Rotating machinery centering data low-latency edge processing method

Abstract

This invention belongs to the field of communication network data transmission and edge computing technology, specifically a low-latency edge processing method for centering data in rotating machinery. The method receives centering data frames, associates them with acquisition sequence number, transposition segment, acquisition timestamp, status version, and measurement quality identifier, and calculates the difference in centering state before and after data frame inclusion using a shadow solving process synchronized with the solution state, generating a solution sensitivity. It determines the semantic deadline by combining the remaining valid time of the transposition segment and the communication queue status, and encodes the data frames as anchor frames, residual incremental frames, or replaceable frames. These are then scheduled for transmission according to the solution sensitivity and semantic deadline, with high-resolution frames arriving later replacing unsent low-resolution replaceable frames. The edge side reconstructs the centering state based on the anchor frames and residual incremental frames, and inserts anchor frames when the accumulated reconstruction deviation meets the conditions. This invention reduces the time slot occupied by low-contribution data, ensures that critical data arrives within the valid period, and limits the accumulation of incremental reconstruction errors.

Description

Technical Field

[0001] This invention belongs to the field of communication network data transmission and edge computing technology, specifically a low-latency edge processing method for centering data of rotating machinery. Background Technology

[0002] Edge processing of centering data in rotating machinery typically employs a method of periodic acquisition, fixed message encapsulation, and edge-centralized solution. The acquisition side generates centering data frames according to a preset sampling period, writing the acquisition sequence number, acquisition timestamp, transposition segment, deviation component, and measurement quality identifier into the data frame before sending it to the edge nodes via the communication link. The sending side generally organizes the communication queue according to a first-in, first-out (FIFO) rule or sets static priorities based on message type. When the link load increases, data with closely spaced adjacent values ​​are often filtered using a fixed variable threshold, or a combination of periodic complete state frames and differential frames is used to reduce transmission volume. After receiving the messages, the edge nodes buffer and sort them according to the acquisition timestamp or arrival order, superimposing subsequent differential data based on the most recent complete state, and calculating the parallel deviation, angular deviation, correction direction, and corresponding correction amount. For lost, late, or out-of-order data, conventional solutions involve waiting for completion, requesting retransmission, expanding the buffer window, or discarding expired messages, and re-exercising the centering solution after obtaining the complete transposition segment data.

[0003] Communication scheduling is based on data generation order, fixed priority, or the magnitude of original numerical changes. This fails to characterize the marginal contribution of a single centering data frame to the current centering solution state and its remaining effective time, thus preventing the alignment of transmission resources with solution requirements. After the rotating machinery enters the fine-tuning phase, a large amount of data with similar numerical changes that do not alter the deviation direction, correction direction, or solution interval continues to enter the queue, occupying transmission time slots and prolonging the queuing time of subsequent messages. Data with minor numerical changes that could alter the centering state boundary may be filtered by a fixed threshold or exceed the validity period of the corresponding transposition segment after low-contribution data. Fixed-period insertion of complete state frames also fails to promptly truncate the state chain based on differential dependency length and reconstruction deviation changes. This makes it difficult for edge nodes to balance the arrival time of critical data with continuous reconstruction of the centering state during network load fluctuations, causing the edge processing latency of centering data to increase with queue backlog and affecting the output state due to expired data and accumulated reconstruction deviations. Summary of the Invention

[0004] The purpose of this invention is to provide a low-latency edge processing method for centering data of rotating machinery, which can solve the problems in the background art mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: a low-latency edge processing method for centering data of rotating machinery, comprising: receiving continuously generated centering data frames and associating each centering data frame with an acquisition sequence number, transposition segment, acquisition time stamp, status version, and measurement quality identifier; inputting each centering data frame into a shadow solving process synchronized with the formal solving state on the edge side, obtaining the centering state difference when including and excluding the corresponding data frame, and generating a solution sensitivity accordingly; generating a semantic deadline based on the solution sensitivity, the remaining effective time of the corresponding transposition segment, and the communication queue status, and encoding each centering data frame as an anchor frame, a residual increment frame, or a replaceable frame; scheduling the transmission order according to the semantic deadline and the solution sensitivity, and replacing the low solution sensitivity replaceable frames with the same status version that have not yet been sent in the queue with later-arriving frames of high solution sensitivity; reconstructing the centering state on the edge side based on the anchor frame and the residual increment frame, and inserting a new anchor frame when the accumulated reconstruction deviation meets the anchor triggering condition.

[0006] Preferably, the shadow solution process includes: reading the current state snapshot of the formal solution state and the state version corresponding to the current state snapshot; establishing an inclusion branch and an exclusion branch that share the current state snapshot for each pair of data frames; updating the parallel deviation, angle deviation, correction direction, and uncertainty interval using the corresponding pair of data frames in the inclusion branch; maintaining the current state snapshot and advancing the state according to the confirmed data within the same transposition segment in the exclusion branch; comparing the deviation component displacement, correction direction change, and solution uncertainty interval boundary displacement of the two branches; constraining the reliability of the comparison results according to the measurement quality identifier to form a multi-component solution sensitivity bound to the state version; and writing the multi-component solution sensitivity and the acquisition sequence number of the corresponding pair of data frames into the transmission scheduling record.

[0007] Preferably, the generation of the semantic deadline includes: establishing a queue state sequence for the communication queue carrying the alignment data frame, the queue state sequence including the number of bytes to be sent, the number of bytes dequeued per unit time, the message arrival interval, and historical queuing jitter; determining the data validity boundary based on the end time stamp and acquisition time stamp of the corresponding transposition segment, and predicting the candidate dequeue time stamp of the corresponding alignment data frame in combination with the queue state sequence; shifting the data validity boundary in stages according to the priority order of correction direction change, uncertain interval boundary displacement, and deviation component displacement in the solution sensitivity, generating a semantic deadline including the latest transmission time stamp, scheduling level, and expired handling identifier; when the queue state sequence changes, only the latest transmission time stamp and scheduling level corresponding to the alignment data frame that has not yet been sent are recalculated.

[0008] Preferably, the encoding and replacement includes: encoding data capable of independently recovering the complete alignment state of the corresponding state version into the anchor frame; encoding the state difference relative to the most recent anchor frame or the confirmed residual increment frame into the residual increment frame; establishing a replacement candidate relationship for residual increment frames with the same state version, adjacent transposition segments, and whose solution sensitivity does not cause a change in the correction direction, and marking the data whose state update range can be covered by subsequent residual increment frames as the replaceable frames; when the subsequent frame has at least one of the deviation component displacement and the uncertain interval boundary displacement higher than the replaceable frame and the remaining items are not lower than the replaceable frame, and the semantic cutoff period of the subsequent frame is not later than the replaceable frame, migrating the dependency start point, acquisition sequence number range, and transposition coverage identifier of the replaceable frame to the subsequent frame, and removing the replaceable frame from the communication queue.

[0009] Preferably, the shadow solution process further includes: establishing a directed graph of state updates according to the data dependencies between the centering state variables, and configuring version identifiers and reversible increment records for each state node in the directed graph of state updates; locating the target node and its successor node affected by the corresponding centering data frame in the directed graph of state updates according to the transposition segment and measurement component carried by each centering data frame; taking the current state snapshot as a common starting point, sequentially applying the state increments formed by the corresponding centering data frames to the target node and its successor node to generate an inclusion branch, and then canceling the state increments and restoring the current state snapshot according to the reversible increment record to generate an exclusion branch; summarizing the state nodes that have changed in the two branches into multi-component solution sensitivities according to the dependency hierarchy, and retaining the state node index corresponding to each sensitivity component.

[0010] Preferably, the scheduling of the transmission order includes: converting the latest transmission time of each pair of data frames into a remaining transmission budget relative to the current queue time, and establishing an ordered table of deadlines by state version partition based on the remaining transmission budget; within each state version partition, forming a transmission chain according to the dependency order between the anchor frame and its associated residual increment frame, and then sorting the head data frames of different transmission chains according to the scheduling level and the desensitization; calculating the expected completion time of each head data frame based on the amount of bytes to be transmitted and the dequeueing amount per unit time; when any expected completion time is later than the corresponding latest transmission time, selecting a target transmission chain with a lower scheduling level and which does not exceed its own latest transmission time after being moved from the transmission chains preceding the corresponding transmission chain, moving the target transmission chain after the corresponding transmission chain, and updating only the expected completion time of each data frame within the moving interval.

[0011] Preferably, the determination of the anchoring trigger condition includes: maintaining a mirror alignment state reconstructed based on the confirmed anchor frame and the confirmed residual increment frame on the transmitting side, and comparing the formal solution state with the mirror alignment state after each generation of residual increment frame to form a reconstruction deviation envelope containing the difference of each deviation component, the quantization residual direction, and the number of unconfirmed dependency layers; when the reconstruction deviation envelope crosses the state partition boundary corresponding to the correction direction, or when the state version switch causes the existing dependency starting point to fail, the formal solution state, the covered acquisition sequence number interval, the previous anchor frame identifier, and the state version are encoded into a new anchor frame; after the new anchor frame enters the communication queue, the unsent residual increment frames that depend on the old anchor frame and whose acquisition sequence number has been included in the covered acquisition sequence number interval are cancelled, and the dependency starting point interval of the subsequent residual increment frames is rewritten as the new anchor frame.

[0012] Preferably, the method further includes: generating a feedback record on the edge side according to the state version, the feedback record containing the identifier of the most recently applied anchor frame, the endpoint of the continuously applied acquisition sequence number, the residual vector between the reconstruction state and the formal verification state, and the edge reception timestamp of each applied data frame; aligning the feedback record and the transmission scheduling record on the transmitting side according to the state version and acquisition sequence number to obtain the correspondence between the resolution sensitivity and the residual vector, and the offset relationship between the semantic deadline and the actual reception timestamp; partitioning and accumulating the correspondence and offset relationships according to the transposition segment and communication queue state to form a sensitivity calibration table and a deadline calibration table; using the sensitivity calibration table to correct the dominance determination boundary of the substitute frame, and using the deadline calibration table to correct the hierarchical forward shift of the data validity boundary, correcting the alignment data frames generated only after the state version corresponding to the feedback record.

[0013] Preferably, the partitioning accumulation includes: dividing the feedback record into complete dependency chain feedback, substitution chain feedback where frame replacement occurs, and truncated chain feedback where semantic deadlines have expired, based on the continuously applied collection sequence number endpoint and transmission scheduling record; updating the sensitivity calibration table using the corresponding data of the latest transmission time stamp and the edge reception time stamp in the complete dependency chain feedback, and updating the deadline calibration table using the offset data of the latest transmission time stamp and the edge reception time stamp in the truncated chain feedback; for the substitution chain feedback, updating the dominance decision boundary based on the state difference encoded by the revoked substituted frame and the later frame, respectively, and using the distance relationship between the two substitution state trajectories and the formal verification state in the feedback record; and applying a monotonic constraint that the latest transmission time stamp is not delayed as the queue load increases to the update results between different communication queue state partitions.

[0014] Preferably, the switching of the new anchor frame includes: generating a switching token packet containing the old state version, the target state version, and the endpoint of the covered acquisition sequence number, and writing the switching token packet into the new anchor frame and the residual increment frames generated thereafter; after receiving the new anchor frame, the edge side retains the first reconstruction context corresponding to the old state version and establishes a second reconstruction context corresponding to the target state version, and writes the state version and dependency start point carried by each residual increment frame into the first reconstruction context or the second reconstruction context respectively; when the second reconstruction context is continuously applied to the endpoint of the covered acquisition sequence number, a switching confirmation record containing the switching token packet and the continuous application endpoint is generated; the sending side cancels the residual increment frames in the communication queue that still depend on the old state version according to the switching confirmation record, and re-encodes the corresponding state increment into a residual increment frame that depends on the target state version.

[0015] Preferably, the load of the anchor frame is sequentially written with the status version, continuous acquisition sequence end point, transposition coverage, parallel deviation, angle deviation, correction direction, undetermined interval, and content verification summary; the residual increment frame records the reference frame identifier and the bitmap of the changed field, and records the difference between the upper and lower boundaries of the undetermined interval and the dependent starting point.

[0016] Preferably, the key value of the deadline ordered table consists of the state version, remaining transmission budget, scheduling level, desensitization and acquisition sequence number; data frames that have been sent from the dependent starting point or are located at the beginning of the same transmission chain are determined as the chain head, and when the corresponding state version has not yet been established, the transmission chain where the anchor frame is located is marked as an uncrossable chain to prevent the residual increment frame from being sent before the state baseline.

[0017] Preferably, the switching token packet is generated by the old state version, the target state version, the end point of the covered collection sequence number, and the content summary of the new anchor frame. The edge side deduplicates the data frames that arrive repeatedly according to the state version, collection sequence number, and content summary, writes the data frames that cannot be located in the first reconstruction context or the second reconstruction context due to the dependent starting point into the set to be continued, and freezes the first reconstruction context after advancing to the end point of the covered collection sequence number in the second reconstruction context.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Before transmission, the shadow solution process, synchronized with the formal solution state, inputs each centering data frame. The parallel deviation, angle deviation, correction direction, and solution uncertainty interval are compared when the corresponding data frame is included versus excluded, forming a solution sensitivity bound to the state version. Then, a semantic deadline is generated by combining the remaining valid time of the transposition segment, the number of bytes to be sent, the dequeue rate, and historical queuing jitter. Based on this, anchor frames, residual increment frames, and replaceable frames are distinguished. The communication queue adjusts the transmission order according to the remaining transmission budget, scheduling level, solution sensitivity, and inter-frame dependencies, allowing high-solution-sensitivity later-arriving frames to replace low-solution-sensitivity replaceable frames that have not yet been sent and whose state update range is covered. Duplicate data that cannot change the current centering solution is thus compressed or canceled before entering the link. Data that may change the deviation direction, correction direction, or solution interval receives an earlier transmission position within the validity period, enabling edge nodes to obtain the data needed to maintain the current solution earlier under the same communication load, reducing the impact of queue backlog on the timing of centering state output.

[0019] 2. A reconstruction chain with state versions and dependency starting points is constructed using anchor frames and residual increment frames. The transmitting side maintains a mirror alignment state. A reconstruction deviation envelope is formed based on the difference in deviation components between the formal solution state and the mirror alignment state, the direction of the quantized residual, and the number of unconfirmed dependency layers. When the reconstruction deviation envelope crosses the state partition boundary corresponding to the correction direction, or when a state version switch causes the original dependency starting point to become invalid, a new anchor frame is generated, unsent residual increment frames already covered by the new anchor frame are revoked, and the dependency starting point of subsequent increment frames is rewritten, ensuring that error accumulation in continuous differential transmission is constrained by the anchor state. The edge side maintains the reconstruction context according to the state version and returns the continuously applied acquisition sequence number, reception timestamp, and residual vector. The transmitting side uses this information to calibrate the alternative frame decision boundary and semantic deadline advance amount, allowing subsequent scheduling to be corrected according to the actual reception deviation and queue state. Local version switching and dependency chain updates also limit the scope of state correction, reducing full replay and repeated solutions caused by anchor updates, message delays, or link fluctuations. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the overall process of low-latency edge processing of rotating machinery centering data in this invention.

[0021] Figure 2 This is a graph showing the change in sensitivity as a function of the acquisition sequence number in this invention.

[0022] Figure 3 This is the frame dominance determination diagram in this invention.

[0023] Figure 4 This is the state increment propagation diagram in this invention.

[0024] Figure 5 This is a timescale diagram showing the expected completion of the transmission chain in this invention.

[0025] Figure 6 This is the reconstructed deviation envelope diagram in this invention. Detailed Implementation

[0026] In one embodiment, please refer to Figure 1 The low-latency edge processing method for centering data of rotating machinery includes: receiving continuously generated centering data frames and associating each centering data frame with acquisition sequence number, transposition segment, acquisition time stamp, status version, and measurement quality identifier; inputting each centering data frame into a shadow solution process synchronized with the formal solution status on the edge side to obtain the centering status difference when the corresponding data frame is included and excluded, and generating a solution sensitivity accordingly; generating a semantic deadline based on the solution sensitivity, the remaining effective time of the corresponding transposition segment, and the communication queue status; encoding each centering data frame as an anchor frame, a residual increment frame, or a replaceable frame; scheduling the transmission order according to the semantic deadline and solution sensitivity, and replacing the low solution sensitivity replaceable frames with the same status version that have not yet been transmitted in the queue with later-arriving frames of high solution sensitivity; reconstructing the centering status on the edge side based on the anchor frame and the residual increment frame, and inserting a new anchor frame when the accumulated reconstruction deviation meets the anchor trigger condition.

[0027] In this embodiment, the alignment data frames are continuously generated during the alignment measurement process of the rotating machinery. The data content is limited to data that can participate in the calculation of parallel deviation, angular deviation, correction direction, and uncertainty interval. No business information unrelated to alignment calculation is introduced. The acquisition end creates a frame record for each set of independently identifiable data generated. The frame record adopts a combination of fixed field area and variable data area. The fixed field area writes the acquisition sequence number, rotation segment, acquisition time stamp, status version, and measurement quality identifier. The variable data area writes the measurement component, the validity bit of the corresponding measurement component, and the change content relative to the previous confirmed status. The sending side completes the integrity verification and establishes the frame index before the frame record enters the communication queue. The frame index uses the status version as the primary key and the acquisition sequence number as the secondary key, so that the data within the same status version maintains a traceable generation order and allows the scheduling process to change the physical transmission order without changing the logical acquisition order.

[0028] Specifically, the acquisition sequence number monotonically increases within the same data source to identify missing, duplicate, and late-arriving data. The transposition segment indicates the current centering measurement position range to which the data belongs. The acquisition timestamp indicates the data formation time point. The status version indicates the formal solution status on which the corresponding data was generated. The measurement quality identifier is jointly encoded by data integrity, acquisition continuity, and solution availability. The sending side does not directly decide to discard data based on the measurement quality identifier, but uses it as a reliability constraint in the shadow solution process. After receiving the data frame, the edge side uses the status version and acquisition sequence number to complete deduplication. If multiple frame records appear with the same acquisition sequence number, the record with the same status version and the more complete measurement quality identifier is retained, and the remaining records are written to the abnormal record area and do not participate in the current reconstruction chain.

[0029] In a preferred embodiment, the formal solution state is saved using a versioned state snapshot. The state snapshot includes at least the current parallel deviation, current angle deviation, current correction direction, current solution uncertainty interval, the end point of the acquisition sequence number of the included data, and the identifier of the most recent anchor frame. The formal solver generates a new state version each time it completes a submitable state update. However, data in the same transposition segment that have not changed the solution basis are allowed to share state versions. The shadow solver reads a read-only state snapshot from the formal solver and does not directly modify the formal solution state. After the shadow solver is completed, it only outputs the solution sensitivity and branch difference records. This ensures that the transmission decision maintains the same data basis as the formal calculation and avoids the scheduling trial calculation from changing the state content that needs to be reconstructed on the edge side.

[0030] Furthermore, the shadow solution process establishes inclusion and exclusion branches for each individual centering data frame to be evaluated. The inclusion branch applies the state increment formed by the corresponding data frame starting from the current state snapshot, while the exclusion branch absorbs only the confirmed data within the same transposition segment with a collection sequence number less than the corresponding data frame, starting from the same state snapshot. Both branches use the same state update rules and numerical precision. After the branch calculation is completed, the changes in parallel deviation, angle deviation, correction direction, and solution uncertainty interval are compared. If the inclusion branch and the exclusion branch only show differences at the end of the numerical values ​​that can be explained by quantization error, then the corresponding differences are included in the residuals and not included in the change of correction direction. If any branch crosses the predefined state partition boundary, then the boundary crossing identifier is retained and the priority of the corresponding sensitivity component is increased.

[0031] In this embodiment, the resolution sensitivity of the i-th centering data frame is calculated according to the following formula: ; in, This represents the solution sensitivity of the i-th paired data frame. and These represent the parallel deviations of the included and excluded branches, respectively. and These represent the angular deviations of the included and excluded branches, respectively. and These represent the correction direction codes for the two branches, respectively. and These represent the widths of the uncertainty intervals for the solutions of the two branches, respectively. , and These represent the normalization references for the corresponding components. to For non-negative weights that sum to 1, This is an indicator function that takes the value 1 when the condition is true and 0 when the condition is false. For example, when the four weights are 0.25, 0.20, 0.40, and 0.15 respectively, and the normalized change of parallel deviation is 0.2, the normalized change of angle deviation is 0.1, and the normalized change of the uncertainty interval is 0.2 when the correction direction changes, the normalized change is 0.2. If the value is set to 0.50, the corresponding data frame is categorized into the high-sensitivity set that needs to be sent in advance.

[0032] Please refer to Figure 2 , Figure 2 The horizontal axis represents the acquisition sequence number of the data frame, displayed as Chinese sequence identifiers such as Acquisition 1, Acquisition 3, Acquisition 5, etc., while the vertical axis represents the normalized score of the solution sensitivity. Figure 2 The solid black line represents the overall sensitivity. The gray dashed line, gray dotted line, black dotted line, and light gray line represent the contributions of parallel deviation, angle deviation, direction change, and interval width, respectively. The legend is located above the plotting area and is used to distinguish the various sensitivity components.

[0033] from Figure 2 As can be seen, the overall sensitivity curve reaches its highest value of approximately 0.507 at acquisition point one, followed by significant peaks near acquisitions nine, twelve, seventeen, and eighteen, all approaching or exceeding 0.47. The data frames corresponding to these peaks all show an increase in contribution to 0.4 due to directional changes, indicating that these data frames not only cause changes in the bias component between the included and excluded branches but also affect the correction direction judgment. These are considered highly sensitive frames in the communication queue that require priority evaluation and transmission.

[0034] The overall sensitivity of frames 2 through 8 and 10 through 16 is generally low, with most scores between 0.04 and 0.14. The curve changes are relatively flat, indicating that these frames mainly reflect small perturbations in parallelism, angle, or interval width, and do not trigger significant state partition changes on their own. The contributions of parallelism and angle deviation fluctuate alternately in the low-sensitivity segment, indicating that the influence components of different transposition segments on the state snapshot are not entirely the same.

[0035] The sensitivity is not simply stored as a single scalar. The transmitting side also retains a component vector consisting of parallel deviation displacement, angular deviation displacement, correction direction change, and uncertainty interval boundary displacement. The scalar value is used for global sorting, and the component vector is used for frame replacement dominance judgment. The measurement quality identifier is used to limit the reliable upper bound of each component. When the data integrity is insufficient but the corresponding data may cause a change in the correction direction, the sensitivity is not directly set to zero. Instead, the corresponding frame is marked as a high-sensitivity frame to be verified, which is retained in the communication queue and the edge side is required to perform a consistency check after receiving the data and combining it with subsequent data. Only when the measurement quality is complete and multiple components are all below the calibration boundary of the current state version is the corresponding frame allowed to be included in the alternative candidate set.

[0036] In one embodiment, the semantic deadline is derived backward from the end point of the corresponding transposed segment. The sending side reads the remaining valid time of the transposed segment, the number of bytes to be sent in the communication queue, the number of dequeued per unit time, the message arrival interval, and the historical queuing jitter. First, it predicts the candidate dequeue time of the centering data frame in the current queue order. Then, it determines the forward shift amount based on the degree of influence of each component in the solution sensitivity on the state boundary. Frames whose correction direction has changed adopt the earlier latest transmission time. Frames that simply reduce the uncertainty interval without changing the deviation direction adopt the later latest transmission time. Frames whose state version has been covered by the new anchor frame and cannot enter the current reconstruction chain are given an expired disposal mark. The expired disposal mark is used to remove the corresponding frame from the queue instead of continuing to occupy reconstruction resources on the edge side.

[0037] In this embodiment, the first The semantic deadline for each pair of data frames is determined according to the following formula: ; in, Indicates the latest transmission time stamp. Indicates the valid data boundary of the corresponding transposed segment. This indicates the maximum solution sensitivity of the evaluated data frames within the current state version. This represents the remaining time from the data collection time stamp to the valid data boundary. This represents the queuing jitter compensation amount obtained based on historical message arrival intervals. This represents the queue occupancy compensation amount obtained based on the amount of bytes to be sent and the amount dequeued per unit time. , and These are forward shift coefficients with values ​​ranging from 0 to 1 and a total value not exceeding 1. For example, when the remaining time is 10 uniform time units, the resolution sensitivity is equal to the current maximum resolution sensitivity, the jitter compensation is 1 uniform time unit, the queue occupancy compensation is 2 uniform time units, and the three forward shift coefficients are 0.4, 0.2, and 0.2 respectively, the latest transmission time stamp is shifted forward by 4.6 uniform time units relative to the valid data boundary, so that the corresponding frame obtains its transmission position before entering the expiration interval.

[0038] Preferably, the anchor frame stores the necessary content to independently recover the complete alignment state of a certain state version, including the state version, parallel deviation, angle deviation, correction direction, undetermined interval, covered acquisition sequence number range, and the previous anchor frame identifier. The residual increment frame stores the state difference, dependency start point, and acquisition sequence number range relative to the most recent anchor frame or the confirmed residual increment frame. The alternative frame is a frame to be sent that has been encoded in residual increment form but whose state update range can be covered by subsequent frames. The sending side determines the frame type and simultaneously binds the desensitization and dependency relationship. It does not force all data to be converted into anchor frames at a fixed period, nor does it allow residual increment frames lacking dependency start points to enter the communication queue alone.

[0039] In this embodiment, the scheduler maintains a queue partitioned by state version. Each partition records the dependency chain consisting of the anchor frame and its subsequent residual increment frames. The scheduler first compares the latest transmission timestamp of the first frame of each chain, and then compares the dissensitivity when the latest transmission timestamps are close. If the later frame and the alternative frame that has not yet been transmitted in the queue belong to the same state version and the later frame covers the state update range of the alternative frame, then the dissensitivity components and semantic deadlines of the two frames are further compared. When the dominance condition is met, the dependency start point, acquisition sequence number range and transposition coverage identifier of the alternative frame are migrated to the later frame, and the alternative frame is removed from the queue. The migration operation synchronously rewrites the predecessor identifier of the subsequent dependency frame, so that the state can still be reconstructed according to a single continuous dependency chain after the edge side receives the data.

[0040] On the edge side, the receiving process selects the corresponding reconstruction context according to the state version. After the anchor frame arrives, the reconstruction context is initialized with the complete state saved in the anchor frame. After the residual increment frame arrives, the dependency start point, the acquisition sequence number range and the predecessor confirmation state are checked. If the check passes, the state difference is applied to the current reconstruction state. If the check fails, the corresponding residual increment frame is temporarily stored in the version waiting area and the data of other state versions is processed. The cumulative reconstruction deviation is jointly determined by the mirror alignment state of the sending side and the verification state fed back by the edge side. When the cumulative reconstruction deviation reaches the anchor trigger condition, the sending side generates a new anchor frame to truncate the existing incremental dependency chain and encodes the subsequent state difference relative to the new anchor state.

[0041] In a preferred embodiment, the shadow solution process includes: reading the current state snapshot of the formal solution state and the state version corresponding to the current state snapshot; establishing an inclusion branch and an exclusion branch that share the current state snapshot for each pair of data frames; updating the parallel deviation, angle deviation, correction direction, and solution uncertainty interval using the corresponding pair of data frames in the inclusion branch; maintaining the current state snapshot and advancing the state according to the confirmed data within the same transposition segment in the exclusion branch; comparing the deviation component displacement, correction direction change, and solution uncertainty interval boundary displacement of the two branches; constraining the reliability of the comparison results according to the measurement quality identifier to form a multi-component solution sensitivity bound to the state version; and writing the multi-component solution sensitivity and the acquisition sequence number of the corresponding pair of data frames into the transmission scheduling record.

[0042] Specifically, the current state snapshot is saved using immutable objects. The inclusion and exclusion branches are established by sharing immutable basic data and recording differences separately, avoiding the copying of the entire historical state for each data frame to be evaluated. The data written in the inclusion branch includes the deviation increment caused by the corresponding frame, the candidate direction value, and the update amount of the uncertain interval boundary. The exclusion branch only advances the state sequence number corresponding to the confirmed data without writing the content of the frame to be evaluated. When the branch ends, the unified comparator reads the state fields with the same name from the two branches. The comparator calculates the signed difference for the numerical field, performs equality judgment on the direction field, and calculates the upper and lower boundary displacements for the interval field. The comparison result and the measurement quality identifier form a confidence mask. The confidence mask limits which components can enter the scheduling and sorting and which components can only be retained as verification information.

[0043] Furthermore, the transmission scheduling record includes at least the state version, acquisition sequence number, transposed segment, solution sensitivity scalar, solution sensitivity component vector, semantic deadline, frame type, dependency start point, alternative candidate identifier, and current queue position. After the shadow solution process is completed, the corresponding record is located by the state version and acquisition sequence number and atomic writing is performed. If the formal solution state has been switched to a new version during the writing process, the current calculation result is not directly used in the new version queue, but is retained in the old version record and the corresponding data frame is reread from the new state snapshot. Only the evaluation result with the same state version and snapshot version enters the transmission scheduling, thereby avoiding the mixing of different solution benchmarks with the same solution sensitivity.

[0044] In a preferred embodiment, the generation of the semantic deadline includes establishing a queue state sequence for the communication queue carrying the centering data frame. The queue state sequence includes the number of bytes to be sent, the number of bytes dequeued per unit time, the message arrival interval, and historical queuing jitter. The effective data boundary is determined based on the end time stamp and acquisition time stamp of the corresponding transposition segment, and the candidate dequeue time stamp of the corresponding centering data frame is predicted in combination with the queue state sequence. The effective data boundary is shifted forward in stages according to the priority order of the change in correction direction, the displacement of the uncertain interval boundary, and the displacement of the deviation component in the solution sensitivity, to generate a semantic deadline that includes the latest transmission time stamp, the scheduling level, and the expiration handling identifier. When the queue state sequence changes, only the latest transmission time stamp and scheduling level corresponding to the centering data frame that has not yet been sent are recalculated.

[0045] The queue state sequence is updated according to each enqueue, dequeue, replacement, cancellation, and acknowledgment event. The number of bytes to be sent is the sum of the encoded lengths of all currently sendable frames. The dequeue quantity per unit time is obtained by dividing the cumulative length of multiple adjacent sent completion events by the corresponding logical time stamp span. The message arrival interval is maintained separately for each data source and summarized within the same state version. The historical queuing jitter is the absolute deviation sequence of the queuing time of sent frames relative to the median value. Using the median value instead of a single maximum value can avoid the long-term expansion of the subsequent deadline advance by a single isolated blockage. The queue state sequence retains the order of event occurrence and state version boundaries. When the state version is switched, the existing sequence is not cleared, but the sample range that can be used for the current prediction is distinguished by the version identifier.

[0046] The valid data boundary is obtained by subtracting the minimum processing interval required for edge reconstruction from the end time of the transposed segment. The minimum processing interval is estimated by recording similar frames that have already been reconstructed. No fixed value related to specific processor models is set. The candidate dequeue time is equal to the sum of the current logical time and the expected queuing time. The expected queuing time is jointly determined by the length of the anchor frame that cannot be crossed before the candidate frame, the length of the dependency chain, and the length of the adjustable transmission chain. The correction direction change component is in the highest priority layer, the uncertain interval boundary displacement component is in the second priority layer, and the deviation component displacement component is in the basic priority layer. When the same data frame contains multiple components, the highest priority layer is used and the remaining components are used as the sorting basis of the same layer.

[0047] Expiration handling indicators include at least three states: forced forward shift, allowed substitution, and stopped transmission. When the candidate dequeue time is later than the latest transmission time and the data frame will change the correction direction, it is marked as forced forward shift. The scheduler will move the corresponding transmission chain to the feasible first position while maintaining the dependency order. When the candidate dequeue time is later than the latest transmission time but the data frame can be covered by the subsequent state, it is marked as allowed substitution. After the scheduler searches for the same version, it performs an overwrite judgment on the frame. When the data frame has been covered by a new anchor frame or its state version has been turned off, it is marked as stopped transmission. Queue state changes only trigger local recalculation of data frames that have not yet been transmitted. Data frames that have already been transmitted retain their original deadline for feedback calibration and do not have their historical judgments rewritten due to subsequent queue changes.

[0048] In a preferred embodiment, the encoding and replacement include encoding data capable of independently restoring the complete alignment state of the corresponding state version as anchor frames, encoding the state difference relative to the most recent anchor frame or the confirmed residual increment frame as residual increment frames; establishing alternative candidate relationships for residual increment frames with the same state version, adjacent transposition segments, and whose solution sensitivity does not cause a change in the correction direction, and marking data whose state update range can be covered by subsequent residual increment frames as replaceable frames; when the subsequent frame has at least one of the deviation component displacement and the uncertain interval boundary displacement higher than the replaceable frame and the other items are not lower than the replaceable frame, and the semantic cutoff period of the subsequent frame is not later than the replaceable frame, the dependency start point, acquisition sequence number range, and transposition coverage identifier of the replaceable frame are migrated to the subsequent frame, and the replaceable frame is removed from the communication queue.

[0049] The load of the anchor frame is encoded in the order of status version information, continuous acquisition sequence end point, transposition coverage range, parallel deviation, angle deviation, correction direction, solution uncertainty interval, and content verification summary. The residual increment frame records the reference frame identifier and bitmaps of each changed field. Fields with a bitmap value of 1 carry the status difference value, while fields with a bitmap value of 0 retain the value from the dependent start point. The interval boundaries record the lower boundary difference value and the upper boundary difference value respectively to prevent the use of a single width change to cover up the overall translation of the interval. The acquisition sequence range is represented by the start sequence number and the end sequence number. The transposition coverage identifier is represented by a segment set. The residual increment load is only applied on the edge side when the reference frame identifier, status version, and continuous acquisition sequence number all satisfy the dependency relationship.

[0050] The replacement candidate relationship is established according to the state update range rather than just the acquisition time. The transmitting side parses each residual increment frame into a set of state fields and a set of transposed segments. The state field set of the later frame contains the state field set of the replaceable frame, and if the acquisition sequence end point covered by the later frame is not earlier than the end point of the replaceable frame, it is considered that there is a coverage basis. If the replaceable frame contains a change in correction direction, the replacement judgment is terminated and the original frame is retained. If neither causes a change in correction direction, the displacement of the deviation component and the boundary displacement of the uncertain interval are compared. The multi-component values ​​after measurement quality constraints are used for comparison to avoid the low-quality later frame covering the more reliable earlier frame based solely on the original amplitude.

[0051] The dominance relationship of a later frame to a substitute frame is determined by the following formula: ; in, This indicates the dominance determination result of the subsequent frame n on the replaceable frame o. and This represents the displacement of the corresponding deviation component. and This represents the boundary displacement of the corresponding uncertain interval. and Use their respective latest transmission timestamps. and This represents the set of state update fields, where all indicator functions are set to 1. Take 1 and allow substitutions, for example Take 0.4, Take 0.3, and If the value is 0.2, and the deadline of the later frame is earlier than that of the original frame and the field set covers the original frame, it is determined to be 1. If the later frame lacks the angle deviation field of the original frame, the set inclusion condition is set to 0 and the original frame is retained.

[0052] Please refer to Figure 3 , Figure 3 The horizontal axis represents the candidate relationship formed by the subsequent frame and the alternative frame, using Chinese sequence identifiers for Candidate 1, Candidate 2, and Candidate 3; the vertical axis represents the decision value. The bar sequences correspond to parallel conditions, interval conditions, gain conditions, time limit conditions, field coverage, and final decision, respectively. The legend is located above the plotting area to illustrate each dominant decision condition.

[0053] from Figure 3 As can be seen, all the decision bars for the three candidate relationships reach 1, indicating that the later-arriving frame satisfies the dominance conditions in terms of deviation component displacement, uncertainty interval boundary displacement, strict gain, latest transmission time stamp, and state field set coverage. Candidate 1 and Candidate 2 are both covered by the same later-arriving frame covering different preceding frames, indicating that when the state update range carried by the later-arriving frame is sufficiently complete, the transmission budget occupied by multiple old residual increment frames can be released simultaneously.

[0054] Candidate 3 corresponds to the coverage relationship of the later acquisition frames, and it also satisfies all the dominance conditions, indicating that the substitution judgment does not only depend on the order of time, but also on the comprehensive relationship between the set of state fields, the transposed coverage range, and the semantic deadline. If any bar is lower than 1, for example, if the field coverage is insufficient or the deadline of the later frame is later than that of the original frame, then the final judgment should be reduced to 0, and the original frame cannot be withdrawn from the communication queue.

[0055] After the replacement is completed, the subsequent frame does not directly use the original frame payload. Instead, it recalculates the cumulative residual within the coverage acquisition sequence number range based on the baseline state corresponding to the original frame's dependency starting point. During the recalculation, it absorbs the intermediate increments after the original frame that have not been canceled according to the acquisition sequence number. The resulting cumulative residual is written into the subsequent frame and the content verification digest is updated. The acquisition sequence number starting point, transposition coverage identifier, and replacement source of the original frame are written into the extended field of the subsequent frame. The communication queue deletes the original frame index and reduces the number of bytes to be sent. Subsequently, only the queue interval affected by the change in the number of bytes to be sent is recalculated for the candidate dequeue timestamp, thereby ensuring that the replaced payload can be continuously reconstructed from the original dependency starting point.

[0056] In a preferred embodiment, the shadow solution process further includes establishing a directed graph of state updates according to the data dependencies between the state variables, and configuring version identifiers and reversible increment records for each state node in the directed graph of state updates; locating the target node and its successor node affected by the corresponding data frame in the directed graph of state updates based on the transposition segment and measurement component carried by each data frame of the ...

[0057] The nodes in the directed graph of state update correspond to the solution state fields, including at least the transposition segment deviation node, parallel deviation summary node, angle deviation summary node, uncertain interval node, and correction direction node. Edges point from the node directly written by the data frame to the node derived from its calculation result. The structure of the directed graph within the same state version remains unchanged. When switching versions, only nodes with dependent changes are generated with new version identifiers. The node record includes the current value, the source collection sequence number range, the predecessor node version, and the reversible increment. The reversible increment is represented by the difference between the value before and after the update and the operation type. For interval merging operations, the source of the merged boundary is also recorded, so that the undo process can restore the original boundary instead of only performing the reverse operation on the interval width.

[0058] The propagation of state increments along a directed graph is described by the following formula: ; in, This represents the state increment vector generated by the i-th pair of data frames in the l-th dependency layer. Let represent the state transfer matrix from layer l to the next layer. The non-zero elements in the matrix correspond to the dependency edges in the directed graph. This represents the direct increment of the data frame to the state of the next layer, for example, the increment of the parallel deviation and angle deviation of layer 0. , This indicates transpose, the pass matrix is ​​the identity matrix and the value is directly incremented. Then the increment of the first layer is taken The node index is marked accordingly. Both the parallel deviation summary node and the angle deviation summary node are affected. If the corresponding element of the angle deviation to the correction direction in the matrix is ​​0, then the correction direction node will not be included in the change set in this propagation.

[0059] Please refer to Figure 4 , Figure 4 The horizontal axis represents the acquisition sequence number of the data frame being centered, using Chinese sequence identifiers; the vertical axis represents the state increment. Solid black lines represent parallel input increments, dashed gray lines represent angular input increments, dotted gray lines represent parallel propagation increments, and dashed black lines represent angular propagation increments. Figure 4 This is used to illustrate how the state increment propagates along the directed graph of state updates to successor nodes after a data frame is directly written to the target node.

[0060] from Figure 4 As can be seen, the parallel input increment reaches approximately 0.25 at acquisition point 5, and after propagation, it forms a parallel propagation increment of approximately 0.26. This indicates that this frame not only changes the directly affected transposition segment deviation nodes, but also affects the parallel deviation aggregation nodes through directed graph dependencies. Acquisition points 1 to 8 are generally in a positive increment range, indicating that the preceding data has a continuous pushing effect on the parallel deviation state.

[0061] After acquisition 12, the parallel input and propagation increments entered the negative range, reaching their lowest point near acquisition 15, with a propagation increment of approximately -0.1699. This indicates that the subsequent data frames produced a reverse correction to the parallel deviation state. The angle propagation increment was highest at acquisition 1, at approximately 0.0774, and lowest near acquisition 14, at approximately -0.053, reflecting that the changes in the angle state and the parallel state were not completely synchronized.

[0062] The included branch only copies the values ​​of successor nodes reachable from the target node and applies the state increments according to the topological order of the directed graph. The excluded branch restores the modified node by reading the reversible increment records in reverse order. The remaining nodes continue to reference the current state snapshot. After the two branches complete the comparison, the temporary node pages are released while the state node indexes are retained. Each component in the multi-component solution sensitivity is associated with a set of node indices. When the scheduler determines frame replacement, it can check whether the set of node indices of the later frame covers the set of indices of the original frame. The feedback calibration process can also locate the source of error at the node level, avoiding the average distribution of the correction direction difference to all deviation components.

[0063] In a preferred embodiment, the scheduling of the transmission order includes converting the latest transmission time of each pair of data frames into a remaining transmission budget relative to the current queue time, and establishing an ordered table of deadlines by state version partition based on the remaining transmission budget; within each state version partition, forming transmission chains according to the dependency order between the anchor frame and its associated residual increment frame, and then sorting the head data frames of different transmission chains according to the scheduling level and desensitization; calculating the expected completion time of each head data frame based on the amount of bytes to be transmitted and the dequeueing amount per unit time; when any expected completion time is later than the corresponding latest transmission time, selecting a target transmission chain with a lower scheduling level and which does not exceed its own latest transmission time after being moved from the transmission chains preceding the corresponding transmission chain, moving the target transmission chain after the corresponding transmission chain, and only updating the expected completion time of each data frame within the moving interval.

[0064] The key values ​​of the deadline ordered table consist of state version, remaining transmission budget, scheduling level, desensitization, and acquisition sequence number. The state version is used to isolate different reconstruction contexts. Data with smaller remaining transmission budgets are placed at the front. Data with higher scheduling levels are placed at the front when the budgets are the same or similar. Desensitization is used to handle contention at the same level. Acquisition sequence number is used to maintain deterministic order. The head of the transmission chain is the data frame that has not yet been transmitted and whose dependent starting point has been transmitted or is located at a higher position on the same chain. The order within the chain does not change due to cross-chain adjustments. The transmission chain where the anchor frame is located is marked as an uncrossable chain when the corresponding state version has not yet been established to prevent residual increment frames from being transmitted before their state baseline.

[0065] The estimated completion time of each chain's head data frame is calculated using the following formula: ; in, Indicates scheduling arrangement The estimated completion time of the m-th chain head data frame. Indicates the current queue logical time stamp. This represents the amount of data that the h-th transmission chain in the permutation must send before releasing the head of the next chain. This represents the amount of data dequeued per unit time given by the queue state sequence. For example, the current logical time stamp is 0, and the data amounts corresponding to the three transmission chains are 12, 8, and 10 code length units respectively. When the value is 2, the expected completion times are 6, 10 and 15 respectively. If the latest transmission time of the third transmission link is 12 and the expected completion time of the second transmission link after the shift is still earlier than its own deadline, then the second and third transmission links are swapped and only the swap interval is recalculated.

[0066] Please refer to Figure 5 , Figure 5 The horizontal axis represents the sequence number of the transmission chain head, using Chinese sequence identifiers such as Chain Head 1, Chain Head 3, Chain Head 5, etc.; the vertical axis represents the unified logical time scale. The black solid line represents the expected completion time scale, and the gray dashed line represents the latest transmission time scale. The two curves together reflect whether the data frame at the head of the chain can be released within the transmission window under the current arrangement.

[0067] from Figure 5 As can be seen, the estimated completion timescale increases continuously with the sequence number of the chain head. The estimated completion timescale of the first chain head is approximately 3.545, corresponding to a latest transmission timescale of approximately 6.769, leaving a transmission margin of approximately 3.223. The second chain head has the largest transmission margin, approximately 6.615, indicating that the leading chains have a wider transmission window under the current dequeue rate and data length conditions.

[0068] As the sequence number increases, the expected completion curve gradually approaches and surpasses the latest sending curve. The expected completion time of the first 18th chain is approximately 30.105, while the latest sending time is approximately 21.436. The sending margin is negative and the gap is approximately 8.669, indicating that the first chain cannot complete the sending on time under the current order. It is necessary to select a lower-level chain with a forward shift margin from the preceding chains to perform the shift, or combine alternative cancellation to reduce the length of the chain to be sent.

[0069] Before executing a chain move, the scheduler creates a feasibility copy. In the copy, it calculates the estimated completion times of all affected chain heads after the target transmission chain is moved. If any chain head exceeds its latest transmission time, the move is canceled and the next lower-level transmission chain is searched. If there are multiple movable targets, the transmission chain with a larger data release volume and a larger move margin is selected. After the move is completed, the queue position and candidate dequeue times in the deadline ordered list are updated. During the communication transmission process, the corresponding transmission chain head is advanced after each data frame is completed. If the dependency starting point of the new chain head has not yet been edge-confirmed but has already been sent, transmission is allowed to continue and it is marked as an in-transit dependency. If the dependency starting point has not yet been sent, the corresponding transmission chain is temporarily frozen.

[0070] In a preferred embodiment, determining the anchoring trigger condition includes maintaining a mirror alignment state reconstructed based on confirmed anchor frames and confirmed residual increment frames on the transmitting side, and comparing the formal solution state with the mirror alignment state after each generation of residual increment frames to form a reconstruction deviation envelope containing the difference of each deviation component, the quantized residual direction, and the number of unconfirmed dependency layers; when the reconstruction deviation envelope crosses the state partition boundary corresponding to the correction direction, or when the state version switch causes the existing dependency starting point to become invalid, the formal solution state, the covered acquisition sequence number interval, the previous anchor frame identifier, and the state version are encoded into a new anchor frame; after the new anchor frame enters the communication queue, the unsent residual increment frames that depend on the old anchor frame and whose acquisition sequence number has been included in the covered acquisition sequence number interval are cancelled, and the dependency starting point of subsequent residual increment frames is rewritten to the new anchor frame.

[0071] The sending side mirror alignment state only absorbs frames that have been confirmed for application in edge feedback, while the formal solution state absorbs the currently available complete data. Therefore, the difference between the two reflects the unconfirmed chain length, quantization residual, and state offset caused by frame substitution. The reconstruction bias envelope is formed according to the normalized difference of each state component and the number of unconfirmed dependency layers. The correction direction state partition is established based on the direction determination boundary already used in the formal solution process, without introducing a boundary inconsistent with the solution logic. When the envelope is still inside the current partition, residual increment frames continue to be generated. When the envelope touches or crosses the boundary of an adjacent partition, a new anchor frame is generated, so that the edge reconstruction state obtains a new complete baseline before the direction determination changes.

[0072] The reconstructed bias envelope is calculated using the following formula: ; in, Indicates the first Reconstructed bias envelope after secondary encoding Indicating the first state in the formal solution process Each state component This indicates the corresponding component in the mirror-aligned state. This indicates the boundary scale of the corresponding state partition on that component. This indicates that the dependency level is not yet confirmed. This represents the envelope added for each additional layer of unconfirmed dependencies. For example, the maximum component difference after normalization is taken as 0.7, and the number of unconfirmed dependency layers is taken as 2. When the value is 0.1, Take 0.9 and continue using the current anchor frame. When the maximum component difference increases to 0.85, Take 1.05 and trigger the generation of a new anchor frame.

[0073] Please refer to Figure 6 , Figure 6The horizontal axis represents the encoding steps, identified by Chinese sequence numbers such as Encoding 1, Encoding 3, Encoding 5, etc.; the vertical axis represents the envelope score. The solid black line represents the envelope score, the dashed gray line represents the trigger threshold, and the dotted gray line represents the maximum component difference. Figure 6 This is used to demonstrate the process of accumulated deviation between the formal solution state and the mirror alignment state, as well as the timing of anchor insertion after the envelope score reaches the trigger threshold.

[0074] from Figure 6 As can be seen, this set of examples corresponds to a high-accumulation-risk scenario where the number of unconfirmed dependency layers continuously increases. The envelope score gradually rises in the early stages of the encoding step and crosses the trigger threshold 1, reaching a maximum value of approximately 2.296 near encoding step 11. Although there is a significant drop at encoding step 12, the minimum value is still approximately 1.599, consistently higher than the threshold 1. Under this scenario, the reconstruction risk remains high, and the transmitting side needs to generate or update anchor frames in a timely manner to truncate old residual chains and prevent further error expansion. Under normal stable operating conditions, the envelope score is mostly below the threshold 1, and anchor frame insertion is only triggered when the deviation accumulates to the boundary, which is consistent with the effect of this invention in limiting the frequency of anchor frame insertion and reducing full playback.

[0075] The maximum component difference shows a slow downward trend between coding one and coding eleven, but the envelope score continues to rise, indicating that the number of unconfirmed dependency layers has a cumulative effect on envelope risk. After coding twelfth, the maximum component difference drops to a lower range, while the envelope score continues to rise with the new coding step, indicating that even if the difference of a single state component decreases, continuous unconfirmed increments may still cause overall reconstruction instability, and an envelope threshold mechanism is needed to achieve on-demand triggering.

[0076] The new anchor frame coverage starts from the earliest acquisition sequence number that has not yet been continuously confirmed by the edge after the old anchor frame and ends at the endpoint of the continuous acquisition sequence number that has been absorbed by the formal solution state. The sending side cancels the unsent residual increment frames within the coverage area, retains the transmission record for the sent but unconfirmed residual increment frames and marks them as covered by the new anchor frame. Subsequent residual increment frames are recalculated with the new anchor frame identifier and the target state version as the dependency starting point. If the acquisition sequence number of the subsequent frame is earlier than the end of the coverage of the new anchor frame, it is merged into the anchor state and no longer sent separately. If it is later than the end of the coverage, its state increment is retained and the dependency chain is updated.

[0077] In a preferred embodiment, a feedback record is generated at the edge according to the state version. The feedback record includes the identifier of the most recently applied anchor frame, the end point of the continuously applied acquisition sequence number, the residual vector between the reconstruction state and the formal verification state, and the edge reception timestamp of each applied data frame. At the transmitting side, the feedback record and the transmission scheduling record are aligned according to the state version and acquisition sequence number to obtain the correspondence between the resolution sensitivity and the residual vector, and the offset relationship between the semantic deadline and the actual reception timestamp. According to the transposition segment and communication queue state, the correspondence and offset relationships are partitioned and accumulated to form a sensitivity calibration table and a deadline calibration table. The sensitivity calibration table is used to correct the dominance decision boundary of the substitute frame, and the deadline calibration table is used to correct the hierarchical forward shift of the data validity boundary, which is only applied to the centering data frames generated after the state version corresponding to the feedback record.

[0078] The edge side generates a feedback record each time it advances to the end of the continuous acquisition sequence number or completes the anchoring switch. The formal verification status is recalculated by the edge side using the data that has been fully received and passed the version verification. The reconstructed status is the status obtained by applying the anchor frame and residual increment frame according to the actual arrival order. The residual vector is written in a fixed order according to the parallel deviation, angle deviation and uncertain interval boundary. The transmitting side looks up the corresponding transmission scheduling record with the status version and the end of the acquisition sequence number, and then restores the actual queuing order according to the reception timestamp of each applied data frame. If the feedback record is missing a certain intermediate acquisition sequence number, the corresponding relationship is only established for the range that has been continuously confirmed, and the calibration table is not updated using the data that still has dependency gaps.

[0079] The sensitivity calibration table uses a combination of transposed segments, measurement quality status, and status node indexes as keys. The table entries store the multi-component sensitivity, edge residual vector, and the number of times they change in the same direction, given by the shadow solution. The deadline calibration table uses the range of data to be sent, the range of data out of queue, and the queuing jitter range as keys. The table entries store the offset sequence between the latest transmission time stamp and the actual reception time stamp. When updating, new feedback is assigned a weight determined by the completeness of the sample. For records with frame substitution, the substitution source is retained and written to the sensitivity calibration table after subsequent replay. For records with anchor switching, only the intervals that have been continuously confirmed before the switching are updated to prevent residual relationships of different state versions from being mixed into the same calibration partition.

[0080] In a preferred embodiment, partitioning accumulation includes dividing feedback records into complete dependency chain feedback, alternative chain feedback where frame replacement occurs, and truncated chain feedback where semantic deadlines have expired, based on the continuously applied acquisition sequence number endpoints and transmission scheduling records; updating the sensitivity calibration table using the corresponding data of residual vectors and solution sensitivity in the complete dependency chain feedback, and updating the deadline calibration table using the offset data of the latest transmission time stamp and the edge reception time stamp in the truncated chain feedback; for alternative chain feedback, replaying the corresponding state update from the same dependency starting point based on the state difference encoded by the revoked alternative frame and the subsequent frame, obtaining two alternative state trajectories, and updating the dominance decision boundary based on the distance relationship between the two alternative state trajectories and the formal verification state in the feedback record; applying a monotonic constraint to the update results between different communication queue state partitions without delaying the latest transmission time stamp as the queue load increases.

[0081] The complete dependency chain feedback requires that the acquisition sequence numbers of the anchor frame and its subsequent residual incremental frames be continuous and that there are no substitution or expiration markers. The substitution chain feedback requires that there be a mapping relationship between the canceled frame and the subsequent frame in the transmission scheduling record. The truncated chain feedback requires that the actual reception time of at least one data frame be later than the latest transmission time or that transmission be stopped due to expiration. The replay of the substitution state trajectory restores the baseline state from the common dependency starting point, absorbs the state difference in the path of the canceled frame and the path of the subsequent frame respectively, and calculates the component distance between it and the formal verification state at the endpoint of the same acquisition sequence number. When the distance of the subsequent frame path is not greater than the distance of the canceled frame path, the current dominant boundary is retained. When the distance is large, the coverage conditions of the corresponding state field combination are tightened.

[0082] Monotonic constraints arrange communication queue state partitions from low to high based on the amount of data to be sent and queuing jitter. If the forward shift of a high-load partition after calibration is less than that of an adjacent low-load partition, the forward shift of the high-load partition is corrected to the value of the adjacent low-load partition. If multiple adjacent partitions violate the monotonic relationship consecutively, a merging correction is performed from the lowest violating partition upwards. The correction process only changes the forward shift of the deadline without changing the recorded actual receiving timestamp. The dominance decision boundary of the sensitivity calibration table is maintained separately according to the state node index and is not directly rewritten due to changes in queue load. This keeps the communication-side calibration and the solution-side calibration separate and allows them to roll back to the previous revision number respectively.

[0083] In a preferred embodiment, the switching of the new anchor frame includes generating a switching token packet containing the old state version, the target state version, and the endpoint of the covered acquisition sequence number, and writing the switching token packet into the new anchor frame and the residual increment frames generated thereafter; after receiving the new anchor frame, the edge side retains the first reconstruction context corresponding to the old state version and establishes a second reconstruction context corresponding to the target state version, and writes the state version and dependency start point carried by each residual increment frame into the first reconstruction context or the second reconstruction context respectively; when the second reconstruction context is continuously applied to the endpoint of the covered acquisition sequence number, a switching confirmation record containing the switching token packet and the continuous application endpoint is generated; the sending side cancels the residual increment frames in the communication queue that still depend on the old state version according to the switching confirmation record, and re-encodes the corresponding state increment into a residual increment frame that depends on the target state version.

[0084] The switching token packet is generated by the old state version, the target state version, the coverage acquisition sequence number endpoint, and the anchor frame content digest. The same switching token packet corresponds to only one version migration. The first reconstruction context continues to receive residual incremental frames that are already in transit and depend on the old state version, but no longer expands their allowed acquisition sequence number range. The second reconstruction context receives residual incremental frames of the target state version with the new anchor frame as the baseline. The edge side deduplicates repeatedly arriving frames according to the state version, acquisition sequence number, and content digest. Frames that depend on the starting point cannot be found in any context are written into the pending continuation set. When the second reconstruction context advances to the coverage endpoint, the first reconstruction context is frozen and a switching confirmation record is generated. After receiving the confirmation, the sending side cancels the old version incremental frames that have not been sent. Increments that still have state contributions and whose acquisition sequence number is later than the coverage endpoint are re-encoded from the target state version baseline. When the confirmation record is lost, the switching state is restored by the identifier of the most recent anchor frame in the subsequent feedback record.

[0085] In a further preferred embodiment, when the initial state version is established, the reconstruction deviation envelope reaches the anchoring trigger condition, or the state version switch causes the original dependent starting point to fail, the sending side reads a state snapshot from the formal solution state that can independently restore the current alignment state, and generates an anchoring frame payload in the order of state version information, continuous acquisition sequence number endpoint, transposition coverage, parallel deviation, angle deviation, correction direction, solution uncertainty interval, and content verification summary. The state version information is used to indicate the solution benchmark corresponding to the anchoring frame, the continuous acquisition sequence number endpoint is used to define the data boundary that has been included in the anchoring state, the transposition coverage uses a segment set to record the transposition segments absorbed by the current anchoring state, the parallel deviation, angle deviation, correction direction, and solution uncertainty interval together form the complete state content required for the edge side to initialize the reconstruction context, and the content verification summary is placed at the end of the payload and corresponds to the aforementioned payload content, so that the edge side can confirm that its state fields have not become inconsistent due to transmission, replacement, or re-encoding before applying the anchoring frame.

[0086] For alignment states that have changed relative to the most recent anchor frame or residual increment frames already confirmed by the edge side, the transmitting side generates a residual increment frame and writes the reference frame identifier, state version, acquisition sequence number range, transpose coverage identifier, and change field bitmaps corresponding to each state field into the residual increment frame. The change field bitmaps are set in a fixed field order of parallel deviation, angle deviation, correction direction, lower boundary of the undetermined interval, and upper boundary of the undetermined interval. Fields with a value of 1 in the bitmap carry the state difference relative to the dependent starting point in the residual increment frame, while fields with a value of 0 in the bitmap do not repeat the state value and use the corresponding value in the dependent starting point during reconstruction. The lower and upper boundaries of the undetermined interval are encoded as boundary differences relative to the dependent starting point, instead of merging them into a single interval width difference, so that the edge side can distinguish between changes in interval width, overall interval translation, and changes in one side boundary of the interval, avoiding masking the actual position changes of the alignment undetermined interval when recovering based solely on the interval width.

[0087] After receiving the residual increment frame at the edge, the corresponding anchor frame or previous residual increment frame is first located based on the reference frame identifier. Then, the state version carried by the residual increment frame is checked to see if it is consistent with the state version of the current reconstruction context. It is also determined whether the endpoint of the continuous acquisition sequence number covered by the dependent starting point can be connected with the range of acquisition sequence numbers that have been continuously applied. When the relationship between the reference frame identifier, state version, and continuous acquisition sequence number is satisfied, the state difference is read item by item according to the change field bitmap. Fields with a bitmap value of 1 are applied to the current reconstruction state, and fields with a bitmap value of 0 are retained in the state value of the dependent starting point. When any dependency condition is not satisfied, the residual increment frame is not directly written into the current reconstruction state, but is temporarily stored in the waiting area of ​​the corresponding state version until the dependent starting point is reached or a new anchor frame establishes the corresponding state baseline before the continuation judgment is performed.

[0088] When a high-resolution sensitive subsequent frame replaces an alternative frame in the communication queue, the transmitting side does not directly copy the original residual payload of the revoked alternative frame. Instead, it restores the baseline state from the dependency starting point of the revoked alternative frame, and sequentially absorbs the intermediate state increments that have not yet been revoked after the original dependency starting point and the state updates corresponding to the subsequent frames according to the acquisition sequence number. It recalculates the cumulative residual within the coverage acquisition sequence number range, and resets the change field bitmap based on the recalculated change field. At the same time, it updates the continuous acquisition sequence number range, the transposition coverage identifier, and the content verification digest. As a result, the subsequent frames after being replaced can still continuously restore the alignment state from the original dependency starting point, and the dependency break between the angle deviation field, interval boundary field, or acquisition sequence number range will not be caused by the frame revocation.

[0089] In a further preferred embodiment, the send scheduler establishes an ordered table with a deadline for each state version, using the state version, remaining send budget, scheduling level, solution sensitivity, and acquisition sequence number as table entry keys. The state version is used to divide data frames dependent on different solution benchmarks into different reconstruction context partitions. The remaining send budget is obtained by subtracting the current queue logical time stamp from the latest send time stamp of the corresponding data frame, with data frames having smaller remaining send budgets prioritized. When multiple data frames have the same remaining send budget or are within the same budget range, data frames with higher scheduling levels are prioritized. When scheduling levels are still the same, they are arranged from highest to lowest solution sensitivity. When solution sensitivity is still the same, a stable and repeatable arrangement order is determined according to the acquisition sequence number, ensuring consistent scheduling results for data frames in the same queue state.

[0090] Within each state version partition, the scheduler forms one or more transport chains based on anchor frames, residual increment frames, and their dependent origins. Data frames that have not yet been sent but satisfy the transmission dependency are designated as the head of the corresponding transport chain. Specifically, when the dependent origin of a data frame has been sent, it is allowed to become the head of the chain and is marked as an in-transit dependency even if it has not yet received confirmation from the edge side. When the dependent origin is located earlier in the same transport chain, the data frame is kept after the dependent origin and is not allowed to cross the chain order. When the dependent origin has not been sent and is not located at the beginning of the same transport chain, the corresponding transport chain is temporarily frozen until the dependent origin is qualified to send. When a state version has not yet established a state baseline through the corresponding anchor frame, the transport chain containing the anchor frame is marked as an uncrossable chain, and other residual increment frames that depend on this state version must not be moved before the anchor frame, thereby preventing the edge side from receiving residual payloads that cannot be applied independently in the absence of a complete state baseline.

[0091] After the scheduler completes the chain head sorting, it calculates the expected completion time of each chain head data frame in turn, taking into account the amount of data that each transmission chain must send before releasing the next chain head and the dequeue amount per unit time in the queue state sequence. For example, when the current logical time is 0, the amount of data to be released for the three transmission chains is 12, 8, and 10 code length units respectively, and the dequeue amount per unit time is 2 code length units, the expected completion times of the three chain heads are 6, 10, and 15 respectively. If the latest transmission time of the third transmission chain head is 12, then the chain head cannot complete the transmission on time under the current order. The scheduler selects a target transmission chain with a lower scheduling level from the transmission chains ahead of it, which, after being moved, still does not exceed its own latest transmission time, and moves the target transmission chain after the third transmission chain while keeping the internal dependency order of each transmission chain unchanged.

[0092] Before executing a transport chain move, the scheduler creates a permutation copy for feasibility assessment only, and recalculates the expected completion timestamps of each chain head within the move interval in the permutation copy. If the move causes any affected chain head to exceed its own latest transmission timestamp, the move is canceled and the search continues for other transport chains with a shift margin. When there are multiple movable target transport chains, the transport chain that can release more data to be transmitted and has a larger shift margin is selected first. After the move is completed, only the queue position, remaining transmission budget, and candidate dequeue timestamps corresponding to the entries in the move interval are updated, without performing a full table recalculation on unaffected state version partitions and transport chains. During the communication transmission process, after each data frame is completed, the transmitted frame is removed from the corresponding transport chain and a new chain head is determined. Then, based on whether the new chain head meets the dependency conditions, the scheduler executes the continue transmission, in-transit dependency marking, or transport chain freezing to ensure that the deadline scheduling and the reconstruction order between anchor frames and residual increment frames remain consistent.

[0093] In a further preferred embodiment, when the reconstructed bias envelope crosses the state partition boundary corresponding to the correction direction, or when the formal solution state undergoes a version switch and the dependent starting point of the existing residual incremental frame becomes invalid, the sending side generates a new anchor frame and generates a switching token packet based on the old state version, the target state version, the acquisition sequence number endpoint covered by the new anchor frame, and the content summary of the new anchor frame. The same set of old state version, target state version, covered acquisition sequence number endpoint, and anchor frame content summary corresponds to only one version migration. The sending side writes the switching token packet into the new anchor frame and writes the same switching token packet into the residual incremental frame generated after the new anchor frame that depends on the target state version. This allows the edge side to classify the new anchor frame and its subsequent incremental frames into the same version switching process based on the switching token packet, without mixing data generated by different anchor updates into the same reconstruction chain.

[0094] Upon receiving a new anchor frame, the edge does not immediately delete the first reconstruction context corresponding to the old state version. Instead, it retains the reconstruction state already formed in the first reconstruction context, the range of continuously applied acquisition sequence numbers, the identifier of the most recently applied anchor frame, and the old version dependency records that are still in transit. At the same time, it establishes a second reconstruction context corresponding to the target state version based on the new anchor frame. The first reconstruction context continues to receive residual incremental frames that were sent before the arrival of the new anchor frame and still depend on the old state version, but it no longer expands the range of acquisition sequence numbers that it is allowed to continue. The second reconstruction context uses the parallel deviation, angle deviation, correction direction, and solution uncertainty interval carried by the new anchor frame as the complete state baseline, and receives residual incremental frames carrying the target state version and the corresponding switching token packet.

[0095] For each data frame received at the edge, its state version, acquisition sequence number, and content summary are read first. The correspondence between these three is used to determine whether the data frame has been registered in the first reconstruction context, the second reconstruction context, or the version waiting area. When the state version, acquisition sequence number, and content summary are all the same as the registered data frame, the subsequent data frame is identified as a duplicate frame and its application is stopped. When the acquisition sequence number is the same but the content summary is different, or the dependency starting point carried by the data frame cannot be located in the first reconstruction context or the second reconstruction context, it is not directly applied to any reconstruction state. Instead, the data frame is written into the set to be continued, and its state version, dependency starting point, acquisition sequence number range, and switching token packet are recorded. The continuation judgment is re-executed after the corresponding dependency starting point arrives or the reconstruction context advances, thereby preventing out-of-order data or cross-version data from destroying the already formed continuous reconstruction state.

[0096] When the second reconstruction context is continuously applied from the new anchor frame to the end point of the coverage acquisition sequence number recorded in the handover token packet, the edge side confirms that the target state version has covered the continuous data range required for this handover, freezes the first reconstruction context and stops extending new acquisition sequence numbers to it, and generates a handover confirmation record containing the end point of the continuous application of the handover token packet and the second reconstruction context. After receiving the handover confirmation record, the sending side cancels the residual increment frames in the communication queue that still depend on the old state version and have not yet been sent, and no longer sends state increments whose acquisition sequence numbers are already covered by the new anchor frame. For increments whose acquisition sequence numbers are later than the coverage end point and still have state contributions, the residual load is recalculated from the new anchor state corresponding to the target state version and encoded as a residual increment frame that depends on the target state version.

[0097] When the handover confirmation record fails to reach the sending side due to link fluctuations, the edge side continues to carry the anchor frame identifier of the most recently applied data and the end point of the continuously applied data collection sequence number in the subsequent feedback record. The sending side aligns the feedback record with the saved handover token packet and transmission scheduling record. When the most recently applied anchor frame identifier in the feedback record corresponds to a new anchor frame and the end point of the continuously applied data is not earlier than the end point of the covered data collection sequence number, the state of the handover is restored to the completed state, and the old version residual incremental frame cancellation and target version re-encoding processing are performed in the same way as when the handover confirmation record is received. In this way, the version migration is completed without requiring the edge side to fall back to the old version state, and the full retransmission and repeated solving caused by the loss of confirmation record are reduced.

Claims

1. A method for low-latency edge processing of centering data in rotating machinery, characterized in that, include: Receive continuously generated alignment data frames and associate each alignment data frame with the acquisition sequence number, transposition segment, acquisition time stamp, status version and measurement quality identifier; Each centering data frame is input into the shadow solution process synchronized with the formal solution state on the edge side to obtain the centering state difference when the corresponding data frame is included and when the corresponding data frame is excluded, and the solution sensitivity is generated accordingly. Based on the resolution sensitivity, the remaining valid time of the corresponding transposition segment, and the communication queue status, a semantic deadline is generated, and each pair of data frames is encoded as an anchor frame, a residual increment frame, or a substitute frame. According to the semantic deadline and the desensitization scheduling order, the low desensitization replaceable frames with the same state version in the queue that have not yet been sent are replaced by later frames with high desensitization. On the edge side, the centering state is reconstructed based on the anchor frame and the residual increment frame, and a new anchor frame is inserted when the cumulative reconstruction deviation meets the anchoring trigger condition.

2. The low-latency edge processing method for centering data in rotating machinery according to claim 1, characterized in that, The shadow solving process includes: reading the current state snapshot of the formal solving state and the state version corresponding to the current state snapshot, and establishing an inclusion branch and an exclusion branch that share the current state snapshot for each pair of data frames; In the included branch, the parallel deviation, angle deviation, correction direction and uncertainty interval are updated using the corresponding data frame. In the excluded branch, the current state snapshot is maintained and the state is advanced according to the confirmed data within the same transposition segment. Compare the deviation component displacement, correction direction change, and solution uncertainty interval boundary displacement of the two branches, and constrain the reliability of the comparison results according to the measurement quality label to form a multi-component solution sensitivity bound to the state version; The multi-component desensitization and the acquisition sequence number of the corresponding data frame are written together into the transmission scheduling record.

3. The low-latency edge processing method for centering data in rotating machinery according to claim 2, characterized in that, The generation of the semantic deadline includes: establishing a queue state sequence for the communication queue carrying the pairing data frame, the queue state sequence including the number of bytes to be sent, the number of bytes dequeued per unit time, the message arrival interval, and historical queuing jitter; The effective boundary of the data is determined based on the end time stamp and acquisition time stamp of the corresponding transposition segment, and the candidate dequeue time stamp of the corresponding data frame is predicted by combining the queue state sequence. Based on the priority order of change in correction direction, displacement of uncertain interval boundary and displacement of deviation component in the solution sensitivity, the effective boundary of data is shifted forward in stages to generate a semantic deadline that includes the latest transmission time stamp, scheduling level and expiration handling identifier. When the queue state sequence changes, only the latest transmission timestamp and scheduling level corresponding to the data frames that have not yet been transmitted are recalculated.

4. The low-latency edge processing method for centering data in rotating machinery according to claim 3, characterized in that, Encoding and replacement include: encoding data capable of independently recovering the complete alignment state of the corresponding state version into an anchor frame, and encoding the state difference relative to the most recent anchor frame or the confirmed residual increment frame into the residual increment frame; For residual incremental frames with the same state version, adjacent transposition segments and no change in correction direction caused by solution sensitivity, establish alternative candidate relationships, and mark the data whose state update range can be covered by subsequent residual incremental frames as the alternative frames. When the subsequent frame has at least one of the deviation component displacement and the uncertain interval boundary displacement that is higher than the alternative frame and the remaining items are not lower than the alternative frame, and the semantic deadline of the subsequent frame is not later than the alternative frame, the dependency start point, acquisition sequence number range and transposition coverage identifier of the alternative frame are migrated to the subsequent frame, and the alternative frame is removed from the communication queue.

5. The low-latency edge processing method for centering data in rotating machinery according to claim 4, characterized in that, The shadow solving process also includes: establishing a directed graph of state updates according to the data dependency relationship between the central state variables, and configuring version identifiers and reversible incremental records for each state node in the directed graph of state updates; Based on the transposition segment and measurement component carried by each pair of data frames, locate the target node and its successor node affected by the corresponding pair of data frames in the directed graph of state update. Starting from the current state snapshot, state increments corresponding to the midpoint data frames are sequentially applied to the target node and its successor nodes to generate an inclusion branch. Then, the state increments are revoked and the current state snapshot is restored based on the reversible increment record to generate an exclusion branch. The state nodes that have changed in the two branches are aggregated into multi-component solution sensitivity according to the dependency hierarchy, and the state node index corresponding to each sensitivity component is retained.

6. The low-latency edge processing method for centering data in rotating machinery according to claim 5, characterized in that, The scheduling of the transmission order includes: converting the latest transmission time of each pair of data frames into the remaining transmission budget relative to the current queue time, and establishing an ordered table of deadlines partitioned by state version based on the remaining transmission budget; Within each state version partition, a transmission chain is formed according to the dependency order between the anchor frame and its associated residual increment frame, and then the head data frames of different transmission chains are sorted according to the scheduling level and desensitization. The estimated completion time of each data frame at the head of each chain is calculated based on the amount of bytes to be sent and the number of frames dequeued per unit time. When any estimated completion time is later than the corresponding latest transmission time, a target transmission chain with a lower scheduling level and which has not exceeded its own latest transmission time after being moved from the transmission chain preceding the corresponding transmission chain is selected. The target transmission chain is then moved after the corresponding transmission chain, and only the estimated completion time of each data frame within the moving interval is updated.

7. The low-latency edge processing method for centering data in rotating machinery according to claim 6, characterized in that, The determination of the anchoring trigger condition includes: maintaining the mirror alignment state reconstructed based on the confirmed anchoring frame and the confirmed residual increment frame on the sending side, and comparing the formal solution state with the mirror alignment state after each generation of residual increment frame to form a reconstruction deviation envelope containing the difference of each deviation component, the quantized residual direction and the number of unconfirmed dependent layers. When the reconstructed deviation envelope crosses the state partition boundary corresponding to the correction direction, or when the state version switch causes the existing dependency starting point to fail, the formal solution state, the covered acquisition sequence number range, the previous anchor frame identifier and the state version are encoded into a new anchor frame. After a new anchor frame enters the communication queue, unsent residual increment frames that depend on the old anchor frame and whose acquisition sequence number has been included in the covered acquisition sequence number range will be cancelled, and the dependency start range of subsequent residual increment frames will be rewritten to the new anchor frame.

8. The low-latency edge processing method for centering data in rotating machinery according to claim 7, characterized in that, Also includes: Feedback records are generated by the edge side according to the state version. The feedback records include the anchor frame identifier of the most recently applied data, the end point of the continuously applied data acquisition sequence number, the residual vector between the reconstruction state and the formal verification state, and the edge reception timestamp of each applied data frame. On the sending side, the feedback record and the transmission scheduling record are aligned according to the status version and the collection sequence number to obtain the correspondence between the solution sensitivity and the residual vector, and the offset relationship between the semantic deadline and the actual reception time stamp. According to the transposition segment and communication queue status, the corresponding relationship and offset relationship are partitioned and accumulated to form a sensitivity calibration table and a deadline calibration table; The sensitivity calibration table is used to correct the dominance decision boundary of the alternative frame, and the cutoff calibration table is used to correct the hierarchical forward shift of the data validity boundary. The correction is only applied to the centering data frames generated after the state version corresponding to the feedback record.

9. The low-latency edge processing method for centering data in rotating machinery according to claim 8, characterized in that, The partitioning accumulation includes: dividing the feedback records into complete dependency chain feedback, alternative chain feedback where frame replacement occurs, and truncated chain feedback where semantic deadlines have expired, based on the continuously applied collection sequence number endpoint and transmission scheduling record. The sensitivity calibration table is updated using the corresponding data of the latest transmission time and the edge reception time in the complete dependency chain feedback, and the deadline calibration table is updated using the offset data of the latest transmission time and the edge reception time in the truncated chain feedback. For the feedback of the substitution chain, the dominance decision boundary is updated based on the state difference encoded by the revoked substituted frame and the subsequent frame, respectively, and the distance relationship between the two substitution state trajectories and the formal verification state in the feedback record. Apply a monotonic constraint to the update results between different communication queue state partitions, which does not delay the latest transmission time stamp as the queue load increases.

10. The low-latency edge processing method for centering data in rotating machinery according to claim 9, characterized in that, The switching of the new anchor frame includes: generating a switching token packet containing the old state version, the target state version, and the endpoint of the overlay acquisition sequence number, and writing the switching token packet into the new anchor frame and the residual increment frame generated thereafter. After receiving a new anchor frame at the edge, the first reconstruction context corresponding to the old state version is retained, and a second reconstruction context corresponding to the target state version is established. The first reconstruction context or the second reconstruction context is written according to the state version and dependency starting point carried by each residual increment frame. When the second reconstruction context is continuously applied to the end of the coverage acquisition sequence number, a switching confirmation record containing a switching token packet and the continuous application end is generated; The sending side cancels residual incremental frames that still depend on the old state version in the communication queue based on the switching confirmation record, and re-encodes the corresponding state increment into residual incremental frames that depend on the target state version.

11. The low-latency edge processing method for centering data in rotating machinery according to claim 10, characterized in that, The load of the anchor frame is sequentially written with the status version, continuous acquisition sequence end point, transposition coverage, parallel deviation, angle deviation, correction direction, solution uncertainty interval and content verification summary; the residual increment frame records the reference frame identifier and change field bitmap, and records the difference between the upper and lower boundaries of the solution uncertainty interval relative to the dependent starting point.

12. The low-latency edge processing method for centering data in rotating machinery according to claim 11, characterized in that, The key value of the deadline ordered table consists of the state version, remaining transmission budget, scheduling level, desensitization, and acquisition sequence number. Data frames that have been sent from the dependent starting point or are located at the beginning of the same transmission chain are identified as the chain head. Before the corresponding state version is established, the transmission chain where the anchor frame is located is marked as an uncrossable chain to prevent the residual increment frame from being sent before the state baseline.

13. The low-latency edge processing method for centering data in rotating machinery according to claim 12, characterized in that, The switching token packet is generated by the old state version, the target state version, the end point of the covered collection sequence number, and the content summary of the new anchor frame. The edge side deduplicates the data frames that arrive repeatedly according to the state version, collection sequence number, and content summary. Data frames that cannot be located in the first reconstruction context or the second reconstruction context due to their dependent starting point are written into the pending continuation set. The first reconstruction context is frozen after the second reconstruction context advances to the end point of the covered collection sequence number.

Images