An equipment abnormality identification method and system applied to an industrial intelligent robot

Abstract

This application provides a method and system for equipment anomaly identification in industrial intelligent robots, relating to the field of data processing. An initial normal shell is constructed based on multiple sets of observation records generated under multi-path inspection and multi-state acquisition conditions of the target equipment. Through observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction, shell fragmentation clues are extracted and fragmentation regions are located. Neighborhood and historical relationships are introduced to perform inward shell filling, screening to form unclosed sets. Multi-condition edge-fitting observation and positional stability tracking are conducted on the unclosed regions to obtain anomaly kernel candidates. Shell backfilling is performed using normal shell fragments from multiple sources, eliminating recoverable areas and forming abnormal shell-shedding areas. Outer edge splicing is performed on the shell-shedding boundaries to construct an abnormal growth contour and output the equipment anomaly identification result. Therefore, it is convenient to effectively distinguish the reverse influence of robot execution behavior on observation results in scenarios where industrial intelligent robots participate in equipment inspection.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method and system for identifying equipment anomalies in industrial intelligent robots. Background Technology

[0002] In the process of industrial intelligent robots participating in equipment inspection and status recognition, it is usually necessary to acquire equipment operating status information through methods such as contact detection, close observation, path scanning, and multimodal acquisition, so as to achieve the identification and location of equipment anomalies. However, in practical applications, when industrial intelligent robots perform detection or operation behaviors, their own movements inevitably cause additional disturbances to the inspected equipment. This means that the collected observation results no longer only reflect the status of the equipment itself, but also include the influence introduced by the robot's behavior.

[0003] During contact detection, the end effector comes into contact with the equipment surface, introducing additional vibration response or altering the original vibration propagation path in localized areas, causing shifts or superpositions in the vibration propagation pattern. During close-range observation, the robot's own heat sources or obstructive behavior may change the temperature distribution on the equipment surface, causing localized increases or blockages in the temperature distribution. During path scanning, airflow disturbances, electromagnetic interference, or structural resonance caused by robot movement may also affect the sound source dwell pattern and signal stability. Furthermore, during repeated observations or multiple revisits, due to changes in robot posture, differences in approach paths, and varying dwell times, the same equipment area may exhibit inconsistent observation results under different observation conditions. Because the disturbances introduced by the robot's actions and equipment anomalies exhibit similar behaviors at the observation level—for example, both may lead to increased vibration, abnormal temperature, or signal fluctuations—existing technologies struggle to effectively distinguish between equipment anomalies and robot-induced disturbances when identifying anomalies in observation data. This can easily lead to misjudging disturbances introduced by the robot as equipment anomalies or masking the original abnormal characteristics of the equipment, thereby reducing the accuracy and stability of anomaly identification results.

[0004] Therefore, how to effectively distinguish the feedback effect of robot performance on observation results in the scenario of industrial intelligent robots participating in equipment inspection has become a key technical problem that urgently needs to be solved. Summary of the Invention

[0005] This application provides a method and system for identifying equipment anomalies in industrial intelligent robots, which facilitates the effective differentiation of the impact of robot performance on observation results in scenarios where industrial intelligent robots participate in equipment detection.

[0006] The first aspect of this application provides a method for identifying equipment anomalies in industrial intelligent robots. The method includes: acquiring multiple sets of observation records formed by the target equipment under the conditions of multi-path inspection and multi-state acquisition of the industrial intelligent robot, and performing fusion and inclusion processing on various observation forms in the multiple sets of observation records to construct an initial normal shell; based on the initial normal shell, performing observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction processing on the multiple sets of observation records to form a shell fracture clue set; around the shell fracture clue set, performing neighborhood expansion and historical relationship introduction processing on the fractured regions to implement inward shell filling and form an unclosed set; for the unclosed set, performing multi-condition edge observation and position stability tracking processing on each unclosed region to form an anomaly kernel candidate set; based on the anomaly kernel candidate set, introducing normal shell fragments from multiple sources to perform shell borrowing and backfilling processing on each anomaly kernel candidate to form an anomaly unshelled region; performing outer edge splicing processing on the boundary between the anomaly unshelled region and the normal shell to form an anomaly growth contour, and outputting equipment anomaly identification results based on the anomaly growth contour.

[0007] A second aspect of this application provides a device anomaly identification system for industrial intelligent robots. The system includes an acquisition module and a processing module. The acquisition module acquires multiple sets of observation records generated by the target device under multi-path inspection and multi-state acquisition conditions of the industrial intelligent robot, and performs fusion and inclusion processing on various observation patterns in the multiple sets of observation records to construct an initial normal shell. The processing module performs observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction processing on the multiple sets of observation records based on the initial normal shell to form a shell fracture clue set. The processing module is also used to... The processing module performs neighborhood expansion and historical relationship introduction processing on the fractured region to implement inward shell filling and form an unclosed set. The processing module is further configured to perform multi-condition edge observation and position stability tracking processing on each unclosed region of the unclosed set to form an abnormal kernel candidate set. The processing module is also configured to introduce normal shell fragments from multiple sources to perform shell borrowing and backfilling processing on each abnormal kernel candidate based on the abnormal kernel candidate set to form an abnormal unshelled region. The processing module is also configured to perform outer edge splicing processing on the boundary between the abnormal unshelled region and the normal shell to form an abnormal growth profile, and output the device anomaly identification result based on the abnormal growth profile.

[0008] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, and both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0009] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing instructions that, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By generating multiple sets of observation records under multi-path inspection and multi-state acquisition conditions, and constructing an initial normal shell, the normal operating status of the equipment no longer depends on a single observation, but is based on the fusion of multi-view, multi-condition, and multi-round observations, thereby improving the integrity and anti-interference capability of the normal state representation from the source. This normal shell possesses spatial continuity, temporal consistency, and neighborhood correlation, ensuring that all subsequent anomaly judgments are based on the complete normal structure, thus avoiding misjudgments caused by local observation biases. Through observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction, the existing closed relationship of the initial normal shell is actively broken, so that potential anomalies are no longer masked by the original observation conditions, but are gradually exposed under various reconstruction conditions. This "shell rupture" mechanism, actively introduced by the system, transforms anomaly identification from passively responding to anomaly features to actively mining unstable points in the normal structure, thereby enabling the discovery of hidden anomalies that are difficult to capture by traditional methods.

[0011] By introducing inward shell filling through neighborhood expansion and historical relationships, each fragmented region is given ample opportunity to be "interpreted as normal." Only when multiple normal expressions fail to complete closure is it retained to form an unclosed set. This shell filling failure judgment mechanism has significant exclusivity, effectively filtering out false anomalies caused by environmental fluctuations, observation errors, or robot perturbations, thereby significantly improving the accuracy of anomaly identification. Based on the unclosed set, multi-condition edge-fitting observation and positional stability tracking are introduced, so that anomaly candidates no longer rely on single observation results, but are screened through repeated verification under multiple distances, directions, time sequences, and operating conditions. Anomaly kernels that are continuously attached in spatial location and stably reproduced under different conditions are selected. This positional stability-based screening method naturally suppresses perturbations introduced by robot execution behavior. Through shell-borrowing backfilling of normal shell fragments from multiple sources, each anomaly kernel candidate undergoes repeated verification from homologous, historical, and analogous normal expressions. Only when all normal interpretations fail to cover it is it confirmed as an anomalous unshelled region. This multi-source backfill verification mechanism enables anomaly identification to have strong exclusivity and high confidence, avoiding misjudging phenomena that can be explained as normal changes as anomalies.

[0012] By performing outer edge splicing processing on the boundary between the abnormal unshelled region and the normal shell, an abnormal growth profile is formed, elevating the anomaly result from a single point or local area to a profile expression with boundary structure and evolution trend. This profile not only reflects the current spatial extent of the anomaly but also characterizes the direction of its expansion and development status, thus providing more valuable information for subsequent operation and maintenance decisions. Therefore, it facilitates the effective differentiation of the feedback effect of robot execution behavior on observation results in scenarios where industrial intelligent robots participate in equipment inspection. Attached Figure Description

[0013] Figure 1 A flowchart illustrating a method for identifying equipment anomalies in industrial intelligent robots, provided in an embodiment of this application; Figure 2 A schematic diagram of a module for an equipment anomaly identification system applied to an industrial intelligent robot, provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0014] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0016] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0017] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. In addition, the terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0018] To address the aforementioned technical problems, this application provides a method for identifying equipment anomalies in industrial intelligent robots, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a method for identifying equipment anomalies in industrial intelligent robots, provided in an embodiment of this application. The method is applied to a server and includes steps S1110 to S160, as follows:

[0019] S110. Acquire multiple sets of observation records formed by the target equipment under the conditions of multi-path inspection and multi-state acquisition of industrial intelligent robots, and perform fusion and inclusion processing on various observation forms in the multiple sets of observation records to construct an initial normal shell.

[0020] Specifically, the server is the core processing entity responsible for unified scheduling, data organization, state modeling, and inference decision-making. Essentially, it is a central processing node with computing and control capabilities, enabling it to control the industrial intelligent robot. The server first reads the target device's outline, functional component distribution, obstruction locations, concentrated heat sources, vibration-sensitive locations, and the industrial intelligent robot's reachability. Based on structural continuity and observational reachability, the server divides the target device into multiple target observation areas. A target observation area refers to a spatially concentrated region with local functional consistency, suitable for continuous data collection from a specific angle by the industrial intelligent robot, rather than a simple geometric block. Subsequently, approach paths, frontal view paths, side view paths, detour paths, and retreat paths are configured around each target observation area. Approach paths are used to gradually bring the industrial robot closer to the target observation area from a distance, in order to first establish an overall outline understanding; frontal paths are used to ensure that the industrial robot faces the target observation area directly, guaranteeing complete frontal detail acquisition; side-view paths are used to observe structural boundary undulations and local occlusion relationships from the side; detour paths are used to form a continuous transitional observation around the target observation area and its adjacent areas; and retreat paths are used to withdraw after acquisition and perform reverse verification. Through the above processing, each target observation area can have a fixed path organization relationship, thus enabling subsequent observation results to be clearly mapped to a specific target observation area and a specific path source.

[0021] After the inspection and passage framework is established, the server issues inspection control commands to the industrial intelligent robot, controlling the robot to sequentially pass through the corresponding target observation area along each inspection and passage framework, maintaining different acquisition states at different path positions to form multi-state acquisition conditions. Multi-state acquisition conditions mean that the industrial intelligent robot does not adopt a single acquisition posture in different path segments, but switches the observation state according to the path position, so that the same target observation area forms multiple rounds of observation under different spatial relationships. The three modes of observation are: **Distant Initial View:** The industrial intelligent robot maintains a relatively long observation distance at the beginning of its approach path to acquire the overall outline, thermal distribution, and boundary relationships of the target equipment. **Frontal Proximity:** The industrial intelligent robot faces the target observation area at a relatively close distance along its frontal view path to enhance the accuracy of acquiring surface details, local thermal anomalies, and local acoustic changes. **Lateral Offset:** The industrial intelligent robot maintains a certain lateral angle along its side-view path to highlight boundary transitions, obscuring trailing edges, and local structural thickness changes. **Continuous Turning:** The industrial intelligent robot adjusts its posture while moving along its detour path to continuously record transition information between the target observation area and adjacent areas. **Retreat and Re-view:** The industrial intelligent robot retreats and observes in the opposite direction while leaving its path, allowing previously acquired areas to be reconfirmed in the opposite direction. Through this processing method, the same target observation area is no longer acquired statically only once, but rather forms a multi-path, multi-state superimposed observation basis, which is beneficial for subsequent identification of stable and sporadic expressions.

[0022] After the industrial intelligent robot completes data collection along each path, the server groups and writes the observations formed at each path location and under each collection state according to the target observation area, path location, and collection state, forming multiple sets of observation records. These multiple sets of observation records are not simply stored separately according to sensor type; instead, a complete collection process is bound into a traceable record unit. Each set of observation records includes a target observation area identifier, path location identifier, collection state identifier, collection time identifier, and the corresponding original observation content. This allows the server to clearly know which target observation area, which path, and which collection state a particular observation result was formed in. The advantage of this approach is that even if inconsistent observation results occur for the same target observation area under different paths, the specific formation conditions can be traced back to determine whether the difference stems from path differences, state differences, or regional differences. This grouping and writing essentially involves conditionally binding and structurally archiving the collection results, ensuring that subsequent fusion and inclusion are based on traceable conditions, rather than directly mixing raw data without source identifiers.

[0023] After multiple sets of observation records are generated, the server extracts temperature distribution patterns, vibration propagation patterns, sound source persistence patterns, and structural boundary patterns from these records. It then performs region-based merging processing on various observation patterns belonging to the same target observation area, resulting in region-based merging results. Temperature distribution patterns refer to the spatial pattern of heat distribution within the target observation area, the locations of local temperature rise accumulation, and the changing state of heat diffusion along the boundaries. Vibration propagation patterns refer to the transmission direction, attenuation trend, and local response concentration relationships of the vibration response in the target observation area and its adjacent locations. Sound source persistence patterns refer to the main attachment locations of acoustic energy in the target observation area, the areas of continuous existence, and the expansion relationships with adjacent areas. Structural boundary patterns refer to the outer contour, connecting boundaries, turning edges, and geometric closure relationships corresponding to the target observation area. Region-based merging processing means that the server does not immediately merge all observation patterns directly, but first aligns similar observation patterns formed in the same target observation area under different paths and acquisition states, allowing multiple rounds of observation results for the same target observation area to be mutually reflected under a unified regional coordinate relationship. By merging observations within the same region, the server can identify which observation patterns appear consistently and stably across multiple paths and various acquisition states, and which observation patterns only appear briefly on a single path or under a single state, thus providing a selection basis for subsequent fusion and inclusion.

[0024] After obtaining the results of the same-area merging process, the server, based on these results, filters out various observation patterns that remain stable under multi-path inspection and multi-state acquisition conditions and performs fusion and inclusion processing to form a local normal shell for each target observation area. Fusion and inclusion processing does not simply superimpose all observation patterns, but rather organizes them around normal interpretation closure relationships. A normal interpretation closure relationship means that multiple observation patterns can support each other, do not conflict with each other, and can collectively demonstrate that the target observation area is in an acceptable normal operating state. For example, if the temperature distribution pattern in a target observation area is smooth and continuous under the frontal, lateral, and retreat paths, and the corresponding structural boundary patterns do not show obvious breaks, while the vibration propagation pattern and sound source residence pattern maintain consistent transmission and attachment relationships with adjacent areas, then the server includes this group of observation patterns as a normal representation of the target observation area. A local normal shell refers to a local normal interpretation shell formed around a single target observation area, indicating that the area can be stably interpreted as normal under multi-path and multi-state observation conditions. If a certain observation pattern appears only under a single path, or if there is an interpretative conflict with other observation patterns, the server will not include it in the local normal shell, but will instead reserve it as a marginal representation for subsequent analysis. After this processing, each target observation region will obtain a local normal shell supported by stable observation patterns.

[0025] After the formation of local normal shells in each target observation area, the server performs neighbor-area splicing processing on each local normal shell to obtain the splicing result, and constructs an initial normal shell covering the entire range of the target device based on the splicing result. Neighbor-area splicing processing means that instead of treating each target observation area as an isolated local unit, it performs boundary docking and continuity verification on the local normal shells of adjacent target observation areas based on the structural connection relationship, energy transfer relationship, and spatial adjacency relationship of the target device. Specifically, the server checks whether the structural boundary morphology between adjacent target observation areas can be continuously closed, whether the temperature distribution pattern can form a reasonable transition, whether the vibration propagation pattern can be normally transmitted along the device structure, and whether the sound source dwelling pattern maintains a reasonable attachment relationship between adjacent areas. If adjacent target observation areas can be continuously connected in the above aspects, the server splices the corresponding local normal shell into a larger-scale normal expression; if there are slight differences at the boundary junction but the overall closure is not broken, the junction position is retained as a boundary retention area. The initial normal shell refers to the overall normal interpretation shell formed by stitching together the local normal shells of the entire target observation area. It covers the entire range of the target device while retaining local boundary difference information. The initial normal shell constructed in this way is not a static snapshot of a single observation, but rather an overall normal state reference gradually formed by the server based on multi-path inspection, multi-state acquisition, multiple sets of observation records, same-area merging processing, and neighboring-area stitching processing. It provides a direct basis for subsequent observation path rearrangement, observation order rearrangement, and neighboring reference reconstruction processing.

[0026] S120. Based on the initial normal shell, perform observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction on multiple sets of observation records to form a set of shell rift clues.

[0027] Specifically, the shell fragmentation clue set refers to the set of local instability clues that still cannot maintain normal closure after actively altering the original path relationships, original observation chain relationships, and original neighborhood reference relationships. When mapping the shell composition information in the initial normal shell to multiple sets of observation records to establish a shell disassembly view corresponding to each target observation area, the server first back-maps the local normal shell of each target observation area in the initial normal shell, the corresponding observation morphology combination, path source, and neighboring region splicing relationship to the multiple sets of observation records that form that shell. The shell composition information includes not only the result that a target observation area is judged to be normal, but also which temperature distribution patterns, vibration propagation patterns, sound source residence patterns, and structural boundary patterns jointly support this result. To quantify the degree of support of a certain observation record for the local normal shell, the support weight of the observation record can be calculated first:

[0028] in, Indicates the first The overall support weight of the group of observation records for the local normal shell of the current target observation area; the larger the value, the stronger the support effect. This indicates the degree to which the temperature distribution pattern in the observation record supports the local normal shell. It is determined based on the similarity between the temperature distribution pattern and the temperature distribution template corresponding to the initial normal shell, and the value can be set to 0 to 1. This indicates the degree of support for the vibration propagation pattern, and its value can be set from 0 to 1. This indicates the degree of support provided by the sound source's dwelling form, and its value can be set from 0 to 1. This indicates the degree of structural boundary support, and its value can be set from 0 to 1. , , , These represent the weighting coefficients of temperature distribution pattern, vibration propagation pattern, sound source residence pattern, and structural boundary pattern in the current target observation area. Each weighting coefficient can be set from 0 to 1, and the sum of the four must be 1. Through these weightings, the server can identify which observation records play a core role in the original normal closure and which play only an auxiliary role. Subsequently, the server organizes the path sources, sequence positions, and neighboring region sources of all high-weighted observation records within the same target observation area to form a shell disassembly view. The shell disassembly view is a normal closure dependency graph oriented towards the target observation area, used to show which type of path, sequence, and neighboring reference are required for a local normal shell to be valid.

[0029] Based on the shell disassembly view, observation path rearrangement is performed on multiple sets of observation records to create cross-referencing relationships between approach paths, frontal paths, side-view paths, bypass paths, and retreat paths used to support the local normal shell. When generating the path rearrangement observation results, the server does not change the original acquisition facts, but rather alters the path referencing relationships when interpreting these observation records. Observation path rearrangement refers to the reconfiguration of path sources that originally only served the normal closure of a specific target observation area. For example, frontal path observations are cross-used with side-view path observations from adjacent target observation areas, retreat path observations are moved to the initial interpretation position, and approach path observations are changed from an overall contour reference to a local interpretation basis. To quantify the degree of preservation of the local normal shell of a target observation area after path rearrangement, a path closure preservation degree can be constructed.

[0030] in, This indicates the path closure preservation degree after path rearrangement. The value can be set from 0 to 1. The larger the value, the closer the path is to the original normal closure after rearrangement. This represents the total number of observation records involved in the interpretation of the current target observation area; Indicates the first Supporting weights of group observation records; Indicates the first The path matching factor, which determines whether the group of observation records still meet the requirements of a local normal shell under path rearrangement conditions, is calculated comprehensively based on the degree of closure consistency of temperature distribution, vibration propagation, sound source residence, and structural boundary morphology after path rearrangement. Its value can be set from 0 to 1. If... The significant decrease indicates that the current local normal shell is highly dependent on the original path relationships; that is, once paths intersect, the original normal interpretation becomes difficult to maintain. Based on this, the server generates path rearrangement observations and provides quantitative evidence for subsequent extraction of path fragments. Path fragments refer to locally unstable segments corresponding to locations where path closure preservation significantly decreases.

[0031] Based on the path rearrangement observation results, the observation order is rearranged for multiple sets of observation records to form an observation chain from the acquisition order of temperature distribution, vibration propagation, sound source dwelling, and structural boundary morphology within each target observation area. When generating the rearranged observation results, the server continues to break the normal mutual verification relationship established by the original time sequence. An observation chain refers to the interpretation sequence formed by the server connecting temperature distribution, vibration propagation, sound source dwelling, and structural boundary morphology according to the new invocation order when interpreting the target observation area. In the original processing, it is common to first use the structural boundary morphology to determine the boundary, then use the temperature distribution morphology to verify the thermal state, and then use the vibration propagation and sound source dwelling morphology to supplement the dynamic features. After the observation order is rearranged, this order can be rewritten as first sound source dwelling morphology, then vibration propagation morphology, then temperature distribution and structural boundary morphology, or it can form an interleaved observation chain across target observation areas. To quantify the impact of the order change on normal closure, the order dependency can be calculated.

[0032] in, This represents the order dependency, and its value can be set from 0 to 1. The larger the value, the more the current local normal shell depends on the original observation order. Indicates the number of observation morphology nodes in the observation chain; Indicates the first The strength of the closed support formed by the observation morphology node and subsequent nodes under the condition of sequential rearrangement is determined by comprehensively considering the consistency, complementarity and continuity between the rearranged nodes, and the value range can be set to 0 to 1. Indicates the first The closed support strength of each observation morphological node under the original sequence conditions can be set to a value range of 0 to 1. Indicates the first The importance coefficient of each observation morphology node in the observation chain can be set to a value between 0 and 1. If A high value indicates that the normal interpretation of the target observation area is highly dependent on the original order. Once the order is changed, it is difficult to maintain normal closure. The server generates rearranged observation results accordingly. Sequence fragmentation refers to the location of instability in the observation chain when the sequence dependency increases and local closure fails.

[0033] Based on the sequential rearrangement of observation results, neighborhood reference reconstruction processing is performed on multiple sets of observation records to generate reference reconstruction observation results. At the spatial reference relationship level, the server further breaks the stability of the initial normal shell. Neighborhood reference refers to the adjacent target observation area or related functional region that provides structural boundary support, thermal transition support, vibration transmission support, and sound source attachment support for the current target observation area within the original normal closure. Neighborhood reference reconstruction refers to replacing, expanding, or removing the original neighborhood source, so that a target observation area no longer relies solely on the original neighboring area for normal interpretation, but instead re-establishes support relationships under new neighborhood conditions. To measure whether the reconstructed neighborhood reference can still maintain normal closure, the neighborhood reference consistency can be calculated.

[0034] in, This represents the neighborhood reference consistency degree, and the value can be set from 0 to 1. The larger the value, the more consistent it is with the original normal support relationship after the neighborhood reference is reconstructed. This indicates the degree of boundary consistency of the structural boundary morphology under the new neighborhood reference condition. It is determined according to the degree of boundary continuity and can be set to 0 to 1. This indicates the degree of consistency in the thermal transition of the temperature distribution pattern under the new neighborhood reference conditions, and its value can be set from 0 to 1. This indicates the degree of consistency in the transmission of vibration propagation patterns under new neighborhood reference conditions, and its value can be set from 0 to 1. This indicates the degree of consistency of the sound source dwell pattern under the new neighborhood reference conditions, and the value can be set from 0 to 1; , , , These represent the weight coefficients of the four observation types in the neighborhood reference reconstruction determination, with each value ranging from 0 to 1, and their sum being 1. If The significant decrease indicates that the current local normal shell is highly dependent on the original neighborhood support relationship, and once the neighborhood source is changed, it can no longer maintain closure. The server generates reference reconstruction observation results based on this. A reference fragment is a fragment corresponding to the location where the neighborhood reference consistency decreases, leading to local support failure.

[0035] Closure comparison processing is performed on path rearrangement observation results, sequence rearrangement observation results, and reference reconstruction observation results. When extracting path fragmentation fragments, sequence fragmentation fragments, and reference fragmentation fragments, the server compares each of the three types of results with the original closure state of the initial normal shell. Closure comparison processing refers to checking whether the local normal shell can still maintain boundary closure, transmission continuity, and attachment stability after changes in path relationship, sequence relationship, and neighborhood reference in the same target observation area. To form a unified criterion for fragmentation judgment, a comprehensive fragmentation intensity can be further defined:

[0036] in, This indicates the overall pyrolysis intensity; a larger value indicates that the local normal closure is less stable. Indicates the path closure preservation degree; Indicates the degree of order dependency; Indicates the degree of neighborhood reference consistency; , , These represent the weight coefficients of path factors, order factors, and neighborhood reference factors in the comprehensive fragmentation determination, respectively. Each value can be set to a range of 0 to 1, and the sum of the three must be 1. If a certain local region... If the threshold for fracturing is exceeded, it indicates that significant instability has occurred in the region after the active dismantling of the normal shell. If the instability is mainly due to... If the value decreases, it is extracted as a path fragmentation segment; if the instability mainly stems from... If the elevation is high, it is extracted as a sequential fragment; if the instability is mainly due to... If the value is reduced, it is extracted as a reference fragment. Path fragments represent normal closed vulnerable positions exposed after path intersection calls, sequence fragments represent sequential dependency positions exposed after changes in the observation chain, and reference fragments represent neighborhood dependency positions exposed after changes in neighborhood support relationships.

[0037] When performing merging and aggregation on path fragments, sequential fragments, and reference fragments to form a shell fragmentation clue set, the server does not treat these fragments as isolated anomalous fragments. Instead, it merges them according to their spatial attachment location, structural boundary location, transmission breakpoint location, and observational morphological breakpoint location. Merging and aggregation refers to associating multiple fragments pointing to the same target observation area, the same structural boundary, the same thermal anomaly location, the same vibration discontinuity location, or the same sound source offset center, transforming them into a comprehensive fragmentation clue that collectively demonstrates the vulnerability of locally normal closure. To quantify the strength of the aggregated clues, a clue aggregation value can be constructed:

[0038] in, Indicates the first The aggregation value of a clustered fragmentation clue indicates that the larger the value, the stronger the evidence of fragmentation for that clue. Indicates being merged into the first The number of fragments in a fragmentation thread; Indicates the first The overall fragmentation intensity corresponding to each fragmentation segment; Indicates the first The contribution coefficient of each fragment in the aggregation is determined based on its recurrence frequency, the importance of its target observation area, and the degree of overlap with the boundary preservation area, and can be set to a range of 0 to 1. The server filters high-intensity fragmentation clues and fragmentation clues to be expanded based on the clue aggregation value, and organizes them into a shell fragmentation clue set. The shell fragmentation clue set is a structured set of clues with source type, spatial location, attachment location, and fragmentation intensity. It is no longer just a collection of scattered instability phenomena, but rather an intermediate result that provides direct input for the next steps of fragmentation region localization, neighborhood expansion and introduction of historical relationships, inward shell filling, and the formation of unclosed sets.

[0039] S130. Around the set of shell fracture clues, perform neighborhood expansion and historical relationship introduction processing on the fracture region to implement inward shell filling and form an unclosed set.

[0040] Specifically, an unclosed set refers to a set of regions where a continuous interpretation link cannot be recovered even after neighborhood compensation and historical compensation. In the process of performing fracture region localization processing on each fracture clue in the shell fracture clue set, and merging fracture clues pointing to the same target observation area, the same structural boundary, or the same observation morphology breakpoint to form a fracture region set, the server first reads the target observation area identifier, fracture source type, attachment location, and corresponding observation morphology type corresponding to each fracture clue in the shell fracture clue set, and then reprojects these discrete clues into the overall spatial representation of the target device. Fracture region localization processing is not simply delineating a geometric range, but rather aligning and aggregating multiple locations that essentially point to the same interpretative instability. For example, if multiple path fracture fragments, sequential fracture fragments, and reference fracture fragments all fall near the same structural boundary of the same target observation area, or all correspond to the same thermal breakpoint in the temperature distribution morphology, the same transmission discontinuity location in the vibration propagation morphology, or the same attachment offset center in the sound source dwelling morphology, then the server merges these fracture clues into the same fracture region. To quantify whether a location has the conditions to form a fracture region, the fracture aggregation intensity can be defined:

[0041] in, Indicates the first The higher the value of the cleavage polymerization intensity of a candidate cleavage region, the greater the degree to which the region is supported by multiple cleavage trajectories; Indicates being merged into the first The number of rift clues in each candidate rift region; Indicates the first The contribution coefficient of the fracture line in the current candidate region is determined based on the degree of spatial overlap, structural boundary overlap and consistency of the observed morphological breakpoint between the fracture line and the current region. The value range can be set to 0 to 1. Indicates the first The rift intensity corresponding to each rift clue directly inherits the comprehensive rift intensity or aggregated rift clue value from the previous step, and is greater than 0. The server uses candidate locations where the rift aggregation intensity exceeds a preset threshold as formal rift regions and organizes all formal rift regions into a rift region set. The rift region set is the direct processing object for subsequent neighborhood expansion and historical relationship introduction. Compared with discrete rift clues, it already has clear attributes such as target observation area affiliation, structural attachment location, and observation morphology breakpoint.

[0042] In the process of constructing a neighborhood expansion framework around each fractured region in the set of fractured regions, and determining directly adjacent and second-adjacent target observation areas based on the neighboring region splicing relationships in the initial normal shell to form a multi-layered neighborhood structure, the server expands the normal expression support range layer by layer outward from each fractured region as the center. The neighborhood expansion framework refers to the multi-layered interpretation compensation structure established around the fractured regions, used to place the observation instability problem, which was originally limited to the fractured region, back into a larger normal closed environment for judgment. The server first reads the neighboring region splicing relationships in the initial normal shell. The neighboring region splicing relationship refers to the local normal shell connection relationship established between different target observation areas during the construction of the initial normal shell through the continuity of structural boundary morphology, the transition of temperature distribution morphology, the transmission of vibration propagation morphology, and the attachment of sound source residence morphology. Subsequently, starting from the target observation area where the fractured region is located, the target observation areas directly connected to it are extracted to form directly adjacent target observation areas; then, starting from these directly adjacent target observation areas, the target observation areas further connected to them are extracted to form second-adjacent target observation areas. Therefore, the server constructs a multi-layered neighborhood structure around each fractured region, including the current target observation area, the directly adjacent target observation area, and the next adjacent target observation area. This multi-layered neighborhood structure is not a simple spatial expansion, but a hierarchical interpretation network that preserves the closed support relationships in the initial normal shell, providing a clear source range for subsequently introducing temperature distribution patterns, vibration propagation patterns, sound source residence patterns, and structural boundary patterns from the neighborhood.

[0043] Within the neighborhood expansion framework, the temperature distribution, vibration propagation, sound source retention, and structural boundary morphology of the neighboring target observation area are introduced into the interpretation system of the fractured region to form the neighborhood introduction results. During this process, the server begins to attempt to repair the fractured region using normal expressions from the neighborhood. Neighborhood introduction is not simply copying neighboring data; rather, it involves introducing observational patterns that have structural continuity, transmission continuity, and attachment continuity with the fractured region according to their corresponding relationships, making them part of the fractured region interpretation system. If the fracture region is mainly characterized by structural boundary fracture, then structural boundary morphology is preferentially introduced from the directly adjacent and second-adjacent target observation areas to restore boundary continuity. If the fracture region is mainly characterized by abnormal temperature distribution, then temperature distribution morphology from the neighborhood is preferentially introduced, and the thermal state is checked in conjunction with the structural boundary morphology to see if a reasonable transition can be formed. If the fracture region is mainly characterized by discontinuous vibration propagation, then vibration propagation morphology from the neighborhood is preferentially introduced, and the vibration transmission path is checked to see if it can extend from the neighborhood into the fracture region. If the fracture region is mainly characterized by sound source attachment offset, then sound source residence morphology from the neighborhood is preferentially introduced, and the current fracture location can be reinterpreted based on the acoustic attachment relationship of the neighborhood. To quantify the supporting effect of the neighborhood introduction on the fracture region, a neighborhood compensation degree can be defined:

[0044] in, This represents the neighborhood compensation degree of the current fracture region. The larger the value, the stronger the support of the neighborhood introduction result for the normal interpretation of the fracture region reconstruction. This indicates the degree of consistency between the temperature distribution pattern in the neighborhood and the temperature distribution pattern in the rift region, and its value can be set from 0 to 1. This indicates the degree of continuity in the transmission between the vibration propagation pattern in the neighborhood and the vibration propagation pattern in the rupture region, and its value can be set from 0 to 1. This indicates the degree of attachment and correspondence between the neighboring sound source residence pattern and the sound source residence pattern in the pyrolysis region, and the value can be set from 0 to 1. This indicates the degree of boundary connection between the boundary morphology of the neighborhood structure and the boundary morphology of the fractured region structure, and its value can be set from 0 to 1. , , , These represent the weight coefficients of the four observation types in neighborhood compensation, with each value ranging from 0 to 1, and their sum being 1. The server then generates neighborhood introduction results based on these coefficients, which characterize whether the fractured region has a tendency to reclose under the support of its spatial neighborhood.

[0045] Based on the historical observation records of the target equipment, the server selects historical observation segments that match the fractured region in terms of target observation area location, structural boundary relationship, and operating conditions. It then extracts the corresponding temperature distribution pattern, vibration propagation pattern, sound source retention pattern, and structural boundary pattern as historical shell fragments to introduce into the fractured region, forming the historical introduction result. In this process, the server further introduces normal expressions in the time dimension into the fractured region. Historical observation records refer to multiple sets of traceable observation records formed during previous inspections of the target equipment. Historical observation segments refer to local observation segments selected from historical observation records that are highly similar to the current fractured region in terms of target observation area location, structural boundary relationship, and operating conditions. Historical shell fragments are local units of normal historical expression extracted and organized from these historical observation segments. During the selection process, the server prioritizes observation segments that correspond to the same target observation area as the current fractured region, are under the same or similar operating conditions, and have historically been determined to be normal. It then extracts the temperature distribution pattern, vibration propagation pattern, sound source retention pattern, and structural boundary pattern from these segments and introduces them into the current fractured region. To quantify the matching degree of historical introduction, a historical matching degree can be defined:

[0046] in, This indicates the historical matching degree between the current historical observation fragment and the fractured region. The larger the value, the more suitable the historical observation fragment is to be introduced as a historical shell fragment. This indicates the degree of matching of the target observation area location. It is determined based on the consistency between the historical observation fragments and the current fragmented region in terms of the target observation area coordinates and location markers. The value range can be set from 0 to 1. This indicates the degree of matching between structural boundary relationships. It is determined based on the consistency between historical observation segments and the current fractured region in terms of structural boundary attachment positions and boundary orientations. The value range can be set from 0 to 1. This indicates the degree of matching of operating conditions. It is determined based on the similarity of state variables such as operating load, operating stage, or environmental conditions, and the value range can be set from 0 to 1. , , These represent the weight coefficients of the three matching factors, with each value ranging from 0 to 1, and their sum equal to 1. The server transforms historical observation fragments with high historical matching degrees into historical shell fragments and organizes them into historical introduction results. Historical introduction results are used to characterize the probability of normal recovery of the fractured region in the time dimension.

[0047] In the process of performing inward shrinking and shelling on the neighboring and historical introduced results to aggregate the various introduced observation forms into the fractured region and reconstruct the interpretative closure relationship, the server truly begins to perform shelling. Inward shrinking and shelling means that instead of allowing the fractured region to directly expand outward to adapt to the neighboring and historical regions, it gradually compresses, gathers, and covers the normal expressions in the neighboring and historical regions into the fractured region, so that the fractured region regains normal interpretative closure. The server first uses the structural boundary morphology as boundary constraint to push the structural boundary morphology in the neighboring and historical introduced results into the fractured region, prioritizing the repair of boundary discontinuities; then, it uses the temperature distribution morphology and vibration propagation morphology as transmission constraints to extend the thermal transition relationship and vibration transmission relationship towards the center of the fractured region; at the same time, it uses the sound source residence morphology as attachment constraint to cover the normal acoustic attachment position into the fractured region. To comprehensively evaluate the degree of closure recovery after shelling, the inward shrinking closure degree can be defined:

[0048] in, This indicates the degree of closure recovery of the current fractured region after the inward shrinkage and shell repair. The larger the value, the closer the fractured region is to regain a continuous interpretation link. Indicates the neighborhood compensation degree; Indicates historical matching degree; This indicates the degree to which the observed morphologies within the fractured region re-establish mutual support relationships after the internal shell is repaired. It is determined based on the overall consistency of temperature distribution morphology, vibration propagation morphology, sound source residence morphology, and structural boundary morphology after the shell repair, and can be set to a value range of 0 to 1. , , These represent the weighting coefficients of neighborhood introduction, historical introduction, and internal reconstruction consistency in the closure recovery determination, respectively. Each value can be set to a range of 0 to 1, and their sum must be 1. Through this inward closure degree, the server can determine whether the fragmented region has recovered to an acceptable normal closure state under the combined effect of multi-source normal expression.

[0049] During the shell-filling process, closure determination is performed on each fractured region. Fractured regions that restore continuous interpretation links are marked as closed regions, while those that cannot restore continuous interpretation links are marked as shell-filling failure regions. The server makes a formal determination based on the degree of closure during the shell-filling process and the restoration of interpretation links. A continuous interpretation link means that after shell-filling, the temperature distribution pattern of the fractured region can maintain a transitional consistency with the normal thermal state in the neighborhood and history, the vibration propagation pattern can restore the normal transmission path, the sound source dwell pattern can reattach to a reasonable position, and the structural boundary pattern can reclose and continuously connect with adjacent target observation areas. If a fractured region reaches a preset closure threshold after shell-filling and a stable mutual verification relationship is re-established between its main observation patterns, the server marks it as a closed region. A closed region indicates that after neighborhood expansion and the introduction of historical relationships, the fractured region can be reintegrated into the interpretation system of the normal shell. If a fractured region still has boundary breaks, thermal discontinuities, vibration discontinuities, or abnormal acoustic attachments after shell-filling, or if a certain dimension is restored but the overall interpretation link is still interrupted, the server marks it as a shell-filling failure region. A region where the patching fails does not necessarily mean it is immediately identified as abnormal. Rather, it means that the region cannot be closed even after receiving adequate and normal compensation, and it is a key target for subsequent screening of abnormal kernel candidates.

[0050] In the process of summarizing the failed shell-patch regions and classifying them according to the location of the target observation area and the type of observation morphology to form an unclosed set, the server organizes all failed shell-patch regions into an unclosed set required for the next step of processing according to a unified rule. Classification and organization is not simply putting the failed regions together, but rather organizing them hierarchically according to their target observation area location, main unstable observation morphology type, main boundary attachment location, and source of shell-patch failure. If some failed shell-patch regions are mainly located near the same structural boundary in the same target observation area, they are grouped into the same location cluster; if some failed shell-patch regions mainly exhibit temperature distribution morphology mismatch, they are grouped into thermal unclosed regions; if they mainly exhibit vibration propagation morphology discontinuity, they are grouped into transmission unclosed regions; if they mainly exhibit sound source dwelling morphology shift, they are grouped into attachment unclosed regions; and if they mainly exhibit structural boundary morphology that cannot be patched, they are grouped into boundary unclosed regions.

[0051] S140. For unclosed sets, perform multi-condition edge observation and position stability tracking on each unclosed region to form an abnormal kernel candidate set.

[0052] Specifically, the candidate set of anomaly kernels refers to the set of candidate regions that, after multi-condition edge observation and position stability tracking, are confirmed to be persistently attached in spatial location, stably reproduced in multiple rounds of observation, and possess boundary convergence characteristics. The entire processing logic is not a single re-judgment, but rather a process of repeated edge verification around the boundary, followed by stability confirmation around the position, thereby further distinguishing transient disturbances from true anomaly kernels. In the process of performing boundary unfolding processing on each unclosed region in the unclosed set to determine the unclosed boundary, the inner boundary region, and the outer boundary region, and to form the boundary structure, the server first performs boundary deconstruction on each unclosed region, no longer treating the region merely as a whole unclosed patch, but explicitly unfolding its boundary relationship with the surrounding normal shell. Boundary unfolding processing refers to jointly depicting the locations of structural boundary morphological changes, abrupt changes in temperature distribution morphology, discontinuities in vibration propagation morphology, and shifts in sound source residence morphology along the transition interface between the unclosed region and the surrounding closed region, forming a continuous unclosed boundary. An unclosed boundary is the boundary between an unclosed region and a region of normal interpretation. It includes not only the geometric edge but also the interpretative switching interface where the observed morphology shifts from normal to unclosed. The inner boundary region refers to the area on the side of the unclosed boundary facing inwards from the unclosed region, primarily used to observe whether anomalous signs are continuously pressing towards the boundary from the inside. The outer boundary region refers to the area on the side of the unclosed boundary facing the normal shell, primarily used to observe whether the normal interpretation is still attempting to fill in the unclosed region. To quantify the clarity of the boundary after it unfolds, we can define boundary visibility:

[0053] in, This indicates the visibility of the boundary of the currently unclosed region. The larger the value, the clearer the unclosed boundary and the more suitable it is for subsequent edge observation. This indicates the clarity of the temperature distribution pattern along the boundary direction. It is determined based on the continuity and abrupt stability of the temperature distribution differences on both sides of the boundary, and its value range can be set from 0 to 1. This indicates the degree of discontinuity and clarity of the vibration propagation pattern along the boundary direction, and its value can be set from 0 to 1. This indicates the clarity of the sound source dwell pattern offset along the boundary direction, and its value can be set from 0 to 1. This indicates the clarity of geometric breakpoints along the boundary direction of the structural boundary morphology, and its value can be set from 0 to 1. , , , These represent the weighting coefficients of the four observation types in the boundary visibility calculation. Each value can be set from 0 to 1, and their sum must be 1. The server uses these coefficients to form the boundary structure. The boundary structure refers to the boundary representation unit composed of the unclosed boundary, the inner boundary region, and the outer boundary region, providing a direct spatial basis for the subsequent construction of the edge-fitting observation channel.

[0054] In the process of constructing edge-fitting observation channels around the boundary structure, including inner edge-fitting channels, boundary-fitting channels, and outer edge-fitting channels, the server no longer performs wide-range scanning along the conventional inspection path. Instead, it establishes dedicated observation channels closely adjacent to the boundary structure, ensuring that the industrial intelligent robot always operates around the unclosed boundary during subsequent observations. Edge-fitting observation channels refer to a set of continuous observation trajectories set up at different lateral positions around the unclosed boundary. The inner edge-fitting channel is deployed in the inner region of the boundary to observe the temperature distribution, vibration propagation, sound source retention, and whether the structural boundary continues to adhere to the boundary within the unclosed area. The boundary-fitting channel is deployed directly along the unclosed boundary to observe the discontinuous state, adhesion state, and local abrupt changes of the boundary itself. The outer edge-fitting channel is deployed in the outer region of the boundary to observe the compressive closing trend of the normal shell on the unclosed boundary and whether the normal shell can still advance towards the boundary. To ensure that the three types of edge-fitting observation channels can spatially cover both sides of the boundary without functional overlap, the server can set inner offset distance, fitting offset distance, and outer offset distance according to the normal direction of the unclosed boundary, and ensure that the edge-fitting observation channels are continuously deployed along the boundary direction. With this processing, the industrial intelligent robot can simultaneously collect data at three levels: the inner and outer sides of the boundary and the boundary body itself, thus avoiding misjudgments of boundary stability based on observations from only one side.

[0055] In the process of controlling an industrial intelligent robot to perform multi-condition edge-fitting observation along an edge-fitting observation channel to generate multi-round edge-fitting observation results, the server issues multiple sets of edge-fitting control commands to the industrial intelligent robot. This causes the robot to repeatedly perform edge-fitting data acquisition around the same unclosed boundary under different observation distances, directions, movement rhythms, dwell times, and working condition slices. Multi-condition edge-fitting observation refers to observation conditions that are not fixed but rather involve various combinations of observation conditions around the same boundary structure to verify whether the unclosed boundary maintains the same attachment behavior under different conditions. Different observation distances are used to distinguish between local boundary details under close-range conditions and overall boundary extension under medium-to-long-range conditions; different observation directions are used to verify whether the boundary attachment position depends on a single viewpoint; different movement rhythms are used to observe whether the boundary can be continuously identified under both slow passage and rapid sweep conditions; different dwell times are used to distinguish between momentarily visible and continuously visible boundary signs; and different working condition slices are used to determine whether the boundary only appears during a specific equipment operation phase. The server records the channel type, observation conditions, acquisition time, and operating status corresponding to each round of edge observation, forming multi-round edge observation results. These multi-round edge observation results are not simply repeated data, but rather a set of conditional observations around the same unclosed boundary under various conditions, providing a spatiotemporal basis for subsequent boundary accompaniment extraction.

[0056] In the process of performing boundary accompaniment extraction processing on multiple rounds of edge-fitting observation results to obtain temperature distribution patterns, vibration propagation patterns, sound source retention patterns, and structural boundary patterns that appear synchronously with and extend along the unclosed boundary, and to form a boundary accompaniment record, the server extracts observation patterns from the multiple rounds of edge-fitting observation results that are not randomly scattered within the unclosed region, but rather always appear synchronously with the unclosed boundary and continuously extend along the boundary direction. Boundary accompaniment extraction processing means that it no longer cares about all observation changes, but only selects observation patterns that have a synchronous spatiotemporal relationship with the unclosed boundary. If a certain temperature distribution pattern only appears isolated within the unclosed region and cannot be continuously distributed along the unclosed boundary, it is not considered a boundary accompaniment expression; if a certain vibration propagation pattern only intensifies briefly at individual points but cannot form a stable discontinuous zone along the boundary direction, it is also not considered a boundary accompaniment expression. Conversely, if the temperature distribution pattern, vibration propagation pattern, sound source retention pattern, or structural boundary pattern can appear synchronously with the unclosed boundary and maintain continuous extension along the boundary direction in multiple rounds of edge-fitting observation, the server includes it in the boundary accompaniment record. To quantify the degree of association between a given observation pattern and an unclosed boundary, we can define the boundary association degree:

[0057] in, This indicates the boundary association degree of the current observation pattern with respect to the unclosed boundary. The larger the value, the more likely the observation pattern is to be a boundary-attached expression. This indicates the degree of synchronous occurrence, determined based on the spatiotemporal synchronization relationship between the observed morphology's location and the location of the unclosed boundary. The value can be set from 0 to 1. It indicates the extent of extension along the boundary, and is determined based on the proportion of continuous coverage along the direction of the unclosed boundary of the observed morphology. The value range can be set to 0 to 1. This indicates the degree of recurrence over multiple rounds, and is determined based on the frequency of the observed pattern repeating in different edge observation rounds. The value range can be set from 0 to 1. , , These represent the weighting coefficients of the degree of synchronous occurrence, the degree of extension along the boundary, and the degree of multiple recurrence in the calculation of boundary adjoint degree, respectively. Each value can be set to a range of 0 to 1, and the sum must be 1. The server writes the observation patterns with high boundary adjoint degree, along with their boundary positions, attachment directions, and extension lengths, into the boundary adjoint record. The boundary adjoint record is the direct input for subsequent position stability tracking.

[0058] In the process of performing position stability tracking on each unclosed region based on boundary accompanying records and obtaining the position stability tracking results, the server continuously tracks the spatial attachment behavior of the same unclosed region under different rounds and multiple edge-fitting conditions, focusing on the attachment position, propagation start position, dwelling center position, and boundary extension position extracted from the boundary accompanying records. Position stability tracking is not simply comparing whether observation points recur, but rather determining whether the boundary accompanying expression consistently attaches to the same structural boundary, the same heat retention position, the same vibration discontinuity position, or the same sound source dwelling center under different conditions. If a boundary accompanying expression attaches to the front section of the boundary in one round of edge-fitting observation and then jumps to the rear section in another round, and this change does not conform to the continuous propagation relationship of the boundary, it indicates that the expression has poor position stability. If a boundary accompanying expression, although exhibiting slight drift under different conditions, consistently clusters around the same structural boundary breakpoint or the same adjacent region splicing breakpoint, it indicates that it possesses position stability. To quantify the position stability of the same unclosed region in multiple rounds of edge-fitting observation, position stability can be defined as follows:

[0059] in, This indicates the stability of the current unclosed region. The value can be set from 0 to 1, with a larger value indicating a more stable attachment position. This indicates the number of edge-fitting observation rounds involved in position stability tracking; Indicates the first The offset distance between the main attachment position and the reference attachment position of the boundary accompanying expression in the wheel edge observation can be calculated based on the boundary parameter coordinates or spatial coordinates, and the value is greater than or equal to 0. This represents the preset maximum allowable offset distance, used to normalize the offset distance, and its value is greater than 0. The server generates a position stability tracking result based on position stability and the clustering relationship of the attached positions. The position stability tracking result not only indicates whether the position is stable, but also retains the trajectory of the attached position changes and the stability level, providing a basis for subsequent cross-condition verification.

[0060] In the process of performing cross-condition correspondence verification on the position stability tracking results to determine the consistency level of attachment positions of each unclosed region under multi-condition edge observation, the server cross-compares the position stability results obtained under different observation distances, observation directions, dwell times, movement rhythms, and working condition slices to determine whether the same unclosed region can establish a consistent attachment position correspondence under different conditions. Cross-condition correspondence verification is not a repetitive comparison, but requires that although the apparent details of the same unclosed region may differ under various observation conditions, its core attachment positions should be able to map to each other. For example, the structural boundary breakpoint position extracted under close-range conditions should correspond to the boundary extension center extracted under medium-range conditions; the temperature distribution attachment center extracted under different working condition slices, if it is a real anomaly, should still fluctuate around the same heat retention location, rather than jumping irregularly under different conditions. To quantify this cross-condition consistency, an attachment position consistency coefficient can be defined:

[0061] in, This represents the consistency coefficient of the attachment position of the currently unclosed region under multi-condition edge observation. The value can be set from 0 to 1. The larger the value, the more consistent the position correspondence across conditions. Indicates the number of observation conditions involved in the verification; Indicates the first The observation condition and the first The degree of correspondence between attachment positions under each observation condition is determined by comprehensively considering the overlap between the attachment center, boundary attachment segment, and propagation initiation position under both conditions, with a value range of 0 to 1. Based on the attachment position consistency coefficient, the server classifies each unclosed region into different attachment position consistency levels: high consistency, medium consistency, and low consistency. The attachment position consistency level is the direct basis for subsequent stability convergence determination, used to distinguish between boundary anomaly signs that only occasionally appear under individual conditions and candidate anomaly cores that stably repeat under multiple conditions.

[0062] In the process of performing stability convergence determination based on the consistency level of attachment positions, extracting unclosed regions that meet the stability convergence conditions as anomalous kernel candidates, and forming an anomalous kernel candidate set, the server no longer treats unclosed regions as open and divergent boundary anomalous areas. Instead, it judges whether, after multi-condition edge observation and cross-condition verification, they have gradually converged into a core region with stable position, stable boundary attachment, stable propagation starting position, and stable residence center. Stable convergence refers to the fact that the unclosed signs that may have been scattered within the unclosed region no longer drift along the boundary over a large area after multiple rounds of edge observation, but gradually concentrate to a relatively fixed attachment center and its neighborhood. To comprehensively evaluate this convergence degree, a stability convergence degree can be defined:

[0063] in, This indicates the stability convergence of the currently unclosed region; a larger value indicates that the region is closer to forming an anomalous kernel candidate. Indicates positional stability; Indicates the adhesion position consistency coefficient; This represents the average value of the boundary adjoint degree within the unclosed region. It is obtained by averaging the boundary adjoint degrees of all boundary adjoint expressions involved in the convergence determination, and the value can be set to 0 to 1. , , These represent the weighting coefficients of position stability, attachment position consistency coefficient, and average boundary adjoint degree in the stability convergence determination, respectively. Each value can be set from 0 to 1, and their sum must be 1. If the stability convergence of an unclosed region exceeds a preset convergence threshold, the server extracts it as an anomalous kernel candidate. An anomalous kernel candidate is not the final anomalous result, but rather a highly reliable candidate core region that has passed boundary attachment, multiple rounds of reproduction, position stability, and cross-condition consistency verification. The server summarizes all anomalous kernel candidates that meet the conditions according to the target observation area location, main attachment boundaries, main observation morphology type, and stability convergence, forming an anomalous kernel candidate set. This anomalous kernel candidate set will serve as the direct input for the next step of shell-borrowing backfilling processing to further confirm which candidate regions still cannot be interpreted as normal after introducing multi-source normal shell fragments, thus gradually forming anomalous shell-breaking regions.

[0064] S150. Based on the abnormal kernel candidate set, introduce normal shell fragments from multiple sources to perform shell backfilling processing on each abnormal kernel candidate to form an abnormal unpacking region.

[0065] Specifically, the anomaly kernel candidate set refers to the set of candidate regions that have obtained high stability through multi-condition edge observation, boundary accompaniment extraction, positional stability tracking, and stability convergence determination; the anomaly unshelled region refers to the set of regions that, after repeated backfilling with normal shell fragments from multiple sources, still cannot be reintegrated into the normal shell interpretation system. The focus of this processing chain is not on directly searching for anomaly evidence, but on continuously introducing normal evidence for coverage verification. Only after the normal evidence fails is the anomaly formally manifested.

[0066] In the process of performing candidate deconstruction processing on each anomalous kernel candidate in the anomalous kernel candidate set, mapping the boundary attachment position, dwelling center position, propagation initiation position, and corresponding observation morphology back to the initial normal shell to form a structural mapping relationship, the server first reads the boundary attachment position, dwelling center position, propagation initiation position, target observation area identifier, boundary accompanying record, and main observation morphology type of each anomalous kernel candidate, and then maps this information back to the local normal expression structure of the initial normal shell. Candidate deconstruction processing refers to decomposing an anomalous kernel candidate from a stable candidate region into multiple structural elements that can correspond to the normal shell item by item. Specifically, the boundary attachment position characterizes the specific location where the anomalous kernel candidate is permanently attached to the boundary of the normal shell; the dwelling center position characterizes the continuous accumulation position of temperature distribution morphology, sound source dwelling morphology, or vibration propagation morphology; the propagation initiation position characterizes the starting breakpoint of anomalous propagation or deviation from the normal propagation path; and the corresponding observation morphology characterizes which type of observation morphology the anomalous kernel candidate mainly relies on for manifestation. To quantify the tightness of the structural mapping between the anomalous kernel candidate and the initial normal shell, a structural mapping composite degree can be constructed.

[0067] in, It represents the structural mapping complexity, and its value ranges from 0 to a continuous value greater than 1. The larger the value, the more complete the structural correspondence between the abnormal kernel candidate and the initial normal shell. The number of elements involved in the structure mapping includes at least four types of elements: boundary attachment location, dwelling center location, propagation start location, and corresponding observation morphology location. Indicates the first The basic weight coefficients of the class mapping elements range from 0 to 1, and the sum of all weight coefficients is greater than 0; Indicates the first abnormal kernel candidate Spatial location vectors of class mapping elements; Indicates the relationship between the initial normal shell and the first The normal reference position vector corresponding to the class mapping element; This represents the spatial offset between the candidate position and the normal reference position. The smaller the offset, the closer the corresponding mapping relationship. Indicates the first The location tolerance parameter for class features, with a value greater than 0, is used to control the rate at which the location deviation decays to the mapping result. The smaller the value, the more sensitive it is to offset. Indicates the first The direction consistency enhancement coefficient for class elements, with a value range from 0 to 1; This indicates that the abnormal kernel candidate and the initial normal shell are in the 1st... The smaller the angle between the boundary directions or the angle between the propagation directions on the class element, the stronger the directional consistency. Indicates the first abnormal kernel candidate Local gradient vector or propagated gradient vector of class element; This represents the local gradient vector or propagation gradient vector of the corresponding element in the initial normal shell. This represents the gradient consistency enhancement coefficient, with a value ranging from 0 to 1; It represents the inner product of two gradient vectors, used to reflect whether the local change trends are consistent; This indicates an extremely small positive number that prevents the denominator from being zero. Through this structural mapping composition, the server can identify where the current abnormal kernel candidate and the initial normal shell maintain a high correspondence and where they have deviated significantly, thus forming the structural mapping relationship for subsequent shell-borrowing backfilling.

[0068] In the process of constructing a backfill reference framework around each abnormal kernel candidate, extracting homologous normal shell fragments from the initial normal shell, extracting historical normal shell fragments from historical observation records, and extracting analogous normal shell fragments from adjacent functional units, the server establishes a multi-source normal reference system around each abnormal kernel candidate. The backfill reference framework refers to a set of callable normal shell fragment sources prepared for the current abnormal kernel candidate. Homologous normal shell fragments refer to local normal fragments extracted from the initial normal shell that are located in the same target observation area or the same structural segment as the current abnormal kernel candidate; historical normal shell fragments refer to normal local fragments selected from the target device's historical observation records that match the current abnormal kernel candidate in terms of location, boundary relationships, and operating conditions; analogous normal shell fragments refer to normal local fragments extracted from adjacent functional units that have an analogous relationship with the current abnormal kernel candidate in terms of structural function, transmission path, and boundary characteristics. To select the most suitable normal shell fragment as a backfill reference from the three types of sources, a fragment reference adaptation index can be constructed:

[0069] in, Indicates the first The reference fit index of a normal shell segment to the current abnormal kernel candidate. The larger the value, the more suitable the normal shell segment is as a reference for shell backfilling. This indicates the structural mapping composition degree between the normal shell segment and the current abnormal kernel candidate; , , , These represent the weight coefficients of structural mapping, location adaptation, working condition adaptation, and observation form adaptation in the comprehensive adaptation, respectively, with values ​​ranging from 0 to 1, and the sum being greater than 0; The number of key location points involved in location adaptation can include at least the boundary attachment location, the dwell center location, and the propagation start location. Indicates the first The weight coefficients for each key location point range from 0 to 1; Indicates the first The first normal shell segment Position vectors of key locations; This represents the position vector of the corresponding key location point in the current abnormal kernel candidate; Indicates the first The position scale parameter of each location point takes a value greater than 0; This indicates the number of operating condition variables involved in operating condition adaptation, such as load level, operating phase, and environmental state. Indicates the first Weighting coefficients for each working condition variable; Indicates the first The first normal shell segment Values ​​of each working condition variable; This indicates the time corresponding to the current abnormal kernel candidate. Values ​​of each working condition variable; Indicates the first The allowable range of differences for each operating condition variable, with values ​​greater than 0; Indicates the first The weight coefficients of the observation morphology adaptation terms can be set by least squares fitting or manual experience, ranging from 0 to 1, and will not be elaborated here; Indicates the first The first normal shell segment in the... The degree of fit in the observation morphology is defined, with four observation morphologies corresponding to structural boundary morphology, vibration propagation morphology, temperature distribution morphology, and sound source residence morphology, respectively, and values ​​ranging from 0 to 1. Using this segment reference fit index, the server can select segments with higher fit from multiple sources of normal shell segments to enter the backfill reference frame.

[0070] In the process of performing alignment and introduction processing on homologous normal shell fragments, historical normal shell fragments, and analogous normal shell fragments to map each source of normal shell fragment to its corresponding abnormal kernel candidate, the server does not directly overwrite the abnormal kernel candidate with each source of normal shell fragment. Instead, it first performs positional alignment, boundary alignment, propagation alignment, and dwell alignment based on structural mapping relationships, so that normal shell fragments from different sources can be mapped to the current abnormal kernel candidate under a unified reference. Alignment and introduction processing refers to aligning homologous normal shell fragments, historical normal shell fragments, and analogous normal shell fragments spatially and semantically with the current abnormal kernel candidate through unified local coordinate relationships and boundary attachment relationships. To quantify the quality of the alignment and introduction, an alignment and introduction comprehensive coefficient can be constructed:

[0071] in, Indicates the first The alignment factor introduced by each normal shell segment to the current abnormal kernel candidate is a comprehensive coefficient. The larger the value, the higher the quality of the alignment introduction. The number of structural elements involved in alignment can include at least the boundary attachment location, the residence center location, the propagation start location, and the dominant location of the observed morphology. Indicates the first Weight coefficients of each structural element; Indicates the first The first normal shell segment Aligned position vectors of each structural element; This represents the position vector of the corresponding structural element in the current abnormal kernel candidate; Indicates the first The alignment scale parameter of each structural element has a value greater than 0; This represents the direction-consistent enhancement coefficient, with a value ranging from 0 to 1; Indicates the first The smaller the angle between the structural elements and the abnormal kernel candidate, the more consistent the alignment direction. This represents the coefficient for enhancing the local change trend, and its value ranges from 0 to 1. Indicates the first The first normal shell segment The local change vector of a structural element can correspond to the boundary change vector, thermal gradient vector, or propagation trend vector. This represents the local change vector of the corresponding structural element in the current abnormal kernel candidate; This represents the inner product of two local change vectors, used to characterize whether the change trends are consistent. By introducing a comprehensive coefficient through this alignment, the server can determine whether the current normal shell segment has met the conditions to enter the shell-borrowing backfilling stage.

[0072] In the process of performing shell-borrowing backfilling on anomalous kernel candidates based on normal shell fragments from various sources, so as to reconstruct the interpretative closure relationship of structural boundary morphology, vibration propagation morphology, temperature distribution morphology, and sound source dwelling morphology within the anomalous kernel candidate region, the server begins to perform core backfilling. Shell-borrowing backfilling refers to pressing the aligned and introduced normal shell fragments into the anomalous kernel candidate region layer by layer according to three levels: boundary, transmission, and attachment, so that the anomalous kernel candidate can re-attempt to obtain a closure interpretation under the influence of normal expressions from different sources. First, the structural boundary morphology is used as the boundary constraint, and the boundary connection relationship in the normal shell fragments is used to repair the boundary discontinuities of the anomalous kernel candidate; then, the vibration propagation morphology and temperature distribution morphology are used as transmission constraints, and normal transmission links are used to cover the anomalous propagation discontinuity and thermal mismatch; at the same time, the sound source dwelling morphology is used as the attachment constraint, and normal dwelling centers and attachment paths are used to cover the dwelling offset of the anomalous kernel candidate. In order to comprehensively measure the closure recovery level of the four types of observation morphologies in the current candidate region after shell-borrowing backfilling, a backfilling closure composite degree can be constructed:

[0073] in, Indicates the first The backfill closure compositeness of the current abnormal kernel candidate under the action of normal shell fragments from one source. The larger the value, the closer it is to restoring the normal interpretation closure after backfilling. The number indicates the observation mode type. The four observation modes correspond to the structural boundary mode, vibration propagation mode, temperature distribution mode, and sound source residence mode, respectively. Indicates the first The weighting coefficient of the observation pattern in the backfill closure determination; Indicates the first Under the influence of a normal shell fragment from a certain source, the first The reconstruction closure strength of the observation morphology in the candidate region of the abnormal kernel is determined comprehensively based on the performance of the morphology in terms of continuity, integrity and consistency after backfilling, and the value is greater than or equal to 0. This indicates the abnormal kernel candidate before backfilling. The original mismatch intensity or notch intensity of the observed morphology, with a value greater than or equal to 0; This represents a very small positive number that prevents the denominator from being zero and ensures the stability of weak values. Indicates the first The recovery magnification index of the observed morphology, with a value greater than 0, is used to adjust the nonlinear contribution of the recovery speed of different morphologies. Indicates the first The differential suppression scaling parameter for observational morphology takes a value greater than 0; Indicates the first The synergistic enhancement coefficient of the observation morphology ranges from 0 to 1; Indicates the first Under the influence of a normal shell fragment from a certain source, the first The degree of cooperative closure between the observation morphology and the other three observation morphologies is determined based on the multi-morphological mutual verification relationship, with a value ranging from 0 to 1. The server performs shell borrowing backfilling on homologous normal shell fragments, historical normal shell fragments, and analogous normal shell fragments respectively to form the backfill closure composite degree under the three sources.

[0074] During the shell-borrowing backfilling process, a backfilling closure determination is performed. Abnormal kernel candidates that restore a continuous interpretation link are marked as backfillable regions, while those that cannot restore a continuous interpretation link under the influence of multiple normal shell segments are marked as backfilling failure regions. In this process, the server makes a unified decision on the backfilling results from all sources. A continuous interpretation link refers to a complete normal interpretation relationship where the structural boundary morphology, vibration propagation morphology, temperature distribution morphology, and sound source residence morphology can re-form a continuous, reasonable, stable, and mutually supportive relationship under the influence of normal shell segments from the same source. If an abnormal kernel candidate, under the influence of at least one normal shell segment, achieves a backfilling closure composite degree reaching a preset closure threshold, and the cross-morphological collaborative closure degree reaches a preset collaborative threshold, then the server marks it as a backfillable region. A backfillable region indicates that the candidate region can still be reinterpreted as normal by some normal shell segment and will not proceed to subsequent abnormal confirmation. If an abnormal kernel candidate, under the influence of multiple normal shell segments, consistently fails to reach the preset closure threshold, or although individual morphologies are restored, the overall collaborative closure relationship remains interrupted, then the server marks it as a backfilling failure region. To quantitatively characterize the degree of backfill failure under multi-source conditions, a composite strength of backfill failure can be constructed:

[0075] in, This indicates the backfill failure composite strength of the current abnormal kernel candidate. The larger the value, the more difficult it is to restore normal closure under the influence of normal shell fragments from multiple sources. Indicates the degree of backfill closure compositeness under the action of homologous normal shell segments; Indicates the degree of backfill closure compositeness under the influence of historical normal shell fragments; This indicates the degree of backfill closure compositeness under the action of a normal shell segment; , , These represent the weighting coefficients of the three sources—homogeneous, historical, and analogous—in the composite intensity of backfill failure, with values ​​ranging from 0 to 1, and the sum being greater than 0. , , These represent the nonlinear adjustment indices for the three types of sources, with values ​​greater than 0, used to adjust the sensitivity of different sources to the overall failure intensity. The higher the composite backfill failure intensity, the more likely it is that the current anomalous kernel candidate cannot be restored to a normal interpretation state regardless of which normal shell fragment is borrowed.

[0076] In the process of summarizing and marking failed backfilling regions as anomalous unshelling regions, the server uniformly organizes and marks all failed backfilling regions according to the target observation area location, boundary attachment location, dominant observation morphology type, and backfilling failure composite intensity. An anomalous unshelling region refers to a region that has undergone repeated backfilling using homologous normal shell fragments, historical normal shell fragments, and analogous normal shell fragments, but still cannot restore a continuous interpretation link. Unshelling means that the region has detached from the structural system that could originally be interpreted by the normal shell; the normal shell no longer has the ability to cover it, therefore its anomalous attributes are highly exclusive. When marking anomalous unshelling regions, the server can simultaneously write the corresponding structural mapping relationship, minimum backfilling closure composite degree, maximum backfilling failure composite intensity, and failure source record for that region, so that subsequent outer edge splicing and anomalous growth contour formation can directly call upon this information.

[0077] S160. Perform outer edge splicing processing on the boundary between the abnormal desquamation area and the normal shell to form an abnormal growth profile, and output the abnormal identification result of the device based on the abnormal growth profile.

[0078] Specifically, the normal shell refers to the overall normal interpretation shell formed based on multi-path inspection, multi-state acquisition, and fusion and inclusion; the abnormal growth contour refers to the outer edge structure expression formed at the boundary between the abnormal unshelled region and the normal shell after revisiting, segmenting, splicing, and extending for confirmation. The focus of the entire processing logic is not simply to output abnormal points or abnormal regions, but to output the boundary structure, extension state, and spatial occupancy relationship of the abnormal outer edge, thereby giving the abnormal results a stronger ability to express structure and evolution.

[0079] In the process of performing boundary extraction processing on the abnormal shelling region to obtain the shelling boundary zone between the abnormal shelling region and the normal shell, the server first puts the abnormal shelling region back into the overall structure of the target device and compares it with the boundary of the normal shell to identify the location where the interpretation switches between the two. Boundary extraction processing is not simply extracting geometric contours, but simultaneously considering the switching positions of structural boundary morphology, temperature distribution morphology, vibration propagation morphology, and sound source dwell morphology between the normal shell and the abnormal shelling region, thus forming a transition zone with interpretative significance. The shelling boundary zone refers to the band-shaped region between the abnormal shelling region and the normal shell where a continuous switch occurs from normal interpretation to abnormal interpretation. It includes both discontinuous positions on the structural boundary and the switching interface in thermal, vibrational, and acoustic performance. To quantify whether a certain location belongs to a stable shelling boundary zone, a boundary salience expression can be constructed as follows:

[0080] in, This indicates the significance of the current candidate boundary position; the larger the value, the more likely the position is to belong to a stable detached boundary zone. The number indicates the observation mode type. The four observation modes correspond to the structural boundary mode, temperature distribution mode, vibration propagation mode, and sound source residence mode, respectively. Indicates the first The weighting coefficient of the observation morphology in boundary extraction has a value range of 0 to 1. Indicates the abnormal unpacking region in the first... Local expression values ​​in the observed morphology; This indicates the normal shell at the corresponding position. Local expression values ​​of observed morphology; This indicates the magnitude of the difference between the abnormal detachment region and the normal shell in the observed morphology; the greater the difference, the more obvious the boundary. Indicates the first The difference scale parameter for observational morphology, with a value greater than 0, is used to control difference normalization; This indicates an extremely small positive number that prevents the denominator from being zero; Indicates the first The enhancement index of the morphological difference item is greater than 0, and the larger the value, the more sensitive it is to significant differences. Indicates the first The coefficient for enhancing inconsistency in the morphological gradient of similar observations, with a value ranging from 0 to 1; This represents the local gradient vector corresponding to the observed morphology of the abnormal uncoating region; This represents the local gradient vector corresponding to the observed morphology of the normal shell. This indicates the magnitude of the gradient difference between the two sides, reflecting whether the changing trends on both sides of the boundary are different. The server extracts continuous highly significant regions based on the boundary saliency, forming a peeled boundary zone.

[0081] In the process of performing boundary segmentation on the shell-shedding boundary zone to form multiple boundary segments, the server does not directly use the shell-shedding boundary zone as a single continuous boundary. Instead, based on local boundary orientation changes, observation mode switching stability, and boundary continuity, it divides the shell-shedding boundary zone into multiple independently verifiable and spliced ​​boundary segments. Boundary segmentation refers to identifying boundary turning points, thermal state switching turning points, vibration discontinuity turning points, and sound source attachment turning points along the extension direction of the shell-shedding boundary zone, and using these turning points as segmentation nodes to divide the entire boundary zone into multiple local boundary units. A boundary segment refers to a boundary fragment with relatively stable boundary orientation, relatively consistent observation mode switching methods, and suitable as the smallest unit for revisiting and splicing within a local area. To quantify the consistency within a given boundary segment, a segment homogeneity expression can be constructed as follows:

[0082] in, This indicates the homogeneity of the current boundary segment; a larger value indicates that the segment is more uniform and stable. This represents the boundary arc length of the current boundary segment, and its value is greater than 0. This represents the arc length parameter along the boundary segment; Indicates the position of the boundary segment Boundary curvature at the location; This represents the rate of change of curvature; the smaller the rate of change, the more stable the boundary orientation. This represents the curvature change suppression parameter, and its value is greater than 0. Indicates the first Observational morphology in position The switching intensity or switching slope at the location; Indicates the th segment within the current boundary segment The average value of the intensity of morphological switching observed; Indicates the first The scale parameter for consistency of observational morphology switching takes a value greater than 0; Indicates the first The weighting coefficient of the observation morphology in the homogeneity of the edge segment ranges from 0 to 1. The server divides continuous regions with high edge segment homogeneity into separate boundary segments, so that subsequent revisits and stability determinations are based at the edge segment level rather than the entire boundary zone level.

[0083] In the process of establishing an outer edge revisit framework around each boundary segment and controlling industrial intelligent robots to perform outer edge revisit data collection at different revisit times and under different working conditions along edge-close revisit paths, cross-edge revisit paths, and along-edge revisit paths to form boundary revisit records, the server establishes a repeat observation mechanism for each boundary segment. The outer edge revisit framework refers to a set of revisit organization methods configured for each boundary segment that are repeatable in time, distinguishable in path, and variable in conditions. Edge-close revisit paths are paths that run close to one side of the boundary segment, used to examine the boundary attachment status in detail; cross-edge revisit paths are paths that traverse the delamination boundary zone and cross both sides of the normal shell and abnormal delamination areas, used to observe the switching relationship between the two sides of the boundary; and along-edge revisit paths are paths that extend parallel to the boundary segment, used to observe the continuity and extension trend of the segment. The revisit time point refers to the moment when the server re-observes the edge segment at different time intervals; the working condition slice refers to the observation condition window captured under different operating stages, load levels, or environmental conditions of the target equipment. The server controls the industrial intelligent robot to repeatedly collect data under the above combination of path, time point, and working condition slice conditions, and writes the boundary position, boundary shape, observation shape switching status, boundary attachment status, and working condition status obtained from each collection into the boundary revisit record. The boundary revisit record is the basis for subsequent edge segment stability determination. Its key is not the result of a single instance, but whether the results of multiple rounds are consistent or show stable extension.

[0084] Based on boundary revisit records, the server performs boundary stability assessment on each boundary segment to determine stable and wandering segments. This process involves comprehensively evaluating the boundary position fluctuations, observational morphology switching relationships, and boundary extension status of the same boundary segment under different revisit times, paths, and working conditions. Segment stability assessment does not simply determine whether a segment reappears, but rather whether it consistently adheres to the same structural position, maintains the same switching relationship, and retains similar segment lengths and orientations across multiple revisits. Stable segments are defined as boundary segments with minimal positional drift, continuous switching relationships, and consistent boundary orientations across multiple revisits. Wandering segments are defined as boundary segments with significant positional drift, fluctuating switching relationships, or unstable segment lengths and orientations across multiple revisits. To quantify segment stability, a segment stability index expression can be constructed as follows:

[0085] in, This represents the edge stability index of the current boundary segment; the larger the value, the more stable the edge segment. This indicates the number of revisit rounds for the current boundary segment to participate in the stability assessment; Indicates the first The center position vector of the boundary segment during the round of revisiting; This represents the average vector of the center position of the boundary segment across all revisit rounds; Indicates the first The offset of the wheel center position relative to the average center position; This represents the position drift scale parameter, and its value is greater than 0. , , These represent the weighting coefficients of the position stability term, orientation stability term, and observation mode switching stability term in the segment stability index, respectively, with values ​​ranging from 0 to 1; Indicates the first The angle between the direction of the side segment and the average direction during the revisit is the smaller the angle, the more stable the direction. Indicates the first The second round of visits to China The intensity of switching of observational morphology on this edge segment; This indicates the first of all follow-up visits. The average value of the intensity of morphological switching observed; Indicates the first The stable scale parameter for the intensity of morphological switching in the observation class takes a value greater than 0; Indicates the first The nonlinear adjustment index of the observed morphology in stability determination takes a value greater than 0. The server divides the boundary segments into stable segments and wandering segments based on the segment stability index and the preset stability threshold.

[0086] In the process of performing edge splicing on stable edges and forming local contour fragments based on structural continuity, observational morphological continuity, and revisit timing, the server attempts to construct a locally continuous contour based solely on stable edges. Edge splicing is not simply connecting spatially adjacent stable edges directly; rather, it requires multiple stable edges to simultaneously meet splicing conditions in terms of structural continuity, observational morphological continuity, and revisit timing. Structural continuity refers to the natural connection between two stable edges in terms of geometric boundary orientation and structural boundary attachment position; observational morphological continuity refers to the consistency or continuous transition of temperature distribution patterns, vibration propagation patterns, sound source residence patterns, and structural boundary morphological switching modes on two stable edges; and revisit timing refers to the consistent sequence or clear temporal extension correlation between two stable edges in multiple revisits. Stable edges satisfying these three relationships are spliced ​​by the server into local contour fragments. These local contour fragments are local anomalous outer edge units formed by splicing multiple stable edges; they are no longer single edges but possess a certain range of outer edge expressive capabilities. To quantify whether two stable edge segments are suitable for splicing, we can construct an edge splicing fitness expression as follows:

[0087] in, Indicates a stable edge segment With stable edge segment The edge splicing fit between the two segments; the larger the value, the better the two segments are spliced. Indicates a stable edge segment The end position vector; Indicates a stable edge segment The starting position vector; Indicates the spatial distance between the endpoints of the two segments; This represents the spatial distance scale parameter for splicing, and its value is greater than 0. , , , These represent the weight coefficients of spatial connectivity, directional consistency, observation form continuity, and temporal connectivity in the edge splicing fit, with values ​​ranging from 0 to 1. Indicates a stable edge segment With stable edge segment The smaller the angle between them, the more consistent their directions are; Indicates a stable edge segment In the Switching characteristic values ​​in the observation morphology; Indicates a stable edge segment In the Switching feature values ​​in the observation morphology; Indicates the first The tolerance parameter for the continuity of the observed morphology, with a value greater than 0; Indicates the first The weighting coefficient of observational patterns in continuity determination; and These represent the primary revisit time sequence positions of the two stable edge segments; This represents the temporal alignment scale parameter, with a value greater than 0. The server stitches highly adapted, stable edges into local contour fragments based on the edge stitching fit.

[0088] In the process of performing contour extension and overall closure determination on local contour segments to form an abnormal growth contour, the server does not directly use the local contour segments as the final contour. Instead, it continues to search for stable edges that have not yet been included along their ends and adjacent positions to confirm whether the local contour segments can continue to extend outward and to determine whether multiple local contour segments can form an overall contour. Contour extension and identification processing refers to starting from the front and back ends of the local contour segments and searching for subsequent edges that are spatially adjacent, structurally and visually continuous, and have continuity in the revisit sequence, thereby confirming whether the local contour segments have a further growth direction. Overall closure determination processing refers to determining whether multiple local contour segments have formed a closed outer edge around the abnormal shedding region, or at least an open outer edge with a clear extension direction. An abnormal growth contour refers to the abnormal outer edge structure formed after splicing, extending, and determining overall closure of local contour segments; it can be a closed contour or an open contour. To quantify the overall contouring degree of a set of local contour segments, a contour formation index expression can be constructed as follows:

[0089] in, This represents the contour formation index of the current local contour segment group. The larger the value, the closer it is to forming a complete abnormal growth contour. , , , These represent the weighting coefficients of the connectivity length term, gap penalty term, open area term, and edge stability term in the contour forming index, respectively, with values ​​ranging from 0 to 1; This indicates the total length of the currently successfully connected local contour segments; This represents the total reference length of the anomaly's outer edge that the theory should cover, and its value is greater than 0. This indicates the number of gaps that are not yet closed between the current local contour segments; Indicates the number of local contour segments involved in the overall closure determination; This represents the quantified value of the open area or open region formed by the current unclosed contour; the smaller the value, the closer it is to closure. This represents a normalized reference value for the open area, and its value is greater than 0. Indicates the first The stability index of each constituent edge segment or fragment. Based on the contour formation index, the server determines the local contour fragment group as either a closed anomalous growth contour or an open anomalous growth contour.

[0090] In the process of performing contour attribute annotation processing based on abnormal growth contours and outputting device anomaly identification results according to the target observation area range, boundary attachment position, and contour status corresponding to the abnormal growth contours, the server performs attribute binding and result generation on the formed abnormal growth contours. Contour attribute annotation processing refers to the unified association between the abnormal growth contour and its covered target observation area range, main boundary attachment positions, dominant observation form switching type, contour status, and contour formation index, so that the abnormal growth contour has interpretable result attributes. Among them, the target observation area range is used to characterize which equipment areas are covered by the abnormal growth contour; the boundary attachment position is used to characterize which structural boundaries, transmission breakpoints, or thermal boundaries the outer edge of the anomaly mainly adheres to; the contour status is used to distinguish whether the abnormal growth contour is closed or open. When outputting device anomaly identification results, the server outputs not only the anomaly location and anomaly boundary, but also whether the anomaly is in a stable enclosed state or continues to extend. In order to comprehensively form the final identification strength, the result confidence index expression can be constructed as follows:

[0091] in, The confidence index represents the current device anomaly identification result; the higher the value, the more reliable the anomaly result. This indicates the number of abnormally uncoated regions involved in the current abnormal growth profile; Indicates the first The contour formation index corresponding to each abnormal unshelled region; Indicates the first The composite strength of backfill failure corresponding to each abnormal desiccation area; Indicates the first The structural mapping composite degree corresponding to each abnormal uncoating region; , , These represent the weighting coefficients of contour formation, backfill failure, and structure mapping in the result confidence index, respectively, ranging from 0 to 1, with a sum greater than 0. Based on the result confidence index and contour attribute annotation results, the server outputs the final device anomaly identification result, ensuring that the result not only indicates the existence of an anomaly but also its spatial range, boundary state, extent of extension, and confirmation credibility.

[0092] This application also provides a device anomaly identification system for industrial intelligent robots, referring to... Figure 2 , Figure 2This application provides a schematic diagram of a device anomaly identification system for industrial intelligent robots. The system is a server, comprising an acquisition module 21 and a processing module 22. The acquisition module 21 acquires multiple sets of observation records generated by the target device under multi-path inspection and multi-state acquisition conditions of the industrial intelligent robot, and performs fusion and inclusion processing on various observation patterns in the multiple sets of observation records to construct an initial normal shell. The processing module 22, based on the initial normal shell, performs observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction processing on the multiple sets of observation records to form a shell fragmentation clue set. The processing module 22 is also used to... The processing module 22 is used to perform neighborhood expansion and historical relationship introduction processing on the fractured regions to implement inward shell filling and form an unclosed set. The processing module 22 is also used to perform multi-condition edge observation and position stability tracking processing on each unclosed region of the unclosed set to form an abnormal kernel candidate set. The processing module 22 is also used to introduce normal shell fragments from multiple sources to perform shell borrowing backfilling processing on each abnormal kernel candidate based on the abnormal kernel candidate set to form an abnormal shell-shedding region. The processing module 22 is also used to perform outer edge splicing processing on the boundary between the abnormal shell-shedding region and the normal shell to form an abnormal growth contour, so as to output the device anomaly identification result according to the abnormal growth contour.

[0093] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0094] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0095] The communication bus 32 is used to enable communication between these components.

[0096] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0097] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0098] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0099] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a device anomaly identification method applied to industrial intelligent robots.

[0100] exist Figure 3In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call an application program stored in the memory 35 for a device anomaly identification method for industrial intelligent robots. When executed by one or more processors, the electronic device executes one or more methods as described in the above embodiments.

[0101] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0102] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0103] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0108] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for identifying equipment anomalies in industrial intelligent robots, characterized in that, The method includes: Acquire multiple sets of observation records formed by the target equipment under the conditions of multi-path inspection and multi-state acquisition of industrial intelligent robots, and perform fusion and inclusion processing on various observation forms in multiple sets of observation records to construct an initial normal shell. Based on the initial normal shell, observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction are performed on multiple sets of observation records to form a shell rift clue set. Around the set of shell fracture clues, neighborhood expansion and historical relationship introduction processing are performed on the fracture regions to implement inward shell filling and form an unclosed set; For the aforementioned unclosed set, multi-condition edge-fitting observation and position stability tracking are performed on each unclosed region to form an abnormal kernel candidate set; Based on the aforementioned abnormal kernel candidate set, normal shell fragments from multiple sources are introduced to perform shell-borrowing backfilling processing on each abnormal kernel candidate to form an abnormal unpacking region; An outer edge splicing process is performed on the boundary between the abnormal desquamation area and the normal shell to form an abnormal growth profile, and the abnormal identification result of the device is output based on the abnormal growth profile.

2. The device abnormality recognition method for an industrial intelligent robot according to claim 1, characterized by, The acquisition of multiple sets of observation records formed by the target device under the conditions of multi-path inspection and multi-state acquisition by the industrial intelligent robot, and the fusion and inclusion processing of various observation modes in the multiple sets of observation records to construct an initial normal shell, specifically includes: Multiple target observation areas are divided around the target device, and approach paths, frontal view paths, side view paths, detour paths, and retreat paths are configured for each target observation area to construct an inspection passage framework; The industrial intelligent robot is controlled to pass through the corresponding target observation area along each of the inspection passage frames in sequence, and maintains a distant initial observation state, a forward close-up state, a lateral offset state, a continuous turning state, and a retraction and re-viewing state at different path positions to form a multi-state acquisition condition. The observations formed at each path location and under each acquisition state are grouped and written according to the target observation area, path location, and acquisition state to form multiple sets of observation records; Temperature distribution pattern, vibration propagation pattern, sound source residence pattern and structural boundary pattern are extracted from multiple sets of observation records. The observation patterns belonging to the same target observation area are merged within the same area to obtain the merged result. Based on the same-area merging processing results, various observation forms that remain stable under multi-path inspection and multi-state acquisition conditions are selected and fusion processing is performed to form local normal shells for each target observation area; Perform neighbor region splicing on each local normal shell layer to obtain the splicing result, and construct an initial normal shell layer covering the entire range of the target device based on the splicing result. 3.The device abnormality recognition method for an industrial intelligent robot according to claim 1, wherein The process of performing observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction on multiple sets of observation records based on the initial normal shell to form a shell rift clue set specifically includes: The shell composition information in the initial normal shell is mapped back to multiple sets of observation records to establish a shell disassembly view corresponding to each target observation area; Based on the shell disassembly view, observation path rearrangement is performed on multiple sets of observation records to make the approach path, frontal path, side view path, detour path and retreat path used to support the local normal shell form a cross-call relationship to generate path rearrangement observation results. Based on the path rearrangement observation results, the observation order is rearranged for multiple sets of observation records so that the acquisition order of temperature distribution, vibration propagation, sound source residence and structural boundary in each target observation area forms an observation chain to generate the sequential rearrangement observation results. Based on the ordered rearrangement of the observation results, neighborhood reference reconstruction processing is performed on the multiple sets of observation records to generate reference reconstruction observation results; Closed-loop control processing is performed on the path rearrangement observation results, the sequential rearrangement observation results, and the reference reconstruction observation results to extract path fragmentation fragments, sequential fragmentation fragments, and reference fragmentation fragments. Merge and aggregate the path fragments, the sequential fragments, and the reference fragments to form the shell fragmentation clue set.

4. The device abnormality recognition method for an industrial intelligent robot according to claim 1, characterized by, The set of cleavage clues surrounding the shell fracture involves performing neighborhood expansion and historical relationship introduction processing on the fractured regions to implement shell shrinkage and form an unclosed set, specifically including: Each fracture clue in the shell fracture clue set is subjected to fracture region localization processing, and fracture clues pointing to the same target observation area, the same structural boundary or the same observation morphology breakpoint are merged to form a fracture region set; A neighborhood expansion framework is constructed around each fracture region in the set of fracture regions, and the directly adjacent target observation region and the second adjacent target observation region are determined based on the neighbor region splicing relationship in the initial normal shell to form a multi-layer neighborhood structure. Within the neighborhood expansion framework, the temperature distribution pattern, vibration propagation pattern, sound source residence pattern, and structural boundary pattern in the neighborhood target observation area are introduced into the interpretation system of the fractured region to form a neighborhood introduction result. Based on the historical observation records of the target device, historical observation segments that match the fractured region in terms of target observation area location, structural boundary relationship and operating conditions are selected, and the corresponding temperature distribution pattern, vibration propagation pattern, sound source residence pattern and structural boundary pattern are extracted as historical shell segments and introduced into the fractured region to form historical introduction results. The neighboring introduction results and the historical introduction results are subjected to inward shrinking and shell filling processing to make the various introduced observation patterns converge into the interior of the fractured region in order to reconstruct the interpretation closure relationship; During the internal shrinking and shell-filling process, a closure determination process is performed on each of the fractured regions. Fractured regions that can restore the continuous interpretation link are marked as closed regions, and fractured regions that cannot restore the continuous interpretation link are marked as shell-filling failure regions. The failed shell-filling regions are summarized and categorized according to the location of the target observation area and the type of observation morphology to form the unclosed set. 5.The device anomaly identification method for an industrial intelligent robot according to claim 1, wherein For the unclosed set, multi-condition edge-fitting observation and position stability tracking are performed on each unclosed region to form an abnormal kernel candidate set, specifically including: Boundary unfolding processing is performed on each unclosed region in the unclosed set to determine the unclosed boundary, the inner boundary region, and the outer boundary region, and to form a boundary structure; An edge-fitting observation channel is constructed around the boundary structure, the edge-fitting observation channel including an inner edge-fitting channel, a boundary fitting channel and an outer edge-fitting channel; The industrial intelligent robot is controlled to perform multi-condition edge observation along the edge observation channel to form multiple rounds of edge observation results; Boundary accompaniment extraction processing is performed on the results of multiple rounds of edge observation to obtain the temperature distribution pattern, vibration propagation pattern, sound source residence pattern and structural boundary pattern that appear synchronously with the unclosed boundary and extend along the unclosed boundary, and to form a boundary accompaniment record; Based on the boundary accompanying record, position stability tracking processing is performed on each unclosed region to obtain the position stability tracking result; Perform cross-condition correspondence verification processing on the position stability tracking results to determine the attachment position consistency level of each unclosed region under multi-condition edge observation; Based on the consistency level of the attachment position, a stability convergence determination process is performed, and the unclosed regions that meet the stability convergence conditions are extracted as abnormal kernel candidates, forming the abnormal kernel candidate set. 6.The device anomaly identification method for an industrial intelligent robot according to claim 1, wherein Based on the set of abnormal kernel candidates, multiple sources of normal shell fragments are introduced to perform shell-borrowing and backfilling processing on each abnormal kernel candidate to form an abnormal unpacking region, specifically including: Candidate deconstruction processing is performed on each abnormal kernel candidate in the abnormal kernel candidate set to map the boundary attachment position, dwelling center position, propagation start position and corresponding observation pattern back to the initial normal shell to form a structural mapping relationship; A backfilling reference framework is constructed around each of the aforementioned abnormal kernel candidates. Homologous normal shell fragments are extracted from the initial normal shell, historical normal shell fragments are extracted from historical observation records, and analogous normal shell fragments are extracted from adjacent functional units. Alignment introduction processing is performed on the homologous normal shell fragment, the historical normal shell fragment, and the analogous normal shell fragment, respectively, so that each source normal shell fragment is mapped to the corresponding abnormal kernel candidate; Based on the normal shell fragments from each source, the abnormal kernel candidate is subjected to shell backfilling processing so that the structural boundary morphology, vibration propagation morphology, temperature distribution morphology and sound source residence morphology can be reconstructed and interpreted within the abnormal kernel candidate region to reconstruct and interpret the closure relationship. During the shell backfilling process, a backfilling closure judgment process is performed. Abnormal kernel candidates that can restore the continuous interpretation link are marked as backfillable regions, and abnormal kernel candidates that cannot restore the continuous interpretation link under the action of normal shell fragments from multiple sources are marked as backfilling failure regions. The areas where backfilling failed are summarized and marked as abnormal unpacking areas. 7.The device anomaly recognition method for an industrial intelligent robot according to claim 1, wherein The process of performing outer edge splicing processing on the boundary between the abnormal unshelled area and the normal shell layer to form an abnormal growth contour, and outputting the device anomaly identification result based on the abnormal growth contour, specifically includes: Perform boundary extraction processing on the abnormal shelling region to obtain the shelling boundary zone between the abnormal shelling region and the normal shell layer; The desquamation boundary zone is segmented to form multiple boundary segments; An outer edge revisit framework is established around each of the aforementioned boundary segments, and the industrial intelligent robot is controlled to perform outer edge revisit data collection at different revisit times and under different working conditions along the edge revisit path, the cross-edge revisit path, and the edge revisit path, so as to form a boundary revisit record. Based on the boundary revisit records, the stability determination process is performed on each boundary segment to obtain stable segments and wandering segments. The stable edge segments are spliced ​​together, and local contour segments are formed based on structural continuity, observation morphology continuity, and revisit timing. The local contour segments are subjected to contour extension and recognition processing and overall closure determination processing to form abnormal growth contours; Based on the abnormal growth contour, contour attribute annotation processing is performed, and the device anomaly identification result is output according to the target observation area range, boundary attachment position and contour status corresponding to the abnormal growth contour.

8. A device anomaly identification system for industrial intelligent robots, characterized in that, The system is used to execute the equipment anomaly identification method for industrial intelligent robots as described in any one of claims 1 to 7, wherein the system includes an acquisition module and a processing module, wherein... The acquisition module is used to acquire multiple sets of observation records formed by the target device under the conditions of multi-path inspection and multi-state acquisition of industrial intelligent robot, and to perform fusion and inclusion processing on various observation forms in the multiple sets of observation records in order to construct an initial normal shell. The processing module is used to perform observation path rearrangement, observation order rearrangement, and neighborhood reference reconstruction processing on multiple sets of observation records based on the initial normal shell, to form a shell rupture clue set. The processing module is also used to perform neighborhood expansion and historical relationship introduction processing on the fracture region around the shell fracture clue set to implement inward shell filling and form an unclosed set; The processing module is also used to perform multi-condition edge observation and position stability tracking processing on each unclosed region of the unclosed set to form an abnormal kernel candidate set; The processing module is also used to introduce normal shell fragments from multiple sources to perform shell backfilling processing on each abnormal kernel candidate based on the abnormal kernel candidate set, so as to form an abnormal unpacking region. The processing module is also used to perform outer edge splicing processing on the boundary between the abnormal desquamation area and the normal shell layer to form an abnormal growth profile, so as to output the device abnormality identification result based on the abnormal growth profile.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, comprising: The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images