Crane hoisting mechanism fault detection method and system based on space-time attention

Abstract

This invention relates to the field of fault data identification technology. It discloses a method and system for fault detection of tower crane hoisting mechanisms based on spatiotemporal attention. The method includes collecting tower crane operation signal data in multi-tower collaborative hoisting scenarios; constructing a synchronous temporal framework based on the spatial arrangement of tower cranes; performing temporal alignment and partitioning of the tower crane operation signal data; extracting and structured encoding the cross-influence interval of multi-tower signals based on the synchronous temporal dataset to generate an inter-tower coupling dataset; constructing a signal topology model using the inter-tower coupling dataset as input and the synchronous temporal framework; outputting a tower crane topology dataset; establishing a spatiotemporal attention framework based on the tower crane topology dataset; performing joint feature mapping on temporal and spatial feature data to generate a tower crane feature dataset; suppressing cross-tower interference, untangling speed, and current coupling distortion; determining the fault detection result based on the tower crane feature dataset; and maintaining independent and stable fault detection boundaries.

Description

Technical Field

[0001] This invention relates to the field of fault data identification technology, and more specifically, to a method and system for fault detection of tower crane hoisting mechanisms based on spatiotemporal attention. Background Technology

[0002] In multi-tower collaborative hoisting environments, tower crane hoisting mechanisms operate in a state of parallel multi-source signal monitoring and dynamic load control. The system needs to complete the acquisition and comparison of multiple types of signals such as current, speed, and displacement under limited computing power and communication latency. Existing technologies mostly adopt approaches such as local feature judgment, transform domain characterization, correlation measurement, and time smoothing screening to achieve fault detection and status identification. The above methods usually rely on the premise of complete signal sampling, single interference source, and stable electromagnetic environment. When the operating environment meets the conditions of high signal-to-noise ratio, smooth load change, and accurate node synchronization, basic detection effect can be maintained.

[0003] In multi-tower parallel construction scenarios, electromagnetic radiation generated by the motors and control units of adjacent towers can cause frequency domain distortion of the hoisting mechanism's sensing signals. At the same time, spatial coupling between tower crane structures causes mutual inductance noise to propagate in multiple signal channels. These unstable factors together lead to coupling distortion between speed and current characteristics, causing overlap and drift in the boundary range of fault detection results. Therefore, the technical problem to be solved is: how to ensure the independence of the fault detection boundary under the conditions of the hoisting mechanism signal being affected by electromagnetic interference from adjacent towers and the mutual coupling of signal characteristics under the state of multi-tower coordinated operation.

[0004] In view of this, the present invention proposes a method and system for fault detection of tower crane hoisting mechanism based on spatiotemporal attention to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, the present invention provides a method and system for fault detection of tower crane hoisting mechanisms based on spatiotemporal attention.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Firstly, a fault detection method for tower crane hoisting mechanisms based on spatiotemporal attention is provided, including: Collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronous timing framework based on the spatial arrangement of tower cranes, and perform timing alignment and partitioning of tower crane operation signal data to obtain a synchronous timing dataset; Based on the synchronous time-series dataset, the cross-influence between signals from multiple towers is extracted and structured to generate an inter-tower coupling dataset. Using the inter-tower coupling dataset as input, a signal topology model is constructed by combining a synchronous timing framework. The correlation attributes between nodes in the signal topology model are expressed in a structured manner to generate a tower crane topology dataset. A spatiotemporal attention framework is established based on the tower crane topology dataset. Based on the spatiotemporal attention framework, joint feature mapping is performed on the temporal and spatial feature data of the tower crane operation signal data to generate a tower crane feature dataset. The fault detection results under the multi-tower collaborative operation state are determined based on the tower crane feature dataset.

[0007] Secondly, a fault detection system for tower crane hoisting mechanisms based on spatiotemporal attention is provided, which is used to implement the aforementioned fault detection method for tower crane hoisting mechanisms based on spatiotemporal attention, including: Synchronization Alignment Module: Used to collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronization timing framework based on the spatial arrangement of tower cranes, perform timing alignment and partitioning of tower crane operation signal data, and obtain a synchronization timing dataset; Coupled coding module: Based on the synchronous time-series dataset, it is used to extract and structure the cross-influence between signals from multiple towers and generate an inter-tower coupled dataset. Topology modeling module: Used to construct a signal topology model by taking the inter-tower coupling dataset as input and combining it with a synchronous timing framework, to express the relationship attributes between nodes in the signal topology model in a structured way, and to generate a tower crane topology dataset; Spatiotemporal determination module: It is used to establish a spatiotemporal attention framework based on the tower crane topology dataset, perform joint feature mapping on the temporal and spatial feature data of the tower crane operation signal data based on the spatiotemporal attention framework, generate a tower crane feature dataset, and determine the fault detection results under the multi-tower collaborative operation state based on the tower crane feature dataset.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention collects tower crane operation signal data in multi-tower collaborative hoisting scenarios and constructs a synchronous timing framework based on the spatial arrangement of tower cranes. It performs timing alignment and partitioning on the tower crane operation signal data to obtain a synchronous timing dataset, enabling fault determination to be based on a unified time caliber and spatial partitioning. Based on the synchronous timing dataset, it extracts intervals and structures the cross-influence between multi-tower signals to generate an inter-tower coupling dataset, explicitly separating the coupling components caused by electromagnetic radiation from neighboring towers and structural mutual inductance. Using the inter-tower coupling dataset as input, it constructs a signal topology model in conjunction with the synchronous timing framework, and structures the node association attributes to generate a tower crane topology dataset, distinguishing between real coupling and pseudo-correlation between channels. Based on the tower crane topology dataset, it establishes a spatiotemporal attention framework, performing joint feature mapping on the temporal and spatial feature data of the tower crane operation signal data to generate a tower crane feature dataset. The spatiotemporal attention suppresses cross-tower interference and enhances single-tower source features under topological constraints. Based on the tower crane feature dataset, it determines the fault detection results under multi-tower collaborative operation conditions. The coupling distortion of speed and current features is untangled, and the fault detection boundary remains independent and does not drift with the operating conditions of neighboring towers. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the fault detection method for tower crane hoisting mechanism based on spatiotemporal attention in this invention. Figure 2 This is a schematic diagram of the tower crane hoisting mechanism fault detection system based on spatiotemporal attention in this invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as “comprising” or “including” mean that the element or object preceding the term covers the element or object listed after the term and its equivalents, without excluding other elements or objects. Terms such as “connection” or “linked” are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0012] In the technical solution of this invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data are all carried out in accordance with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. They do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0013] Example 1 Figure 1 This disclosure illustrates a method for fault detection of tower crane hoisting mechanisms based on spatiotemporal attention, provided in at least one embodiment, including: S10: Collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronous timing framework based on the spatial arrangement relationship of tower cranes, perform timing alignment and partitioning of tower crane operation signal data, and obtain a synchronous timing dataset. Methods for constructing a synchronous temporal framework based on the spatial arrangement of tower cranes include: Extract tower crane identification marks and spatial coordinates from the spatial arrangement of tower cranes, and compile spatial coordinate records according to tower crane identification marks to obtain an arrangement table; Extract the spatial connection relationships between tower crane identifiers from the layout table, and write the spatial connection relationships into the connection record set to obtain the connection table; The tower crane identifier and sampling time are extracted from the tower crane operation signal data, and the sampling chain is formed by sorting the sampling times to obtain the time sequence table; Using the connection table as a reference condition, filter the same segment sampling records of adjacent tower cranes in the time series table, and register the same segment sampling records as reference pairs to obtain the reference set; A time mapping table is generated based on the reference set, and the time mapping table is associated with the tower crane identifier and written into the frame record to obtain the synchronous time sequence frame.

[0014] In this embodiment, the processing objective in the above steps is to organize the operating signals of each tower crane in a multi-tower collaborative hoisting scenario under the same time reference, so that subsequent comparison and zoning are only carried out within the scope of "adjacent tower cranes," thereby avoiding the amplification of the coupling propagation range due to the mis-spontaneous splicing of samples from unrelated tower cranes into the same signal segment. The "arrangement table" is a set of records that aggregate spatial coordinates using tower crane identifiers as keys; the "connection table" is a set of tower crane identifier pairs formed by spatial connection relationships; the "time sequence table" is a set of sampling chains organized by tower crane identifiers and sorted by sampling time; and the "comparison set"... It is a set of comparison pairs of adjacent tower cranes within the same sampling segment. The "time mapping table" is a set of mapping records that align the sampling times of adjacent tower cranes to a unified time axis. For example, the set of tower crane identifiers is {T1,T2,T3}, the set of spatial coordinates is {(T1,(0,0)),(T2,(30,0)),(T3,(30,40))}, and the set of spatial connection relationships is {(T1,T2),(T2,T3)}. This set is used to limit the timing comparison to only the two sets of tower cranes (T1,T2) and (T2,T3) and generate a synchronous timing framework.

[0015] In this embodiment, when constructing the synchronous timing framework, the "tower pairs that need to be aligned" are first fixed by the layout table and the connection table, and then the sampling chain of each tower crane is fixed by the timing table. Subsequently, the connection table is used as a comparison condition to filter the same segment sampling records of adjacent tower cranes. The so-called "same segment sampling record" refers to the sampling time chains of two adjacent tower cranes having valid records at the same time within a certain continuous time period and corresponding time order, thereby allowing the establishment of a stable time mapping table and writing the mapping results into the framework record. For example, if the sampling time chain for T1 is {10:00:00, 10:00:01, 10:00:02, 10:00:03} and the sampling time chain for T2 is {10:00:01, 10:00:02, 10:00:03, 10:00:04}, then the reference set can be registered as {((T1,T2),(10:00:01↔10:00:01),(10:00:02↔10:00:02),(10:00:03↔10:00:03))}. This reference set is used to generate a time mapping table and form a synchronous timing framework, so that the subsequent timing alignment and partitioning of tower crane operation signal data can output a synchronous timing dataset under a unified time reference.

[0016] In this embodiment, when performing time-series alignment and partitioning of tower crane operation signal data based on the synchronous timing framework, a time mapping table is used to cut the speed signal segments and current signal segments of adjacent tower cranes to the same sampling segment boundary and write them into the synchronous timing dataset. Thus, in the context of multiple towers operating in parallel and electromagnetic interference between adjacent towers, the "signal segments within the same sampling segment of adjacent tower cranes" are used as the input basis for subsequent extraction of inter-tower coupling segments and topology modeling. This allows the determination of fault detection boundaries to be based on comparable synchronous segments, reducing the risk of speed and current characteristic coupling distortion caused by misalignment of sampling times of each tower spreading to non-adjacent tower pairs.

[0017] The method for selecting the same segment sampling records of adjacent tower cranes in the time series table using the connection table as a reference condition, and registering the same segment sampling records as reference pairs to obtain the reference set includes: The sampling chains are grouped according to the tower crane identification in the time sequence table, and the sampling segments are divided according to the sampling time to obtain the segment table; Extract the start and end times of the sampling segments from the segmentation table, and write the start and end times of the segments into the segment boundary record to obtain the segment boundary table. The segment boundary table is retrieved using the tower crane identifier pairs defined in the join table as an index, and the segment boundary records corresponding to the tower crane identifier pairs are paired and aggregated to obtain candidate pairs. Perform time overlap test on the segment boundary records in the candidate pairs, and register the segment boundary records that pass the test as the same segment sampling records to obtain the same segment set; Write the tower crane identification pairs and the sampling records of the same segment into the comparison pair record, and output the comparison set.

[0018] In this embodiment, the same-segment sampling record is used to limit the sampling segments that can be compared within the same continuous time range of adjacent tower cranes, thereby limiting the subsequent processing to the comparable time period to avoid misjudging the frequency domain distortion caused by electromagnetic interference from neighboring towers as a fault of the hoisting mechanism itself. The segment table is used to record the continuous sampling segments after the sampling chain of each tower crane is divided, the segment boundary table is used to record the start time and end time of each sampling segment, the candidate pair is used to collect the segment boundary record pairs corresponding to adjacent tower cranes, the same-segment set is used to store the segment boundary record pairs that have passed the time overlap test, and the comparison set is used to bind the tower crane identification pair with its same-segment sampling record to form a comparison pair for subsequent time mapping table generation.

[0019] In this embodiment, the purpose of grouping the sampling chain according to the tower crane identifier and dividing the sampling segment is to separate "continuous available sampling" from "sampling gaps," so that the segment boundary table can express the effective sampling coverage range of each tower crane on the time axis. Then, using the tower crane identifier pairs defined by the connection table as indexes, segment boundary pairing and time overlap checks are performed only on adjacent tower cranes to avoid meaningless segment boundary combinations for non-adjacent tower cranes, which would expand the comparison range. The purpose of the time overlap check is to confirm that the effective sampling segments of two adjacent tower cranes have a common time interval, so that speed signal segments and current signal segments can be trimmed within this common interval while maintaining the corresponding time order.

[0020] In this embodiment, the set of tower crane identifiers {T1,T2,T3} and the set of spatial connection relationships {(T1,T2),(T2,T3)} are used, and the example of (T1,T2) in the sampling time chain is used; The sampling time chain T1 is {10:00:00, 10:00:01, 10:00:02, 10:00:03}, and the sampling time chain T2 is {10:00:01, 10:00:02, 10:00:03, 10:00:04}. This forms sampling segments and segment boundary records, and completes candidate pair pairing and time overlap verification to obtain the same-segment sampling records and the control set. Specifically, it can be represented by the following table. In the table, "Sampling Segment" corresponds to the segment table, "Segment Start Time / Segment End Time" corresponds to the segment boundary table, and "Candidate Pair / Common Interval" corresponds to the candidate pair and the same-segment sampling records. The control set is used to limit the mapping to only occur within the common interval when generating the time mapping table later and output the control pair records.

[0021] Table 1 Table 2 In this embodiment, the reference range of adjacent tower cranes is fixed as "sampling records of the same segment within the common interval" by the reference set, so that the subsequent extraction of the interval of cross-influence between multiple tower signals is based on the segment covered by the same time. In the context of multiple towers operating in parallel and the existence of electromagnetic radiation and spatial coupling between adjacent towers, the risk of speed and current characteristic coupling distortion caused by sampling gaps or sampling time misalignment is amplified to outside the segment is reduced, and the determination of fault detection boundary has a time boundary that can be compared.

[0022] The method for retrieving the segment boundary table using tower crane identifier pairs defined in the join table as an index, and then pairing and aggregating the segment boundary records corresponding to the tower crane identifier pairs to obtain candidate pairs includes: Read the tower crane identifier pairs from the join table and generate index entries based on the tower crane identifier pairs to obtain the join index; Read the tower crane identifier, segment start time and segment end time from the segment boundary table, and generate segment index entries according to the tower crane identifier to obtain the segment index table; The tower crane identifier in the segment index table is matched with the concatenation index, and the start and end times of the matched segment are extracted to obtain the matched segment. The matching segments are grouped into segment boundary record pairs according to the tower crane identification pairs, and the segment boundary record pairs are written into the candidate pair record to obtain the candidate pairs.

[0023] In this embodiment, the crane identifier pairs defined by the join table are used to fix the "objects that need to be paired" first, so that the retrieval scope of the segment boundary table converges from all cranes to adjacent crane pairs, thereby ensuring that the pairing action of segment boundary records falls on the verifiable index matching, avoiding the introduction of irrelevant time intervals due to cross-tower mixing of segment boundary records in multi-tower collaborative scenarios. The role of the join index is to organize crane identifier pairs such as (T1,T2) and (T2,T3) into directly searchable key values. The role of the segment index table is to organize segment boundary records such as (T1, segment start time, segment end time) into searchable items with crane identifiers as keys. After the join index and the segment index table complete the matching, the matching segment is obtained. The matching segment is then combined according to the crane identifier pairs, and candidate pairs are thus formed and used as input for subsequent time overlap verification.

[0024] For example, using the previous example, the set of tower crane identifier pairs {(T1,T2),(T2,T3)} can be read from the join table and written into the join index {(T1,T2),(T2,T3)}; From the segment boundary table, the tower crane identifier and the segment start time and end time can be read to form a segment index table {(T1,(10:00:00,10:00:03)),(T2,(10:00:01,10:00:04))}. When connecting the indexes to match the segment index table, for the tower crane identifier pair (T1,T2), the matching segment {(T1,10:00:00,10:00:03),(T2,10:00:01,10:00:04)} can be extracted and combined into a single segment. The segment boundary record pair ((T1, 10:00:00, 10:00:03), (T2, 10:00:01, 10:00:04)) is written into the candidate pair record to obtain the candidate pair {((T1, 10:00:00, 10:00:03), (T2, 10:00:01, 10:00:04))}. This candidate pair maintains a one-to-one correspondence between the tower crane identification pair and the segment boundary record, which facilitates the subsequent generation of a reference set according to the common interval and supports the generation of the time mapping table.

[0025] S20: Based on the synchronous time-series dataset, the cross-influence between signals from multiple towers is extracted and structured to generate an inter-tower coupling dataset. Based on synchronous time-series datasets, methods for generating inter-tower coupling datasets by performing interval extraction and structured encoding of the cross-influence between multi-tower signals include: Extract the tower crane identifier pairs and time mapping table from the synchronous timing framework, and collect the time mapping table records according to the tower crane identifier pairs to obtain the tower crane pair table; Speed ​​and current signal segments corresponding to each tower crane identifier are extracted from the synchronous time-series dataset and sorted according to the sampling time to form a signal chain, thus obtaining a signal table; Using the tower pair table as an index, the signal table is retrieved and the speed signal segment and the current signal segment are paired to obtain the paired segment group. Based on the paired segment group, the cross-influence section is divided to obtain the influence section table. Extract the start and end times of the cross-influenced sections from the affected section table, and combine the start and end times of the sections with the tower crane identification pair into the coding record to obtain the coupling code table. The coupling code table is compiled into a collection of coded records according to the tower crane identification, and the compiled coded records are registered as an inter-tower coupling dataset to obtain the inter-tower coupling dataset.

[0026] In this embodiment, the cross-influence between multiple tower signals originates from electromagnetic radiation introduced by the motors and control units of adjacent towers, as well as spatial coupling between tower crane structures. In speed signal segments and current signal segments, this manifests as alignment of state switching points within the same time interval, synchronous reversal of polarity signs, or simultaneous occurrence of local distortions. Therefore, the synchronous timing framework first provides a tower crane identification pair and time mapping table, fixing the comparable time range and forming a tower pair table. Then, speed signal segments and current signal segments are extracted from the synchronous timing dataset according to the tower crane identification and organized into signal chains according to the sampling time to form a signal table. This allows subsequent retrieval and pairing to be performed on the same time axis. Subsequently, the signal table is paired using the tower pair table as an index to obtain paired segment groups. Within the paired segment groups, cross-influence segments are identified according to the common interval defined by the same segment sampling record to form an influence segment table. The influence segment table extracts the segment start time and segment end time and combines them with the tower crane identification pair to write into a coupling code table. The coupling code table is collected according to the tower crane identification pair and registered as an inter-tower coupling dataset, thereby solidifying "when the cross-influence occurs and between which pair of adjacent tower cranes" into an coded record that can be directly used for subsequent topology modeling.

[0027] In this embodiment, to facilitate a clear and intuitive view of the data connection relationships between the "tower pair table, signal table, affected segment table, coupling code table, and inter-tower coupling dataset"; Using the tower crane identifier set {T1,T2,T3} and the spatial connection relationship set {(T1,T2),(T2,T3)}, and the reference set {((T1,T2),(10:00:01,10:00:03))}, the tower crane identifier pairs (T1,T2) and their time mapping table records read from the synchronous time series framework are compiled into a tower pair table. At the same time, speed signal segments and current signal segments within the common interval are extracted from the synchronous time series dataset and formed into signal chains according to the sampling time, which are registered into a signal table. Then, the signal table is retrieved using the tower pair table index to form paired segment groups. Based on the paired segment groups, 10:00:02 to 10:00:03 is registered as an inter-influence segment and written into the influence segment table. Furthermore, the segment start time and segment end time are combined with the tower crane identifier pairs and written into the coupling code table and compiled into an inter-tower coupling dataset. This ensures that when the inter-tower coupling dataset is subsequently mapped to node connection edges, the interval and tower pair source are traceable.

[0028] Table 3 The method of retrieving signal tables using tower pairing tables as indexes and pairing velocity signal segments with current signal segments to obtain paired segment groups, and then dividing cross-influence zones based on these paired segment groups to obtain an influence segment table, includes: Read tower crane identifier pairs from the tower crane pair table and generate search terms. At the same time, sort the search terms according to the tower crane identifier pairs to form a search chain and obtain the search table. Read the tower crane identification and speed signal segments from the signal table, and simultaneously read the tower crane identification and current signal segments from the signal table, and write them into the segment index record according to the tower crane identification to obtain the segment index table; The tower crane identifier pairs in the segment index table are matched with the search table, and the matched speed signal segments and current signal segments are paired and aggregated according to the sampling time to obtain paired segment groups. Calculate the first-order difference sequence of velocity signal segments in the paired segment group, compare the first-order difference sequence with the preset acceleration threshold to identify the running state switching point, generate the initial segment boundary by combining the polarity sign of the current signal segment, and remove the small segment boundary with a time length less than the preset dejitter window to obtain the segment boundary table. The segment boundary records in the segment boundary table are combined with the corresponding tower crane identifiers and written into the affected segment records. The affected segment records are then aggregated to generate the affected segment table.

[0029] In this embodiment, the above method is used to pair the tower crane identification pairs defined by the tower pair table with the speed signal segments and current signal segments in the signal table on the same sampling time axis. After pairing, a segment boundary table of the cross-influenced segment is generated from the two clues of "speed change segment boundary" and "current polarity indication". The segment boundary table is then organized into an influence segment table, thereby constraining the coupling distortion caused by "frequency domain distortion caused by electromagnetic radiation from neighboring towers and the superposition and propagation of mutual inductance noise" in the background technology within a locatable time interval. This facilitates the subsequent mapping of this interval into the encoded record of the inter-tower coupling dataset and further construction of the signal topology model.

[0030] In this embodiment, the process of reading the tower pair table and forming a retrieval table is used to transform the tower crane identification pairs from a set to a retrieval chain that can be traversed one by one. This ensures that subsequent segment retrieval of the signal table has a fixed index entry and avoids full scans without indexes in multi-tower operation scenarios. The process of reading the signal table and generating a segment index table is used to register speed signal segments and current signal segments as segment index records according to tower crane identification, so that speed signal segments and current signal segments with the same tower crane identification can be hit by the same index and enter the pairing process. The process of matching the retrieval table with the segment index table and generating paired segment groups is used to align and aggregate speed signal segments and current signal segments under the same tower crane identification pair in the sampling time dimension, so that subsequent segment boundary identification no longer relies on single-channel features but is oriented towards joint information of the same time slice.

[0031] In this embodiment, the processing action of generating a segment boundary table based on paired segment groups is used to separate the boundaries of the cross-influenced segments from the continuous signal. The first-order difference sequence of the velocity signal segment is used to characterize the velocity change amplitude between adjacent sampling times. The preset acceleration threshold is used to screen out candidate switching points of velocity change from general fluctuations. The polarity sign of the current signal segment is used to supplement the constraint on the candidate switching points to exclude velocity jumps caused only by noise. The jitter window is used to remove small segment boundaries with short durations that are difficult to correspond to the actual working condition switching, so that the segment boundary record is more in line with the actual operating rhythm of multi-tower collaborative hoisting and facilitates the subsequent collection of segment boundary records into an influence segment table.

[0032] For example, using a record from the aforementioned tower crane pairing table, the tower crane identification pairs are T1-T2, with the time mappings as follows: 10:00:01 corresponds to 10:00:01, 10:00:02 corresponds to 10:00:02, and 10:00:03 corresponds to 10:00:03. Simultaneously, using the two signal chains corresponding to T1-T2 from the signal table, the speed signal chains are (10:00:01, 1.2m / s) and (10:00:02, 1.1m / s) respectively. / s), (10:00:03, 0.6m / s), the current signal chains are (10:00:01, 18A), (10:00:02, 19A), (10:00:03, 27A), then the T1-T2 are read from the tower matching table to form a search item, and the speed signal chain and current signal chain corresponding to the same tower crane identifier are located in the segment index table. Then, the two are aligned and grouped into a paired segment group according to the sampling time, and the records are (10:00:03, 0.6m / s). The speed signals (0:01, 1.2 m / s, 18 A), (10:00:02, 1.1 m / s, 19 A), and (10:00:03, 0.6 m / s, 27 A) were used to calculate the first-order difference of the velocity signal chain, resulting in (10:00:02, -0.1 m / s) and (10:00:03, -0.5 m / s). These differences were compared with the acceleration threshold of 0.3 m / s to determine that the differential amplitude corresponding to 10:00:03 met the switching point identification criteria. Simultaneously, the polarity sign of the current signal chain at 10:00:03 is read and combined with the switching point record to form a segment boundary record (T1,T2,10:00:02,10:00:03). When the duration of the segment boundary record meets the retention condition at 1s in the de-jitter window, it is written into the segment boundary table. The segment boundary table is then aggregated according to the tower crane identification to generate an influence segment table and output the cross-influence segment (10:00:02,10:00:03) corresponding to T1-T2.

[0033] S30: Using the inter-tower coupling dataset as input, a signal topology model is constructed by combining the synchronous timing framework. The correlation attributes between nodes in the signal topology model are expressed in a structured manner to generate a tower crane topology dataset. Methods for generating tower crane topology datasets by structurally representing the association attributes between nodes in a signal topology model include: Parse the inter-tower coupling dataset, map the encoded records in the inter-tower coupling dataset to node connection edges to construct a signal topology model, extract node identifier pairs and node connection records from the signal topology model, and aggregate the node connection records according to the node identifier pairs to obtain a connection table; Extract the node identifier and time mapping table from the synchronous timing framework, and write the mapping index record according to the node identifier to obtain the mapping table; Using the connection table as a reference condition, extract the sampling time chain corresponding to the node identifier pair in the mapping table, and extract the tower crane operation signal data according to the sampling time chain to obtain the segment table; In the fragment table, each node is identified as having a combination of velocity signal fragments and current signal fragments, and the velocity signal fragments and current signal fragments are written into the associated attribute record to obtain the attribute table; The connection table and attribute table are merged by node identifier, and the merged field results are written into the topology record set to obtain the tower crane topology dataset.

[0034] In this embodiment, the signal topology model is used to transcribe the already encoded cross-influence segments in the "inter-tower coupling dataset" into a searchable node connection structure. This allows for the subsequent extraction of velocity signal segments and current signal segments under the same time reference from the synchronization timing framework based on node identifiers. The two types of segments are then written as association attributes into the same topology record. This confines the coupling distortion caused by multi-tower electromagnetic interference to specific node connection edges and corresponding time periods, avoiding boundary overlap and drift caused by local signals from a single tower. Specifically, when parsing the inter-tower coupling dataset, each encoded record in the coupling code table is mapped to a node connection. The nodes are connected to form node identifier pairs and node connection records, and then the connection table is obtained by aggregating the node identifier pairs. At the same time, the node identifier and time mapping table is extracted from the synchronous timing framework and a mapping table is formed according to the node identifier. Then, the sampling time chain corresponding to the node identifier pair is read from the mapping table using the connection table as a reference condition, and the tower crane operation signal data is extracted to obtain the segment table. The speed signal segment and current signal segment under the same node identifier pair in the segment table are aggregated and written into the associated attribute record to form the attribute table. Finally, the connection table and the attribute table are merged according to the node identifier pair fields and written into the topology record set to obtain the tower crane topology dataset.

[0035] For example, using the aforementioned tower crane identifier set {T1, T2, T3} and spatial connection relationship set {(T1, T2), (T2, T3)}, and using a coded record (T1, T2, 10:00:02, 10:00:03) from the aforementioned inter-tower coupling dataset, a node connection edge record can be obtained after parsing (node ​​identifier pair is T1-T2; segment is 10:00:02 to 10:00:03) and written into the connection table; at the same time, using the time mapping table record of T1 and T2 in the synchronization time sequence framework (10:00:01→10:00:01; 10:00:02→10:00:02; 10:00:03→10:00:03) and writing it into the mapping table, under the constraint of the connection table, the sampling time chain (10:00:02; 10:00:03) corresponding to T1-T2 is read from the mapping table. The tower crane operation signal data is extracted based on the time interval 10:00:03 to obtain a segment table. The speed signal segment corresponding to T1-T2 in the segment table can be written as (10:00:02, 1.1m / s; 10:00:03, 0.6m / s), and the current signal segment can be written as (10:00:02, 19A; 10:00:03, 27A). Then, the above two types of segments are collected and written into the same associated attribute record to form an attribute table. Finally, the node connection records in the connection table and the associated attribute records in the attribute table are merged according to the node identifier to generate a topology record and registered as a tower crane topology dataset. This tower crane topology dataset can directly provide a consistent input caliber of "connection structure + same segment signal segment" when it is used to establish a spatiotemporal attention framework, so that the fault detection results can maintain distinguishable boundaries in the multi-tower collaborative operation state.

[0036] The method for extracting the sampling time chain corresponding to the node identifier pair from the mapping table using the connection table as a reference condition, and then extracting the tower crane operation signal data according to the sampling time chain to obtain the segment table includes: Read the node identifier pairs from the join table and generate a join index based on the node identifier pairs to obtain the join index table; Read the node identifier and time mapping table records from the mapping table, and generate a mapping index according to the node identifier to obtain the mapping index table; The node identifier is extracted from the node mapping table by matching the node mapping table with the node identification table. The valid sampling time of the time mapping table record is directly read to obtain the candidate set of time chains. In the candidate set of time chains, the sampling times are collected by node identifier, and the valid sampling times in the candidate set of time chains are directly arranged in chronological order to obtain the sampling time chain. Using the sampling time chain as an index, velocity signal records and current signal records are extracted from the synchronous time-series dataset, and written to the segment records according to the node identifier to obtain the segment table.

[0037] In this embodiment, the connection table is used to provide the node identifier pairs that need to be processed; the mapping table is used to provide the time mapping table record corresponding to each node identifier; the sampling time chain is used to organize the common available sampling times of the node identifier pairs into a time-series index that can be directly retrieved; the fragment table is used to store the velocity signal records and current signal records extracted from the synchronous time-series dataset according to the sampling time chain, which is convenient for subsequent aggregation in the attribute table to form associated attribute records. The above processing is used to implement the "common sampling interval that can be compared" to the signal interception action at the node identifier pair level in the scenario of multiple towers operating in parallel and electromagnetic interference, so that the subsequent structured expression of the associated attributes between nodes has a unified time-series entry and a traceable data source.

[0038] In this embodiment, reading the connection table and generating a connection index based on node identifier pairs extracts node identifier pairs from the connection record set as searchable index items, allowing subsequent matching operations to revolve around the node identifier pairs. Reading the mapping table and generating a mapping index based on node identifiers establishes an index relationship between node identifiers and time mapping table records, enabling subsequent quick location of valid sampling times based on node identifiers. Matching the connection index table with the mapping index table and reading valid sampling times synchronously obtains the sources of valid sampling times for both ends of the nodes under the constraint of node identifier pairs, forming a candidate set of time chains. The sampling times are then aggregated in the candidate set of time chains based on node identifier pairs and sorted by time. The sequential arrangement organizes available sampling times into a continuous and searchable chain of sampling times, avoiding unclear segment boundaries caused by time jumps during subsequent truncation. Extracting velocity signal records and current signal records using the sampling time chain as an index and writing them into the segment table involves writing multiple types of signal records corresponding to the common sampling times of node identifier pairs into the segment records using the same index, thereby ensuring that each record in the segment table can be traced back to the corresponding index item in the connection table and mapping table. The above steps enable the segment table to reflect the common sampling window of node identifier pairs and provide a consistent data truncation standard for subsequently writing velocity signal segments and current signal segments into the associated attribute records.

[0039] In this embodiment, in order to facilitate understanding the connection between the fragment table and the subsequent associated attribute records from the perspective of "data organization - index relationship - subsequent data retrieval path", after completing the construction description of the fragment table, it is necessary to summarize the field composition, index scope and corresponding purpose of each data object, so that the listed "data object - data content - formation method and purpose" can correspond to the connection table, mapping table, sampling time chain and fragment table in the aforementioned text description, as explained in Table 4 below.

[0040] Table 4 In this embodiment, the “Tower Pair Table, Connection Table, Mapping Table, Sampling Time Chain, and Segment Table” in Table 4 are organized in a connected manner according to the same node identifier. The sampling time chain fixes the comparable sampling times. The segment table writes the velocity signal record and the current signal record at the same sampling time into the same segment record. Based on this, when generating the attribute table later, the segment record can be directly read and associated attribute record can be formed, avoiding repeated time matching between the velocity signal record and the current signal record.

[0041] S40: Establish a spatiotemporal attention framework based on the tower crane topology dataset. Perform joint feature mapping on the temporal and spatial feature data of the tower crane operation signal data based on the spatiotemporal attention framework to generate a tower crane feature dataset. Determine the fault detection results under the multi-tower collaborative operation state based on the tower crane feature dataset.

[0042] In this embodiment, the tower crane topology dataset is a collection of topology records that merges the fields of the connection table and attribute table mentioned above. Its core function is to place node identifier pairs, node connection records, and corresponding speed signal segments and current signal segments in the same data object, which facilitates the simultaneous reading of time series information and node connection information under the same index caliber. The spatiotemporal attention framework can be understood as a model structure that uses time segments and node connections as input constraints. The time feature data corresponds to the sequence features formed by the arrangement of speed signal segments and current signal segments according to the sampling time, and the spatial feature data corresponds to the adjacency constraints formed by the connection relationship of node identifier pairs in the connection table. The two are simultaneously involved in feature extraction and fusion within the same framework, thereby avoiding boundary overlap and drift caused by the superposition and propagation of electromagnetic radiation from neighboring towers and mutual inductance noise when interpreting based solely on single tower or single channel signals.

[0043] In this embodiment, the joint feature mapping does not simply concatenate time series and spatial connections. Instead, it first generates an alignable set of input segments using node identifier pairs as a unified index. Then, within the model structure, it performs time vectorization representation on velocity signal segments and current signal segments according to sampling time. At the same time, it introduces the representation of adjacent node identifier pairs as spatial context according to node connection records, so that time vectors at the same sampling time can complete information aggregation and rearrangement within their adjacent range. This results in a tower crane feature dataset on the output side. Each record in the tower crane feature dataset still uses node identifier pairs as keys and carries the feature vector sequence on the corresponding sampling time chain to ensure continuity and consistency with the aforementioned segment table, attribute table, and connection table in terms of index relationship.

[0044] Methods for establishing a spatiotemporal attention framework based on tower crane topology datasets include: Extract node identifier pairs and associated attribute records from the tower crane topology dataset, and aggregate the associated attribute records by node identifier pairs to obtain the topology table; The node identifier and time mapping table are extracted from the synchronous timing framework, and the sampling time chain of node identifier pairs is extracted with the topology table as a reference condition to obtain the time chain list; Using the sampling time defined by the time-linked list as an index, extract velocity signal segments and current signal segments from the associated attribute records of the topology table to obtain the input segment set; Construct a model network structure that includes an attention mechanism, register the set of input fragments as the input tensor of the spatiotemporal attention framework and load it into the model network structure, and associate node identifiers with node identifier pairs and write them into the framework record to obtain the initialized spatiotemporal attention framework.

[0045] In this embodiment, the key to establishing a spatiotemporal attention framework based on the tower crane topology dataset lies in organizing the "node identifier pair - sampling time chain - velocity signal segment and current signal segment - node connection record" into a directly readable input object under the same index. This allows the subsequent model network structure to simultaneously obtain time series constraints and adjacency constraints without having to backtrack across tables to match relationships when reading in data. The topology table is used to carry the associated attribute records corresponding to the node identifier pair and maintains the same node connection scope as the connection table. The time chain list is used to carry the sampling time chain available for the node identifier pair under the synchronous temporal framework and maintains... The model maintains the same time mapping caliber as the mapping table. The input fragment set is used to carry velocity signal fragments and current signal fragments aligned on the same sampling time chain and maintains the same fragment recording caliber as the fragment table. The model network structure is used to carry the computation path of the attention mechanism and maintain the joint constraint relationship between the input fragment set and the node connection record. This enables the spatiotemporal attention framework to have the data entry conditions of "data acquisition aligned by sampling time and adjacency context acquisition by node connection" in the initialization stage, thereby providing stable data support for subsequent joint feature mapping and reducing the risk of feature mismatch caused by inconsistent indexes under multi-tower electromagnetic interference conditions.

[0046] In this embodiment, the purpose of extracting node identifier pairs and associated attribute records from the tower crane topology dataset and aggregating them to obtain a topology table is to group the associated attribute records scattered in the topology record set into directly indexable record units according to node identifier pairs. This enables the subsequent stable reading of speed signal segments and current signal segments using node identifier pairs as keys and synchronous acquisition of node connection records, avoiding the problem of inconsistent indexing caused by repeated retrieval of the topology record set during the model construction stage.

[0047] In this embodiment, the purpose of extracting the node identifier and time mapping table from the synchronous timing framework and extracting the sampling time chain under the condition of the topology table is to ensure that the data retrieval window of each node identifier pair is strictly limited to the common sampling interval allowed by the synchronous timing framework, thereby ensuring that each segment record in the input segment set can be aligned and arranged on the same sampling time chain, providing a sampling time sequence that can be directly traversed for the subsequent attention mechanism to expand the calculation by time step.

[0048] In this embodiment, the purpose of extracting velocity signal segments and current signal segments and forming an input segment set by using the sampling time defined by the time chain as an index is to organize multiple types of signal records under the same sampling time chain into a unified segment input unit, so that the model network structure can obtain the correspondence between velocity and current at the same time in a single reading, thereby maintaining a consistent entry of "same input unit at the same sampling time" at the time feature data level.

[0049] In this embodiment, the purpose of constructing the model network structure and registering the input fragment set as an input tensor and loading it is to first determine the computation path of the attention mechanism in the time dimension and node connection dimension, and then write the input fragment set as an input sequence consistent with the sampling time chain into the model entry point. At the same time, the node identifier and node identifier pair are written into the frame record to retain the mapping relationship of "input fragment - node identifier pair - node identifier", so that when generating the tower crane feature dataset and outputting the fault detection results in the future, the corresponding time segment and adjacency constraint source can still be traced back according to the node identifier pair.

[0050] Example 2 Please see Figure 2 As shown, based on the same inventive concept, this embodiment discloses a tower crane hoisting mechanism fault detection system based on spatiotemporal attention. For details not covered in this embodiment, please refer to the relevant sections of Embodiment 1. The system includes: Synchronization Alignment Module: Used to collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronization timing framework based on the spatial arrangement of tower cranes, perform timing alignment and partitioning of tower crane operation signal data, and obtain a synchronization timing dataset; Coupled coding module: Based on the synchronous time-series dataset, it is used to extract and structure the cross-influence between signals from multiple towers and generate an inter-tower coupled dataset. Topology modeling module: Used to construct a signal topology model by taking the inter-tower coupling dataset as input and combining it with a synchronous timing framework, to express the relationship attributes between nodes in the signal topology model in a structured way, and to generate a tower crane topology dataset; Spatiotemporal determination module: It is used to establish a spatiotemporal attention framework based on the tower crane topology dataset, perform joint feature mapping on the temporal and spatial feature data of the tower crane operation signal data based on the spatiotemporal attention framework, generate a tower crane feature dataset, and determine the fault detection results under the multi-tower collaborative operation state based on the tower crane feature dataset.

[0051] The accompanying drawings of the embodiments of this invention only involve the structures involved in the embodiments of this invention. Other structures can refer to the general design. In the absence of conflict, the features of the same embodiment and different embodiments of this invention can be combined with each other. The above are only specific implementations of this invention, but the protection scope of this invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this invention should be included within the protection scope of this invention. Therefore, the protection scope of this invention should be determined by the protection scope of the claims.

Claims

1. A method for fault detection of tower crane hoisting mechanism based on spatiotemporal attention, characterized in that, include: Collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronous timing framework based on the spatial arrangement of tower cranes, and perform timing alignment and partitioning of tower crane operation signal data to obtain a synchronous timing dataset; Based on the synchronous time-series dataset, the cross-influence between signals from multiple towers is extracted and structured to generate an inter-tower coupling dataset. Using the inter-tower coupling dataset as input, a signal topology model is constructed by combining a synchronous timing framework. The correlation attributes between nodes in the signal topology model are expressed in a structured manner to generate a tower crane topology dataset. A spatiotemporal attention framework is established based on the tower crane topology dataset. Based on the spatiotemporal attention framework, joint feature mapping is performed on the temporal and spatial feature data of the tower crane operation signal data to generate a tower crane feature dataset. The fault detection results under the multi-tower collaborative operation state are determined based on the tower crane feature dataset.

2. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 1, characterized in that, Methods for constructing a synchronous temporal framework based on the spatial arrangement of tower cranes include: Extract tower crane identification marks and spatial coordinates from the spatial arrangement of tower cranes, and compile spatial coordinate records according to tower crane identification marks to obtain an arrangement table; Extract the spatial connection relationships between tower crane identifiers from the layout table, and write the spatial connection relationships into the connection record set to obtain the connection table; The tower crane identifier and sampling time are extracted from the tower crane operation signal data, and the sampling chain is formed by sorting the sampling times to obtain the time sequence table; Using the connection table as a reference condition, filter the same segment sampling records of adjacent tower cranes in the time series table, and register the same segment sampling records as reference pairs to obtain the reference set; A time mapping table is generated based on the reference set, and the time mapping table is associated with the tower crane identifier and written into the frame record to obtain the synchronous time sequence frame.

3. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 2, characterized in that, The method for selecting the same segment sampling records of adjacent tower cranes in the time series table using the connection table as a reference condition, and registering the same segment sampling records as reference pairs to obtain the reference set includes: The sampling chains are grouped according to the tower crane identification in the time sequence table, and the sampling segments are divided according to the sampling time to obtain the segment table; Extract the start and end times of the sampling segments from the segmentation table, and write the start and end times of the segments into the segment boundary record to obtain the segment boundary table. The segment boundary table is retrieved using the tower crane identifier pairs defined in the join table as an index, and the segment boundary records corresponding to the tower crane identifier pairs are paired and aggregated to obtain candidate pairs. Perform time overlap test on the segment boundary records in the candidate pairs, and register the segment boundary records that pass the test as the same segment sampling records to obtain the same segment set; Write the tower crane identification pairs and the sampling records of the same segment into the comparison pair record, and output the comparison set.

4. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 3, characterized in that, The method for retrieving the segment boundary table using tower crane identifier pairs defined in the join table as an index, and then pairing and aggregating the segment boundary records corresponding to the tower crane identifier pairs to obtain candidate pairs includes: Read the tower crane identifier pairs from the join table and generate index entries based on the tower crane identifier pairs to obtain the join index; Read the tower crane identifier, segment start time and segment end time from the segment boundary table, and generate segment index entries according to the tower crane identifier to obtain the segment index table; The tower crane identifier in the segment index table is matched with the concatenation index, and the start and end times of the matched segment are extracted to obtain the matched segment. The matching segments are grouped into segment boundary record pairs according to the tower crane identification pairs, and the segment boundary record pairs are written into the candidate pair record to obtain the candidate pairs.

5. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 1, characterized in that, Based on synchronous time-series datasets, methods for generating inter-tower coupling datasets by performing interval extraction and structured encoding of the cross-influence between multi-tower signals include: Extract the tower crane identifier pairs and time mapping table from the synchronous timing framework, and collect the time mapping table records according to the tower crane identifier pairs to obtain the tower crane pair table; Speed ​​and current signal segments corresponding to each tower crane identifier are extracted from the synchronous time-series dataset and sorted according to the sampling time to form a signal chain, thus obtaining a signal table; Using the tower pair table as an index, the signal table is retrieved and the speed signal segment and the current signal segment are paired to obtain the paired segment group. Based on the paired segment group, the cross-influence section is divided to obtain the influence section table. Extract the start and end times of the cross-influenced sections from the affected section table, and combine the start and end times of the sections with the tower crane identification pair into the coding record to obtain the coupling code table. The coupling code table is compiled into a collection of coded records according to the tower crane identification, and the compiled coded records are registered as an inter-tower coupling dataset to obtain the inter-tower coupling dataset.

6. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 5, characterized in that, The method of retrieving signal tables using tower pairing tables as indexes and pairing velocity signal segments with current signal segments to obtain paired segment groups, and then dividing cross-influence zones based on these paired segment groups to obtain an influence segment table, includes: Read tower crane identifier pairs from the tower crane pair table and generate search terms. At the same time, sort the search terms according to the tower crane identifier pairs to form a search chain and obtain the search table. Read the tower crane identification and speed signal segments from the signal table, and simultaneously read the tower crane identification and current signal segments from the signal table, and write them into the segment index record according to the tower crane identification to obtain the segment index table; The tower crane identifier pairs in the segment index table are matched with the search table, and the matched speed signal segments and current signal segments are paired and aggregated according to the sampling time to obtain paired segment groups. Calculate the first-order difference sequence of velocity signal segments in the paired segment group, compare the first-order difference sequence with the preset acceleration threshold to identify the running state switching point, generate the initial segment boundary by combining the polarity sign of the current signal segment, and remove the small segment boundary with a time length less than the preset dejitter window to obtain the segment boundary table. The segment boundary records in the segment boundary table are combined with the corresponding tower crane identifiers and written into the affected segment records. The affected segment records are then aggregated to generate the affected segment table.

7. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 1, characterized in that, Methods for generating tower crane topology datasets by structurally representing the association attributes between nodes in a signal topology model include: Parse the inter-tower coupling dataset, map the encoded records in the inter-tower coupling dataset to node connection edges to construct a signal topology model, extract node identifier pairs and node connection records from the signal topology model, and aggregate the node connection records according to the node identifier pairs to obtain a connection table; Extract the node identifier and time mapping table from the synchronous timing framework, and write the mapping index record according to the node identifier to obtain the mapping table; Using the connection table as a reference condition, extract the sampling time chain corresponding to the node identifier pair in the mapping table, and extract the tower crane operation signal data according to the sampling time chain to obtain the segment table; In the fragment table, each node is identified as having a combination of velocity signal fragments and current signal fragments, and the velocity signal fragments and current signal fragments are written into the associated attribute record to obtain the attribute table; The connection table and attribute table are merged by node identifier, and the merged field results are written into the topology record set to obtain the tower crane topology dataset.

8. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 7, characterized in that, The method for extracting the sampling time chain corresponding to the node identifier pair from the mapping table using the connection table as a reference condition, and then extracting the tower crane operation signal data according to the sampling time chain to obtain the segment table includes: Read the node identifier pairs from the join table and generate a join index based on the node identifier pairs to obtain the join index table; Read the node identifier and time mapping table records from the mapping table, and generate a mapping index according to the node identifier to obtain the mapping index table; The node identifier is extracted from the node mapping table by matching the node mapping table with the node identification table. The valid sampling time of the time mapping table record is directly read to obtain the candidate set of time chains. In the candidate set of time chains, the sampling times are collected by node identifier, and the valid sampling times in the candidate set of time chains are directly arranged in chronological order to obtain the sampling time chain. Using the sampling time chain as an index, velocity signal records and current signal records are extracted from the synchronous time-series dataset, and written to the segment records according to the node identifier to obtain the segment table.

9. The tower crane hoisting mechanism fault detection method based on spatiotemporal attention according to claim 1, characterized in that, Methods for establishing a spatiotemporal attention framework based on tower crane topology datasets include: Extract node identifier pairs and associated attribute records from the tower crane topology dataset, and aggregate the associated attribute records by node identifier pairs to obtain the topology table; The node identifier and time mapping table are extracted from the synchronous timing framework, and the sampling time chain of node identifier pairs is extracted with the topology table as a reference condition to obtain the time chain list; Using the sampling time defined by the time-linked list as an index, extract velocity signal segments and current signal segments from the associated attribute records of the topology table to obtain the input segment set; Construct a model network structure that includes an attention mechanism, register the set of input fragments as the input tensor of the spatiotemporal attention framework and load it into the model network structure, and associate node identifiers with node identifier pairs and write them into the framework record to obtain the initialized spatiotemporal attention framework.

10. A tower crane hoisting mechanism fault detection system based on spatiotemporal attention, used to implement the tower crane hoisting mechanism fault detection method based on spatiotemporal attention as described in any one of claims 1-9, characterized in that, include: Synchronization Alignment Module: Used to collect tower crane operation signal data in multi-tower collaborative hoisting scenarios, construct a synchronization timing framework based on the spatial arrangement of tower cranes, perform timing alignment and partitioning of tower crane operation signal data, and obtain a synchronization timing dataset; Coupled coding module: Based on the synchronous time-series dataset, it is used to extract and structure the cross-influence between signals from multiple towers and generate an inter-tower coupled dataset. Topology modeling module: Used to construct a signal topology model by taking the inter-tower coupling dataset as input and combining it with a synchronous timing framework, to express the relationship attributes between nodes in the signal topology model in a structured way, and to generate a tower crane topology dataset; Spatiotemporal determination module: It is used to establish a spatiotemporal attention framework based on the tower crane topology dataset, perform joint feature mapping on the temporal and spatial feature data of the tower crane operation signal data based on the spatiotemporal attention framework, generate a tower crane feature dataset, and determine the fault detection results under the multi-tower collaborative operation state based on the tower crane feature dataset.

Images