Early fault monitoring method and system for liquid-cooled instrument based on dynamic knowledge graph

Abstract

The application discloses a liquid-cooled instrument early fault monitoring method and system based on a dynamic knowledge graph. The method acquires the latest set of historical operation data of the liquid-cooled instrument and generates a historical state event set, performs local incremental updating on related state event nodes, fault precursor nodes and associated edge weights in a preset fault knowledge graph, and obtains a dynamic knowledge graph. Multidimensional real-time operation data in a current period are acquired and current state events are generated, candidate fault links are recalled based on the dynamic knowledge graph, and to-be-verified conditions are generated according to hit nodes and missed nodes in the links; evidence support degrees are determined according to real-time data, counter-evidence suppression systems are determined according to working condition changes, and early fault judgment results are obtained accordingly, and the method is suitable for early fault monitoring in the operation process of the liquid-cooled instrument.

Description

Technical Field

[0001] This invention relates to the field of liquid-cooled instrument fault monitoring technology, specifically to a method and system for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs. Background Technology

[0002] Early fault monitoring for liquid-cooled instruments is primarily used in instruments and equipment with high heat flux density, long-term continuous operation, or high temperature stability requirements. It detects operational anomalies in the cooling circuit before they occur, such as blockages, micro-leaks, pump degradation, decreased heat exchange capacity, or abnormal coolant conditions. Existing monitoring methods typically collect data such as inlet temperature, outlet temperature, flow rate, pressure, pump speed, valve opening, instrument load, and ambient temperature. These data are then used to determine if any abnormalities exist in the liquid cooling system through fixed thresholds, trend analysis, alarm rules, or model recognition.

[0003] However, early faults in liquid-cooled instruments often manifest as weakly coupled deviations between multiple parameters. A single parameter may not yet have exceeded its limits, and load changes, pump speed adjustments, valve opening variations, and ambient temperature fluctuations can also cause similar data changes. Existing methods, employing static rules or fixed models, struggle to update fault correlations in a timely manner based on the instrument's recent operating status. They also fail to distinguish between genuine early faults and transient anomalies caused by changes in normal operating conditions, easily leading to false alarms, missed alarms, or unclear fault causes. Therefore, a method and system for monitoring early faults in liquid-cooled instruments based on dynamic knowledge graphs is needed to address these issues. Summary of the Invention

[0004] To address the above problems, this invention provides the following technical solution: an early fault monitoring method for liquid-cooled instruments based on dynamic knowledge graphs, comprising:

[0005] S1. Obtain the latest set of historical operating data for the liquid cooling instrument, and generate a set of historical status events based on the historical operating data;

[0006] S2. Based on the historical state event set, locally incrementally update the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph to obtain a dynamic knowledge graph.

[0007] S3. Obtain multi-dimensional real-time operating data of the liquid cooling instrument for the current period, and generate a current status event based on the multi-dimensional real-time operating data;

[0008] S4. Based on the current state event, recall the fault precursor node and trace the fault type node in the dynamic knowledge graph to obtain the candidate fault link, and generate the verification condition based on the hit state event node, the missing state event node and the missing sensor parameter node in the candidate fault link.

[0009] S5. Match the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time running data, and determine the evidence support degree of the candidate fault link based on the matching results;

[0010] S6. Determine the degree of counter-evidence suppression of the candidate fault link based on the changes in instrument load, pump speed, valve opening and ambient temperature, and generate the early fault judgment result of the liquid-cooled instrument based on the comparison result of the evidence support degree and the degree of counter-evidence suppression.

[0011] Furthermore, the historical operating data includes liquid coolant inlet temperature, liquid coolant outlet temperature, flow rate, pressure, pump speed, valve opening, coolant conductivity, instrument load, and ambient temperature.

[0012] Furthermore, generating a historical state event set based on the historical operational data includes:

[0013] The historical operating data is segmented into operating condition segments based on instrument load, pump speed, and valve opening to obtain historical operating condition segments.

[0014] Temperature, flow rate, pressure, and coolant conductivity within the same historical operating condition segment are time-aligned to obtain an alignment parameter sequence;

[0015] A set of historical state events is generated based on the direction, magnitude, and duration of parameter changes in the alignment parameter sequence.

[0016] Furthermore, the preset fault knowledge graph includes component nodes, sensor parameter nodes, operating condition nodes, status event nodes, fault precursor nodes, and fault type nodes, and the associated edge weights include the associated edge weights between status event nodes and fault precursor nodes, and the associated edge weights between fault precursor nodes and fault type nodes.

[0017] Furthermore, the step of generating current status events based on the multidimensional real-time operating data includes generating parameter offset events, coupling anomaly events, and response hysteresis events based on the response relationship between temperature, flow rate, pressure, pump speed, and instrument load within the current time period.

[0018] Furthermore, the step of recalling fault precursor nodes and tracing fault type nodes in the dynamic knowledge graph based on the current state event to obtain candidate fault links, and generating verification conditions based on the hit state event nodes, miss state event nodes, and miss sensor parameter nodes in the candidate fault links, includes:

[0019] The current state event is matched with the state event nodes in the dynamic knowledge graph to obtain the matched state event nodes.

[0020] Based on the hit state event node, recall the fault precursor nodes that have associated edges with it;

[0021] Tracing along the associated edges between the fault precursor nodes and the fault type nodes, a candidate fault link is obtained, consisting of hit state event nodes, fault precursor nodes, and fault type nodes.

[0022] From the candidate fault links, the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes are determined, and the conditions to be verified are generated based on the parameter changes corresponding to the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes.

[0023] Furthermore, the step of matching the parameter change direction, change duration, and parameter coupling relationship in the conditions to be verified based on the multidimensional real-time operational data, and determining the evidence support degree of the candidate fault link based on the matching results, includes:

[0024] When the multidimensional real-time running data satisfies the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is improved.

[0025] When the multidimensional real-time running data does not satisfy at least one of the parameter change direction, change duration, and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is reduced.

[0026] Furthermore, determining the counter-evidence suppression degree of the candidate fault link based on changes in instrument load, pump speed, valve opening, and ambient temperature includes:

[0027] When the parameter change corresponding to the current state event is synchronously related to at least one of the changes in instrument load, pump speed, valve opening degree and ambient temperature, the counter-evidence suppression degree of the candidate fault link is increased.

[0028] If the parameter change corresponding to the current state event persists after at least one of the changes in instrument load, pump speed, valve opening, and ambient temperature has ended, the counter-evidence suppression degree of the candidate fault link is reduced.

[0029] Furthermore, the step of generating an early fault judgment result for the liquid-cooled instrument based on the comparison result of the evidence support degree and the rebuttal suppression degree includes:

[0030] When the evidence support is higher than the disproving degree, an early fault judgment result representing the existence of early faults is generated based on the candidate fault links;

[0031] When the evidence support is not higher than the disproving evidence suppression, the candidate fault link is suppressed and an early fault judgment result representing an unconfirmed early fault is generated.

[0032] The early fault judgment results include fault existence, suspected fault type, suspected component, triggering state event, candidate fault link, evidence support, counter-evidence suppression degree, and verification conditions.

[0033] This invention also provides an early fault monitoring system for liquid-cooled instruments based on dynamic knowledge graphs, comprising:

[0034] The historical event generation module is used to acquire the latest set of historical operating data of the liquid cooling instrument and generate a set of historical status events based on the historical operating data.

[0035] The graph update module is used to perform local incremental updates on the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph according to the historical state event set, so as to obtain a dynamic knowledge graph.

[0036] The real-time event generation module is used to acquire multi-dimensional real-time operating data of the liquid cooling instrument in the current period and generate current status events based on the multi-dimensional real-time operating data.

[0037] The fault link generation module is used to recall fault precursor nodes and trace fault type nodes in the dynamic knowledge graph based on the current state event to obtain candidate fault links, and generate verification conditions based on the hit state event nodes, miss state event nodes and miss sensor parameter nodes in the candidate fault links.

[0038] The evidence support determination module is used to match the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time running data, and determine the evidence support degree of the candidate fault link based on the matching results.

[0039] The fault diagnosis module is used to determine the degree of disproving evidence suppression of the candidate fault links based on changes in instrument load, pump speed, valve opening and ambient temperature, and to generate an early fault diagnosis result for the liquid-cooled instrument based on the comparison result between the degree of evidence support and the degree of disproving evidence suppression.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The method and system for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs generate a historical state event set from the latest set of historical operating data of the liquid-cooled instrument. Based on this, the relevant state event nodes, fault precursor nodes, and their associated edge weights in the preset fault knowledge graph are locally incrementally updated, so that the fault association can be adjusted with the recent changes in the instrument's operating status. By recalling candidate fault links in the dynamic knowledge graph for the current state event, and generating verification conditions based on the hit state event nodes, miss state event nodes, and miss sensor parameter nodes in the candidate fault links, the real-time data no longer relies solely on a single threshold or fixed model for judgment, but is verified and confirmed around the link between fault precursors and fault types. At the same time, by comparing the evidence support degree with the disinformation suppression degree determined by changes in instrument load, pump speed, valve opening, and ambient temperature, the interference of transient anomalies caused by changes in normal operating conditions on fault judgment can be suppressed, thereby more clearly distinguishing between real early faults and data fluctuations caused by normal operation adjustments, and outputting early fault judgment results including suspected fault types, suspected components, triggering state events, and candidate fault links. Attached Figure Description

[0042] Figure 1 A schematic diagram illustrating the steps of the early fault monitoring method for liquid-cooled instruments based on dynamic knowledge graphs provided by this invention;

[0043] Figure 2 A schematic diagram of the early fault judgment process based on candidate fault link generation and bidirectional judgment provided by the present invention;

[0044] Figure 3 This is a schematic diagram of the structure of the liquid-cooled instrument early fault monitoring system based on dynamic knowledge graph provided by the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Please see Figure 1 This invention provides a method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs, including:

[0048] S1. Obtain the latest set of historical operating data for the liquid cooling instrument, and generate a set of historical status events based on the historical operating data;

[0049] Specifically, liquid-cooled instruments can be precision instruments, testing instruments, computing devices, or thermal control devices with liquid-cooled circulation loops. The latest set of historical operating data can be continuous sampling data within a preset historical time window prior to the current monitoring time, or it can be a set of operating data generated from the last maintenance to the current monitoring time. During data acquisition, historical data is organized according to a unified device number, sampling timestamp, and sensor number, and data segments with communication interruptions, sensor offline, or significantly exceeding the measurement range are removed. For data with different sampling frequencies, a timestamp alignment method is used to map them to a unified time axis, so that subsequently generated historical state events can reflect the changing relationships between multiple parameters under the same operating condition. The historical state event set is used to describe state changes that have occurred in the historical operating data but may not have reached the alarm threshold, such as slow temperature rise, slow flow decrease, insufficient pressure response, enhanced pump speed compensation, or abnormal changes in coolant conductivity.

[0050] Furthermore, in one embodiment provided in this application, the historical operating data includes liquid coolant inlet temperature, liquid coolant outlet temperature, flow rate, pressure, pump speed, valve opening, coolant conductivity, instrument load, and ambient temperature.

[0051] Specifically, the liquid cooling inlet temperature indicates the temperature of the coolant before it enters the cooling circuit of the liquid-cooled instrument, and the liquid cooling outlet temperature indicates the temperature of the coolant after it flows out of the heating components. Both are used to form the temperature difference change characteristics. Flow rate and pressure are used to describe the coolant circulation capacity, pump speed and valve opening are used to describe the active adjustment state of the liquid cooling system, coolant conductivity is used to reflect changes in the coolant state, and instrument load and ambient temperature are used to distinguish the normal effects of heat source changes or external environmental changes on the liquid cooling parameters.

[0052] Furthermore, in one embodiment provided in this application, generating a historical state event set based on the historical operational data includes:

[0053] The historical operating data is segmented into operating condition segments based on instrument load, pump speed, and valve opening to obtain historical operating condition segments.

[0054] Temperature, flow rate, pressure, and coolant conductivity within the same historical operating condition segment are time-aligned to obtain an alignment parameter sequence;

[0055] A set of historical state events is generated based on the direction, magnitude, and duration of parameter changes in the alignment parameter sequence.

[0056] Specifically, during the segmentation of operating conditions, the moments when the instrument load change exceeds the preset load change condition, the pump speed undergoes a step adjustment, or the valve opening is adjusted are used as segment boundaries to obtain several historical operating condition segments. Within each historical operating condition segment, the load, pump speed, and valve opening are relatively stable, facilitating the distinction between parameter changes caused by normal control actions and continuous deviations caused by fault precursors. For time alignment, an alignment time point is generated using a uniform sampling period, mapping temperature, flow rate, pressure, and coolant conductivity to the alignment time point. If a short-term missing value exists at a certain time point, it is interpolated using adjacent valid sampled values. If the continuous missing duration exceeds the preset missing duration, the segment is marked as an unusable segment and does not participate in the generation of historical state events.

[0057] The magnitude of parameter change can be expressed as a relative change, for example, for any parameter The relative change within the historical operating condition segment is:

[0058]

[0059] in, The parameter value represents the start time of a historical operating condition segment. The parameter value represents the end time of the historical operating condition segment. To prevent extremely small positive numbers with a denominator of zero, This is used to determine the direction and magnitude of the parameter's change within the specified operating condition segment. If the relative change of a parameter reaches the event trigger condition, and the duration of the change exceeds the minimum duration, a corresponding historical state event is generated.

[0060] Historical state events can include event identifiers, associated parameters, direction of change, duration of change, associated operating conditions, coupling relationships, and purpose of results. Event identifiers are generated by combining device number, time period, and event type, used to distinguish different historical state events or current state events; associated parameters record the temperature, flow rate, pressure, pump speed, coolant conductivity, or instrument load parameters that triggered the event, and are used to determine the corresponding sensor parameter node; direction of change indicates whether the parameter is rising, falling, fluctuating, or has insufficient response, and is used to match with fault precursor nodes; duration of change indicates the length of time the parameter change has continuously existed, and is used to eliminate transient noise; associated operating conditions indicate the load, pump speed, and valve opening status at the time the event occurred, and are used to determine whether the event was caused by a change in normal operating conditions; coupling relationships indicate whether there are synchronous, lagging, or opposite changes between two or more parameters, and are used to generate candidate fault chains and conditions to be verified; purpose of results identifies the stage in which the event participates in spectrum updates, real-time recall, or counter-evidence judgment. Through the above fields, the historical state event set not only records whether a single parameter has changed, but also records the operating conditions, duration, and relationships between parameters when the parameter changes occur, making it an input for subsequent local incremental updates of the map.

[0061] S2. Based on the historical state event set, locally incrementally update the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph to obtain a dynamic knowledge graph.

[0062] Specifically, a pre-defined fault knowledge graph stores the relationships between components, sensing parameters, operating conditions, state events, fault precursors, and fault types of the liquid-cooled instrument. After inputting the historical state event set, each historical state event is first matched against state event nodes in the graph. The matching criteria include association parameters, direction of change, duration of change, and coupling relationship. If a historical state event matches an existing state event node, the weight of the association edge between that state event node and the corresponding fault precursor node is updated. If a historical state event is similar to an existing state event node but not completely matched, it can be merged into a similar state event node and the event description recorded, without changing the main structure of the graph. Local incremental updates are only performed on the local subgraphs matched by historical state events; the graph regions not matched remain in their original state, thus avoiding the need to reconstruct the entire graph with each update. When multiple historical state events repeatedly hit the same fault precursor node in different historical operating condition segments, and the fault type corresponding to the fault precursor node is confirmed in the historical alarm record, maintenance record, or review record associated with the historical operating data, the association edge weight between the fault precursor node and the corresponding fault type node is updated; when the fault type corresponding to the fault precursor node is not confirmed in the subsequent review, the association edge weight between the fault precursor node and the corresponding fault type node is reduced.

[0063] Furthermore, in one embodiment provided in this application, the preset fault knowledge graph includes component nodes, sensor parameter nodes, operating condition nodes, status event nodes, fault precursor nodes, and fault type nodes, and the associated edge weights include the associated edge weights between status event nodes and fault precursor nodes, and the associated edge weights between fault precursor nodes and fault type nodes.

[0064] Specifically, component nodes can include pumps, valves, cold plates, heat exchangers, pipelines, liquid storage units, and sensors; sensor parameter nodes can include inlet temperature, outlet temperature, flow rate, pressure, pump speed, valve opening, conductivity, load, and ambient temperature; operating condition nodes can include stable load conditions, increased load conditions, pump speed adjustment conditions, and valve regulation conditions; status event nodes can include slow flow rate decrease, abnormal pressure difference, slow temperature difference increase, delayed pressure recovery, abnormal conductivity, and enhanced pump speed compensation; fault precursor nodes can include precursors to partial blockage, micro-leakage, pump performance degradation, decreased heat exchange efficiency, and abnormal coolant condition; fault type nodes can include pipeline blockage, coolant leakage, circulating pump degradation, decreased heat exchanger efficiency, and coolant deterioration.

[0065] When updating the weights of associated edges, the update amount can be determined based on the matching degree between historical state events and graph nodes, as well as the persistence of the event in historical work segment. The weights of associated edges can be updated according to the following formula:

[0066]

[0067] in, This represents the updated weight of the associated edges. This indicates the weight of the associated edges before the update. Represents the historical forgetting coefficient, and It takes a value between 0 and 1; This indicates the degree of matching between historical state events and the nodes at both ends of the associated edge, and It takes a value between 0 and 1; Indicates the intensity of the persistence of historical state events, and The values ​​range from 0 to 1. The updated associated edge weights are used to recall fault precursor nodes and trace fault type nodes in subsequent real-time status events, enabling the dynamic knowledge graph to reflect recurring weak anomaly relationships in the recent operation of the liquid-cooled instrument.

[0068] S3. Obtain multi-dimensional real-time operating data of the liquid cooling instrument for the current period, and generate a current status event based on the multi-dimensional real-time operating data;

[0069] Specifically, the current time period can be a short sliding window preceding the current monitoring moment, such as several recent sampling cycles. The data collected from multidimensional real-time operational data remains consistent with historical operational data, but the processing objectives differ: historical operational data is used to update the dynamic knowledge graph, while the multidimensional real-time operational data for the current time period is used to determine if the liquid-cooled instrument has early-stage faults. After real-time data enters the system, it first undergoes timestamp verification, sampling integrity checks, and outlier filtering. Then, based on the direction of parameter changes, short-term change amplitude, duration, and response relationships between parameters within the current time period, a current state event is generated. If a single sampling point in the real-time data experiences a jump but does not persist, a current state event is not generated to avoid mistaking transient noise as a fault precursor.

[0070] Furthermore, in one embodiment provided in this application, the step of generating the current state event based on the multidimensional real-time operating data includes generating parameter offset events, coupling anomaly events, and response hysteresis events based on the response relationship between temperature, flow rate, pressure, pump speed, and instrument load within the current time period.

[0071] Specifically, parameter offset events indicate that a single parameter is continuously rising, continuously falling, or continuously deviating from the normal range within the current time period, such as a slow increase in outlet temperature or a continuous decrease in flow rate; coupling anomaly events indicate that the relationship between two or more parameters does not conform to the normal response pattern of the liquid cooling system, such as the flow rate not recovering synchronously after the pump speed increases, or the temperature difference continuously increasing even without an increase in load; response hysteresis events indicate that after a change in operating conditions occurs, the liquid cooling parameters do not recover to the corresponding state within a preset response time, such as a delayed pressure change or slow flow recovery after valve opening adjustment. These current state events are used to subsequently recall fault precursor nodes in the dynamic knowledge graph, enabling real-time data to participate in the judgment in the form of structured events, rather than simply comparing thresholds in raw numerical form.

[0072] S4. Based on the current state event, recall the fault precursor node and trace the fault type node in the dynamic knowledge graph to obtain the candidate fault link, and generate the verification condition based on the hit state event node, the missing state event node and the missing sensor parameter node in the candidate fault link.

[0073] For details, please refer to Figure 2After a current state event enters the dynamic knowledge graph, it is first matched with state event nodes. If a match is successful, it is recalled along the association edges between the state event node and the fault precursor nodes, resulting in one or more fault precursor nodes. Then, it is traced along the association edges between the fault precursor nodes and the fault type nodes, forming a candidate fault link. The candidate fault link is used to express the knowledge reasoning path from the current state event to the suspected fault type. For example, when the current state event is "increased pump speed compensation and insufficient flow recovery," it can recall "partial blockage precursor" or "pump performance degradation precursor," and further trace to fault types such as "pipeline blockage" or "circulating pump degradation."

[0074] Furthermore, in one embodiment provided by this application, the step of recalling fault precursor nodes and tracing fault type nodes in the dynamic knowledge graph based on the current state event to obtain candidate fault links, and generating verification conditions based on the hit state event nodes, miss state event nodes, and miss sensor parameter nodes in the candidate fault links, includes:

[0075] The current state event is matched with the state event nodes in the dynamic knowledge graph to obtain the matched state event nodes.

[0076] Based on the hit state event node, recall the fault precursor nodes that have associated edges with it;

[0077] Tracing along the associated edges between the fault precursor nodes and the fault type nodes, a candidate fault link is obtained, consisting of hit state event nodes, fault precursor nodes, and fault type nodes.

[0078] From the candidate fault links, the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes are determined, and the conditions to be verified are generated based on the parameter changes corresponding to the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes.

[0079] Specifically, when matching the current state event with state event nodes, the event type, associated parameters, direction of change, and duration can be compared. If the current state event matches multiple state event nodes, the corresponding fault precursor nodes are recalled respectively, and the associated edge weights of each recalled path are retained. For each candidate fault link composed of state event nodes, fault precursor nodes, and fault type nodes, the system determines the matched state event nodes in the link and identifies the unmatched state event nodes and unmatched sensor parameter nodes in the link. Unmatched state event nodes can be state event nodes in the candidate fault link that have not yet been triggered by the current state event, such as "pressure recovery hysteresis" or "continuous increase in temperature difference"; unmatched sensor parameter nodes can be sensor parameter nodes in the candidate fault link that have not yet been covered by the current state event, such as "pressure", "flow rate", or "conductivity". When generating verification conditions based on hit state event nodes, miss state event nodes, and miss sensor parameter nodes, the parameter changes corresponding to the hit state event nodes are used as the basic verification conditions. The miss state event nodes and miss sensor parameter nodes are converted into parameter change directions, durations of change, and parameter coupling relationships that can be verified by real-time data. This allows the graph to request supplementary evidence from real-time data based on the hit evidence. In this way, candidate fault links are not directly confirmed by a single current state event, but are verified through the basic verification conditions formed by hit state event nodes and the supplementary verification conditions formed by miss state event nodes and miss sensor parameter nodes.

[0080] S5. Match the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time running data, and determine the evidence support degree of the candidate fault link based on the matching results;

[0081] Specifically, after the conditions to be verified are generated, the system returns to the multi-dimensional real-time operational data of the current time period to search for parameter changes corresponding to the conditions to be verified. The direction of parameter change is used to determine whether the real-time parameters change in the direction required by the candidate fault link; the duration of change is used to determine whether the change meets the persistence requirement; and the parameter coupling relationship is used to determine whether multiple parameters jointly support the candidate fault link. For example, for a candidate fault link suspected of partial blockage, the conditions to be verified may include a decrease in flow rate, insufficient flow recovery after pump speed compensation enhancement, and an abnormal response between pressure changes and flow rate changes; if the real-time data simultaneously meets the above conditions, the candidate fault link obtains a high degree of evidence support.

[0082] Furthermore, in one embodiment provided in this application, the step of matching the parameter change direction, change duration, and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time operating data, and determining the evidence support degree of the candidate fault link based on the matching results, includes:

[0083] When the multidimensional real-time running data satisfies the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is improved.

[0084] When the multidimensional real-time running data does not satisfy at least one of the parameter change direction, change duration, and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is reduced.

[0085] Specifically, the strength of evidence support can be determined by the hit rate of the condition to be verified and the weights of the associated edges of the candidate faulty link. For the th The evidence support for each candidate faulty link can be expressed as:

[0086]

[0087] in, Indicates the first The degree of evidence support for each candidate faulty link Indicates the first The hit value of each condition to be verified is 1 when the multidimensional real-time running data meets the condition to be verified, 0 when the condition to be verified is not met, or between 0 and 1 depending on the degree of satisfaction. Indicates the first The weight of each condition to be verified; Indicates the first The normalized composite value of the weights of the associated edges in the candidate faulty links; This represents the influence coefficient of the correlation strength of the graph on the support of the evidence, where each... and satisfy ,make The result is within the same normalized scale as the suppression of counter-evidence. This calculation result is used for subsequent comparison with the suppression of counter-evidence. If there are missing data in the current real-time data, the conditions to be verified corresponding to the missing data will not be treated as hit conditions, and the evidence integrity of the candidate fault link will be reduced to avoid false confirmation due to missing data.

[0088] S6. Determine the degree of counter-evidence suppression of the candidate fault link based on the changes in instrument load, pump speed, valve opening and ambient temperature, and generate the early fault judgment result of the liquid-cooled instrument based on the comparison result of the evidence support degree and the degree of counter-evidence suppression.

[0089] Specifically, the disproportionate suppression ratio indicates whether the current state event corresponding to a candidate fault link can be explained by changes in normal operating conditions. Normal operating condition changes include increases or decreases in instrument load, active pump speed adjustment, valve opening changes, and ambient temperature changes. When the current state event occurs synchronously with the above-mentioned operating condition changes and gradually recovers after the change, the current state event is more likely to be caused by a normal adjustment process, and the disproportionate suppression ratio of the candidate fault link is increased. When the current state event persists after the change, or its direction of change is inconsistent with the normal adjustment pattern, the disproportionate suppression ratio of the candidate fault link is decreased. By introducing the disproportionate suppression ratio, the system can avoid misjudging sudden load increases, ambient temperature fluctuations, or normal pump speed compensation as early faults.

[0090] Furthermore, in one embodiment provided in this application, determining the counter-evidence suppression degree of the candidate fault link based on changes in instrument load, pump speed, valve opening, and ambient temperature includes:

[0091] When the parameter change corresponding to the current state event is synchronously related to at least one of the changes in instrument load, pump speed, valve opening degree and ambient temperature, the counter-evidence suppression degree of the candidate fault link is increased.

[0092] If the parameter change corresponding to the current state event persists after at least one of the changes in instrument load, pump speed, valve opening, and ambient temperature has ended, the counter-evidence suppression degree of the candidate fault link is reduced.

[0093] Specifically, the synchronization relationship can be determined by the time difference between the occurrence of the current state event and the occurrence of the change in operating conditions, or by whether the direction of parameter change conforms to the normal response relationship after the change in operating conditions. For example, a short-term increase in outlet temperature after a load increase is a possible normal response. If this temperature increase disappears as the load recovers, the degree of resistance to contradiction increases; if the outlet temperature continues to rise after the load recovers, and there is a simultaneous decrease in flow rate or an increase in pump speed compensation, the degree of resistance to contradiction decreases. The degree of resistance to contradiction can be expressed as:

[0094]

[0095] in, Indicates the first The degree of disproven suppression of candidate fault links, Indicates the first The degree to which each operating condition variation factor explains the candidate fault link, and It takes a value between 0 and 1; when the... When a change in operating condition and a current state event occur within a preset synchronization time threshold, and the direction of parameter change corresponding to the current state event can be explained, Take the higher value; when the current state event persists after the end of the change factor in the operating condition, or when the direction of parameter change does not conform to the normal response relationship, Take the lower value; Indicates the first The weights of each operating condition variation factor are assigned, and the sum of the weights of all operating condition variation factors is 1, so that the degree of counter-evidence suppression is achieved. With the degree of support of evidence They are within the same normalization scale. If the evidence support... Higher than the degree of suppression of counter-evidence This indicates that real-time data provides stronger support for candidate fault links than it can explain changes in normal operating conditions. Based on this candidate fault link, the system generates an early fault judgment result indicating the presence of an early fault; if the evidence support... Not higher than the degree of suppression of counter-evidence This indicates that the candidate fault link is more likely to be caused by changes in normal operating conditions. The system suppresses the candidate fault link and generates an early fault judgment result representing an unconfirmed early fault. If multiple candidate fault links exist, the evidence support and disproven suppression ratios of each link are calculated separately, and the candidate fault links with higher evidence support than disproven suppression ratios and larger differences are prioritized for output.

[0096] Furthermore, in one embodiment provided in this application, the early fault judgment result includes fault existence, suspected fault type, suspected component, triggering state event, candidate fault link, evidence support degree, counter-evidence suppression degree, and verification conditions.

[0097] Specifically, the "fault presence" field indicates whether an early fault is confirmed; the "suspected fault type" field originates from the fault type node in the candidate fault chain; the "suspected component" field originates from the component node associated with that fault type node; the "triggered state event" field is the current state event that first recalls the candidate fault chain; the "candidate fault chain" field represents the reasoning path between state events, fault precursors, and fault types; the "evidence support" and "counter-evidence suppression" fields explain the basis for fault judgment; and the "review conditions" field indicates which parameters and components should be reviewed manually or automatically, such as reviewing flow sensors, checking pump operating status, checking pipeline differential pressure, or confirming coolant status. By outputting these fields, the early fault judgment result not only indicates whether a fault exists but also explains the source of the fault judgment and the objects requiring review.

[0098] In one embodiment provided in this application, a set of exemplary operating data is used to illustrate the above-mentioned early fault judgment process. During the current monitoring period, the liquid coolant inlet temperature changed from 25.2℃ to 25.5℃, the liquid coolant outlet temperature changed from 32.8℃ to 34.6℃, the flow rate decreased from 6.8L / min to 5.9L / min, the pressure changed from 0.21MPa to 0.27MPa, the pump speed was adjusted from 2800r / min to 3300r / min, the valve opening remained within the range of 72% to 75%, the coolant conductivity changed from 7.6μS / cm to 7.9μS / cm, the instrument load changed from 78% to 80%, and the ambient temperature changed from 24.8℃ to 25.0℃. Based on the above data, the system generates current status events of "outlet temperature continues to rise," "flow rate continues to decrease," and "insufficient flow rate recovery after pump speed compensation enhancement."

[0099] Specifically, after inputting the aforementioned current state events into a dynamic knowledge graph, the system recalls fault precursor nodes related to the precursors of local blockage and traces them back to the fault type node corresponding to the pipeline blockage, generating candidate fault links. Based on the hit state event nodes, miss state event nodes, and miss sensor parameter nodes in the candidate fault links, the system generates conditions to be verified. These conditions include the direction of flow rate decrease, the duration of flow rate decrease, the flow rate recovery relationship after pump speed increase, and the coupling change relationship between pressure and flow rate. Since the flow rate continuously decreases during the current monitoring period, and the flow rate does not recover to the expected flow rate range under the corresponding operating conditions after the pump speed increases, and there is a corresponding relationship between pressure increase and flow rate decrease, the evidence support for this candidate fault link is improved.

[0100] Furthermore, the system determines the degree of disprovenance based on changes in instrument load, pump speed, valve opening, and ambient temperature. In this exemplary operating data, the instrument load only changed from 78% to 80%, the ambient temperature only changed from 24.8℃ to 25.0℃, the valve opening did not change significantly, while the decrease in flow rate and the increase in outlet temperature persisted even after the pump speed was increased. Therefore, the changes in these parameters cannot be explained solely by changes in load, ambient temperature, or valve adjustment. Based on this, the system reduces the degree of disprovenance for candidate fault links, and when the evidence support is higher than the degree of disprovenance, it generates an early fault judgment result indicating the presence of an early fault. In this early fault judgment result, the suspected fault type is pipeline blockage, the suspected component is liquid-cooled pipeline or cold plate flow channel, the triggering state events include continuous increase in outlet temperature, continuous decrease in flow rate, and insufficient flow recovery after pump speed compensation enhancement, and the verification conditions include checking the pressure difference of liquid-cooled pipeline, the blockage of cold plate flow channel, and the status of flow sensor.

[0101] During another current monitoring period, if the liquid cooling outlet temperature rises synchronously with the increase of instrument load and falls back after the instrument load recovers, and there is no continuous abnormal coupling relationship between flow rate, pressure and pump speed, then the system increases the counter-evidence suppression degree of the corresponding candidate fault link and outputs the early fault judgment result characterizing the unconfirmed early fault.

[0102] Please see Figure 3 An early fault monitoring system for liquid-cooled instruments based on dynamic knowledge graphs includes:

[0103] The historical event generation module is used to acquire the latest set of historical operating data of the liquid cooling instrument and generate a set of historical status events based on the historical operating data.

[0104] The graph update module is used to perform local incremental updates on the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph according to the historical state event set, so as to obtain a dynamic knowledge graph.

[0105] The real-time event generation module is used to acquire multi-dimensional real-time operating data of the liquid cooling instrument in the current period and generate current status events based on the multi-dimensional real-time operating data.

[0106] The fault link generation module is used to recall fault precursor nodes and trace fault type nodes in the dynamic knowledge graph based on the current state event to obtain candidate fault links, and generate verification conditions based on the hit state event nodes, miss state event nodes and miss sensor parameter nodes in the candidate fault links.

[0107] The evidence support determination module is used to match the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time running data, and determine the evidence support degree of the candidate fault link based on the matching results.

[0108] The fault diagnosis module is used to determine the degree of disproving evidence suppression of the candidate fault links based on changes in instrument load, pump speed, valve opening and ambient temperature, and to generate an early fault diagnosis result for the liquid-cooled instrument based on the comparison result between the degree of evidence support and the degree of disproving evidence suppression.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs, characterized in that, include: S1. Obtain the latest set of historical operating data for the liquid cooling instrument, and generate a set of historical status events based on the historical operating data; The historical operating data includes liquid coolant inlet temperature, liquid coolant outlet temperature, flow rate, pressure, pump speed, valve opening, coolant conductivity, instrument load, and ambient temperature. S2. Based on the historical state event set, locally incrementally update the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph to obtain a dynamic knowledge graph. S3. Obtain multi-dimensional real-time operating data of the liquid cooling instrument for the current period, and generate a current status event based on the multi-dimensional real-time operating data; S4. Based on the current state event, recall fault precursor nodes and trace fault type nodes in the dynamic knowledge graph to obtain candidate fault links. Generate verification conditions based on the hit state event nodes, missed state event nodes, and missed sensor parameter nodes in the candidate fault links, including: The current state event is matched with the state event nodes in the dynamic knowledge graph to obtain the matched state event nodes. Based on the hit state event node, recall the fault precursor nodes that have associated edges with it; Tracing along the associated edges between the fault precursor nodes and the fault type nodes, a candidate fault link is obtained, consisting of hit state event nodes, fault precursor nodes, and fault type nodes. From the candidate fault links, the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes are determined, and the conditions to be verified are generated based on the parameter changes corresponding to the hit state event nodes, the miss state event nodes, and the miss sensor parameter nodes. S5. Match the parameter change direction, change duration, and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time operating data, and determine the evidence support degree of the candidate fault link based on the matching results, including: When the multidimensional real-time running data satisfies the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is improved. When the multidimensional real-time running data does not satisfy at least one of the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified, the evidence support of the candidate fault link is reduced. S6. Determine the degree of disproving evidence suppression of the candidate fault link based on the changes in instrument load, pump speed, valve opening and ambient temperature, and generate the early fault judgment result of the liquid cooling instrument based on the comparison result of the evidence support degree and the degree of disproving evidence suppression. The determination of the counter-evidence suppression degree of the candidate fault link based on changes in instrument load, pump speed, valve opening, and ambient temperature includes: When the parameter change corresponding to the current state event is synchronously related to at least one of the changes in instrument load, pump speed, valve opening degree and ambient temperature, the counter-evidence suppression degree of the candidate fault link is increased. If the parameter change corresponding to the current state event persists after at least one of the changes in instrument load, pump speed, valve opening, and ambient temperature has ended, the counter-evidence suppression degree of the candidate fault link is reduced.

2. The method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs according to claim 1, characterized in that, The step of generating a historical state event set based on the historical operational data includes: The historical operating data is segmented into operating condition segments based on instrument load, pump speed, and valve opening to obtain historical operating condition segments. Temperature, flow rate, pressure, and coolant conductivity within the same historical operating condition segment are time-aligned to obtain an alignment parameter sequence; A set of historical state events is generated based on the direction, magnitude, and duration of parameter changes in the alignment parameter sequence.

3. The method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs according to claim 1, characterized in that, The preset fault knowledge graph includes component nodes, sensor parameter nodes, operating condition nodes, status event nodes, fault precursor nodes, and fault type nodes. The associated edge weights include the associated edge weights between status event nodes and fault precursor nodes, and the associated edge weights between fault precursor nodes and fault type nodes.

4. The method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs according to claim 1, characterized in that, The process of generating current status events based on the multidimensional real-time operating data includes generating parameter offset events, coupling anomaly events, and response hysteresis events based on the response relationship between temperature, flow rate, pressure, pump speed, and instrument load within the current time period.

5. The method for early fault monitoring of liquid-cooled instruments based on dynamic knowledge graphs according to claim 1, characterized in that, The step of generating an early fault diagnosis result for the liquid-cooled instrument based on the comparison result of the evidence support degree and the counter-evidence suppression degree includes: When the evidence support is higher than the disproving degree, an early fault judgment result representing the existence of early faults is generated based on the candidate fault links; When the evidence support is not higher than the disproving evidence suppression, the candidate fault link is suppressed and an early fault judgment result representing an unconfirmed early fault is generated. The early fault judgment results include fault existence, suspected fault type, suspected component, triggering state event, candidate fault link, evidence support, counter-evidence suppression degree, and verification conditions.

6. A liquid-cooled instrument early fault monitoring system based on dynamic knowledge graph, used to implement the liquid-cooled instrument early fault monitoring method according to any one of claims 1 to 5, characterized in that, include: The historical event generation module is used to acquire the latest set of historical operating data of the liquid cooling instrument and generate a set of historical status events based on the historical operating data. The graph update module is used to perform local incremental updates on the state event nodes, fault precursor nodes and their associated edge weights that match the historical state events in the preset fault knowledge graph according to the historical state event set, so as to obtain a dynamic knowledge graph. The real-time event generation module is used to acquire multi-dimensional real-time operating data of the liquid cooling instrument in the current period and generate current status events based on the multi-dimensional real-time operating data. The fault link generation module is used to recall fault precursor nodes and trace fault type nodes in the dynamic knowledge graph based on the current state event to obtain candidate fault links, and generate verification conditions based on the hit state event nodes, miss state event nodes and miss sensor parameter nodes in the candidate fault links. The evidence support determination module is used to match the parameter change direction, change duration and parameter coupling relationship in the conditions to be verified based on the multi-dimensional real-time running data, and determine the evidence support degree of the candidate fault link based on the matching results. The fault diagnosis module is used to determine the degree of disproving evidence suppression of the candidate fault links based on changes in instrument load, pump speed, valve opening and ambient temperature, and to generate an early fault diagnosis result for the liquid-cooled instrument based on the comparison result between the degree of evidence support and the degree of disproving evidence suppression.

Images