return fluid and the power distribution subsystem 104 is sufficient to cool the power distribution subsystem 104.

[0058] Furthermore, in some embodiments, both the cold side and the warm side of the LC subsystem 106 may be thermally coupled to different parts of the power distribution subsystem 104 and / or different power distribution subsystems 104.

[0059] Figure 1B Also depicted is a configuration where the LC subsystem 106 and the power distribution subsystem 104 are in thermally coupled separate housings. For example, the LC subsystem 106 may be located in the LC housing 140, and the power distribution subsystem 104 may be located in the power housing 142 physically attached to the LC housing 140. In some embodiments, the power distribution system 100 includes a thermal interface material (TIM) 144 between the LC housing 140 and the power housing 142. Any suitable TIM may be used, including but not limited to thermal grease, thermal pads, or phase change materials (PCMs). In some embodiments, the LC housing 140 and the power housing 142 have thermally conductive portions (e.g., metal portions) fixed together with bolts or any other suitable fastening mechanism.

[0060] The LC housing 140 and the power housing 142 may also have any shape or design. In some embodiments, the LC housing 140 and the power housing 142 have complementary shapes to facilitate contact and high thermal coupling. For example, the LC housing 140 and the power housing 142 may each have at least one flat side (e.g., as Figure 1B depicted) that provides a continuous contact interface. As another example, the LC housing 140 and the power housing 142 may each have side surfaces with complementary curved shapes to provide a continuous contact interface. As another example, the LC housing 140 and the power housing 142 may be joined using thermally conductive mounts or interface materials. Such a configuration may be applicable to, but not limited to, configurations where the LC housing 140 and the power housing 142 have thermally incompatible shapes and / or materials. As another example, the LC housing 140 and the power housing 142 are joined using a hinged mechanism. Such a configuration may allow access to the contact interface and thus may enable the selection, modification, or replacement of the TIM. For example, access to the contact point and / or the thermal interface material between the LC housing 140 and the power housing 142 may be provided by rotating one or both of these housings.

[0061] It is contemplated herein that a configuration providing separate housings for the LC subsystem 106 and the power distribution subsystem 104 may permit the use of off-the-shelf components and / or may naturally mitigate the risk of electrical short circuits caused by fluid leakage by separating these components. Additionally, although Figure 1B the supply nozzle 126 and the return nozzle 128 are depicted on a common face with the outlet 110, this is for illustration only and should not be construed as a limitation on the scope of the present disclosure. In some embodiments, the supply nozzle 126 and the return nozzle 128 may be located on a different face from the corresponding outlet 110 to mitigate the risk of electrical short circuits caused by fluid leakage while maintaining thermal contact between the LC subsystem 106 and the power distribution subsystem 104.

[0062] In some embodiments, although not explicitly shown, the LC subsystem 106 and the power distribution subsystem 104 may share a common housing. Such a configuration may advantageously permit greater thermal coupling and thus permit more active cooling of the power distribution subsystem 104. Additionally, it should be noted that while encapsulating liquid cooling components and electrical components into a common housing may present some risk of electrical short circuits in the event of fluid leakage, it is contemplated herein that the components of the LC subsystem 106 may be manufactured with high tolerances such that the risk may be negligible or at least acceptable for certain applications. As an illustration, components for cold plate cooling of the load 102 (e.g., GPU, CPU, etc.) typically also require close proximity of cooling fluid and sensitive components.

[0063] The LC subsystem 106 and / or the power distribution subsystem 104 may also include various additional components adapted to facilitate heat transfer. For example, one or more housings of the LC subsystem 106 and / or the power distribution subsystem 104 may include fins to facilitate radiative heat transfer. As another example, one or more housings of the LC subsystem 106 and / or the power distribution subsystem 104 may include surface enhancements on the inner and / or outer surfaces to facilitate radiative heat transfer. For example, surface enhancements may include, but are not limited to, painted surfaces, anodized surfaces, or structured surfaces. As an illustration, portions of the cold side outer surface adjacent to the LC housing 140 may include surface enhancements to improve heat transfer relative to non-enhanced materials. As another illustration, the inner and / or outer surfaces of the power housing 142 may include any elements adapted to improve natural convection heat transfer relative to non-enhanced materials, such as, but not limited to, perforated ventilation devices, heat sinks, cavities, or chambers.

[0064] Now referring Figures 1C to 1E to, additional aspects of the power distribution subsystem 104 are described in more detail in accordance with one or more embodiments of the present disclosure.

[0065] Figure 1Cis a block diagram view of PIM 114 according to one or more embodiments of the present disclosure. PIM 114 may generate various measurements of input power 108, including but not limited to input voltage and input current, and may also calculate its energy metering data. For example, PIM 114 may include PIM current sensing circuitry 146 and / or PIM voltage sensing circuitry 148. PIM 114 may also include one or more PIM microcontrollers 150, PIM power supply 152, and PIM memory 154. PIM memory 154 may be embedded memory (e.g., internal memory) and / or external memory.

[0066] Figure 1D is a block diagram view of POM 116 according to one or more embodiments of the present disclosure. POM 116 may generate various measurements of load signal 156 (see Figure 1A ) associated with power transfer to load 102 through outlet 110, including but not limited to load voltage and load current, and may also calculate its energy metering data. For example, POM 116 may include POM current sensing circuitry 158 and / or POM voltage sensing circuitry 160. POM 116 may also include one or more POM microcontrollers 162, POM power supply 164, and POM memory 166. POM memory 166 may be embedded memory (e.g., internal memory) and / or external memory. POM 116 may also control bistable relay 168 to selectively provide power (or not provide power) to each outlet in outlet 110.

[0067] Figure 1E is a block diagram view of IMD 118 according to one or more embodiments of the present disclosure. IMD 118 may include but not limited to one or more IMD microprocessors 170, IMD power supply 172, display interface 174 (which may be associated with Figure 1Athe same as or different from the visual display device 120 shown, or non-volatile (NV) memory in the form of flash memory 176 and / or DDR memory 178, which may be embedded or external. The IMD 118 may also include a plurality of ports, including but not limited to one or more LAN Ethernet ports 180, one or more single-wire sensor ports 182, RS-232 / RS-485 ports 184, USB ports 186, or microSD slots 188. The IMD may be coupled to a reset switch 190 to enable a user to initiate a hard reset of the IMD 118, the power distribution system 100, and / or other subsystems within the power distribution system 100. The IMD 118 may also communicate with other components via the RS-485 physical layer. In some embodiments, the IMD 118 is provided as a hot-swappable network card in the power distribution system 100.

[0068] The IMD 118 may act as a monitoring host controller for the power distribution system 100 and may communicate continuously with the POM 116. By communicating with the POM 116, the IMD 118 may provide a means for a user to enable or disable one or more features of the power distribution system 100 and to obtain and display status information.

[0069] In an embodiment, the IMD 118, PIM 114, POM 116, and / or the LC subsystem 106 are communicatively coupled (e.g., via one or more communication buses). In this way, each may command any of the others, send data to any of the others, and / or receive data from any of the others. For example, the IMD 118 may command the PIM 114, which in turn may command the POM 116 to configure the relay state of one or more of the bistable relays 168. As another example, the IMD 118 may act as a bus master that is connected to both the PIM 114 and the POM 116, where the PIM 114 and the POM 116 do not interact directly with each other. As another example, the PIM 114 and / or the POM 116 are capable of autonomous behavior without a command from the IMD 118. As another example, the LC subsystem 106 may be communicatively coupled to any one of the IMD 118, POM 116, or PIM 114 to send and / or receive data. For example, the IMD 118 may receive at least some load diagnostic data from either the LC subsystem 106 or the POM 116.

[0070] Generally referring to Figures 1A to 1E it should be understood that Figures 1A to 1E and the associated description are provided for illustrative purposes only and should not be construed as restrictive. For example, although Figures 1A to 1EThe IMD 118, POM 116, and PIM 114 of the power distribution system 100 are presented as separate components, but this is merely illustrative and should not be construed as limiting the scope of the present disclosure. In some embodiments, any of the IMD 118, POM 116, or PIM 114 may be integrated together in a common housing or as a common component. Additionally, the IMD 118, POM 116, and / or PIM 114 may generally include any type of processor known in the art, including but not limited to a microprocessor, a microcontroller, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processor (DSP), a CPU, or a GPU. The IMD 118, POM 116, and / or PIM 114 may also include any type of memory (e.g., non-transitory media), including but not limited to read-only memory, random access memory, solid state drives, etc. Additionally, such memory may be embedded or external. In this way, the power distribution system 100 as a whole may utilize any combination of embedded memory and external memory. Some embodiments of the power distribution system 100 are generally described in U.S. Patent Application Serial No. 18 / 198,504, entitled "OUTLET IN-RUSH CURRENT LIMITER FOR INTELLIGENT POWER STRIP," filed on May 17, 2023, by Kevin Ferguson, Casey Gilson, Scott Cooper, and Jason Armstrong, which is incorporated herein by reference in its entirety. Thus, any of the IMD 118, PIM 114, or one or more POM 116 may execute program instructions (or a subset of program instructions) individually or in combination to implement the various processing steps disclosed herein. As another example, the power distribution system 100 may have any combination of controllers and does not require at least one of the PIM 114, POM 116, or IMD 118 as depicted in Figures 1A to 1E at least one of the PIM 114, POM 116, or IMD 118 as depicted in

[0071] In some embodiments, various processing steps, combinations of processing steps, or portions of processing steps are distributed among different components of the power distribution system 100. Such a configuration can provide for efficient use of computing resources and / or minimize costs. Additionally, the IMD 118, POM 116, and / or PIM 114 can have different architectures to provide for efficient allocation of computing capabilities for different tasks and to further manage the costs of the overall power distribution system 100. For example, the POM 116 can have a relatively simpler architecture than the IMD 118. By way of illustration, the POM 116 can include a microcontroller and / or a math accelerator adapted to receive data and perform computational tasks on the data, while the IMD 118 can include more advanced processing units such as, but not limited to, a DSP, a microprocessor, a GPU, a faster CPU, etc. In this way, the IMD 118 can perform more computationally intensive tasks, interface with the user, interface with additional devices, etc.

[0072] By way of illustration, each POM 116 can generate or receive load diagnostic data associated with the load 102 and / or classify any of the loads in the load 102 based on the load diagnostic data. Each POM 116 can then send the results of the classification and / or the underlying load diagnostic data to the IMD 118. The IMD 118 can then provide further processing and / or communication with an external system (e.g., a server, other power distribution systems (e.g., PDUs), etc.). For example, the IMD 118 can manage and collect data from multiple POM 116s and / or direct actions based on data from any of the POM 116s. As another example, the IMD 118 can receive user commands and / or instructions and can direct components such as the PIM 114 or POM 116 to comply with such instructions. As another example, the IMD 118 can provide instructions and / or executable operations to the POM 116 that are adapted to classify the load signal 156 based on the load diagnostic data. For example, the IMD 118 can train a machine learning model based on load diagnostic data from any of the POM 116s, the LC subsystem 106, and / or an external source and then provide the trained machine learning model to the POM 116 for implementation. As another example, the IMD 118 can receive a trained machine learning model and provide the trained machine learning model to the POM 116 for implementation. In this way, the machine learning model can generally be trained by any internal or external component of the power distribution system 100.

[0073] Figure 2FIG. 0 is a flow chart showing steps performed in a method 200 for classifying a load 102 according to one or more embodiments of the present disclosure. Embodiments and enabling technologies previously described herein in the context of a power distribution system 100 should be interpreted as extending to method 200. For example, the IMD 118, PIM 114, and / or POM 116 may execute program instructions such that the associated processor(s) individually or in combination implement (or direct the implementation of) any of the processing steps in method 200 or portions thereof. However, it should also be noted that method 200 is not limited to the architecture of power distribution system 100. In this manner, any number or type of components may be used to implement the various processing steps of method 200.

[0074] In an embodiment, method 200 includes a processing step 202 of capturing load diagnostic data for a load 102 (e.g., for any load 102 connected to any outlet in outlet 110).

[0075] Load diagnostic data may generally include any data that may indicate the operating condition of load 102 or that is affected by the operating condition of load 102. For example, load diagnostic data may include temperature data such as, but not limited to, the temperature of load 102, the ambient temperature around load 102, the temperature of the fluid leaving the respective supply nozzle 126, the temperature of the fluid entering the respective return nozzle 128, or the ambient conditions (e.g., temperature, humidity, moisture, etc.) of any load in the power distribution system 100 (e.g., in or around one or more enclosures of the power distribution system 100) and / or load 102. As another example, load diagnostic data may include a load signal 156 (or data derived therefrom) associated with an electrical connection. For example, load diagnostic data may include instantaneous or time series data or root mean square (RMS) values of current and / or voltage data (e.g., current and / or voltage drawn by any load in load 102) associated with the load signal 156 between load 102 and the associated outlet 110. As another example, load diagnostic data may include utilization data associated with load 102 (e.g., utilization data of at least one of the CPU or GPU of load 102).

[0076] Load diagnostic data may be generated by any combination of components associated with or external to the power distribution system 100. For example, load diagnostic data may be captured by the environmental sensor 136, POM 116, PIM 114, or provided to the power distribution system 100 via the data port 138.

[0077] In an embodiment, method 200 includes a processing step 204 of classifying load 102 into two or more categories based on load diagnostic data. For example, the two or more categories include one or more normal categories associated with one or more acceptable operating conditions and one or more atypical categories associated with one or more atypical operating conditions.

[0078] In some embodiments, load 102 is classified using a binary category classification system that includes a single normal category and a single atypical category. Such a configuration can be well-suited but not limited to characterizing a single type of load 102, where an alarm for atypical behavior is desired. In some embodiments, load 102 is classified using a multi-category classification system that includes a single normal category and multiple atypical categories. Such a configuration can be well-suited for but not limited to distinguishing different atypical categories. For example, different atypical categories can be associated with different failure mechanisms and / or different pre-failure conditions, such as but not limited to cold plate failure, manifold fouling, or power failure.

[0079] Referring again to Figure 2 , in some embodiments, step 204 of classifying load 102 into two or more categories based on load diagnostic data includes using a machine learning model that receives the load diagnostic data as input to classify load 102 into two or more categories.

[0080] The machine learning model can utilize any type of learning or combination of learning types, including but not limited to supervised learning, unsupervised learning, or reinforcement learning. Additionally, any type or structure of machine learning model known in the art can be utilized in step 204, such as but not limited to a support vector machine classifier, a nearest neighbor classifier, a perceptron, a logistic regression classifier, or a Bayesian classifier.

[0081] Figure 3 is a flowchart depicting classification using a machine learning model according to one or more embodiments of the present disclosure.

[0082] As Figure 3 depicted in, various load diagnostic data 302 can be preprocessed (block 304) into a form suitable for use with a machine learning model structure 306. For example, the machine learning model structure 306 can include but not limited to a multi-layer perceptron (MLP), a convolutional neural network (CNN), etc.

[0083] The preprocessing step (block 304) can include any processing step suitable for providing the load diagnostic data 302 (or a portion thereof) in a form suitable for use with the machine learning model structure 306. For example, the load diagnostic data 302 can be provided in the form of a tensor. As another example, the load diagnostic data 302 can be normalized.

[0084] Then, the machine learning model structure 306 can be trained using any number of training epochs with a loss function 308, an optimization function 310, and an activation function 312. Any suitable loss function 308 can be utilized, such as but not limited to a binary cross-entropy loss function or a hinge loss function. Any suitable optimization function 310 can be utilized, such as but not limited to a gradient descent function (e.g., a stochastic gradient descent function, etc.). Any suitable activation function 312 can be utilized, such as but not limited to tanh for hidden layers, a sigmoid function for output layers (e.g., for binary class classification), or a softmax function (e.g., for multi-class classification).

[0085] The model configuration, weights, and biases can be stored in the memory 314 and updated through the training process.

[0086] The training data can generally include labeled payload diagnostic data associated with two or more classes and other labeled inputs (e.g., payload diagnostic data associated with known classes). The training data can also be provided by any source. For example, the training data can be synthetically generated locally or remotely. As another example, the training data can be associated with historical payload diagnostic data with known classes, which can be generated locally or received from a remote source (e.g., a data lake, etc.).

[0087] Once trained, the machine learning model structure 306 can be used to classify the payload 102 into one of any number of classes 316 (e.g., prediction labels) based on the associated payload diagnostic data 302. In some embodiments, the machine learning model structure 306 provides the probability that a particular payload 102 belongs to each class. In this way, the class with the highest probability can correspond to the assigned class.

[0088] Note that Figure 3 depicts a non-limiting configuration for implementing multi-class classification. For example, Figure 3 depicts classifying the payload 102 into four classes 316, including a first class associated with normal operating conditions and three atypical classes. In particular, the atypical classes include a second class associated with cold plate failure, a third class associated with fouling of the manifold 124, and a fourth class associated with power failure. As another example, Figure 3 depicts a classification based on the payload diagnostic data 302, which includes GPU current RMS value, GPU voltage RMS value, manifold 124 supply temperature, manifold 124 return temperature, GPU temperature, ambient temperature, and CPU / GPU utilization percentage. Note here that Figure 3The associated description is provided for illustrative purposes only and should not be construed as limiting the scope of the present disclosure. For example, any combination of the load diagnostic data 302, the number of categories 316, or the type of categories 316 is within the spirit and scope of the present disclosure.

[0089] It is further contemplated herein that the actual implementation of machine learning models such as, but not limited to Figure 3 the machine learning models depicted in may be computationally intensive. Accordingly, it may be desirable to distribute various tasks among different computing elements to balance throughput and system cost.

[0090] In some embodiments, as Figures 1A to 3 the power distribution system 100 depicted in may distribute various implementations among different hardware components.

[0091] For example, the IMD 118 may perform computationally intensive tasks of training a machine learning model and then send the trained model to the POM 116. By way of illustration, the IMD 118 may compute weights and biases, the activation for each layer for forward propagation, and the optimization / loss for backpropagation. This data may then be sent to the POM116 along with the structure of the machine learning model. Each POM 116 may then classify the load 102 based on the load diagnostics using the trained machine learning model.

[0092] As another example, the IMD 118 may receive a trained machine learning model and then send the trained model to the POM 116. In this manner, the machine learning model may be trained by an external system, which may significantly reduce the computational requirements of the power distribution system 100.

[0093] The IMD 118 may also provide additional tasks such as, but not limited to, logging, communication with external sources (e.g., data lakes) for training and / or archival purposes, commanding the POM 116 (or other components) to capture data at selected times for training and / or classification, or selecting a particular machine learning model for implementation (e.g., selecting a binary class model, a multi-class model, etc.).

[0094] Referring again to Figure 2 , the method 200 may also include a processing step 206 of generating one or more alert signals. For example, when one or more alert conditions are met, an alert signal may be generated based on the classified load signal 156. Any alert condition may be utilized such as, but not limited to, when the load signal 156 is classified as one of one or more atypical conditions.

[0095] The alarm signal can also trigger any action. For example, the alarm signal can trigger the automatic disconnection of the associated load 102. As another example, the alarm signal can trigger at least one of a visual indicator (e.g., via one or more visual display devices 120) or an audio indicator (e.g., via one or more speakers 122). In this way, any combination of visual or audio text, speech, graphics, or other indicators can be provided to the user. By way of illustration, the alarm signal can trigger visual and / or audio indications that one or more loads 102 may need to be disconnected, repaired, or maintained. As another example, the alarm signal can trigger a remote indicator using any selected protocol, such as but not limited to email, Simple Network Management Protocol (SNMP), Modbus, or the application programming interface (API) of a selected service.

[0096] The subject matter described herein sometimes shows different components that are included within or connected to other components. It should be understood that such depicted architectures are merely exemplary, and in fact, many other architectures can be implemented that achieve the same functionality. In a conceptual sense, any arrangement of components that achieve the same functionality is effectively "associated" such that the desired functionality is achieved. Thus, any two components combined herein to achieve a particular function can be considered to be "associated" with each other such that the desired functionality is achieved, regardless of the architecture or intermediate components. Similarly, any two components so associated can also be considered to be "connected" or "coupled" to each other to achieve the desired functionality, and any two components that can be so associated can also be considered to be "couplable" to each other to achieve the desired functionality. Specific examples of couplable include but are not limited to components that physically interact and / or physically interact and / or wirelessly interact and / or wirelessly interact and / or logically interact and / or logically interact.

[0097] It is believed that the present disclosure and its many attendant advantages will be understood from the foregoing description, and it will be apparent that various changes in the form, construction, and arrangement of the components can be made without departing from the disclosed subject matter or sacrificing all of its material advantages. The described forms are merely illustrative, and the appended claims are intended to cover and include such changes. Additionally, it should be understood that the invention is defined by the appended claims.

Claims

1. A power distribution system, comprising: a power distribution subsystem comprising power distribution circuitry configured to provide power from an input source to a plurality of loads through a plurality of outlets, wherein each outlet of the plurality of outlets is configured to provide an electrical connection to any connected load of the plurality of loads; a liquid cooling subsystem configured to provide active cooling to at least some of the plurality of loads, wherein the liquid cooling subsystem comprises one or more manifolds providing one or more supply nozzles and one or more return nozzles for directing fluid for the active cooling; and One or more housings configured to at least partially enclose the power distribution subsystem and the liquid cooling subsystem, wherein the one or more housings also provide thermal coupling between the power distribution subsystem and the liquid cooling subsystem for at least partially cooling at least a portion of the power distribution subsystem.

2. The power distribution system according to claim 1, wherein: The one or more housings include a single housing at least partially surrounding the power distribution subsystem and the liquid cooling subsystem.

3. The power distribution system according to claim 1, wherein: The one or more housings include: a first housing at least partially enclosing the liquid cooling subsystem; and A second housing at least partially encloses the power distribution subsystem, wherein the first housing and the second housing are thermally coupled.

4. The power distribution system of claim 3, further comprising one or more hinges to provide access to a thermal interface material providing the thermal coupling via rotation of at least one of the first housing or the second housing.

5. The power distribution system according to claim 1, further comprising: one or more controllers, wherein each of the plurality of outlets is coupled to at least one of the one or more controllers, wherein a respective controller of the one or more controllers comprises one or more processors configured to execute program instructions stored on a memory device, wherein the program instructions are configured to cause the one or more processors to: receiving load diagnostic data for any load connected to any outlet of the plurality of outlets; and The plurality of loads connected to at least one of the plurality of outlets are classified into two or more categories based on the load diagnostic data.

6. The power distribution system according to claim 5, wherein: The load diagnostic data for a corresponding load among the multiple loads includes at least one of the following: a temperature of the corresponding load, a temperature of a fluid leaving a corresponding supply nozzle among the one or more supply nozzles, a temperature of a fluid leaving any of the one or more manifolds, a temperature of a fluid entering a corresponding return nozzle of the one or more return nozzles, a temperature of a fluid entering any of the one or more manifolds, an ambient temperature of the power distribution system, an ambient temperature of the corresponding load, a die temperature associated with a processor of the corresponding load, utilization data of at least one of a central processing unit or a graphics processing unit of the corresponding load, a current drawn by the corresponding load, or a voltage drawn by the corresponding load.

7. The power distribution system according to claim 5, wherein: The two or more categories include one or more normal categories associated with one or more acceptable operating conditions, wherein the two or more categories further include one or more atypical categories associated with one or more atypical operating conditions.

8. The power distribution system according to claim 7, wherein: The two or more categories comprise a set of binary categories, wherein the one or more normal categories comprise a single normal category, and wherein the one or more atypical categories comprise a single atypical category.

9. The power distribution system according to claim 7, wherein: The two or more categories include: Two or more category sets, wherein each of the two or more category sets corresponds to a different load type, wherein each of the two or more category sets includes at least one normal category of the one or more normal categories and at least one atypical category of the one or more atypical categories.

10. The power distribution system according to claim 7, wherein: The one or more controllers are also configured to execute a subset of the program instructions, which causes the one or more controllers to generate one or more alarm signals when at least one load of the plurality of loads is classified as one of the one or more atypical categories.

11. The power distribution system according to claim 7, wherein: The one or more controllers are also configured to execute a subset of the program instructions, which causes the one or more controllers to disconnect power to at least one of the multiple loads when the at least one load among the multiple loads is classified as one of the one or more atypical categories.

12. The power distribution system according to claim 5, wherein: The one or more controllers include: one or more first controllers configured to be communicatively coupled to the plurality of outlets, wherein the one or more first controllers are configured to execute a subset of the program instructions that cause the one or more first controllers to classify the plurality of loads into the two or more categories using a machine learning model; and One or more second controllers configured to be communicatively coupled to the one or more first controllers, wherein the one or more second controllers are configured to execute a subset of the program instructions that cause the one or more second controllers to train the machine learning model using training data that includes labeled load diagnostic data associated with the two or more categories.

13. The power distribution system according to claim 12, wherein: The one or more first controllers and the one or more second controllers are located within the one or more housings.

14. The power distribution system according to claim 12, wherein: At least one of the one or more first controllers or the one or more second controllers is located outside the one or more housings.

15. The power distribution system according to claim 12, wherein: The one or more first controllers include one or more power output modules, wherein the one or more second controllers include one or more interchangeable monitoring devices.

16. The power distribution system according to claim 12, wherein: The one or more first controllers utilize embedded memory, wherein the one or more second controllers utilize external memory.

17. The power distribution system according to claim 12, wherein: At least some of the tagged load diagnostic data are associated with historical load diagnostic data.

18. The power distribution system according to claim 17, wherein: The historical load diagnostic data is provided by at least one of: the power distribution system, one or more other power distribution systems, known historical faults of the plurality of loads, or a data lake.

19. The power distribution system according to claim 5, wherein: The one or more controllers are also configured to execute a subset of the program instructions, which causes the one or more controllers to display information associated with at least one of the multiple loads on a display device based on an associated classification according to the load diagnostic data.

20. A method for distributing electricity, comprising: capturing load diagnostic data for a plurality of loads connected to a plurality of outlets to receive power from an input power source connected to the plurality of outlets using a power distribution system, wherein the power distribution system includes a liquid cooling subsystem configured to provide active cooling to at least some of the plurality of loads, wherein the liquid cooling subsystem includes one or more manifolds, the one or more manifolds providing one or more supply nozzles and one or more return nozzles for directing fluid for the active cooling; categorizing the plurality of loads connected to at least one of the plurality of outlets into two or more categories based on the load diagnostic data, wherein the two or more categories include one or more normal categories associated with one or more acceptable operating conditions, wherein the two or more categories further include one or more atypical categories associated with one or more atypical operating conditions; and When a specific load among the plurality of loads is classified into one of the one or more atypical categories, performing at least one of the following: generating one or more alarm signals; or Power to the particular load is disconnected.

Images