A robot chip microchannel liquid cooling abnormal grading disposal method and system
By generating expected cooling states and pump/valve response deviations, anomaly judgment states are formed. Cooling requirements and handling capabilities are matched, and action preconditions and mutual exclusion relationships are set. This solves the problem of the disconnect between anomaly judgment and handling range in robot chip liquid cooling control, and achieves accurate anomaly handling and system consistency.
Patent Information
- Application Number
- CN202611337350.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies in robot chip thermal management and liquid cooling control have difficulty distinguishing between changes in thermal load, failure of circulating pumps or valves to respond to the target state, flow rate decrease due to changes in microchannel cooling branch flow resistance, pressure difference abnormalities, and chip temperature rise. This leads to a disconnect between the anomaly judgment results and the system's executable handling range, and the control basis for branch isolation, activation of backup cooling branches, and chip load reduction is inconsistent.
By acquiring the chip thermal load status, branch hydraulic status, and the target and actual status of pumps and valves, the expected cooling status is generated. The pump and valve response deviation is compared to form an anomaly judgment status. The cooling demand gap is matched with the available handling capacity status to generate an anomaly level and handling target. Preconditions and mutual exclusion relationships for actions are set, and the handling sequence is executed to update the effective branch set and regenerate the branch target flow.
It enables accurate identification and effective handling of anomaly types, reduces anomaly type confusion, ensures that handling actions match the actual system capabilities, maintains the continuity and consistency of liquid cooling control, avoids ineffective adjustment actions, and ensures that subsequent control is consistent with the actual branch status.
Smart Images

Figure CN123124288A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robot chip thermal management and liquid cooling control technology, and in particular to a method and system for graded handling of microchannel liquid cooling anomalies in robot chips. Background Technology
[0002] In the field of robot chip thermal management and liquid cooling control technology, high-power chips typically dissipate heat through multiple microchannel cooling branches. Each microchannel cooling branch is connected to a circulating pump, flow regulating valve, or isolation valve, and the coolant supply is adjusted based on the flow rate, differential pressure, and temperature detection status. Existing solutions generally identify anomalies based on the relationship between chip temperature, branch flow rate, branch differential pressure, circulating pump speed, or valve opening and a set threshold, and then perform circulating pump adjustment, valve adjustment, branch isolation, switching to a backup cooling branch, chip load reduction, or shutdown protection.
[0003] Existing solutions often rely on single-state out-of-limits conditions, fixed risk levels, or preset fault types as the basis for handling. When the chip's thermal load changes, the circulating pump or valve fails to respond according to the target state, or the flow resistance of the microchannel cooling branch changes, the decrease in flow rate, abnormal pressure difference, and increase in chip temperature may present similar external states. It is difficult to distinguish between changes in thermal load, pump failure, valve failure, and branch failure based on a single measurement value. Although some solutions combine multiple operating parameters for fault identification, there is a lack of continuous matching between the anomaly judgment results and the adjustable margins of the circulating pump, valves, backup cooling capacity, and chip de-load capacity. The anomaly level is prone to disconnect from the actual executable handling range of the current system.
[0004] Furthermore, branch isolation, activation of backup cooling branches, chip load reduction, and shutdown protection involve different controlled objects, and there are execution sequences and state constraints between each action. If a backup cooling state is not established or the corresponding chip operating state is not adjusted before branch isolation, the isolation action may change the total loop flow and the hydraulic state of other microchannel cooling branches; if the original branch target flow is still used after the branch connectivity changes, it is also easy to cause inconsistencies in the control basis before and after the handling. Therefore, it is necessary to address the problems of expected cooling state, pump and valve response deviation, cooling demand gap, handling availability, and difficulty in continuous matching between handling actions in the case of abnormal multi-branch microchannel liquid cooling of robot chips. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a method for graded handling of microchannel liquid cooling anomalies in robot chips, comprising: S100: Obtain the chip thermal load status, chip temperature status, branch hydraulic status, pump and valve target and actual status, and the correspondence between the branch and the chip area; Based on the chip thermal load status, match the normal operation benchmark and generate the expected cooling status; S200: Compare the hydraulic state of the branch and the chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve to generate hydraulic deviation, thermal state deviation, pump response deviation and valve response deviation; generate anomaly judgment state based on the deviation combination relationship and cross-branch correlation relationship; S300: Determine the cooling demand gap based on the anomaly determination status and the expected cooling status; generate the available handling capacity status based on the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the backup cooling capacity, and the chip deload capacity; match the cooling demand gap with the available handling capacity status to generate the anomaly level and handling target. S400. Generate candidate handling actions based on the anomaly level and handling target, and generate a handling sequence according to the action preconditions and action mutual exclusion relationships; execute the handling sequence, and when the handling sequence includes a branch isolation action or a backup cooling branch activation action, update the effective branch set and regenerate the branch target flow, and update the anomaly level according to the pump and valve status, branch hydraulic status and chip operating status after handling.
[0006] Furthermore, the branch hydraulic condition includes branch flow rate and branch pressure difference; The target and actual status of the pump and valve include the target operating volume of the circulating pump, the actual speed of the circulating pump, the target status of the valve, and the valve position; Based on the branch identifier, chip area, time identifier, and the correspondence between the branch and the chip area, the chip thermal load status, chip temperature status, branch flow rate, branch pressure difference, circulating pump target operating volume, circulating pump actual speed, valve target status, and valve position are associated. The expected cooling state is generated by matching the expected branch flow rate, expected branch pressure difference, and expected chip temperature change state with the chip thermal load state.
[0007] Furthermore, the total loop flow rate and the supply-return hydraulic pressure difference are obtained; within the pump response time window after the target operating amount of the circulating pump changes, the changes in the actual speed of the circulating pump, the total loop flow rate, and the supply-return hydraulic pressure difference are compared with the changes in the target operating amount of the circulating pump; when the actual speed of the circulating pump, the total loop flow rate, or the supply-return hydraulic pressure difference does not form a change that matches the target operating amount of the circulating pump, the pump response deviation is generated; Within the valve response time window after a change in the valve target state, the changes in valve position, branch flow rate, and branch pressure difference are compared with the changes in the valve target state; if the valve position, branch flow rate, or branch pressure difference does not form a change that matches the valve target state, a valve response deviation is generated.
[0008] Furthermore, the anomaly determination status includes anomaly type and anomaly impact range, and its generation includes: When the flow rate of a single microchannel cooling branch decreases, the differential pressure increases, the valve response deviation is not formed, and the corresponding chip area forms the thermal state deviation, a branch failure state and a single branch range are generated. When the flow rates of multiple microchannel cooling branches decrease in the same direction and the pump response deviates from the formation, a pump failure state and the total liquid supply circuit range are generated. When the valve response deviation is formed, a valve fault state and the corresponding microchannel cooling branch range are generated. When the hydraulic deviation does not occur but the thermal deviation does occur, an abnormal temperature state and a corresponding chip region range are generated.
[0009] Furthermore, the cooling demand gap is generated based on the difference between the expected branch flow and the current branch flow within the range of the anomaly's influence. The adjustable margin of the circulating pump is generated based on the current operating rate of the circulating pump and the allowable operating range of the pump. The valve's adjustable margin is generated based on the valve's current opening degree and its opening range; The branch isolation status and backup cooling capacity are generated based on the branch isolation conditions and the operating status of the backup cooling branch. The chip deload capability is generated based on the current power consumption of the affected chip region and the chip's allowable power consumption range.
[0010] Furthermore, a chip thermal risk is generated based on the chip temperature state and temperature change trend; When the adjustable margin of the circulating pump and the adjustable margin of the valve match the cooling demand gap, a first anomaly level and a liquid cooling compensation target are generated. When the branch isolation status and backup cooling capacity match the cooling demand gap, a second anomaly level and a branch reconfiguration target are generated. When the chip's deload capacity matches the cooling demand gap, a third anomaly level and a chip deload target are generated. When the cooling demand gap does not match the available processing capacity, or when the chip thermal risk reaches the shutdown protection condition, a fourth abnormality level and a shutdown protection target are generated.
[0011] Furthermore, the generation of the disposal sequence includes: When the anomaly type is pump failure, the liquid cooling compensation action that increases the target operating rate of the faulty circulating pump is removed from the candidate handling actions. When the anomaly type is valve failure, the liquid cooling compensation action that adjusts the target state of the faulty valve is removed from the candidate handling actions. When the candidate handling action includes branch isolation action, the prerequisite for the branch isolation action is that the backup cooling branch reaches the branch target flow or the affected chip area reaches the chip deload target. Set the liquid cooling compensation action and the branch isolation action for the same microchannel cooling branch to be mutually exclusive.
[0012] Furthermore, the updating of the effective branch set and the branch target traffic includes: The handling sequence includes removing the isolated microchannel cooling branch from the set of effective branches when the branch isolation action is performed; The processing sequence includes adding the activated backup cooling branch to the set of valid branches when the backup cooling branch is activated. Based on the set of effective branches, the total loop flow rate, and the expected cooling state of each chip region, the target flow rate of each effective microchannel cooling branch is generated. Stop generating the branch target flow for the isolated microchannel cooling branch.
[0013] Furthermore, the update of the anomaly level includes: Within the treatment verification time window, the actual speed of the circulating pump, valve position, branch flow rate, branch pressure difference, chip power consumption and chip temperature after treatment are compared with the verification conditions corresponding to the treatment target. If the verification conditions are not met, the scope of the anomaly does not change, and the current anomaly level is lower than the fourth anomaly level, the anomaly level is increased; if the current anomaly level is the fourth anomaly level, the shutdown protection target is maintained. When hydraulic or thermal deviations occur in other microchannel cooling branches, update the scope of the anomaly and regenerate the anomaly level and handling target. When the verification conditions are continuously met and the recovery and maintenance conditions are met, the hydraulic state and pump valve state of the branch in the effective branch set are first confirmed to meet the expected cooling state, and then the chip power consumption is gradually restored; when the isolated microchannel cooling branch meets the recovery conditions, its operating state is restored.
[0014] This invention also provides a graded handling system for microchannel liquid cooling anomalies in robot chips, comprising: The expected cooling state generation module is used to obtain the chip thermal load state, chip temperature state, branch hydraulic state, pump valve target and actual state, and the correspondence between the branch and the chip area, and generate the expected cooling state based on the chip thermal load state and the normal operation benchmark. An anomaly determination module, connected to the expected cooling state generation module, is used to compare the branch hydraulic state and chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve, generating hydraulic deviation, thermal state deviation, pump response deviation and valve response deviation, and generating anomaly determination states based on the deviation combination relationship and cross-branch correlation relationship. The handling capacity and level generation module is connected to the anomaly determination module. It is used to determine the cooling demand gap based on the anomaly determination status and the expected cooling status, generate the handling availability status based on the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the backup cooling capacity and the chip deload capacity, and match the cooling demand gap with the handling availability status to generate the anomaly level and handling target. The disposal sequence generation module, connected to the disposal capability and level generation module, is used to generate candidate disposal actions based on the anomaly level and disposal target, and to generate a disposal sequence according to the action preconditions and action mutual exclusion relationships. The execution feedback module is connected to the disposal sequence generation module and is used to execute the disposal sequence. When the disposal sequence includes branch isolation or activation of backup cooling branch, the effective branch set is updated and the branch target flow is regenerated. The abnormality level is updated according to the pump and valve status, branch hydraulic status and chip operating status after disposal.
[0015] The key innovations of this invention include: (1) The expected cooling state corresponding to the chip thermal load state is used as the comparison benchmark between the branch hydraulic state and the chip temperature state. The response relationship between the target operating amount of the circulating pump, the target state of the valve and the actual speed, valve position and corresponding hydraulic changes is included in the abnormal evidence. Based on the combination relationship of hydraulic deviation, thermal state deviation, pump response deviation, valve response deviation and cross-branch correlation, an abnormal judgment state including abnormal type and abnormal influence range is formed.
[0016] (2) Based on the expected cooling status within the scope of the abnormality and the current branch hydraulic status, a cooling demand gap is formed, and the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the standby cooling capacity and the chip deload capacity are organized into the available capacity status for handling. When the pump response deviation or valve response deviation occurs, the theoretical adjustment margin of the corresponding fault execution object is excluded, and the abnormality level and handling target are formed based on the matching relationship between the cooling demand gap and the available capacity status for handling.
[0017] (3) Candidate handling actions are defined by the anomaly type, anomaly level and handling target. The execution order of liquid cooling compensation, branch reconstruction, chip load reduction and shutdown protection is organized by the action preconditions and action mutual exclusion relationship. After the branch isolation or the backup cooling branch is activated, the effective branch set is updated and the branch target flow is regenerated. Then, the anomaly level is updated according to the pump and valve status, branch hydraulic status and chip operating status after handling.
[0018] The following are its main beneficial effects: (1) By combining the expected cooling state corresponding to the chip thermal load with the actual response after pump and valve control for anomaly determination, normal thermal load changes, unresponsive circulating pumps, unresponsive valves and abnormal microchannel cooling branch status can be distinguished, reducing confusion of anomaly types caused by exceeding limits of temperature, flow rate or pressure difference alone, and making the scope of anomaly impact correspond to the actual branch and chip area where deviation occurs.
[0019] (2) By matching the cooling demand gap with the current available liquid cooling regulation capacity, branch reconfiguration capacity and chip load reduction capacity, the anomaly level reflects the relationship between the anomaly handling demand and the remaining handling capacity; after the corresponding actuator forms a response deviation, its theoretical adjustment margin no longer participates in the capacity matching, reducing the situation where the anomaly level is inconsistent with the actual executable handling range.
[0020] (3) By setting preconditions and mutual exclusion relationships for candidate handling actions, it is possible to avoid continuing to output invalid adjustment actions for faulty circulating pumps or faulty valves, and to isolate the branch until the backup cooling branch reaches the branch target flow or the affected chip area reaches the chip deload target; after the effective branch set changes, the branch target flow is regenerated so that the subsequent liquid cooling control is consistent with the actual branch connection status, and the abnormal level is maintained, upgraded or restored according to the handling feedback from the liquid cooling side and the chip side. Attached Figure Description
[0021] Figure 1 A flowchart illustrating a method for graded handling of microchannel liquid cooling anomalies in a robot chip, provided in this application embodiment; Figure 2 This is a structural block diagram of a robot chip microchannel liquid cooling anomaly classification and treatment system provided in an embodiment of this application. Detailed Implementation
[0022] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for graded handling of microchannel liquid cooling anomalies in a robot chip according to an embodiment of the present invention. The process may include at least steps S100-S400: S100: Obtain the chip thermal load status, chip temperature status, branch hydraulic status, pump and valve target and actual status, and the correspondence between the branch and the chip area; Based on the chip thermal load status, match the normal operation benchmark and generate the expected cooling status; S200: Compare the hydraulic state of the branch and the chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve to generate hydraulic deviation, thermal state deviation, pump response deviation and valve response deviation; generate anomaly judgment state based on the deviation combination relationship and cross-branch correlation relationship; S300: Determine the cooling demand gap based on the anomaly determination status and the expected cooling status; generate the available handling capacity status based on the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the backup cooling capacity, and the chip deload capacity; match the cooling demand gap with the available handling capacity status to generate the anomaly level and handling target. S400. Generate candidate handling actions based on the anomaly level and handling target, and generate a handling sequence according to the action preconditions and action mutual exclusion relationships; execute the handling sequence, and when the handling sequence includes a branch isolation action or a backup cooling branch activation action, update the effective branch set and regenerate the branch target flow, and update the anomaly level according to the pump and valve status, branch hydraulic status and chip operating status after handling.
[0023] S100: Obtain the chip thermal load status, chip temperature status, branch hydraulic status, target and actual pump and valve status, and the correspondence between branches and chip areas; based on the chip thermal load status, match the normal operation benchmark to generate the expected cooling status: This implementation is applied to a liquid cooling system with multiple robot chip regions and multiple microchannel cooling branches. Each chip region forms heat exchange contact with a microchannel liquid cooling plate, which is connected between the supply manifold and the return manifold, forming a corresponding microchannel cooling branch. The liquid cooling circuit includes a circulation pump, branch flow regulating valves, and branch isolation valves. The circulation pump is controlled by the liquid cooling controller, the branch flow regulating valves adjust their target states according to the branch's target flow rate, and the branch isolation valves shut off the corresponding microchannel cooling branch during abnormal operation. The chip power management unit provides the liquid cooling controller with the chip thermal load status and chip operating status, and the robot safety controller receives shutdown protection commands. Flow sensors are installed in each microchannel cooling branch, pressure sensors are installed at the inlet and outlet of the microchannel cooling branch, or differential pressure sensors directly detect the branch pressure difference; the chip temperature status is obtained by the chip's built-in temperature sensor or a temperature sensor corresponding to the chip region.
[0024] The liquid cooling controller reads the chip thermal load status, chip temperature status, branch flow rate, branch differential pressure, circulating pump target operating rate, circulating pump actual speed, valve target status, and valve position according to a set acquisition cycle. When the circulating pump target operating rate, valve target status, chip power consumption, or effective branch set changes, the liquid cooling controller does not wait for the next complete acquisition cycle but initiates a status acquisition after the corresponding status change. The chip thermal load status can be represented by the chip's current power consumption, chip task load, or the thermal load output by the chip power management unit. The chip temperature status includes the current temperature and the temperature change within the continuous acquisition cycle. The branch hydraulic status includes the branch flow rate and branch differential pressure; the pump and valve target and actual status includes the circulating pump target operating rate, circulating pump actual speed, valve target status, and valve position.
[0025] The correspondence between branches and chip regions is pre-configured in the liquid cooling controller. This configuration indicates the chip region cooled by each microchannel cooling branch, and the connection relationships between branches when a chip region is cooled by multiple microchannel cooling branches. The liquid cooling controller associates the chip thermal load status, chip temperature status, branch flow rate, branch pressure difference, and pump / valve status within the same acquisition period according to branch identifiers, chip regions, and time identifiers. Since the acquisition cycles for chip status, flow rate status, and pump / valve feedback may differ, the liquid cooling controller selects the most recent valid status within the time alignment window; statuses outside the time alignment window are not directly combined with the branch status at the target time. If a sensor does not return a valid status within the time alignment window, the liquid cooling controller retains the missing status for that acquisition location and uses the status from the previous acquisition cycle that has not exceeded the valid hold time. Missing statuses do not directly trigger branch isolation or shutdown protection, but are checked together with other hydraulic states and execution responses in S200.
[0026] The normal operating baseline is matched to the chip thermal load state and the current liquid cooling circuit structure. The normal operating baseline can be formed during the liquid cooling system commissioning phase by observing the operating conditions of the circulating pump and branch flow control valves when they are in normal operation; alternatively, it can be formed from historical normal operating conditions within the same chip thermal load range, the same set of effective branches, and the same coolant inlet temperature. The normal operating baseline includes the expected branch flow rate, the expected branch pressure difference, and the expected chip temperature change state. In one embodiment, the chip thermal load state is divided into continuous thermal load ranges, and the liquid cooling controller calls the corresponding baseline based on the range in which the current chip thermal load falls. In another embodiment, the liquid cooling controller interpolates the expected states from two adjacent thermal load baselines to obtain the expected cooling state corresponding to the current chip thermal load.
[0027] The expected branch flow rate represents the coolant flow rate required for the corresponding microchannel cooling branch to operate normally under the current chip thermal load; the expected branch pressure difference represents the pressure difference range corresponding to the expected branch flow rate passing through the current microchannel cooling branch; the expected chip temperature change state represents the range of chip temperature change within a set thermal response time during the period when the current chip thermal load and expected branch flow rate are maintained. The matching of the normal operation baseline also considers the effective branch set. When branch isolation or the activation of a backup cooling branch changes the liquid cooling loop connection state, the normal operation baseline before the topology change is no longer used. Instead, the expected branch flow rate and expected branch pressure difference for each chip region are rematched according to the updated effective branch set.
[0028] The liquid cooling controller combines the matched expected branch flow rate, expected branch pressure difference, and expected chip temperature change state to form an expected cooling state, and associates this expected cooling state with the actual operating state under the same branch, the same chip region, and the same time marker. This association serves as the basis for comparison in S200 for hydraulic deviation, thermal state deviation, pump response deviation, and valve response deviation, respectively. When the normal operating reference does not have a state that matches the current chip thermal load and effective branch set, the liquid cooling controller maintains the current pump and valve target state and does not directly generate irreversible actions; simultaneously, it forms a temporary expected cooling state according to the normal operating reference adjacent to the current chip thermal load, and corrects the temporary expected cooling state based on the actual stable operating state in subsequent acquisition cycles.
[0029] S200: Compare the branch hydraulic pressure state and chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve to generate hydraulic deviation, thermal state deviation, pump response deviation, and valve response deviation; generate anomaly judgment states based on the deviation combination relationship and cross-branch correlation relationship: The anomaly evidence and scope determination module receives the expected cooling state generated by S100, as well as the branch hydraulic pressure state, chip temperature state, and target and actual pump and valve states associated with the expected cooling state. The module first checks whether the states being compared correspond to the same microchannel cooling branch, the same chip region, and overlapping acquisition periods; if the states belong to different objects, no deviation combination is performed. When the target operating flow of the circulating pump or the target state of the valve changes, there is a transition process in the branch flow rate and branch pressure difference of the liquid cooling system. Therefore, the module evaluates the actual pump and valve response within the response time window after the control action is applied, rather than directly using short-term fluctuations at the moment the action is applied as evidence of a fault.
[0030] Hydraulic deviation is caused by changes in the actual flow rate and actual differential pressure of a branch relative to the expected cooling state. A flow rate decrease is defined as the actual flow rate of a branch is lower than the allowable range of the expected branch flow rate; a differential pressure abnormality is defined as the actual differential pressure of a branch is higher or lower than the expected range. The anomaly evidence and range determination module generates hydraulic deviation based on the combination of the flow rate decrease and differential pressure abnormalities. A decrease in flow rate accompanied by an increase in differential pressure indicates an increase in flow resistance in the corresponding microchannel cooling branch; a decrease in flow rate accompanied by a decrease in both supply and return hydraulic pressure differentials indicates a change in the fluid supply capacity; when the flow sensor output decreases but the branch differential pressure, total loop flow rate, and corresponding chip temperature do not change accordingly, this flow rate state is retained as a pending verification state and is not directly defined as a branch failure state.
[0031] Thermal state deviation is defined as the deviation of the chip temperature from the expected chip temperature change. When the chip temperature changes in the expected direction within the thermal response time after an increase in thermal load, and the branch flow rate and branch pressure difference remain within the expected cooling range, this temperature change is considered a normal temperature response under thermal load changes. Thermal state deviation occurs when the chip temperature continues to rise without an increase in thermal load, or when the chip temperature fails to enter the corresponding expected range within the thermal response time even though the chip power consumption has decreased. If a hydraulic deviation first forms in the microchannel cooling branch, and subsequently the corresponding chip area experiences a temperature rise within the thermal response time, the hydraulic deviation and thermal state deviation are considered continuous evidence of the same branch anomaly. If a hydraulic deviation does not form but the thermal state deviation persists, the anomaly determination indicates a temperature anomaly in the chip area, and it is not directly attributed to a liquid cooling branch failure.
[0032] Pump response deviation occurs after a change in the target operating rate of the circulating pump. The anomaly evidence and range determination module records the actual speed of the circulating pump, total loop flow rate, and supply-return hydraulic pressure difference before the change in the target operating rate, and acquires the corresponding state after the change within the pump response time window. When the target operating rate of the circulating pump increases, the actual speed of the circulating pump, total loop flow rate, and supply-return hydraulic pressure difference should change according to the response relationship in the normal operating baseline; if one of these states does not change in a matching manner, a pump response pending verification state is first formed. When the actual speed of the circulating pump, total loop flow rate, or supply-return hydraulic pressure difference still does not change in a matching manner within a continuous acquisition period, and multiple microchannel cooling branches show a decrease in flow rate in the same direction, a pump response deviation is formed. When the actual speed of the circulating pump has reached the target state, but the total loop flow rate and supply-return hydraulic pressure difference still deviate from the expected range, the module retains the pump response deviation and, in conjunction with the branch pressure difference distribution, checks whether there is an abnormal common flow resistance in the total fluid supply circuit.
[0033] Valve response deviation is formed based on the changes in valve position and branch hydraulic pressure after the valve's target state changes. When the target opening of the branch flow regulating valve increases, the abnormal evidence and range determination module reads the valve position, branch flow, and branch pressure difference within the valve response time window; if the valve position does not change according to the target state, or if the valve position has changed but the branch flow and branch pressure difference do not change accordingly, a valve response deviation is formed. In embodiments without valve position feedback, the module uses the combined changes in branch flow and branch pressure difference to characterize the actual valve state; if the valve's target state changes continuously while the branch flow and branch pressure difference remain unchanged, a valve response deviation is formed. If the valve position and branch flow change according to the target state, but the branch pressure difference continues to rise, this state is not classified as a valve failure, but is checked together with the thermal state deviation to see if there is an increase in flow resistance or a decrease in heat exchange capacity in the microchannel cooling branch.
[0034] Anomaly determination states are formed based on the combination relationships between multiple deviations and cross-branch correlations. When a single microchannel cooling branch experiences a decrease in flow rate and an increase in differential pressure, but no valve response deviation occurs, and a thermal state deviation occurs in the corresponding chip area, the anomaly determination state includes the branch failure state and the single branch range. When the flow rate of multiple microchannel cooling branches decreases in the same direction within the same time period, accompanied by pump response deviation, the anomaly determination state includes the pump failure state and the range of the total liquid supply circuit. When a valve response deviation occurs only in one branch, the anomaly determination state includes the valve failure state and the range of the corresponding microchannel cooling branch. When no hydraulic deviation occurs but a thermal state deviation occurs in a chip area, the anomaly determination state includes the temperature anomaly state and the range of the corresponding chip area. When there are conflicts between deviation evidence, the anomaly evidence and range determination module retains the current liquid cooling control state and extends the verification cycle; the chip temperature state that reaches the independent shutdown protection condition is not limited by this extended verification process and is directly transmitted to S300 to calculate the chip thermal risk.
[0035] S300: Determine the cooling demand gap based on the anomaly determination status and the expected cooling status; generate a handling availability status based on the adjustable margin of the circulating pump, the adjustable margin of the valves, the branch isolation status, the backup cooling capacity, and the chip's deload capability; match the cooling demand gap with the handling availability status to generate anomaly level and handling target: The handling capacity and level generation module receives the anomaly determination status output by S200 and reads the expected cooling status and current actual operating status of each chip area within the anomaly's impact range. The cooling demand gap represents the difference between the expected branch flow rate and the currently sustainable branch flow rate within the anomaly's impact range. In a single branch range, the handling capacity and level generation module calculates the difference between the expected branch flow rate and the current branch flow rate for that microchannel cooling branch; in multiple branches or a total liquid supply loop range, it calculates the flow rate difference for each affected branch separately and combines it with the total loop flow rate of the shared circulation pump to form a cooling demand gap. If the current branch flow rate is higher than the expected branch flow rate, no positive cooling demand gap is formed; however, if the branch pressure difference or chip temperature is still in an abnormal state, the anomaly determination status remains, and matching is performed by branch reconstruction or chip deload conditions.
[0036] The adjustable margin of the circulating pump is determined based on the current operating rate and the pump's allowable operating range. The current operating rate of the circulating pump can be represented by the current target operating rate and the actual speed. When the current target operating rate has not reached the pump's allowable operating limit and the pump response deviation has not yet occurred, the adjustable range between the target operating rate and the operating limit is included in the circulating pump's adjustable margin. When the pump response deviation has occurred, the remaining target operating range of the faulty circulating pump is not included in the available capacity status. When a standby circulating pump is set up in the liquid cooling circuit, if the standby circulating pump is in a startable state and its inlet and outlet valves meet the switching conditions, the liquid supply capacity corresponding to the standby circulating pump is included in the standby cooling capacity, but not in the adjustable margin of the faulty circulating pump.
[0037] Valve adjustable margin is determined based on the current valve opening and its range. For microchannel cooling branches requiring increased flow, the range between the current valve opening and the maximum allowable opening is used as the positive adjustment margin. For cases requiring coolant redistribution to abnormal branches, it is also necessary to check whether there is available flow in the remaining microchannel cooling branches. Valves whose response deviates from the established parameters are not included in the valve adjustable margin. When a microchannel cooling branch is controlled by multiple valves, the handling capacity and level generation module checks the status of each valve according to the flow direction. If one of the series valves is closed or has a deviated response, the opening range of the remaining valves is not directly converted into the available flow of that branch.
[0038] Branch isolation status is determined based on whether the abnormal branch is equipped with a branch isolation valve, whether a cooling continuity path exists for the corresponding chip area after isolation, and whether the isolation action affects the formation of a shared liquid supply circuit. A branch is marked as executable when it has independent shut-off conditions and the isolation action does not simultaneously shut down a normal branch. Backup cooling capacity is determined based on the current valve position, flow rate, differential pressure, and connection relationships of the backup cooling branch. When a backup cooling branch is not yet activated, its rated state is not directly considered as available capacity; the handling capacity and level generation module first checks the inlet and outlet connection status of the backup branch, the valve status of the backup branch, and the status of the corresponding circulating pump before generating the callable backup cooling capacity. When a backup cooling branch is activated but its flow rate does not reach the target flow rate of the corresponding branch, the branch is in the establishment process, and its cooling capacity is not considered a condition for completing the isolation of the abnormal branch.
[0039] Chip de-load capability is determined by the current power consumption of the affected chip region and the chip's allowable power consumption range. The chip power management unit provides the current power consumption, current task load, and allowable de-load lower limit; the range between the current power consumption and the allowable de-load lower limit is converted into a reducible thermal load and mapped to a cooling demand gap. When the chip is already at the allowable de-load lower limit, its chip de-load capability is zero. If the chip power management unit does not return a de-load receipt, or if the chip's actual power consumption does not change with the de-load command, the theoretical power consumption difference corresponding to the de-load command is not included in the available handling capability status. When multiple chip regions share a single microchannel cooling branch, the handling capability and level generation module reads the allowable power consumption range of each chip region and summarizes the chip de-load capability within that branch range according to the correspondence between the branch and the chip region.
[0040] Chip thermal risk is determined based on the chip's temperature status and temperature change trend. When the chip's current temperature is within the allowable range and the temperature change trend is gradual, the chip thermal risk remains within the liquid cooling compensation range. When the chip's current temperature has not reached the shutdown protection condition but continues to rise, the chip thermal risk enters the load reduction verification range. When the chip's current temperature reaches the shutdown protection condition, or is expected to reach the shutdown protection condition within the response time based on the current temperature change trend, the chip thermal risk enters the shutdown protection range. When there are conflicting temperature sensor states, the handling capability and level generation module checks the chip power consumption, coolant temperature difference, and adjacent temperature sensor states. If a consistent state still cannot be obtained, a higher chip thermal risk is retained, and the already executed protection actions are not lifted based on a single abnormal temperature reading.
[0041] The handling capacity and level generation module sequentially matches the cooling demand gap with the adjustable margins of the circulating pump, valves, branch isolation status, backup cooling capacity, and chip deload capacity. When the increased flow corresponding to the adjustable margins of the circulating pump and valves covers the cooling demand gap, and the chip thermal risk has not entered the deload verification range, a first anomaly level and liquid cooling compensation target are generated. When ordinary pump and valve adjustments cannot cover the cooling demand gap, but the expected cooling state can be maintained through the activation of backup cooling branches, isolation of abnormal branches, and redistribution of effective branch flow, a second anomaly level and branch reconstruction target are generated. When the liquid cooling side handling capacity still cannot cover the cooling demand gap, but the reduction in thermal load corresponding to the chip deload capacity matches the remaining cooling capacity, a third anomaly level and chip deload target are generated. When the cooling demand gap does not match the handling capacity status, or the chip thermal risk reaches the shutdown protection conditions, a fourth anomaly level and shutdown protection target are generated. The anomaly level and the handling target are jointly transmitted to S400; the handling target retains the target branch flow, target chip power consumption, target valve status or shutdown status, for subsequent generation of action sequence and verification conditions.
[0042] S400: Generate candidate handling actions based on the anomaly level and handling target; generate a handling sequence according to action preconditions and action mutual exclusion relationships; execute the handling sequence; when the handling sequence includes a branch isolation action or a backup cooling branch activation action, update the effective branch set and regenerate the branch target flow; update the anomaly level based on the pump and valve status, branch hydraulic status, and chip operating status after handling. The response sequence generation module selects candidate response actions from liquid cooling compensation actions, branch reconfiguration actions, chip load reduction actions, and shutdown protection actions based on the anomaly level, response target, anomaly type, and anomaly impact range generated by S300. Candidate response actions corresponding to the first anomaly level include changing the target operating volume of the circulating pump, changing the target state of the valve, and reallocating the target flow rate of the branch. Candidate response actions corresponding to the second anomaly level include activating the backup circulating pump, activating the backup cooling branch, reducing the current power consumption of the affected chip area, and isolating the abnormal branch. Candidate response actions corresponding to the third anomaly level include limiting the target power consumption of the chip, limiting the chip's workload, and maintaining the operation of the currently effective liquid cooling branch. Candidate response actions corresponding to the fourth anomaly level include stopping the operation of the affected chip area, shutting down the corresponding microchannel cooling branch, or sending a shutdown protection command to the robot safety controller. These candidate response actions only represent control actions to be verified and are not issued directly before the action constraint verification is completed.
[0043] The anomaly type is first considered in the candidate handling action screening. When the anomaly status is pump failure, the handling sequence generation module deletes the liquid cooling compensation action that increases the target operating rate of the failed circulating pump, and retains the actions of activating the backup circulating pump, limiting branch flow, chip deload, and shutdown protection. When the anomaly status is valve failure, the action of continuing to adjust the target state of the failed valve is deleted; if the branch has an independent isolation valve, the isolation action can still be retained. When the anomaly status is single branch failure, the deload action of unrelated chip areas is not added to the current handling sequence; only after the anomaly's impact extends to the main liquid supply circuit will the handling sequence reference the chip deload capacity of multiple affected chip areas.
[0044] The preconditions for actions are used to determine the execution order during branch reconfiguration. When the handling sequence includes actions such as activating the backup cooling branch and isolating the abnormal branch, the target valve status and target operating volume of the circulating pump for the backup cooling branch are first issued. Only after the execution feedback module confirms that the branch flow rate of the backup cooling branch has reached the target flow rate and its branch pressure difference has entered the corresponding expected range is the execution of the abnormal branch isolation action permitted. If the backup cooling branch does not exist or has not reached its target flow rate, the affected chip area must first reach the chip deload target before the branch isolation action can enter the executable state. If the chip deload command has been sent but the actual chip power consumption has not decreased, the preconditions for actions are still not met. The handling sequence maintains the current isolation state of the abnormal branch and includes the chip execution state in subsequent abnormality level updates.
[0045] Action mutual exclusion restricts the same execution object from receiving instructions with opposite directions or conflicting states during the same handling phase. For example, liquid cooling compensation actions that increase the target flow rate of a microchannel cooling branch are mutually exclusive with actions that isolate that microchannel cooling branch; actions that increase the target operating rate of a faulty circulating pump are mutually exclusive with actions that activate a backup circulating pump; actions that restore chip power consumption are mutually exclusive with unresolved chip load reduction targets; when a shutdown protection action has already been executed, ordinary liquid cooling compensation and automatic recovery actions remain locked. The handling sequence generation module forms a directed execution order based on action preconditions and appends corresponding verification conditions after each action. If a precondition is not met within a set waiting time, the subsequent isolation action is not skipped; instead, the incomplete state of the precondition is transmitted to the execution feedback module.
[0046] After the action sequence is formed, the execution feedback module issues control commands to the circulating pump controller, branch valve controller, chip power management unit, and robot safety controller in the order of execution. After each action object returns a control acknowledgment, the execution feedback module continues to verify the actual status to avoid simply accepting the acknowledgment as the completion of the action. The circulating pump action is verified by the actual speed of the circulating pump, the total loop flow rate, and the supply and return pressure difference; the valve action is verified by the valve position, branch flow rate, and branch pressure difference; the chip unloading action is verified by the chip power consumption and chip task load; and the shutdown protection action is verified by the chip operating status and the status returned by the robot safety controller. If the control acknowledgment times out or the actual status does not change, the execution feedback module transmits the corresponding pump response deviation or valve response deviation to the anomaly evidence and range determination module to re-form the anomaly judgment status.
[0047] When the handling sequence includes branch isolation, the execution feedback module, after confirming the isolation status, removes the isolated microchannel cooling branch from the effective branch set and stops generating the branch target flow rate for that branch. When the handling sequence includes backup cooling branch activation, the backup cooling branch is added to the effective branch set after the valve position, branch flow rate, and branch differential pressure of the backup cooling branch meet the operating conditions. The liquid cooling controller then regenerates the branch target flow rate for each effective microchannel cooling branch according to the updated effective branch set, the current total loop flow rate, and the expected cooling status of each chip region. If the total loop flow rate is insufficient to cover the expected flow rate of each branch, the liquid cooling controller synchronously corrects the branch target flow rate and chip deload target according to the chip temperature status and allowable power consumption range of each chip region, and does not continue to use the flow allocation results before branch isolation.
[0048] After the action is executed, a verification time window is entered. The execution feedback module compares the actual speed of the circulating pump, valve position, branch flow rate, branch pressure difference, chip power consumption, and chip temperature with the verification conditions corresponding to the action target. If the branch flow rate and branch pressure difference reach the target range, the chip power consumption reaches the load reduction target, the chip temperature change enters the expected state, and no new hydraulic or thermal deviations occur in the other effective microchannel cooling branches, the current anomaly level remains unchanged until the recovery and maintenance conditions are continuously met. If the verification conditions are not met but the scope of the anomaly's impact does not change, the anomaly level is increased, and the execution status of the original action is used as the input for the next round of action capacity calculation; once the pump and valve have reached the operating boundary, the corresponding adjustment margin is no longer included in the available action capacity status.
[0049] When other effective microchannel cooling branches experience hydraulic or thermal deviations within the handling verification time window, the execution feedback module does not directly use the original anomaly impact range but transmits the newly added deviation status to S200 to re-verify the cross-branch correlation. After the anomaly impact range is updated, S300 recalculates the cooling demand gap, handling availability status, anomaly level, and handling target. When the handling target continuously meets the recovery and maintenance conditions, the liquid cooling controller exits the handling state in the order of first restoring the liquid cooling state and then restoring the chip power consumption. Specifically, it first verifies that the flow rate and differential pressure in the effective branch set are in the expected cooling state, and then gradually restores the chip power consumption; when the chip temperature re-deviates from the thermal state during the recovery process, it stops power consumption recovery and maintains the current pump and valve status. When an isolated branch needs to be put back into operation, it is first added to the verification state, and the flow rate, differential pressure, and valve response are verified before being added to the effective branch set. The branch's target flow rate is not directly restored due to the maintenance completion flag.
[0050] Example 2: Figure 2 A structural block diagram of a robotic chip microchannel liquid-cooled anomaly classification and handling system according to an embodiment of the present invention is shown. Figure 2As shown, the structure may include: The expected cooling state generation module 01 is used to acquire the chip thermal load state, chip temperature state, branch hydraulic pressure state, target and actual pump and valve states, and the correspondence between branches and chip areas. Based on the chip thermal load state, it matches a normal operating benchmark to generate the expected cooling state. The chip thermal load state is provided by the chip power management unit; the chip temperature state is provided by the chip's built-in temperature detection unit or a temperature detection unit located on the microchannel liquid cooling plate; the branch hydraulic pressure state is provided by the flow detection unit and differential pressure detection unit in the microchannel cooling branch; and the target and actual pump and valve states are read from the liquid cooling controller, circulating pump controller, and valve controller, respectively. The correspondence between branches and chip areas is pre-configured in the liquid cooling controller to represent the heat exchange relationship between each microchannel cooling branch and its corresponding chip area. This module correlates chip thermal load status, chip temperature status, branch flow rate, branch pressure difference, circulating pump target operating rate, circulating pump actual speed, valve target status, and valve position according to branch identifier, chip area, and time identifier. When the acquisition cycles of different detection units are inconsistent, the module selects the valid status corresponding to the current control cycle within a preset time alignment range. Statuses outside the time alignment range are not included in the matching for this cycle. The normal operation benchmark includes the expected branch flow rate, expected branch pressure difference, and expected chip temperature change status under different chip thermal load states and different sets of valid branches. This module calls the corresponding normal operation benchmark according to the current chip thermal load status and liquid cooling circuit connection status, and outputs the matching result as the expected cooling status to the anomaly judgment module. When some detection statuses are missing, the status of the previous control cycle that has not exceeded the valid holding time is temporarily used. When a valid status cannot be obtained and no historical status is available, the corresponding branch is set to a pending verification status. This status is transmitted to the anomaly judgment module along with the existing detection results, without directly triggering branch isolation or shutdown protection.
[0051] The anomaly determination module 02, connected to the expected cooling state generation module, compares the branch hydraulic state and chip temperature state with the expected cooling state, and compares the changes in the target and actual states of the pump and valve, generating hydraulic deviation, thermal state deviation, pump response deviation, valve response deviation, and anomaly determination states. After receiving the expected cooling state, this module first checks whether the branch hydraulic state, chip temperature state, and pump and valve states involved in the comparison correspond to the same microchannel cooling branch, the same chip area, and overlapping acquisition periods. Hydraulic deviation occurs when the actual branch flow rate deviates from the expected branch flow rate, or when the actual branch pressure difference deviates from the expected branch pressure difference. Specifically, when the branch flow rate decreases and is accompanied by an increase in branch pressure difference, the hydraulic deviation is marked as a single-branch flow resistance change state; when the flow rates of multiple microchannel cooling branches decrease in the same direction within the same control cycle, the hydraulic deviation is marked as a cross-branch common change state. Thermal state deviation occurs when the chip thermal load does not change accordingly but the chip temperature continuously deviates from the expected chip temperature change state; thermal state deviation does not occur when the chip thermal load changes and the chip temperature is within the range corresponding to the normal operating baseline. After a change in the target operating volume of the circulating pump, this module compares the changes in the actual pump speed, total circuit flow rate, and supply / return hydraulic pressure difference within the pump response time window. A pump response deviation is identified when the actual state does not match the target operating volume. Similarly, after a change in the target valve state, the module compares the valve position, branch flow rate, and branch pressure difference within the valve response time window. A valve response deviation is identified when the valve position or hydraulic pressure state does not match the target valve state. This module distinguishes between pump failure, valve failure, branch failure, and temperature anomaly based on the combination of multiple deviations. It then determines the range based on the number of branches where deviations occur, the shared circulating pump status, and the supply / return fluid status, forming a single branch range, multiple branch ranges, or a total supply circuit range. When detection states conflict, the module maintains the current pump and valve control status and extends the verification cycle by one period. If the chip temperature status already meets the shutdown protection conditions, the module does not wait for the conflict to be resolved and directly provides the corresponding anomaly judgment status to the handling capability and level generation module.
[0052] The handling capacity and level generation module 03, connected to the anomaly determination module, is used to determine the cooling demand gap based on the anomaly determination status and the expected cooling status. It generates a handling availability status based on the adjustable margin of the circulating pump, the adjustable margin of the valves, the branch isolation status, the backup cooling capacity, and the chip's deload capability. It then matches the cooling demand gap with the handling availability status to generate anomaly level and handling target. The cooling demand gap is formed based on the difference between the expected branch flow rate and the current branch flow rate within the anomaly's influence range. For a single branch anomaly, the flow rate difference of the corresponding microchannel cooling branch is used; for multiple branches or the main liquid supply circuit anomalies, the flow rate difference of each affected branch and the overall circuit flow rate status are used together. The adjustable margin of the circulating pump is calculated based on the current operating volume of the circulating pump and its allowable operating range. When a pump response deviation has occurred, the theoretical operating margin of the corresponding faulty circulating pump is not included in the handling availability status. The adjustable margin of the valve is formed based on the current valve opening and the valve opening range. When a valve response deviation has occurred, the opening margin of the corresponding faulty valve is not included in the cooling demand gap matching. Branch isolation status is determined based on whether the abnormal branch has independent isolation conditions, whether the isolation action affects the normal branch, and whether a cooling continuity path exists in the corresponding chip area. Backup cooling capacity is determined based on the valve position, flow rate, differential pressure, and operating status of the backup circulating pump or backup cooling branch. If the backup cooling branch has not yet reached the target flow rate of the corresponding branch, its rated flow rate is not included in the current backup cooling capacity. Chip de-load capacity is determined based on the current power consumption of the affected chip area and the chip's allowable power consumption range. If the chip power management unit has not returned to the execution state, or if the chip's actual power consumption has not changed with the control command, the corresponding power difference is not included in the available handling capacity status. This module sequentially matches the cooling demand gap with the adjustable margins of the circulating pump, valves, backup cooling capacity, and chip de-load capacity. When the pump and valve adjustment range covers the cooling demand gap, an anomaly level and handling target are generated corresponding to liquid cooling compensation. When the normal adjustment range is insufficient but branch isolation and backup cooling capacity meet the current cooling demand, an anomaly level and handling target are generated corresponding to branch reconstruction. When the liquid cooling side handling capacity is insufficient but the chip de-load capacity matches the remaining cooling demand, an anomaly level and handling target are generated corresponding to chip de-load. When the handling available capacity status does not match the cooling demand gap or the chip temperature status meets the shutdown protection conditions, an anomaly level and handling target are generated corresponding to shutdown protection. The anomaly level and handling target are provided to the handling sequence generation module, wherein the handling target includes the target branch flow rate, target chip power consumption, target valve status, or shutdown status.
[0053] The disposal sequence generation module 04, connected to the disposal capacity and level generation module, is used to generate candidate disposal actions based on the anomaly level and disposal target, and to generate a disposal sequence according to the action preconditions and action mutual exclusion relationships. The candidate disposal actions include adjusting the target operating volume of the circulating pump, adjusting the target state of the valve, redistributing the target flow of the branch, activating the backup cooling branch, isolating the abnormal branch, limiting chip power consumption, and providing shutdown protection. Before forming the disposal sequence, this module reads the anomaly type and the scope of anomaly impact from the anomaly determination status. When the anomaly type is pump failure, the action to increase the operating volume of the faulty circulating pump is removed from the candidate disposal actions; when the anomaly type is valve failure, the action to adjust the target state of the faulty valve is removed; when a single branch anomaly does not extend to the shared liquid supply circuit, no chip power consumption limiting action is generated for unrelated chip areas. When the disposal sequence includes isolating an abnormal branch, reaching the branch target flow of the backup cooling branch or reaching the chip deload target in the affected chip area are set as preconditions for the isolation action; when the backup cooling branch has not established a stable flow and the chip power consumption has not reached the deload target, the abnormal branch isolation action remains in a waiting state. Flow compensation and branch isolation actions for the same microchannel cooling branch are set to mutually exclusive states. Operational adjustment actions for a faulty circulating pump and switching actions for a backup circulating pump are not issued in the same execution phase. After the shutdown protection action enters the execution state, ordinary liquid cooling compensation and chip power consumption recovery actions remain locked. The filtered and sorted processing sequence includes the execution object, target state, action order, preconditions, and processing verification conditions, and is transmitted to the execution feedback module. If the action preconditions are not met within the set waiting time, this module does not skip the current action and directly execute the subsequent branch isolation; instead, it maintains the current liquid cooling state and passes the incomplete precondition status to the execution feedback module for verification.
[0054] The execution feedback module 05, connected to the disposal sequence generation module, executes the disposal sequence. When the disposal sequence includes branch isolation or activation of a backup cooling branch, it updates the effective branch set and regenerates the branch target flow. It also updates the anomaly level based on the pump / valve status, branch hydraulic pressure status, and chip operating status after the disposal. This module issues control commands to the circulating pump controller, valve controller, chip power management unit, and robot safety controller according to the disposal sequence, and continues to read the actual execution status after receiving the control receipt. The circulating pump action is checked against the actual circulating pump speed, total loop flow, and supply / return hydraulic pressure difference; the valve action is checked against the valve position, branch flow, and branch pressure difference; and the chip unloading action is checked against the chip's actual power consumption and chip operating status. When the controlled object returns a receipt but the actual status has not changed accordingly, the execution feedback module re-provides this status as a pump response deviation, valve response deviation, or chip execution anomaly to the anomaly determination module. After the branch isolation action is completed, the isolated microchannel cooling branch is removed from the effective branch set, and this branch no longer receives the branch target flow. After the valve position, branch flow, and branch differential pressure of the standby cooling branch meet the operating conditions, the standby cooling branch is added to the effective branch set. When the effective branch set changes, this module transmits the updated results to the expected cooling state generation module, and regenerates the branch target flow of each effective microchannel cooling branch based on the current total loop flow, the expected cooling state of each chip area, and the updated branch connectivity. During the handling verification time window, the actual speed of the circulating pump, valve position, branch flow, branch differential pressure, chip power consumption, and chip temperature are compared with the verification conditions corresponding to the handling target. If the verification conditions are not met and the abnormal influence range does not change, the abnormality level is updated to the next level. If other effective microchannel cooling branches experience hydraulic deviation or thermal state deviation, the new state is transmitted to the abnormality judgment module to re-form the abnormal influence range. When the verification conditions are continuously met and the recovery and maintenance conditions are met, the effective branch set and pump and valve status are maintained first, and then the chip load reduction is gradually cancelled according to the chip power consumption recovery status. If the thermal state deviation occurs again during the recovery process, the current recovery action is stopped and the state is re-included in the anomaly level update.
Claims
1. A robot chip microchannel liquid cooling abnormality hierarchical treatment method, characterized in that, include: S100: Obtain the chip thermal load status, chip temperature status, branch hydraulic status, pump and valve target and actual status, and the correspondence between the branch and the chip area; Based on the chip thermal load status, match the normal operation benchmark and generate the expected cooling status; S200: Compare the hydraulic state of the branch and the chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve to generate hydraulic deviation, thermal state deviation, pump response deviation and valve response deviation; generate anomaly judgment state based on the deviation combination relationship and cross-branch correlation relationship; S300: Determine the cooling demand gap based on the anomaly determination status and the expected cooling status; generate the available handling capacity status based on the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the backup cooling capacity, and the chip deload capacity; match the cooling demand gap with the available handling capacity status to generate the anomaly level and handling target. S400. Generate candidate handling actions based on the anomaly level and handling target, and generate a handling sequence according to the action preconditions and action mutual exclusion relationships; execute the handling sequence, and when the handling sequence includes a branch isolation action or a backup cooling branch activation action, update the effective branch set and regenerate the branch target flow, and update the anomaly level according to the pump and valve status, branch hydraulic status and chip operating status after handling.
2. The method for graded handling of abnormal liquid cooling in robot chip microchannels according to claim 1, characterized in that: The hydraulic status of the branch circuit includes the branch flow rate and the branch pressure difference; The target and actual status of the pump and valve include the target operating volume of the circulating pump, the actual speed of the circulating pump, the target status of the valve, and the valve position; Based on the branch identifier, chip area, time identifier, and the correspondence between the branch and the chip area, the chip thermal load status, chip temperature status, branch flow rate, branch pressure difference, circulating pump target operating volume, circulating pump actual speed, valve target status, and valve position are associated. The expected cooling state is generated by matching the expected branch flow rate, expected branch pressure difference, and expected chip temperature change state with the chip thermal load state.
3. The method for graded handling of abnormalities in robot chip microchannel liquid cooling according to claim 2, characterized in that: Obtain the total loop flow rate and the supply and return hydraulic pressure difference; Within the pump response time window after the target operating quantity of the circulating pump changes, the changes in the actual speed of the circulating pump, the total loop flow rate, and the supply and return pressure difference are compared with the changes in the target operating quantity of the circulating pump. When the actual speed of the circulating pump, the total circuit flow rate, or the supply-return pressure difference does not change in a way that matches the target operating rate of the circulating pump, a pump response deviation is generated. Within the valve response time window after a change in the valve target state, the changes in valve position, branch flow rate, and branch pressure difference are compared with the changes in the valve target state; if the valve position, branch flow rate, or branch pressure difference does not form a change that matches the valve target state, a valve response deviation is generated.
4. The robotic chip microchannel liquid cooling anomaly classification and handling method of claim 3, wherein, The anomaly determination status includes the anomaly type and the scope of the anomaly's impact, and its generation includes: When the flow rate of a single microchannel cooling branch decreases, the differential pressure increases, the valve response deviation is not formed, and the corresponding chip area forms the thermal state deviation, a branch failure state and a single branch range are generated. When the flow rates of multiple microchannel cooling branches decrease in the same direction and the pump response deviates from the formation, a pump failure state and the total liquid supply circuit range are generated. When the valve response deviation is formed, a valve fault state and the corresponding microchannel cooling branch range are generated. When the hydraulic deviation does not occur but the thermal deviation does occur, an abnormal temperature state and a corresponding chip region range are generated.
5. The method for graded handling of abnormal liquid cooling in robot chip microchannels according to claim 4, characterized in that: The cooling demand gap is generated based on the difference between the expected branch flow and the current branch flow within the range of the anomaly's influence. The adjustable margin of the circulating pump is generated based on the current operating rate of the circulating pump and the allowable operating range of the pump. The valve's adjustable margin is generated based on the valve's current opening degree and its opening range; The branch isolation status and backup cooling capacity are generated based on the branch isolation conditions and the operating status of the backup cooling branch. The chip deload capability is generated based on the current power consumption of the affected chip region and the chip's allowable power consumption range.
6. The method for graded handling of abnormal liquid cooling in robot chip microchannels according to claim 5, characterized in that: Chip thermal risk is generated based on the chip temperature state and temperature change trend; When the adjustable margin of the circulating pump and the adjustable margin of the valve match the cooling demand gap, a first anomaly level and a liquid cooling compensation target are generated. When the branch isolation status and backup cooling capacity match the cooling demand gap, a second anomaly level and a branch reconfiguration target are generated. When the chip's deload capacity matches the cooling demand gap, a third anomaly level and a chip deload target are generated. When the cooling demand gap does not match the available processing capacity, or when the chip thermal risk reaches the shutdown protection condition, a fourth abnormality level and a shutdown protection target are generated.
7. The robotic chip microchannel liquid cooling anomaly classification and handling method of claim 6, wherein, The generation of the disposal sequence includes: When the anomaly type is pump failure, the liquid cooling compensation action that increases the target operating rate of the faulty circulating pump is removed from the candidate handling actions. When the anomaly type is valve failure, the liquid cooling compensation action that adjusts the target state of the faulty valve is removed from the candidate handling actions. When the candidate handling action includes branch isolation action, the prerequisite for the branch isolation action is that the backup cooling branch reaches the branch target flow or the affected chip area reaches the chip deload target. Set the liquid cooling compensation action and the branch isolation action for the same microchannel cooling branch to be mutually exclusive.
8. The robotic chip microchannel liquid cooling anomaly classification and handling method of claim 7, wherein, The update of the effective branch set and the branch target traffic includes: The handling sequence includes removing the isolated microchannel cooling branch from the set of effective branches when the branch isolation action is performed; The processing sequence includes adding the activated backup cooling branch to the set of valid branches when the backup cooling branch is activated. Based on the set of effective branches, the total loop flow rate, and the expected cooling state of each chip region, the target flow rate of each effective microchannel cooling branch is generated. Stop generating the branch target flow for the isolated microchannel cooling branch.
9. The robotic chip microchannel liquid cooling anomaly classification and handling method of claim 8, wherein, The update of the anomaly level includes: Within the treatment verification time window, the actual speed of the circulating pump, valve position, branch flow rate, branch pressure difference, chip power consumption and chip temperature after treatment are compared with the verification conditions corresponding to the treatment target. If the verification conditions are not met, the scope of the anomaly does not change, and the current anomaly level is lower than the fourth anomaly level, the anomaly level is increased; if the current anomaly level is the fourth anomaly level, the shutdown protection target is maintained. When hydraulic or thermal deviations occur in other microchannel cooling branches, update the scope of the anomaly and regenerate the anomaly level and handling target. When the verification conditions are continuously met and the recovery and maintenance conditions are met, the hydraulic state and pump valve state of the branch in the effective branch set are first confirmed to meet the expected cooling state, and then the chip power consumption is gradually restored; when the isolated microchannel cooling branch meets the recovery conditions, its operating state is restored.
10. A robot chip microchannel liquid cooling abnormality hierarchical treatment system applied to the method in any one of claims 1 to 9, characterized in that, include: The expected cooling state generation module is used to obtain the chip thermal load state, chip temperature state, branch hydraulic state, pump valve target and actual state, and the correspondence between the branch and the chip area, and generate the expected cooling state based on the chip thermal load state and the normal operation benchmark. An anomaly determination module, connected to the expected cooling state generation module, is used to compare the branch hydraulic state and chip temperature state with the expected cooling state, and compare the changes in the target and actual states of the pump and valve, generating hydraulic deviation, thermal state deviation, pump response deviation and valve response deviation, and generating anomaly determination states based on the deviation combination relationship and cross-branch correlation relationship. The handling capacity and level generation module is connected to the anomaly determination module. It is used to determine the cooling demand gap based on the anomaly determination status and the expected cooling status, generate the handling availability status based on the adjustable margin of the circulating pump, the adjustable margin of the valve, the branch isolation status, the backup cooling capacity and the chip deload capacity, and match the cooling demand gap with the handling availability status to generate the anomaly level and handling target. The disposal sequence generation module, connected to the disposal capability and level generation module, is used to generate candidate disposal actions based on the anomaly level and disposal target, and to generate a disposal sequence according to the action preconditions and action mutual exclusion relationships. The execution feedback module is connected to the disposal sequence generation module and is used to execute the disposal sequence. When the disposal sequence includes branch isolation or activation of backup cooling branch, the effective branch set is updated and the branch target flow is regenerated. The abnormality level is updated according to the pump and valve status, branch hydraulic status and chip operating status after disposal.
