A method and system for inferring latent device faults by combining multi-source IoT data

Abstract

This invention provides a method and system for inferring latent device faults by combining multi-source IoT data. First, it acquires a multi-source IoT data set containing operating parameters of core components of smart home devices, ambient environmental perception data, and data on interactions between components. Next, it extracts temporal evolution imprints from the data set to generate a multi-dimensional data evolution imprint set. Then, it constructs an initial structure for a fault tracing inference tree based on the device's physical operating logic. Next, it dynamically matches the multi-dimensional data evolution imprint set with the nodes of the initial inference tree structure, adjusting the branch growth direction and node weights to generate an adaptive fault tracing inference tree. Finally, it mines the latent correlation logic between imprints and nodes based on this inference tree to generate a conclusion on the latent device fault tracing. This invention can comprehensively and accurately infer latent faults in smart home devices, improving system operational stability.

Description

Technical Field

[0001] This invention relates to the field of IoT device operation and maintenance technology, and more specifically, to a method and system for inferring latent device faults by combining multi-source IoT data. Background Technology

[0002] In the field of IoT-enabled smart homes, with the widespread adoption and deep interconnection of various smart devices, the number of devices in the home environment has increased dramatically, and the amount of data generated has also exploded. Smart home systems aim to provide users with a convenient, comfortable, and secure living experience; however, the stable operation of devices is fundamental to achieving this goal.

[0003] Currently, fault diagnosis and handling of smart home devices face numerous challenges. On the one hand, traditional fault diagnosis methods largely rely on manual inspection. For simple, obvious faults, such as devices failing to power on, manual inspection may quickly pinpoint the problem. However, smart home devices are often highly integrated and functionally complex, and many hidden faults are not readily apparent. For example, abnormal fluctuations in smart sensor data may not trigger obvious alarms, or communication failures between devices may cause intermittent malfunctions of some functions. Manual inspection is not only inefficient but also struggles to accurately locate the source of the fault. On the other hand, existing data analysis-based fault diagnosis methods mostly focus on a single device or a single type of data, such as analyzing only the operating parameters of a smart air conditioner to determine if it is faulty. This ignores the close interrelationships between devices in a smart home system and the impact of environmental factors on device operation. A smart home is an organic whole; devices work collaboratively through data interaction, and changes in environmental conditions also affect device operation. Analysis of a single data source cannot comprehensively reflect the actual operating status of the devices, significantly reducing the accuracy and comprehensiveness of fault diagnosis and failing to meet users' demands for stable and reliable operation of smart home systems. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for inferring latent device faults by combining multi-source IoT data, the method comprising:

[0005] Acquire a multi-source IoT data set of the device. The multi-source IoT data set includes operating parameter data of the device's core components, environmental perception data of the device's surrounding environment, and interaction behavior data between components. The operating parameter data records the state change information of the components during operation. The environmental perception data records the state change information of the environment in which the device is located. The interaction behavior data records the state change information of data transmission and response between components.

[0006] The time-series evolution imprint is extracted from the multi-source IoT data set to generate a multi-dimensional data evolution imprint set. The multi-dimensional data evolution imprint set includes the change pattern imprint, mutation node imprint, and correlation diffusion imprint of various types of data in different time segments.

[0007] Based on the physical operation logic of the equipment, an initial structure of the fault tracing and inference tree is constructed. The initial structure of the fault tracing and inference tree includes a root node, a first-level branch node, and a second-level branch node. The root node corresponds to the overall operating status of the equipment, the first-level branch node corresponds to the core system of the equipment, and the second-level branch node corresponds to the key components under the core system.

[0008] The multi-dimensional data evolution imprint set is dynamically matched with the nodes of the initial structure of the fault source inference tree. The branch growth direction and node weight of the inference tree are adjusted according to the matching result to generate an adaptive fault source inference tree.

[0009] Based on the adaptive fault tracing and inference tree, the implicit correlation logic between imprints and nodes is mined to generate equipment implicit fault tracing conclusions. The equipment implicit fault tracing conclusions include potential source components of the fault, fault evolution paths, and fault-related imprint features.

[0010] In another aspect, embodiments of the present invention also provide a device latent fault reasoning system that combines multi-source IoT data, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, by acquiring a multi-source IoT data set containing operating parameter data of core components of smart home devices, environmental perception data of the surrounding environment, and interaction behavior data between components, the core component operating parameter data can accurately reflect the device's own operating status, the environmental perception data can capture the impact of environmental factors on the device, and the interaction behavior data between components reflects the collaborative work between devices. The combination of these three provides rich and comprehensive data support for accurately inferring hidden device faults. Extracting temporal evolution imprints from the multi-source IoT data set generates a multi-dimensional data evolution imprint set, which can deeply explore the changing patterns, abrupt change nodes, and correlation diffusion of data at different time segments, accurately capturing the dynamic changes in the device's operating status and providing highly valuable clues for subsequent fault reasoning. Based on the physical operating logic of smart home devices, an initial structure for fault tracing and inference is constructed, clarifying the logical framework for fault reasoning and making the reasoning process more systematic and organized. By dynamically matching the multi-dimensional data evolution imprint set with the nodes of the initial structure of the fault tracing inference tree, and adjusting the branch growth direction and node weights of the inference tree according to the matching results, an adaptive fault tracing inference tree is generated. This enables real-time dynamic monitoring of the operating status of smart home devices and automatic adjustment of fault reasoning, greatly improving the accuracy and flexibility of reasoning and allowing for timely adaptation to changes in device operating status and environment. Finally, based on the adaptive fault tracing inference tree, the implicit correlation logic between imprints and nodes is mined to generate implicit fault tracing conclusions for devices, including potential fault source components, fault evolution paths, and fault-related imprint features. This helps to quickly locate and resolve faults, effectively improving the stability and reliability of smart home systems. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the device latent fault reasoning method that combines multi-source IoT data provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the device latent fault reasoning system that combines multi-source IoT data, provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for inferring device latent faults by combining multi-source IoT data, provided in one embodiment of the present invention. The following is a detailed description of this method for inferring device latent faults by combining multi-source IoT data.

[0015] Step S110: Obtain the multi-source IoT data set of the device. The multi-source IoT data set includes the operating parameter data of the device's core components, the environmental perception data of the device's surrounding environment, and the interaction behavior data between components. The operating parameter data records the state change information of the components during operation. The environmental perception data records the state change information of the environment in which the device is located. The interaction behavior data records the state change information of data transmission and response between components.

[0016] In this embodiment, a multi-source IoT data set is acquired using a central air conditioning unit in an IoT smart home system. The core components of the central air conditioning unit include a compressor, condenser, evaporator, and fan. Operating parameter data is collected through sensors deployed on each core component, covering the compressor's operating speed, output power, and operating current; the condenser's heat dissipation temperature and pressure; the evaporator's evaporation temperature and heat exchange efficiency; and the fan's wind speed and direction. This data records the real-time status changes of each component during operation.

[0017] The surrounding environment data of the equipment is collected by environmental sensors installed in the room where the central air conditioning is located and in the surrounding area. This includes indoor temperature, humidity, air quality index, carbon dioxide concentration, outdoor temperature, humidity, wind speed, etc. These data reflect the changes in the state of the environment in which the equipment is located.

[0018] Inter-component interaction data is collected through the internal communication module of the central air conditioning system. This includes data transmission frequency, data transmission volume, and response delay between the compressor and condenser; command interaction content and interaction interval between the condenser and evaporator; and status feedback time and feedback information integrity between the fan and the control module. These data record the status changes of data transmission and response between each component.

[0019] During the data collection process, environmental perception data involving user privacy, such as temperature change data collected by indoor sensors that may indirectly reflect user activities, are anonymized to remove identifying information that may be associated with user identity. At the same time, the collected data is transmitted to the data processing terminal through an encrypted transmission channel to prevent data leakage during transmission.

[0020] Step S120: Extract time-series evolution imprints from the multi-source IoT data set to generate a multi-dimensional data evolution imprint set. The multi-dimensional data evolution imprint set includes the change pattern imprints, mutation node imprints, and correlation diffusion imprints of various types of data in different time segments.

[0021] In this embodiment, time-series evolution imprints are extracted from the acquired multi-source IoT data set of central air conditioning. First, the multi-source IoT data set is divided according to data type. Then, corresponding imprints are extracted for different data subsets. Finally, the data is integrated to form a multi-dimensional data evolution imprint set.

[0022] Step S121: Divide the multi-source IoT data set into a subset of operating parameter data, a subset of environmental perception data, and a subset of interactive behavior data according to data type. Each data subset is divided into multiple consecutive time segments according to time order, and each time segment corresponds to the same time span.

[0023] The collected central air conditioning multi-source IoT data set is divided into categories: the operating parameter data subset includes operating parameter data of core components such as compressors, condensers, evaporators, and fans; the environmental perception data subset includes environmental data such as indoor and outdoor temperature, humidity, and air quality index; and the interaction behavior data subset includes relevant data on data transmission and response between various components.

[0024] Subsequently, each data subset is divided into consecutive time segments in chronological order, with each time segment having a fixed duration, such as one hour. The daily operating parameter data subset is divided into twenty-four consecutive time segments, with each time segment covering all operating parameter data collected within that hour. Other data subsets are also processed according to the same time span and division method.

[0025] Step S122: For each time segment of each data subset, extract the data change pattern imprint. The data change pattern imprint is obtained by analyzing the change trend, change frequency and change amplitude of the data within the time segment. If the data shows a periodic upward trend, the change frequency is stable and the change amplitude is within a preset range within the time segment, then a periodic upward pattern imprint is formed.

[0026] For each time segment of each data subset after division, data change pattern imprints are extracted. Taking a certain time segment of compressor operating speed data in the operating parameter data subset as an example, the changing trend, frequency, and amplitude of compressor operating speed within that time segment are analyzed. If the analysis finds that the compressor operating speed exhibits a periodic upward trend within that time segment, and the frequency of change is stable, with the amplitude of change within the preset normal operating range, then a periodic upward pattern imprint of compressor operating speed is formed.

[0027] Step S1221: For each time segment of each data subset, obtain all data points within that time segment to form a data point sequence.

[0028] For each time segment of each data subset, retrieve all data points of the corresponding data type collected within that time segment from the data storage terminal. For example, for a certain time segment of indoor temperature in the environmental sensing data subset, obtain the indoor temperature data points collected every minute within that time segment, and arrange these data points in chronological order of collection time to form a sequence of indoor temperature data points.

[0029] Step S1222: Calculate the changing trend of the data point sequence. Use a linear fitting method to fit the data point sequence and obtain the slope of the fitted line. If the slope is positive and the absolute value is greater than the preset slope threshold, the changing trend is determined to be an upward trend; if the slope is negative and the absolute value is greater than the preset slope threshold, the changing trend is determined to be a downward trend; if the absolute value of the slope is less than the preset slope threshold, the changing trend is determined to be a stable trend.

[0030] Taking a time segment data point sequence of the data transmission frequency between the compressor and condenser in the interactive behavior data subset as an example, a linear fitting method is used to fit this data point sequence. A linear fitting tool is used to fit the data points of the data transmission frequency in the data point sequence to obtain a fitted straight line, and the slope of this fitted line is calculated.

[0031] The calculated slope is compared with a preset slope threshold. If the slope is positive and its absolute value is greater than the preset slope threshold, it indicates that the data transmission frequency between the compressor and the condenser is generally on an upward trend within this time segment. If the slope is negative and its absolute value is greater than the preset slope threshold, the trend is determined to be downward. If the absolute value of the slope is less than the preset slope threshold, it indicates that the data transmission frequency changes little within this time segment, and the trend is determined to be stable.

[0032] Step S1223: Calculate the change frequency of the data point sequence. The change frequency is calculated by dividing the number of times the data values ​​in the statistical data point sequence change by the time span of the time segment. The result is the number of data changes per unit time.

[0033] Taking a data point sequence of evaporator heat exchange efficiency over a specific time segment from the subset of operating parameter data as an example, we count the number of times the evaporator heat exchange efficiency value changes within this data point sequence, i.e., the number of times the value differs between two adjacent data points. Dividing the counted number of changes by the time span of this time segment yields the number of changes in evaporator heat exchange efficiency per unit time, which is the frequency of change in the data point sequence.

[0034] Step S1224: Calculate the change range of the data point sequence. The change range is obtained by finding the maximum and minimum values ​​in the data point sequence and calculating the difference between the maximum and minimum values.

[0035] For a specific time segment of outdoor humidity data points within a subset of environmental sensing data, iterate through all data points in the sequence to find the maximum and minimum values. Subtract the minimum value from the maximum value to obtain the difference, which represents the magnitude of the change in the outdoor humidity data point sequence within that time segment.

[0036] Step S1225: Compare the change trend, change frequency and change magnitude with the preset corresponding threshold intervals, and generate the data change pattern imprint for the time segment based on the comparison results.

[0037] The calculated trend, frequency, and magnitude of data changes within a specific time segment are compared with preset trend threshold intervals, frequency threshold intervals, and magnitude threshold intervals, respectively. For example, for a specific time segment of the fan speed within the subset of operating parameter data, if its trend is stable and falls within a preset trend threshold interval, and the frequency and magnitude of changes are also within their respective preset threshold intervals, a stable fan speed change pattern imprint is generated. This imprint contains information that the fan speed change trend, frequency, and magnitude within that time segment conform to normal operating standards.

[0038] Step S1226: Compare the data change pattern imprints of adjacent time segments. If the difference in imprint characteristics between two adjacent time segments is less than a preset difference threshold, then merge the imprints of the two time segments to form a change pattern imprint over a longer time span. If the difference in imprint characteristics is greater than the preset difference threshold, then retain their respective independent imprints.

[0039] The variation patterns of compressor output power in two adjacent time segments from the selected subset of operating parameter data are compared. The differences between the two patterns are analyzed in terms of variation trend, frequency, and magnitude. If the calculated difference is less than a preset threshold, it indicates that the compressor output power variation patterns in these two adjacent time segments are similar, and the patterns from these two time segments are merged to form a single compressor output power variation pattern covering both time spans. If the difference is greater than the preset threshold, it indicates that the variation patterns of the two time segments are significantly different, and each pattern is retained independently.

[0040] Step S123: For each data subset, identify mutation nodes in the data sequence. The mutation node is a node whose data value deviates from the data change pattern of adjacent time segments. Extract the occurrence time of the mutation node, the difference between the data before and after the mutation, and the stable duration of the data after the mutation to form a mutation node imprint.

[0041] For each data subset, identify abrupt change nodes within the overall data sequence. Taking the indoor temperature data sequence of the environmental perception data subset as an example, continuously monitor the indoor temperature change patterns of each time segment. If the indoor temperature data value of a certain time segment deviates significantly from the temperature change patterns of the previous and subsequent time segments—for example, if the indoor temperature showed a stable trend in the previous time segment, suddenly rose sharply in this time segment, and then returned to stability in the next time segment—then the node corresponding to this time segment is the abrupt change node.

[0042] Extract the occurrence time of the mutation node, i.e., the start time of the time segment; calculate the data difference before and after the mutation, i.e., the difference between the highest temperature in the time segment and the average temperature of the previous time segment; record the duration of data stability after the mutation, i.e., the time it takes for the indoor temperature to return to a stable state after the end of the time segment. Based on this information, form an indoor temperature mutation node imprint.

[0043] Step S124: Analyze the correlation between different data subsets. If the change pattern imprint of any time segment of the running parameter data subset appears, the change pattern imprint of the corresponding time segment of the environmental perception data subset will appear, and the two change trends are consistent. Then, extract the time difference of the occurrence of the correlation, the duration of the correlation, and the correlation strength to form a correlation diffusion imprint.

[0044] Analyze the correlation between different data subsets, taking the operational parameter data subset and the environmental perception data subset as examples. If the condenser heat dissipation temperature in the operational parameter data subset shows an upward trend during a certain time segment, the indoor temperature in the environmental perception data subset also shows an upward trend during the corresponding time segment, and the upward trends of the two are consistent.

[0045] At this point, the time difference of the correlation is extracted, that is, the interval between the time when the condenser heat dissipation temperature rise imprint appears and the time when the indoor temperature rise imprint appears; the duration of the correlation is determined, that is, the length of time that the two imprints coexist; the correlation strength is calculated by analyzing the degree of fit between the two imprints in terms of change amplitude and change frequency. Based on the extracted and calculated information, a correlation diffusion imprint is formed.

[0046] Step S125: Classify and organize the extracted change pattern imprints, mutation node imprints, and related diffusion imprints according to data type and time segment, remove duplicate imprint content, and form a multi-dimensional data evolution imprint set.

[0047] The change pattern imprints, mutation node imprints, and correlation diffusion imprints extracted from each data subset will be divided into operation parameter imprints, environmental perception imprints, and interactive behavior imprints according to data type. Each type will then be arranged in chronological order according to time segments.

[0048] During the data processing, duplicate imprints are checked for. For example, if multiple imprints of the same data type and within the same time segment exhibit identical patterns of change, the duplicates are removed, and only one is retained. After classification, organization, and deduplication, a multi-dimensional data evolution imprint set is formed, encompassing various imprint types and arranged in an orderly manner according to data type and time segment.

[0049] Step S126: Add identification information to each imprint in the multi-dimensional data evolution imprint set. The identification information includes the data source type, time segment range, and imprint feature description corresponding to the imprint.

[0050] Add identification information to each imprint in the multi-dimensional data evolution imprint set. For example, for the imprint of the periodic increase pattern of compressor operating speed in the operating parameter class, its identification information indicates that the data source type is compressor operating parameters, the time segment range is a certain period of a certain day, and the imprint characteristic is described as the compressor operating speed increasing periodically within this period, with a stable frequency of change and a change amplitude within the normal range. Add complete identification information to each imprint in the above manner for subsequent querying and matching.

[0051] Step S130: Based on the physical operation logic of the equipment, construct the initial structure of the fault tracing and inference tree. The initial structure of the fault tracing and inference tree includes a root node, a first-level branch node, and a second-level branch node. The root node corresponds to the overall operating status of the equipment, the first-level branch node corresponds to the core system of the equipment, and the second-level branch node corresponds to the key components under the core system.

[0052] In this embodiment, an initial fault tracing tree structure is constructed based on the physical operation logic of the central air conditioning system. The physical operation logic of the central air conditioning system is as follows: the refrigerant is compressed by the compressor, liquefied by the condenser, evaporated and absorbed heat by the evaporator, and the cooling capacity is delivered to the room by the fan. The core systems work together to achieve temperature regulation. Based on this logic, an initial structure containing a root node, first-level branch nodes, and second-level branch nodes is constructed.

[0053] Step S131: Analyze the physical composition and structure of the equipment to determine the core systems included in the equipment. The core systems include a power supply system, a control and regulation system, a data transmission system, and a heat dissipation system. Each core system corresponds to a key functional module for the operation of the equipment.

[0054] Analyzing the physical structure of a central air conditioning system identifies its core systems. The power supply system, composed of compressors and motors, provides power for the system's operation. The control and regulation system, including controllers and sensors, receives commands and adjusts the operating status of each component. The data transmission system, consisting of communication modules and data cables, enables data exchange between components. The heat dissipation system, including condensers and cooling fans, dissipates heat generated during operation. Each core system corresponds to a key functional module for the central air conditioning system's operation, supporting its normal functioning.

[0055] Step S132: Set the overall operating status of the equipment as the root node of the fault tracing and deduction tree. The attribute information of the root node includes the current overall operating mode and running time of the equipment.

[0056] The overall operating status of the central air conditioning system is set as the root node of the fault tracing and inference tree. The attribute information of the root node records the current overall operating mode of the equipment, such as cooling mode, heating mode, air supply mode, etc., as well as the running time of the equipment from startup to the current moment.

[0057] Step S133: Set each core system as a first-level branch node in the fault tracing and deduction tree. Establish a connection between each first-level branch node and the root node. The connection relationship marks the degree of influence of the core system on the overall operating status of the equipment. The degree of influence is determined based on the functional importance of the core system.

[0058] The power supply system, control and regulation system, data transmission system, and heat dissipation system are each designated as a first-level branch node in the fault tracing tree. Each first-level branch node is connected to the root node, and the degree of influence of each core system on the overall operation of the central air conditioning system is marked in the connection relationship. For example, the power supply system, as the core system providing power, has the highest degree of influence on the overall operation, followed by the control and regulation system. The data transmission system and heat dissipation system are marked with their respective degrees of influence according to their functional importance.

[0059] Step S134: For the core system corresponding to each first-level branch node, identify the key components contained in the system, set each key component as a second-level branch node of the fault tracing and deduction tree, establish a connection between each second-level branch node and the corresponding first-level branch node, and mark the degree of support of the component for the realization of the core system functions.

[0060] For each primary branch node, the key components of its core system are identified. Taking the power supply system as an example, its key components include compressors, motors, and frequency converters; the key components of the control and regulation system include the main controller, temperature sensors, and humidity sensors; the key components of the data transmission system include communication modules and data interfaces; and the key components of the heat dissipation system include condensers, cooling fans, and heat dissipation pipes.

[0061] The aforementioned key components are designated as secondary branch nodes under their respective primary branch nodes. Each secondary branch node is connected to its primary branch node, and the degree of support that component provides to the core system's functions is indicated in the connection relationships. For example, the compressor, as the core component of the power supply system, provides the highest degree of support for the system's functions, followed by the motor, and the frequency converter is labeled with its corresponding degree of support based on its role.

[0062] Step S135: Add an initial weight to each node. The initial weight of the root node is set to the highest. The initial weight of the first-level branch nodes is allocated according to the degree of influence of the core system. The initial weight of the second-level branch nodes is allocated according to the degree of support of the components.

[0063] Initial weights are added to each node in the initial structure of the fault tracing tree. The root node is given the highest initial weight, representing its core position in the entire tree. For first-level branch nodes, initial weights are assigned based on the impact of each core system on the overall operating status of the equipment; the higher the impact, the greater the initial weight. For example, the initial weight of the first-level branch node corresponding to the power supply system is higher than that of nodes corresponding to other core systems. The initial weights of second-level branch nodes are assigned based on the degree of support each key component provides for the core system's functionality; the greater the support, the greater the initial weight.

[0064] Step S1351: Determine the evaluation index for node weight. The evaluation index includes the functional importance of the corresponding component or system, the probability of failure, and the scope of failure impact. Functional importance reflects the degree of support of the component or system for the overall operation of the equipment. The probability of failure reflects the frequency of failure of the component or system in historical operation. The scope of failure impact reflects the degree of impact of the failure of the component or system on other components or systems.

[0065] The evaluation metrics for determining node weights are functional importance, failure probability, and failure impact range. Functional importance measures the degree to which the corresponding component or system supports the overall operation of the central air conditioning system; for example, the compressor's functional importance is higher than that of the cooling fan. Failure probability is determined based on the frequency of failures of the component or system during historical operation; components or systems with more historical failures have a higher failure probability. Failure impact range refers to the degree to which the failure of this component or system affects the normal operation of other components or systems; for example, the impact range of a power supply system failure is greater than the impact range of a single sensor failure.

[0066] Step S1352: Set a weight coefficient for each evaluation indicator. The weight coefficient for functional importance is the highest, followed by the probability of failure, and the weight coefficient for the scope of failure impact is the lowest. The sum of the weight coefficients of each evaluation indicator is 1.

[0067] Weighting coefficients were assigned to three evaluation indicators: functional importance, failure probability, and failure impact range. Functional importance was assigned the highest weighting coefficient because its functional importance directly determines its core role in equipment operation. Failure probability received the second highest weighting coefficient, reflecting the reference value of historical failure scenarios for the current weighting allocation. Failure impact range received the lowest weighting coefficient. The weighting coefficients of the three indicators were summed to 1 to ensure the rationality of the weighting allocation.

[0068] Step S1353: For each node, use the expert scoring method to score its functional importance, failure probability and failure impact range. The higher the score, the more significant the corresponding indicator performance.

[0069] An expert scoring method was used to score three evaluation indicators for each node. Technical experts in the field of central air conditioning were invited to score the functional importance based on the actual situation of the corresponding components or systems of each node. The more important the function, the higher the score. The probability of failure was scored by referring to the historical failure records of the component or system. The node with more frequent historical failures scored higher. The scope of components or systems that may be affected after a failure was analyzed, and the scope of failure impact was scored. The node with a wider scope of impact scored higher.

[0070] Step S1354: Calculate the initial weight of the node. The initial weight is calculated by the sum of the product of the functional importance score and the functional importance weight coefficient, the product of the failure probability score and the failure probability weight coefficient, and the product of the failure impact range score and the failure impact range weight coefficient.

[0071] For example, for a first-level branch node corresponding to a power supply system, if its functional importance score is a certain value, the functional importance weight coefficient is the highest set coefficient, the failure probability score is the corresponding value, the failure probability weight coefficient is the next highest coefficient, the failure impact range score is the corresponding value, and the failure impact range weight coefficient is the lowest coefficient, then the initial weight of the node is the sum of these three products.

[0072] Step S1355: Normalize the initial weights of all nodes, record the initial weight calculation process of each node, including the scores of each evaluation index, weight coefficients and calculation results, and form a node initial weight calculation report.

[0073] After calculating the initial weights of all nodes, these initial weights are normalized to ensure that the weight values ​​of all nodes fall within the same numerical range, facilitating subsequent comparison and analysis. Normalization is achieved by dividing the initial weight of each node by the sum of the initial weights of all nodes.

[0074] Simultaneously, the initial weight calculation process for each node is meticulously recorded, including specific scores for functional importance, failure probability, and failure impact scope; the weight coefficients corresponding to each evaluation indicator; and the results of product and summation calculations. This information is compiled into a node initial weight calculation report, which can be used for subsequent traceability and verification of node weight adjustments.

[0075] Step S136: Record the node hierarchy, connection relationship and initial node weight of the initial structure of the fault tracing inference tree to form the initial structure document of the inference tree. The initial structure document of the inference tree also includes the physical location information and functional description of the equipment corresponding to each node.

[0076] The hierarchical relationships of each node in the initial structure of the fault tracing tree, namely the subordinate relationships between the root node, first-level branch nodes, and second-level branch nodes, as well as the connection relationships between each node and the information such as the degree of influence and support of the labels, are recorded together with the initial weight of each node.

[0077] In addition, the physical location information of the equipment corresponding to each node should be added to the record. For example, the secondary branch node corresponding to the compressor should specify its specific installation location in the central air conditioning equipment, as well as the functional description of each node. For example, the primary branch node corresponding to the power supply system should explain its function of providing operating power to the equipment.

[0078] Step S140: Dynamically match the multi-dimensional data evolution imprint set with the nodes of the initial structure of the fault source inference tree, adjust the branch growth direction and node weight of the inference tree according to the matching result, and generate an adaptive fault source inference tree.

[0079] In this embodiment, based on preset matching rules, each imprint in the multi-dimensional data evolution imprint set is dynamically matched with nodes in the initial structure of the fault source tracing tree. The node weights are adjusted according to the number of imprints matched to each node and their correlation, and whether to add or prune branches is determined based on the matching results, ultimately generating an adaptive fault source tracing tree.

[0080] Step S141: Establish imprint node matching rules. The matching rules are determined based on the correlation between imprint features and the physical attributes of the equipment corresponding to the node. If the parameter changes reflected by the change pattern imprint are related to the operating parameters of any secondary branch node's corresponding component, then the change pattern imprint matches the secondary branch node.

[0081] Based on the correlation between imprint features and the physical attributes of the equipment corresponding to the nodes, imprint node matching rules are established. For imprints with a change pattern, if the parameter changes they reflect, such as the periodic changes in the compressor's operating speed, are related to the operating parameters of the component corresponding to a certain secondary branch node, i.e., the compressor, then the imprint with the change pattern is determined to match that secondary branch node.

[0082] Step S1411: Collect the standard range of operating parameters for each core system and key component of the equipment. The standard range of operating parameters includes the range of parameter values, the type of change trend, and the range of change magnitude when the component is operating normally.

[0083] Collect the standard ranges of operating parameters for each core system and key component of the central air conditioning system. These standard ranges are determined by the technical manuals and industry standards provided by the equipment manufacturers. For example, the standard range of operating parameters for the compressor in the power supply system includes the speed range during normal operation, the type of allowable change trend (such as stable or slow fluctuation), and the range of change amplitude; the standard range of operating parameters for the condenser in the heat dissipation system includes the normal operating temperature range, the type of pressure change trend, and the range of change amplitude, etc.

[0084] Step S1412: For each imprint type in the multidimensional data evolution imprint set, analyze the correlation between imprint characteristics and the standard range of operating parameters. If the trend of change of the change pattern imprint is consistent with the trend of change of the standard range of operating parameters of any component, then it is determined that there is a potential matching relationship between the imprint type and the node corresponding to the component.

[0085] For each type of imprint in the multi-dimensional data evolution imprint set, such as change pattern imprints, abrupt change node imprints, and correlation diffusion imprints, the correlation between their imprint characteristics and the standard range of operating parameters of each component is analyzed one by one. Taking change pattern imprints as an example, if the change trend of a certain change pattern imprint, such as a slow upward trend, is consistent with the change trend type allowed in the standard range of condenser operating parameters, then it is determined that there is a potential matching relationship between this change pattern imprint type and the corresponding secondary branch node of the condenser.

[0086] Step S1413: For the mutation node imprint, analyze the correspondence between the parameter type corresponding to the mutation node and the component operating parameters. If the parameter type corresponding to the mutation node belongs to the core operating parameters of any component, then it is determined that there is a potential matching relationship between the mutation node imprint and the node corresponding to the component.

[0087] For abrupt change node imprints, the focus is on analyzing the corresponding parameter types, such as current, temperature, and pressure, and their correspondence with the operating parameters of each key component. If the parameter type corresponding to a certain abrupt change node imprint is the compressor's operating current, and the operating current is a core operating parameter of the compressor, then it is determined that there is a potential matching relationship between this abrupt change node imprint and the corresponding secondary branch node of the compressor.

[0088] Step S1414: For the correlation diffusion imprint, analyze the correspondence between the data source types involved in the correlation diffusion and the core system. If the correlation diffusion involves both the running parameter data source and the environmental awareness data source, and both are related to any core system, then it is determined that there is a potential matching relationship between the correlation diffusion imprint and the first-level branch node corresponding to the core system.

[0089] Analyze the data source types involved in the correlation diffusion imprint, such as operational parameter data sources, environmental perception data sources, and interactive behavior data sources, and their correspondence with each core system. If a correlation diffusion imprint involves both operational parameter data sources and environmental perception data sources, and both data sources are related to the control and regulation system—that is, the operational parameter data reflects the state of the internal components of the system, and the environmental perception data provides the basis for the system's regulation—then it is determined that there is a potential matching relationship between this correlation diffusion imprint and the corresponding first-level branch node of the control and regulation system.

[0090] Step S1415: Set the correlation calculation method for imprint node matching. The correlation is calculated by multiplying the fit between the imprint features and the standard node parameters and the proximity between the time segment corresponding to the imprint and the current running time of the node. The fit is calculated based on the degree of overlap between the imprint features and the standard features, and the proximity is calculated based on the reciprocal of the time difference.

[0091] The specific calculation method for correlation is defined as follows: First, the fit is calculated, which is the degree of overlap between the imprint feature and the standard feature of the corresponding running parameters of the node. The higher the degree of overlap, the larger the fit value. Next, the proximity is calculated, which is the reciprocal of the time difference between the time segment corresponding to the imprint and the current running time of the node. The smaller the time difference, the larger the proximity value. The correlation is then the product of the fit and the proximity.

[0092] Step S1416: Organize the above potential matching relationships and correlation calculation methods into an imprint node matching rule document. The imprint node matching rule document also includes matching priorities corresponding to different imprint types. The matching priority of mutation node imprints is higher than that of change pattern imprints, and the matching priority of association diffusion imprints is higher than that of mutation node imprints.

[0093] The identified potential matching relationships and correlation calculation methods are compiled into an imprint node matching rule document. The document clearly defines the matching priority of different imprint types, with association diffusion imprints having the highest priority, followed by mutation node imprints, and change pattern imprints having the lowest priority. When multiple imprints of different types have potential matching relationships with the same node, the matching result is determined based on the imprint with the highest priority.

[0094] Step S142: Traverse each imprint in the multi-dimensional data evolution imprint set, find the corresponding matching node in the initial structure of the fault tracing inference tree according to the matching rules, and record the number of imprints matched by each node and the correlation between each imprint and the node.

[0095] Following the rules in the imprint node matching document, each imprint in the multi-dimensional data evolution imprint set is traversed. For each imprint, a matching node is found in the initial structure of the fault tracing tree based on its type and characteristics. For example, if the parameter type corresponding to a certain mutation node imprint is the wind turbine's rotational speed, and it belongs to the core operating parameter of the wind turbine, then it is matched with the secondary branch node corresponding to the wind turbine.

[0096] After matching is completed, the number of imprints matched by each node and the correlation value between each matched imprint and the node are recorded to form a node imprint matching record table.

[0097] Step S143: Calculate the matching score for each node, which is obtained by summing the correlation of all imprints matched by the node.

[0098] For each node, the relevance values ​​of all matched imprints are summed, and the sum is the matching score for that node. For example, if a second-level branch node matches three imprints with three different relevance values, the matching score for that node is the sum of these three values.

[0099] Step S144: Adjust the node weight based on the node matching score. If the node matching score is higher than the preset score threshold, increase the node weight. The increase is positively correlated with the portion of the matching score that exceeds the preset score threshold. If the node matching score is lower than the preset score threshold, decrease the node weight. The decrease is positively correlated with the portion of the matching score that is lower than the preset score threshold.

[0100] The matching score of each node is compared with a preset score threshold. If the matching score of a node is higher than the preset score threshold, it indicates that the component or system corresponding to that node has a high degree of correlation with the data evolution imprint, and its weight needs to be increased. The increase increases as the matching score exceeds the threshold.

[0101] If a node's matching score is lower than a preset score threshold, it indicates that the node has a low correlation with the data evolution imprint and its weight needs to be reduced. The reduction increases as the number of matching scores below the threshold increases.

[0102] Step S145: For secondary branch nodes where the number of matched imprints exceeds a preset threshold, add a tertiary branch node. The tertiary branch node corresponds to the key sub-component of the secondary branch node component. The newly added branch node establishes a connection with the original secondary branch node, and the connection relationship indicates the degree of influence of the sub-component on the function of the original component.

[0103] The number of marks matched for each second-level branch node is counted. If the number of marks matched for a certain second-level branch node exceeds the preset threshold, it indicates that the component corresponding to that node may have more granular fault correlation points, and a new third-level branch node needs to be added.

[0104] Step S1451: For each second-level branch node, count the number of imprints in the multi-dimensional data evolution imprint set that it matches. If the number of imprints exceeds the preset threshold, it is determined that the second-level branch node needs to add a third-level branch node.

[0105] The number of imprints matched from the multi-dimensional data evolution imprint set for each second-level branch node is counted one by one, and the results are compared with a preset threshold. If the number of imprints for a second-level branch node exceeds the threshold, it is determined that the node needs to add a third-level branch node.

[0106] Step S1452: Analyze the composition structure of the component corresponding to the secondary branch node, and determine the key sub-components contained in the component. The key sub-components are those that play a core role in the realization of the component's function or those that have experienced frequent historical failures.

[0107] Taking the components corresponding to this secondary branch node, such as a compressor, as an example, we can analyze its structural composition, including sub-components such as crankshaft, piston, cylinder, and valve assembly. From this, we can identify key sub-components, namely the crankshaft and piston, which play a core role in realizing the compressor's function, and the valve assembly, which has a history of frequent failures.

[0108] Step S1453: Set each key sub-component as a third-level branch node. The name of each third-level branch node is consistent with the name of the key sub-component. The attribute information of the third-level branch node includes the physical location, functional description and operating parameter type of the sub-component.

[0109] The selected key sub-components, such as crankshaft, piston, and valve assembly, are designated as three-level branch nodes, named "crankshaft," "piston," and "valve assembly," respectively. The attribute information of each three-level branch node specifies its physical location, such as the crankshaft being located in the core transmission part of the compressor; its function description, such as the crankshaft driving the piston movement; and its operating parameters, such as the crankshaft's rotation frequency and wear level.

[0110] Step S1454: Establish the connection relationship between the third-level branch node and the corresponding second-level branch node. The attribute of the connection relationship includes the degree of influence of the sub-component on the function of the original component. The degree of influence is determined by analyzing the degree of damage to the function of the original component when the sub-component fails.

[0111] Establish the connection relationships between the three-level branch nodes (crankshaft, piston, valve assembly, etc.) and the two-level branch node (compressor). Specify the degree of impact of each sub-component on the compressor's function in the connection relationship attributes. For example, crankshaft failure will cause the compressor to completely stop operating, having the highest impact; valve assembly failure will cause a decrease in compressor efficiency, having the second highest impact. This degree of impact is determined by analyzing the damage to the compressor's function when each sub-component fails.

[0112] Step S1455: Assign initial weights to the newly added third-level branch nodes. The initial weights are determined based on the functional importance of the sub-components and their correlation with the second-level branch nodes.

[0113] Based on the functional importance of the sub-components corresponding to the third-level branch nodes (e.g., the crankshaft is more important than the valve assembly), and the degree of correlation between the sub-components and the second-level branch nodes (e.g., the piston is more related to the compressor than to other auxiliary sub-components), an initial weight is assigned to each newly added third-level branch node. The higher the functional importance and the stronger the correlation of the sub-component, the greater the initial weight of its corresponding third-level branch node.

[0114] Step S1456: Add the newly added third-level branch nodes and their corresponding connection relationships to the fault tracing and deduction tree, update the node hierarchy structure of the deduction tree, and record the relevant information of the newly added nodes, including the reason for addition, sub-component information and initial weight.

[0115] The newly added third-level branch nodes and their connections with the second-level branch nodes are added to the fault tracing and deduction tree, expanding the node hierarchy of the deduction tree from three levels to four levels. Simultaneously, relevant information for each newly added node is recorded, including the reason for addition (i.e., the number of matching imprints of the original second-level branch node exceeds a threshold), the physical location and functional description of the sub-component, and the assigned initial weight, thus completing the adjustment of the branch growth of the deduction tree.

[0116] Step S146: For nodes that do not match any marks, check whether there is a data acquisition gap in the corresponding device component. If there is a data acquisition gap, keep the node and mark the missing status; if there is no data acquisition gap, prune the node and its corresponding branch.

[0117] For nodes that do not match any trace during dynamic matching, first check if there is any missing data acquisition for the corresponding device component, such as a sensor malfunction causing the failure to collect the component's operating parameter data. If missing data acquisition exists, it means that the lack of matching traces does not indicate that the component has no abnormal association. Keep the node and mark it with the "missing data acquisition" status in its attribute information.

[0118] If there is no missing data collection, meaning that the relevant data for this component has been collected normally, but no imprint has been matched, it indicates that this component is not related to the current data evolution imprint. Prune this node and its branches and remove it from the fault tracing and inference tree.

[0119] Step S147: Integrate the adjusted node weights, newly added branch nodes, and pruning results to generate an adaptive fault tracing inference tree. The adaptive fault tracing inference tree includes the updated node hierarchy, connection relationships, node weights, and imprint matching records.

[0120] The nodes with adjusted weights, the newly added third-level branch nodes, and the nodes retained after pruning are integrated to form an adaptive fault tracing tree. This adaptive fault tracing tree includes the updated node hierarchy structure, namely the hierarchical relationships of the root node, first-level branch nodes, second-level branch nodes, and third-level branch nodes, the connection relationships between each node and the degree of influence and support of the labeled nodes, the adjusted node weights, and the imprint records matched to each node.

[0121] Step S150: Based on the adaptive fault tracing inference tree, mine the implicit correlation logic between imprints and nodes, and generate the equipment implicit fault tracing conclusion. The equipment implicit fault tracing conclusion includes the potential source component of the fault, the fault evolution path, and the fault association imprint features.

[0122] In this embodiment, based on the adaptive fault tracing and inference tree, the core focus nodes are extracted, the sequence of imprint appearance and the relationship between nodes are analyzed, the implicit relationship logic between imprints and nodes is mined, and finally integrated to form the equipment implicit fault tracing conclusion that includes the potential source components of the fault, the fault evolution path and the characteristics of the fault-related imprints.

[0123] Step S151: Extract the nodes with the highest weights in the adaptive fault tracing and inference tree, which are ranked according to a preset proportion, as core attention nodes. The core attention nodes reflect the components or systems in the equipment that are most closely related to the data evolution imprint.

[0124] The weights of all nodes in the adaptive fault tracing tree are sorted, and the nodes with the highest weight rankings are extracted as core focus nodes according to a preset ratio. For example, if the preset ratio is the top 30%, then the nodes with the top 30% weight rankings are extracted. The components or systems corresponding to these nodes are most closely related to the multi-dimensional data evolution imprints and are the key analysis objects for fault tracing.

[0125] Step S152: For each core focus node, collect all the data evolution imprints it matches, analyze the chronological order of the data evolution imprints, and determine the occurrence sequence of the data evolution imprints. If the mutation node imprint appears first, followed by the associated diffusion imprint, and finally the change pattern imprint, then a mutation association pattern imprint occurrence sequence is formed.

[0126] For each core node of interest, all data evolution imprints matched with it are collected from the imprint matching records of the adaptive fault tracing and inference tree, including imprints of change patterns, imprints of abrupt change nodes, and imprints of related diffusion.

[0127] Analyzing the temporal order of these imprints, that is, determining the order of their appearance based on the time segments corresponding to the imprints, forms an imprint appearance sequence. For example, if a core node of interest first matches a mutation node imprint, then matches an association diffusion imprint, and finally matches a change pattern imprint, then the imprint appearance sequence of this node is "mutation node imprint - association diffusion imprint - change pattern imprint", that is, the imprint appearance sequence of mutation association pattern.

[0128] Step S153: Based on the imprint occurrence sequence and combined with the physical operation logic of the equipment, deduce the state change process of the component or system corresponding to the core focus node. If the mutation node imprint corresponds to the parameter mutation of the component, the associated diffusion imprint corresponds to the parameter mutation of the component spreading to other components, and the change pattern imprint corresponds to the stable abnormal state formed after diffusion. Then it is deduced that the component first undergoes parameter mutation, then triggers the abnormality of the associated components, and finally forms a stable abnormal state.

[0129] Based on the sequence of imprint appearances and combined with the physical operation logic of central air conditioning, the state change process of the components or systems corresponding to the core focus nodes is deduced. For example, if a core focus node is a secondary branch node corresponding to a compressor, its imprint appearance sequence is: mutation node imprint — related diffusion imprint — change pattern imprint.

[0130] Among them, the mutation node imprint corresponds to a sudden increase in the compressor's operating current, the associated diffusion imprint corresponds to the diffusion of this current mutation to the condenser, and the change pattern imprint corresponds to the continuous abnormal current state formed by the compressor and condenser after diffusion. Combining the collaborative working logic of the compressor and condenser in the central air conditioning, it is deduced that the compressor first experiences a sudden change in operating current, then the abnormality spreads to the condenser, and finally leads to a stable abnormal current state for both.

[0131] Step S154: Analyze the imprint association relationship between different core concern nodes. If the association diffusion imprint of the first core concern node and the mutation node imprint of the second core concern node are continuous in time, and the association diffusion direction points to the component corresponding to the second core concern node, then it is determined that there is a state influence relationship between the two core concern nodes. The abnormal state of the first core concern node triggers the abnormal state of the second core concern node.

[0132] Analyze the imprint relationships between different core concern nodes to determine whether there are state influence relationships between nodes, that is, whether the abnormal state of one node triggers the abnormal state of another node.

[0133] For example, step S1541: For any two core focus nodes, extract the time information of all data evolution imprints matched by each node to form the imprint time series of the two nodes.

[0134] Select any two core nodes of interest, such as the secondary branch node corresponding to the compressor and the secondary branch node corresponding to the condenser, and extract the time information of all data evolution imprints matched by these two nodes, that is, the time segment corresponding to each imprint, and arrange them in chronological order to form their respective imprint time series.

[0135] Step S1542: Compare the imprint time series of the two nodes and find temporally continuous imprint combinations. If the time difference between the end time of any imprint of the first core focus node and the start time of any imprint of the second core focus node is less than a preset time difference threshold, then mark the two imprints as a potential associated imprint pair.

[0136] The imprint time series of the secondary branch nodes corresponding to the compressor and the secondary branch nodes corresponding to the condenser are compared, and the time range of each imprint in the two series is examined one by one. If the time difference between the end time of a certain associated diffusion imprint of the secondary branch node corresponding to the compressor and the start time of a certain mutation node imprint of the secondary branch node corresponding to the condenser is less than a preset time difference threshold, then these two imprints are identified as a potential associated imprint pair and marked in the imprint matching record.

[0137] Step S1543: Analyze the feature association relationship of potential related imprint pairs. If the imprint of the first core focus node is a related diffusion imprint and the diffusion direction points to the component corresponding to the second core focus node, and the imprint of the second core focus node is a mutation node imprint, and the mutation parameter is consistent with the parameter type involved in the diffusion imprint of the first core focus node, then it is determined that the two imprints have a feature association.

[0138] For the identified potential correlation imprint pairs, their characteristic correlation relationships are further analyzed. Taking the potential correlation imprint pair consisting of the correlation diffusion imprint of the secondary branch node corresponding to the compressor and the abrupt node imprint of the secondary branch node corresponding to the condenser as an example, we first confirm whether the diffusion direction of the correlation diffusion imprint corresponding to the compressor points to the condenser. Then, we check the abrupt parameter type of the abrupt node imprint corresponding to the condenser, such as whether it is temperature, pressure, etc. If the abrupt parameter type is consistent with the parameter type involved in the correlation diffusion imprint of the compressor, then we determine that the potential correlation imprint pair has a characteristic correlation.

[0139] Step S1544: Calculate the association strength of the potential associated imprint pair. The association strength is calculated by multiplying the temporal continuity and the feature fit, wherein the temporal continuity is the ratio of the time difference threshold to the actual time difference, and the feature fit is the degree of overlap between the two imprint features.

[0140] When calculating the association strength of a potentially related imprint pair, the temporal continuity is first calculated by dividing a preset time difference threshold by the actual time difference between the two imprints. The smaller the actual time difference, the larger the temporal continuity value. Next, the feature fit is calculated by comparing the degree of overlap between the two imprints in features such as the direction of parameter change and the range of influence. The higher the degree of overlap, the larger the feature fit value. Finally, the temporal continuity and the feature fit are multiplied together to obtain the association strength of the potentially related imprint pair.

[0141] Step S1545: If the association strength is higher than the preset association strength threshold, it is determined that there is a state influence relationship between the two core attention nodes, the first core attention node is the source node of influence, and the second core attention node is the affected node.

[0142] The calculated correlation strength of the potential association imprint pairs is compared with a preset correlation strength threshold. If the correlation strength is higher than the preset correlation strength threshold, it indicates that the correlation between the two core attention nodes has reached the set standard. In this case, it is determined that the two core attention nodes have a state influence relationship, where the first core attention node is the source node, and its abnormal state will affect the second core attention node, which is the affected node.

[0143] Step S1546: Perform pairwise analysis on all core nodes of interest, record the node pairs with state influence relationships and their corresponding association imprint pairs and association strengths, and form a list of node association relationships. The list of node association relationships also includes the derivation basis and time order of the association relationships.

[0144] Following the steps outlined above, perform pairwise analysis on all core nodes of interest in the adaptive fault tracing tree to investigate whether any state-related relationships exist between each pair of core nodes of interest. Record each node pair with a state-related relationship, its corresponding association imprint pair, and the association strength value, noting the basis for deriving the association relationship, such as the specific calculation process for temporal continuity and feature fit, and the temporal order of the association imprints. Compile and summarize this information to form a list of node association relationships.

[0145] Step S155: Based on the state change process of the core concern nodes and the state influence relationship between nodes, analyze the fault evolution path. The fault evolution path starts from the core concern node with the earliest abnormal imprint and extends to other core concern nodes along the state influence relationship.

[0146] Based on the state change process of each key concern node—that is, the component or system state changes derived from the sequence of anomaly occurrences—and combined with the state influence relationships between nodes recorded in the node association list, the fault propagation path is traced. First, the key concern node with the earliest anomaly mark is identified, serving as the starting point of the fault evolution path. Then, according to the state influence relationships between nodes, the key concern nodes affected by it are connected sequentially to form a complete fault evolution path. For example, if the earliest anomaly mark appears at the secondary branch node corresponding to the compressor, this node affects the secondary branch node corresponding to the condenser, and the secondary branch node corresponding to the condenser in turn affects the secondary branch node corresponding to the evaporator. Therefore, the fault evolution path is "compressor—condenser—evaporator".

[0147] Step S156: Determine the component corresponding to the core focus node where the earliest abnormal imprint appears as the potential source component of the fault, and extract the key imprint features matched by the potential source component of the fault as the fault association imprint features.

[0148] In the analyzed fault evolution path, the core node of concern with the earliest abnormal imprint is identified. The equipment component corresponding to this core node of concern is the potential source component of the fault. For example, if the core node of concern with the earliest abnormal imprint is the secondary branch node corresponding to the compressor, then the compressor is the potential source component of the fault. From all data evolution imprints matched to this potential source component of the fault, key imprint features that can reflect its abnormal state are extracted, such as the mutation parameter type of the mutation node imprint and the diffusion range of the associated diffusion imprint. These features are identified as fault association imprint features.

[0149] Step S157: Integrate potential source components of the fault, fault evolution path and fault association imprint features to form a conclusion on the source of hidden equipment faults. The conclusion on the source of hidden equipment faults also includes the abnormal state description and imprint matching basis of each core focus node.

[0150] The identified potential source components of the fault, the traced fault evolution paths, and the extracted fault association features are integrated. Simultaneously, descriptions of abnormal states at each key focus point are added, such as sudden changes in compressor operating current and abnormal increases in condenser heat dissipation temperature, along with the matching criteria for the abnormal states of each key focus point, i.e., the corresponding association feature types and characteristics. This information is then organized logically to form a complete conclusion for tracing the source of latent equipment faults.

[0151] Figure 2 The illustration shows exemplary hardware and software components of a device latent fault reasoning system 100 incorporating multi-source IoT data, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 may be used in the device latent fault reasoning system 100 incorporating multi-source IoT data and to perform the functions described in this application.

[0152] The device latent fault reasoning system 100, which combines multi-source IoT data, can be a general-purpose server or a special-purpose server. Both can be used to implement the device latent fault reasoning method combining multi-source IoT data of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0153] For example, a device latent fault reasoning system 100 incorporating multi-source IoT data may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the device latent fault reasoning system 100 incorporating multi-source IoT data may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The device latent fault reasoning system 100 incorporating multi-source IoT data also includes an I / O interface 150 between the computer and other input / output devices.

[0154] For ease of explanation, only one processor is described in the device latent fault reasoning system 100 that incorporates multi-source IoT data. However, it should be noted that the device latent fault reasoning system 100 in this application may also include multiple processors, and therefore the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the device latent fault reasoning system 100 incorporating multi-source IoT data performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0155] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the device latent fault reasoning method combining multi-source IoT data is implemented as described above.

[0156] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for inferring latent device faults by combining multi-source IoT data, characterized in that, The method includes: Acquire a multi-source IoT data set of the device. The multi-source IoT data set includes operating parameter data of the device's core components, environmental perception data of the device's surrounding environment, and interaction behavior data between components. The operating parameter data records the state change information of the components during operation. The environmental perception data records the state change information of the environment in which the device is located. The interaction behavior data records the state change information of data transmission and response between components. The time-series evolution imprint is extracted from the multi-source IoT data set to generate a multi-dimensional data evolution imprint set. The multi-dimensional data evolution imprint set includes the change pattern imprint, mutation node imprint, and correlation diffusion imprint of various types of data in different time segments. Based on the physical operation logic of the equipment, an initial structure of the fault tracing and inference tree is constructed. The initial structure of the fault tracing and inference tree includes a root node, a first-level branch node, and a second-level branch node. The root node corresponds to the overall operating status of the equipment, the first-level branch node corresponds to the core system of the equipment, and the second-level branch node corresponds to the key components under the core system. The multi-dimensional data evolution imprint set is dynamically matched with the nodes of the initial structure of the fault source inference tree. The branch growth direction and node weight of the inference tree are adjusted according to the matching result to generate an adaptive fault source inference tree. Based on the adaptive fault tracing and inference tree, the implicit correlation logic between imprints and nodes is mined to generate equipment implicit fault tracing conclusions. The equipment implicit fault tracing conclusions include potential source components of the fault, fault evolution paths, and fault-related imprint features.

2. The device latent fault reasoning method combining multi-source IoT data according to claim 1, characterized in that, The step of extracting time-series evolution imprints from the multi-source IoT data set to generate a multi-dimensional data evolution imprint set includes: The multi-source IoT data set is divided into a subset of operating parameter data, a subset of environmental perception data, and a subset of interactive behavior data according to data type. Each data subset is divided into multiple consecutive time segments according to time order, and each time segment corresponds to the same time span. For each time segment of each data subset, data change pattern imprints are extracted. These imprints are obtained by analyzing the change trend, frequency, and magnitude of the data within the time segment. If the data shows a periodic upward trend, a stable frequency of change, and a magnitude of change within a preset range within the time segment, then a periodic upward pattern imprint is formed. For each data subset, identify mutation nodes in the data sequence. The mutation node is a node where the data value deviates from the data change pattern of adjacent time segments. Extract the occurrence time of the mutation node, the difference between the data before and after the mutation, and the stable duration of the data after the mutation to form a mutation node imprint. The correlation between different data subsets is analyzed. If the change pattern imprint of any time segment of the running parameter data subset appears, the change pattern imprint of the corresponding time segment of the environmental perception data subset will appear, and the change trends of the two are consistent, then the time difference of the occurrence of the correlation, the duration of the correlation, and the correlation strength are extracted to form the correlation diffusion imprint. The extracted change pattern imprints, mutation node imprints, and associated diffusion imprints are classified and organized according to data type and time segment, and duplicate imprint content is removed to form a multi-dimensional data evolution imprint set. Add identification information to each imprint in the multi-dimensional data evolution imprint set. The identification information includes the data source type, time segment range, and imprint feature description corresponding to the imprint.

3. The device latent fault reasoning method combining multi-source IoT data according to claim 2, characterized in that, The extraction of data change patterns for each time segment of each data subset includes: For each time segment of each data subset, obtain all data points within that time segment to form a data point sequence; The trend of the data point sequence is calculated by fitting the data point sequence with a linear fitting method to obtain the slope of the fitted line. If the slope is positive and the absolute value is greater than the preset slope threshold, the trend is determined to be an upward trend; if the slope is negative and the absolute value is greater than the preset slope threshold, the trend is determined to be a downward trend; if the absolute value of the slope is less than the preset slope threshold, the trend is determined to be a stationary trend. The frequency of change of the data point sequence is calculated by dividing the number of times the data values ​​in the data point sequence change by the time span of the time segment, and the result is the number of times the data changes per unit time. The magnitude of change in the data point sequence is calculated by finding the maximum and minimum values ​​in the data point sequence and calculating the difference between the maximum and minimum values. The trend, frequency, and magnitude of change are compared with the corresponding preset threshold ranges, and a data change pattern imprint for the time segment is generated based on the comparison results. The data change patterns of adjacent time segments are compared. If the difference in the characteristics of the patterns of two adjacent time segments is less than a preset difference threshold, the patterns of the two time segments are merged to form a change pattern pattern over a longer time span. If the difference in the characteristics of the patterns is greater than the preset difference threshold, each pattern is retained independently.

4. The device latent fault reasoning method combining multi-source IoT data according to claim 1, characterized in that, The initial structure for constructing a fault tracing and deduction tree based on the physical operation logic of the equipment includes: Analyze the physical composition and structure of the equipment to determine the core systems it contains. The core systems include a power supply system, a control and regulation system, a data transmission system, and a heat dissipation system. Each core system corresponds to a key functional module for the operation of the equipment. The overall operating status of the equipment is set as the root node of the fault tracing and inference tree. The attribute information of the root node includes the current overall operating mode and running time of the equipment. Each core system is set as a first-level branch node in the fault tracing and inference tree. Each first-level branch node is connected to the root node. The connection relationship marks the degree of influence of the core system on the overall operating status of the equipment. The degree of influence is determined based on the functional importance of the core system. For each primary branch node, identify the key components of the core system and set each key component as a secondary branch node in the fault tracing tree. Establish a connection between each secondary branch node and the corresponding primary branch node, and use the connection relationship to indicate the degree of support of the component for the realization of the core system's functions. Add an initial weight to each node. Set the root node to the highest initial weight. The initial weights of first-level branch nodes are allocated according to the degree of influence on the core system, and the initial weights of second-level branch nodes are allocated according to the degree of support of the components. Record the node hierarchy, connection relationships, and initial node weights of the initial structure of the fault tracing tree to form an initial structure document of the tracing tree. The initial structure document of the tracing tree also includes the physical location information and functional description of the equipment corresponding to each node.

5. The device latent fault reasoning method combining multi-source IoT data according to claim 4, characterized in that, The process of adding initial weights to each node includes: Evaluation metrics for determining node weights include the functional importance of the corresponding component or system, the probability of failure, and the scope of failure impact. Functional importance reflects the degree of support of the component or system for the overall operation of the equipment. The probability of failure reflects the frequency of failure of the component or system in historical operation. The scope of failure impact reflects the degree of impact of the failure of the component or system on other components or systems. Each evaluation indicator is assigned a weight coefficient, with the functional importance having the highest weight coefficient, followed by the probability of failure, and the scope of failure impact having the lowest. The sum of the weight coefficients of each evaluation indicator is 1. For each node, an expert scoring method is used to score its functional importance, failure probability, and failure impact range. The higher the score, the more significant the corresponding indicator performance. The initial weight of the node is calculated by summing the products of the functional importance score and the functional importance weight coefficient, the failure probability score and the failure probability weight coefficient, and the failure impact range score and the failure impact range weight coefficient. The initial weights of all nodes are normalized, and the initial weight calculation process of each node is recorded, including the scores of each evaluation indicator, weight coefficients and calculation results, forming a node initial weight calculation report.

6. The device latent fault reasoning method combining multi-source IoT data according to claim 1, characterized in that, The step of dynamically matching the multi-dimensional data evolution imprint set with the nodes of the initial structure of the fault source tracing inference tree, and adjusting the branch growth direction and node weights of the inference tree according to the matching results to generate an adaptive fault source tracing inference tree includes: Establish a matching rule for imprint nodes. The matching rule is determined based on the correlation between imprint features and the physical attributes of the equipment corresponding to the node. If the parameter change reflected by the change pattern imprint is related to the operating parameters of the component corresponding to any secondary branch node, then the change pattern imprint is matched with that secondary branch node. Traverse each imprint in the multi-dimensional data evolution imprint set, find the corresponding matching node in the initial structure of the fault tracing inference tree according to the matching rules, and record the number of imprints matched by each node and the correlation between each imprint and the node. Calculate the matching score for each node, which is obtained by summing the relevance of all imprints matched by the node; The node weight is adjusted based on the node matching score. If the node matching score is higher than the preset score threshold, the node weight is increased, and the increase is positively correlated with the portion of the matching score that exceeds the preset score threshold. If the node matching score is lower than the preset score threshold, the node weight is decreased, and the decrease is positively correlated with the portion of the matching score that is lower than the preset score threshold. For secondary branch nodes where the number of matched imprints exceeds a preset threshold, a new tertiary branch node is added. The tertiary branch node corresponds to a key sub-component of the corresponding secondary branch node component. The new branch node establishes a connection with the original secondary branch node, and the connection relationship indicates the degree of influence of the sub-component on the function of the original component. For nodes that do not match any traces, check if there is any missing data acquisition in the corresponding device component. If there is missing data acquisition, keep the node and mark the missing status; if there is no missing data acquisition, prune the node and its corresponding branches. By integrating the adjusted node weights, newly added branch nodes, and pruning results, an adaptive fault tracing tree is generated. The adaptive fault tracing tree includes the updated node hierarchy, connection relationships, node weights, and imprint matching records.

7. The device latent fault reasoning method combining multi-source IoT data according to claim 6, characterized in that, The rules for establishing imprint node matching include: Collect the standard range of operating parameters for each core system and key component of the equipment. The standard range of operating parameters includes the range of parameter values, the type of change trend, and the range of change amplitude when the component is operating normally. For each imprint type in the multidimensional data evolution imprint set, analyze the correlation between imprint characteristics and standard range of operating parameters. If the trend of change of the change pattern imprint is consistent with the trend of change of the standard range of operating parameters of any component, then it is determined that there is a potential matching relationship between the imprint type and the node corresponding to the component. For mutation node imprints, analyze the correspondence between the parameter type corresponding to the mutation node and the component operating parameters. If the parameter type corresponding to the mutation node belongs to the core operating parameters of any component, it is determined that there is a potential matching relationship between the mutation node imprint and the node corresponding to the component. For the correlation diffusion imprint, analyze the correspondence between the data source types involved in the correlation diffusion and the core system. If the correlation diffusion involves both the running parameter data source and the environmental awareness data source, and both are related to any core system, then it is determined that there is a potential matching relationship between the correlation diffusion imprint and the first-level branch node corresponding to the core system. The correlation calculation method for imprint node matching is defined. The correlation is calculated by multiplying the fit between the imprint features and the standard node parameters and the proximity between the time segment corresponding to the imprint and the current running time of the node. The fit is calculated based on the degree of overlap between the imprint features and the standard features, and the proximity is calculated based on the reciprocal of the time difference. The above potential matching relationships and correlation calculation methods are compiled into an imprint node matching rule document. The imprint node matching rule document also includes the matching priority corresponding to different imprint types. The matching priority of mutation node imprints is higher than that of change pattern imprints, and the matching priority of association diffusion imprints is higher than that of mutation node imprints.

8. The device latent fault reasoning method combining multi-source IoT data according to claim 6, characterized in that, For second-level branch nodes where the number of matched imprints exceeds a preset threshold, a new third-level branch node is added, including: For each second-level branch node, count the number of imprints in the set of multi-dimensional data evolution imprints that it matches. If the number of imprints exceeds the preset threshold, it is determined that the second-level branch node needs to add a third-level branch node. The composition structure of the component corresponding to the secondary branch node is analyzed to identify the key sub-components contained in the component. The key sub-components are those that play a core role in the realization of the component's function or those that have experienced frequent historical failures. Each key sub-component is set as a third-level branch node. The name of each third-level branch node is consistent with the name of the key sub-component. The attribute information of the third-level branch node includes the physical location, functional description and operating parameter type of the sub-component. Establish the connection relationship between the third-level branch node and the corresponding second-level branch node. The attribute of the connection relationship includes the degree of influence of the sub-component on the function of the original component. The degree of influence is determined by analyzing the degree of damage to the function of the original component when the sub-component fails. Assign initial weights to newly added third-level branch nodes. The initial weights are determined based on the functional importance of the sub-components and their correlation with the second-level branch nodes. The newly added third-level branch nodes and their corresponding connections are added to the fault tracing and inference tree, the node hierarchy of the inference tree is updated, and the relevant information of the newly added nodes is recorded. The relevant information includes the reason for the addition, sub-component information and initial weight.

9. The device latent fault reasoning method combining multi-source IoT data according to claim 1, characterized in that, The process of mining the implicit correlation logic between imprints and nodes based on the adaptive fault tracing inference tree to generate implicit fault tracing conclusions for equipment includes: Extract the nodes with the highest weights in the adaptive fault tracing and inference tree, which are then selected as core focus nodes. These core focus nodes reflect the components or systems in the equipment that are most closely associated with the data evolution imprint. For each core focus node, collect all the data evolution imprints it matches, analyze the chronological order of the data evolution imprints, and determine the occurrence sequence of the data evolution imprints. If the mutation node imprint appears first, followed by the associated diffusion imprint, and finally the change pattern imprint, then a mutation association pattern imprint occurrence sequence is formed. Based on the sequence of imprint occurrences and combined with the physical operation logic of the equipment, the state change process of the component or system corresponding to the core focus node is deduced. If the parameter of the component corresponding to the mutation node imprint changes abruptly, the associated diffusion imprint corresponds to the parameter mutation of the component spreading to other components, and the change pattern imprint corresponds to the stable abnormal state formed after diffusion. Then it is deduced that the component first undergoes a parameter mutation, which then triggers the abnormality of the associated components, and finally forms a stable abnormal state. Analyze the imprint association relationship between different core concern nodes. If the association diffusion imprint of the first core concern node and the mutation node imprint of the second core concern node are continuous in time, and the association diffusion direction points to the component corresponding to the second core concern node, then it is determined that there is a state influence relationship between the two core concern nodes. The abnormal state of the first core concern node triggers the abnormal state of the second core concern node. Based on the state change process of the core concern nodes and the state influence relationship between nodes, the fault evolution path is analyzed. The fault evolution path starts from the core concern node with the earliest abnormal imprint and extends to other core concern nodes along the state influence relationship. The component corresponding to the core focus node where the earliest abnormal imprint appears is identified as the potential source component of the fault, and the key imprint features matched by this potential source component of the fault are extracted as fault association imprint features. By integrating potential source components of the fault, the fault evolution path, and the fault association characteristics, a conclusion on the source of hidden equipment faults is formed. The conclusion on the source of hidden equipment faults also includes the abnormal state description and imprint matching basis of each core focus node.

10. A device latent fault reasoning system combining multi-source IoT data, characterized in that, The device includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the device latent fault reasoning method combining multi-source IoT data as described in any one of claims 1-9.

Images