A digital twin assembly and correction method for scaffolding and roof steel frame
By using a digital twin assembly topology model, the problems of difficult hole insertion and insufficient ridge closure during the assembly of the scaffolding and roof steel frame were solved, enabling synchronous updates of on-site data and control of response errors, and providing feasible correction instructions.
Patent Information
- Application Number
- CN202611288205.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-10
AI Technical Summary
The existing construction model is difficult to update on-site geometric measurements, component orientation, member strain, support displacement and wind load data in a synchronous manner, which leads to problems such as difficulty in inserting holes, exceeding the limit of vertical coordinates at mid-span, or insufficient closure of the ridge during the assembly of the scaffold and roof steel frame.
A digital twin model of the assembly topology is established. Through time alignment and sensitivity propagation of multi-source data, a local influence domain is formed. The set of feasible poses with six degrees of freedom is calculated, and the ridge closure margin is calculated in reverse. Mixed candidate actions that do not meet the constraints are eliminated, and the posture adjustment, support adjustment and installation sequence instructions are output.
It enables the simultaneous correction of component positions, loads, and support boundaries during assembly, reduces response error boundaries, predicts stress in adjacent members and future interface poses, avoids difficulties in hole insertion and insufficient ridge closure, and generates repeatable correction instructions.
Smart Images

Figure CN123069575A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of steel structure assembly construction and digital twin correction, and particularly to a digital twin assembly correction method for scaffolding and roof steel frame. Background Technology
[0002] Segmented scaffolding and roof steel frames are typically constructed by hoisting, temporarily supporting, bolting, and fastening multiple components in stages according to predetermined connections. Errors in component manufacturing and measurement, changes in hoisting posture, settlement of temporary supports, on-site wind loads, and the order of insertion and tightening of double-bolt joints all affect the orientation of component interfaces and structural stress, which are then transmitted to subsequent interfaces and the ridge closure position as assembly progresses.
[0003] Existing construction models are mostly based on design geometry and fixed boundaries, making it difficult to synchronously update node constraints and structural status using on-site geometric measurements, component orientation, member strain, support displacement, wind speed and direction, and bolt tightening data. At the same time, when adjusting the orientation of a single component or changing the tightening and installation sequence, there is usually a lack of linkage prediction of the stress of adjacent members, the sweep space of components and tools, the feasible orientation of future interfaces, and the ridge closure margin. It is easy to find that the hole cannot be inserted, the vertical coordinate at the mid-span exceeds the limit, or the final closure is insufficient only after the node is finally tightened.
[0004] Therefore, a digital twin assembly and correction method for scaffolding and roof steel frames is needed to address the shortcomings of the existing technologies. Summary of the Invention
[0005] One objective of this invention is to propose a digital twin assembly and correction method for scaffolding and roof steel frames. Addressing the problems in existing technologies such as the inability to synchronously update the construction model through on-site observation, the difficulty in tracking the impact of local adjustments on adjacent forces and subsequent installation space, and the lack of ridge closure prediction before tightening, this invention proposes establishing an assembly topology digital twin model containing component geometric and mechanical parameters, temporary support variable displacement boundaries, and double-bolt staged constraints. It utilizes time-aligned multi-source data to update the model state, and forms a local influence domain based on event sensitivity, port convergence, and error expansion. It treats attitude adjustment, support, double-bolt operation, and installation sequence as mixed candidate actions, and constructs a set of feasible six-degree-of-freedom poses for each action, intersecting them in the opposite direction towards the ridge. Based on hard constraints and comprehensive cost, it releases segmented correction commands. This invention achieves the technical effect of simultaneously considering hole insertion, sweep space, member stress, mid-span vertical coordinates, and ridge closure accuracy during the still-adjustable stage.
[0006] This invention provides a digital twin assembly and correction method for scaffolding and roof steel frame, comprising:
[0007] S1. Based on the design, measured geometry and node relationships, establish an assembly topology digital twin model that characterizes the geometry and mechanics of the components, the temporary support boundary and the state of the double bolt nodes;
[0008] S2. Time-align the on-site geometry, mechanics, boundary, environment, and fastening data, estimate the status of components, supports, and double-bolt nodes based on observation and prediction residuals, and update the model.
[0009] S3. The sensitivity of the structural response and interface pose is propagated along the assembly topology by responding to changes in the assembly state of the component, temporary support boundary, double bolt node state or wind load. Based on the sensitivity, a local influence domain is formed, and the external structure is condensed into port stiffness and equivalent load. The local influence domain is then expanded according to the port error.
[0010] S4. Form a mixed candidate action that includes component orientation adjustment, support adjustment, double bolt operation and installation sequence. Calculate the stress of adjacent members and the predicted pose of the future interface for each action based on the port stiffness and equivalent load. Based on the predicted pose of the future interface and the geometric, spatial and mechanical constraints in the model, roll to construct a set of six-degree-of-freedom feasible poses for subsequent installation and intersect them in the opposite direction of the ridge to determine the ridge closure margin.
[0011] S5. Eliminate mixed candidate actions that are not feasible in the future or violate assembly accuracy, spatial collision avoidance, stress and closure constraints. Determine the comprehensive cost based on the closure margin, stress and amplitude of the remaining actions. Select and output attitude adjustment, support adjustment, double bolt operation and installation sequence instructions. After retesting, return to step S2.
[0012] Optionally, S1 includes: establishing a local coordinate system for each component and recording the cross-sectional parameters, material parameters, designed interface pose, and measured interface pose;
[0013] The components and double-bolt nodes are respectively used as component nodes and connection edges in the assembly topology digital twin model;
[0014] Establish a displacement boundary for the temporary support that can be updated by measured settlement or jacking along the direction of the support action;
[0015] For each double-bolt node, record the measured hole center, hole axis direction, hole diameter, bolt diameter, hole-bolt clearance, preload parameters, and tightening stage of the two connecting holes.
[0016] Optionally, S2 includes: estimating the state of the double-bolt node by: identifying the state of the double-bolt node as an uninserted state, a first bolt positioning state, a two-bolt inserted state, an initial tightening state, or a final tightening state based on the insertion signals of the two bolts, the torque-rotation sequence, the strain changes of the members adjacent to the node, and the relative poses of the members on both sides of the node.
[0017] According to the identified double bolt node state, switch the hole-bolt contact degree of freedom and sliding boundary, and update the equivalent preload and node tangential stiffness. The first bolt positioning state corresponds to retaining the relative rotational degree of freedom around the first bolt axis, the two bolt insertion state corresponds to retaining the relative sliding degree of freedom within the hole-bolt gap range, and the initial tightening state and the final tightening state correspond to the equivalent preload range determined by their respective torque-angle intervals.
[0018] Furthermore, updating the assembly topology digital twin model includes: registering the on-site geometric measurement data with the measurement reference points in the corresponding component local coordinate system to form the geometric residual;
[0019] The strain residual is formed by calculating the strain based on the strain data of the member and the strain calculated by the assembly topology digital twin model;
[0020] The support residual is formed based on the temporary support displacement data and the current displacement boundary.
[0021] The tightening event residual is formed by comparing the state transition time corresponding to the bolt tightening data with the state transition time predicted by the model.
[0022] The inverse matrix of the measurement covariance of each residual is used as the weight to perform joint estimation of each residual, and the joint estimation result is written into the model state version with the corresponding timestamp.
[0023] Optionally, S3 includes: forming the local influence domain includes: taking the location where the change occurs as the propagation starting point, and calculating the sensitivity of the port state to the changed state quantity step by step along the component edge connected to the propagation starting point;
[0024] Propagation stops at a port when the sensitivity norm at a port is not greater than the sensitivity cutoff threshold and the condensation response error at that port in the previous model state version is not greater than the port error cutoff threshold.
[0025] The component-node sub-graph enclosed by the stop propagation port is defined as the local influence domain. The external degrees of freedom are eliminated by static condensation into the port stiffness and equivalent load acting on the stop propagation port. The allowable stress calculation error, allowable interface pose error and allowable ridge closure error are divided by the upper bound of the state quantity of the corresponding sensitivity propagation path, and the three ratios are arranged in ascending order of value. The first ratio after the arrangement is used as the sensitivity cutoff threshold.
[0026] Furthermore, extending the local influence domain includes: under the same state change and the same model state version, using the port stiffness and equivalent load to solve the local mechanical response after condensation, and taking the difference between the displacement back substitution value of the port that stops propagating in the local mechanical response and the corresponding port displacement value of the verification solution obtained after incorporating the adjacent layer of components outside the port as the port error.
[0027] When the error at any port exceeds the port error cutoff threshold, at least one layer of adjacent components outside the port is incorporated into the local influence domain and then re-condensed and solved until the error at each port is no greater than the port error cutoff threshold or the local influence domain is expanded to the complete assembly topology. The displacement, internal force, stress utilization rate and future interface prediction pose of adjacent members are obtained from the final local mechanical response.
[0028] Optionally, S4 includes: forming the mixed candidate action includes: discretely generating three-dimensional translation and three-dimensional rotation within the allowable orientation range of the component; generating lifting or unloading within the stroke limit and bearing limit of the temporary support; generating two sequences for the two bolts to be tightened, namely, the first bolt first and the second bolt first; and generating an installation sequence for subsequent components that meet the current connection prerequisite relationship.
[0029] A wind load scenario set is formed based on the current wind speed and direction and the predicted wind speed and direction during the execution period of the candidate actions. Based on the measurement covariance of geometric measurement data, component pose data and member strain data, pose uncertainty intervals and stress uncertainty intervals are formed according to the coverage factor corresponding to the preset confidence probability. For each hybrid candidate action, the intersection of feasible pose sets and the stress envelope of adjacent members within the wind load scenario set are calculated. When the predicted wind speed and direction are unavailable, the current wind speed component is superimposed with the wind speed safety increment determined by the upper bound of the current wind speed measurement uncertainty to form an alternative wind load scenario.
[0030] The following constraints are defined as hard constraints: the hole insertion constraint that does not allow the introduction of relaxation, the component and tool sweep body non-intersection constraint, the stress limit constraint, the allowable range constraint of the vertical coordinate at mid-span, and the lower limit constraint of the ridge closure margin. The motion amplitude preference that allows the introduction of relaxation and whose relaxation is included in the comprehensive cost is defined as soft constraints. The hard constraint screening is performed according to the boundary values of the intersection of the feasible pose set and the stress envelope of the adjacent members.
[0031] Furthermore, constructing the six-degree-of-freedom feasible pose set includes: taking the predicted pose of the future interface as the set center, and performing an intersection operation on the insertable translational and rotational intervals corresponding to the hole-bolt gap, the pose intervals when the component and tool sweep bodies do not intersect, the allowable pose adjustment range, and the elastic displacement intervals determined by the stress limit of the rod.
[0032] Based on the connection relationship of subsequent components, the intersection obtained in the current installation step is transformed to the interface coordinate system of the next installation step, and the intersection with the corresponding intersection of the next installation step is continued to be found until the ridge interface.
[0033] The three translational degrees of freedom are normalized by the allowable positional deviation of the ridge interface, and the three rotational degrees of freedom are normalized by the allowable attitude deviation of the ridge interface. In the normalized six-degree-of-freedom pose space, the shortest distance from the predicted closed state to the boundary of the set obtained by intersecting the ridge interface is assigned a positive or negative sign according to whether it is inside or outside the set. The lower limit of the closure margin is determined by the allowable positional deviation of the ridge interface, the allowable attitude deviation, and the confidence interval boundary determined by the measurement uncertainty according to the preset confidence probability and coverage factor.
[0034] Optionally, S5 includes: determining the comprehensive cost includes: normalizing the negative number of the ridge closure margin, the stress increment of adjacent members, the three-way translation, the three-way rotation, the temporary support adjustment, and the installation sequence change, and then performing a weighted summation, wherein the installation sequence change is the number of inversion pairs of the candidate installation sequence relative to the design installation sequence.
[0035] When the number of subsequent components is not less than 2, the installation sequence change amount is normalized by half of the product of the number of subsequent components and the number minus one. When the number of subsequent components is less than 2, the installation sequence change amount is zero and the normalization is not performed.
[0036] The allowable interval width of each of the other normalization scales is taken respectively, and the weights are determined according to the distribution ratio of the allowable closure error of the current assembly stage among the evaluation quantities.
[0037] When the combined cost of two action sequences is the same, first select the one with the largest ridge closure margin; if they are still the same, select the one with the smallest stress increment of adjacent members; if they are still the same, select the one with the fewest members that need to be adjusted.
[0038] Furthermore, outputting and executing the posture adjustment, support adjustment, double bolt operation and installation sequence instructions includes: when there are no other mixed candidate actions after eliminating mixed candidate actions, stopping the release of the final tightening instruction of the current double bolt node and subsequent installation instructions, maintaining the current adjustable state, collecting retest data and returning to step S2 to update the model and regenerate mixed candidate actions;
[0039] When other mixed candidate actions exist, the selected action sequence is decomposed into component posture adjustment instructions before final tightening, temporary support adjustment instructions, double bolt operation instructions, and subsequent component installation sequence instructions. Each instruction records the component identification, field coordinate system, point of action, direction, amplitude, execution stage, retest conditions, and corresponding model state version. The subsequent component installation sequence instructions also record the connection prerequisite relationship and sequence position.
[0040] After each instruction is executed, collect the geometric measurement data, strain data of the members, displacement data of the temporary supports and bolt tightening data specified by the retest conditions, and determine the interface geometric deviation, stress increment of adjacent members and reduction of ridge closure margin, respectively.
[0041] The corresponding component interface tolerance is used as the geometric allowable deviation, the difference between the stress limit of the member and the estimated stress of the member in the model state version before the command is executed is used as the stress allowable increment, and the difference between the ridge closing margin and the lower limit of the closing margin before the command is executed is used as the closing margin allowable reduction.
[0042] When the interface geometric deviation exceeds the allowable geometric deviation, the stress increment of adjacent members exceeds the allowable stress increment, or the reduction in the ridge closure margin exceeds the allowable reduction in the closure margin, the retest result is marked as an execution anomaly, subsequent instructions are stopped and the process returns to S2; otherwise, the next instruction is released.
[0043] The beneficial effects of this invention are:
[0044] 1. By jointly estimating the residuals of multi-source field data and the assembly topology digital twin model, the component position, load, support boundary and tangential stiffness of double bolt nodes can be corrected simultaneously, so that subsequent action evaluation is based on the current construction state.
[0045] 2. By using sensitivity propagation, out-of-domain port aggregation, and port error-triggered domain expansion, the mechanical solution range required for each event and candidate action can be reduced while maintaining the response error boundary, and the stress of adjacent members and the predicted pose of future interfaces can be obtained.
[0046] 3. By finding the intersection of the feasible pose set of the six degrees of freedom with the ridge in the opposite direction and evaluating the directional closure margin, actions that could cause damage such as hole insertion, installation sweep space, mid-span vertical coordinates, member stress, or final closure conditions can be eliminated before the final tightening of the node, and a repeatable attitude adjustment-support-tightening closed-loop command can be formed. Attached Figure Description
[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0048] Figure 1 This is a flowchart of the digital twin assembly and correction method for the canopy frame and roof steel skeleton of the present invention.
[0049] Figure 2 This is a flowchart of the formation and expansion of the local influence domain in S3 of the present invention. Detailed Implementation
[0050] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0051] refer to Figures 1-2 A digital twin assembly and correction method for scaffolding and roof steel frame, comprising:
[0052] S1. Based on the design, measured geometry and node relationships, establish an assembly topology digital twin model that characterizes the geometry and mechanics of the components, the temporary support boundary and the state of the double bolt nodes;
[0053] S2. Time-align the on-site geometry, mechanics, boundary, environment, and fastening data, estimate the status of components, supports, and double-bolt nodes based on observation and prediction residuals, and update the model.
[0054] S3. The sensitivity of the structural response and interface pose is propagated along the assembly topology by responding to changes in the assembly state of the component, temporary support boundary, double bolt node state or wind load. Based on the sensitivity, a local influence domain is formed, and the external structure is condensed into port stiffness and equivalent load. The local influence domain is then expanded according to the port error.
[0055] S4. Form a mixed candidate action that includes component orientation adjustment, support adjustment, double bolt operation and installation sequence. Calculate the stress of adjacent members and the predicted pose of the future interface for each action based on the port stiffness and equivalent load. Based on the predicted pose of the future interface and the geometric, spatial and mechanical constraints in the model, roll to construct a set of six-degree-of-freedom feasible poses for subsequent installation and intersect them in the opposite direction of the ridge to determine the ridge closure margin.
[0056] S5. Eliminate mixed candidate actions that are not feasible in the future or violate assembly accuracy, spatial collision avoidance, stress and closure constraints. Determine the comprehensive cost based on the closure margin, stress and amplitude of the remaining actions. Select and output attitude adjustment, support adjustment, double bolt operation and installation sequence instructions. After retesting, return to step S2.
[0057] In this specific embodiment, S1 includes:
[0058] This specific implementation takes a steel frame consisting of twelve portal frames, two rows of roof longitudinal tie rods, and a ridge connecting section as the implementation object. The digital twin server reads the design model, component factory measurement documents, on-site total station point documents, and node assembly list, and writes the component identification, batch number, installation level, design prerequisite relationship, and measurement timestamp into the assembly object registration form. The component identification serves as the unified primary key for subsequent geometric records, mechanical records, support records, and fastening records. Records lacking component identification or timestamps are marked as pending review and are not written into the current model state version.
[0059] The server establishes a right-handed local coordinate system for each component, with the component axis as the local first axis and the principal axes of the section as the local second and third axes, and then... Complete the transformation from local coordinates to field coordinates. Indicates the first Within the first component The three-dimensional coordinate vectors of each measuring point in the field coordinate system, in mm, are derived from total station measurement records. Indicates the first The orthogonal rotation matrix from the local coordinate system of a component to the field coordinate system is a dimensionless quantity and satisfies... , Indicates the same measuring point on the 1st The three-dimensional coordinate vectors in the local coordinate system of each component, in mm, are derived from the component's factory measurement documents. Indicates the first The translation vector of the origin of the coordinate system of each component, in mm, is obtained from the rigid body registration results of no less than three non-collinear reference points. This represents the component serial number, with values ranging from 1 to the total number of currently registered components. Indicates the first The serial number of the measuring point within each component ranges from 1 to the total number of valid measuring points for that component. This represents the matrix transpose operation; it is a mathematical notation and units are not applicable. It represents a third-order identity matrix and is a dimensionless quantity. When the root mean square error of registration exceeds 2.0 mm, the original design pose is retained and the component is marked as needing to be remeasured.
[0060] The component node record includes component identification, cross-sectional type, cross-sectional area, two principal moments of inertia, torsional constant, elastic modulus, Poisson's ratio, density, yield strength, designed interface orientation, measured interface orientation, and current installation status. The cross-sectional area is recorded in mm², and the principal moments of inertia and torsional constant are recorded in mm. 4 Records are kept in MPa for elastic modulus and yield strength, and in kg / m³ for density. Sectional parameters are from the design model and factory quality inspection documents, and material parameters are from the re-inspection report of the same batch of materials. When the value falls outside the valid range of the quality inspection report, the record will not be used to cover the design value, and a parameter conflict flag will be generated.
[0061] The server constructs an assembly topology diagram with component nodes as vertices, connection relationships between components, and double-bolt nodes as connection edges. The connection edges record the starting component identifier, ending component identifier, interface number, connection type, current fastening stage, allowable relative degrees of freedom, contact state, and previous installation relationships. When constructing the diagram, duplicates are removed using the composite key composed of component identifier and interface number. For connection edges that exist in the design model but are missing in the site list, the design record is retained and set to an unassembled state. For connection edges added on site but not in the design model, they are prevented from entering the solution diagram and a site verification task is generated.
[0062] For each temporary support, the support record saves the support identifier, the identifier of the connected component, the on-site point of action, the unit direction of action, the initial displacement boundary, the current settlement, the current jacking amount, the travel limit, the bearing limit, the sampling time, and the sensor calibration version. The server updates the initial displacement boundary along the support's direction of action by subtracting the current settlement from the current jacking amount. All displacements are recorded in mm, with positive values indicating the jacking direction. The direction of action is obtained by fitting the layout coordinates and the support axis and normalizing it into a dimensionless unit vector. Update values exceeding the travel limit are truncated to the corresponding boundary and an out-of-bounds flag is set. Supports with out-of-bounds flags are only used for state estimation and not for generating new adjustment actions.
[0063] For each double-bolt node, the node record saves the measured three-dimensional coordinates of the hole center, the unit vector of the hole axis, the hole diameter, the corresponding bolt diameter, and the radial clearance between the hole and the bolt according to the first and second connecting holes respectively. It also saves the preload parameters of the two bolts, the torque-angle reference range, the insertion signal channel, the tightening stage, and the data validity mark. The coordinates and diameter are recorded in mm, the preload is recorded in kN, the torque is recorded in N·m, and the angle is recorded in degrees. The hole center and hole axis are obtained by cylindrical fitting and the fitting residual is not greater than 0.5 mm. When the hole diameter is smaller than the bolt diameter or the hole axis angle exceeds the design allowable value, the node is set to a geometrically non-insertable state.
[0064] This implementation uses a model parameter library to uniformly maintain the cross-section material table, measurement covariance table, support capacity table, and bolt tightening calibration table. Each table records the version number, effective time, data source, applicable component or node type, field unit, and effective range. The cross-section material table is generated by merging the design documents and the re-inspection report. The measurement covariance table is calculated by equipment calibration and at least thirty sets of samples of repeated measurements at the same point. The support capacity table is formed by no-load zeroing and graded loading tests. The bolt tightening calibration table is formed by axial force-torque-rotation angle tests of the same specification connection pairs. When the field equipment is recalibrated, the material batch changes, or the connection pair specifications change, a new version is created without overwriting the historical version.
[0065] After the topology and parameters are loaded, the server performs integrity checks based on the reference relationships of component nodes, connection edges, support boundaries, and double bolt status records. Only records with primary keys, consistent units, effective versions, and registered at both ends of the connection are written into the initial model state. The initial model state, along with coordinate transformations, parameter versions, and data validity flags, is saved as the baseline of the assembly topology digital twin model, and its model state version number and timestamp are passed to S2 as the benchmark for multi-source data alignment and status updates on site.
[0066] In this specific embodiment, S2 includes:
[0067] In this digital twin assembly correction method for the scaffolding and roof steel frame, the state estimation server uses the model state version generated by S1 as a reference and establishes a fusion time every 200ms. It receives the coordinates of the total station interface, the readings of the component strain acquisition device, the displacement of the temporary support, the wind speed and direction, the bolt insertion switch quantity, and the torque-angle sequence of the electric torque wrench. The original records are uniformly written into the observation buffer with the device identifier, sampling time, reception time, quality code and model state version number. The timestamp is corrected by the field clock synchronization service. The time difference between the corrected devices is no more than 20ms.
[0068] For continuous quantities, the server selects adjacent valid samples within a 500ms window before and after the fusion time and interpolates them linearly over time. For inserted signals and the consolidation phase, the most recent valid state before the fusion time is retained. For wind speed and direction, the vector average value is calculated using a 2s sliding window. Records that exceed the equipment calibration range, have invalid quality codes, or have a rate of change from the previous valid sample that exceeds the physical upper limit are removed. When there are no valid samples in a window for a certain data channel, the estimated value at the previous fusion time is retained and the measurement variance of that channel is increased, so that the weight of missing information in the joint estimation is automatically reduced.
[0069] Linear interpolation is calculated first. Recalculate , The dimensionless interpolation coefficients at the fusion time are restricted to between 0 and 1. The fusion time is expressed in milliseconds (ms). This indicates the time of the previous valid sample within the window, in milliseconds. This indicates the time of the last valid sample within the window, in milliseconds (ms). Represents continuous observations at the fusion time, with units consistent with the original channel. Represents the previous valid sample value and its unit is . Consistent, Indicates the next valid sample value and the unit is the same as the previous valid sample value. If the time of the preceding and following samples is the same or either sample is missing, the division is not performed and the branch that preserves the most recent valid value is entered.
[0070] The server calculates standardized residuals for each type of observation. , Indicates the fusion moment No. The standardized residuals of the observations are dimensionless and truncated between -6 and 6. This represents the corresponding field observation value, with units consistent with the source record for that channel. This represents the model's predicted value for the same physical quantity at the same time, with units equal to or equal to the predicted value. same, Indicates the standard deviation of measurement and is The measurement covariance table, from the same unit and sourced from S1, is amplified at time intervals when missing measurements are maintained. This indicates the valid sequence number at the current fusion moment. The fusion time index is indicated. The variance of the geometry and support channels is in mm² and the standard deviation is in mm. The variance of the strain channel is in με² and the standard deviation is in με. The variance of the fastening event channel is in s² and the standard deviation is in s.
[0071] The geometric residuals are obtained by rigidly registering the field interface measurement points with the corresponding local measurement reference points of the component after S1 coordinate transformation. The geometric residuals are recorded in mm and the corresponding measurement covariance is recorded in mm². The strain residuals are obtained by subtracting the measured strain of the member after temperature compensation from the finite element predicted strain. The strain residuals are recorded in με and the corresponding measurement covariance is recorded in με². The support residuals are obtained by subtracting the measured settlement or jacking amount from the current displacement boundary of the model. The support residuals are recorded in mm and the corresponding measurement covariance is recorded in mm². The fastening event residuals are obtained by subtracting the measured transition time of the insertion, initial tightening, or final tightening state from the model predicted transition time. The fastening event residuals are recorded in s and the corresponding measurement covariance is recorded in s². Each residual carries the observation source, the corresponding component or node identifier, and the covariance version.
[0072] Joint estimates The objective function is defined, and the Gaussian-Newton iteration is used to obtain the model state increment. Indicates the fusion moment The weighted sum of squared residuals is a dimensionless, non-negative quantity, derived from all standardized residuals of the current iteration. Indicates the fusion moment The total number of valid observations used in the estimation is a positive integer and at least 4. The aforementioned first Standardized residuals and The meaning is consistent with the aforementioned residual formula. In each iteration, the observation function is re-linearized and the component translation increment is limited to no more than 10 mm, the rotation increment to no more than 0.5 degrees, and the support boundary increment to no more than 5 mm. The iteration ends when the relative decrease of the objective function is less than one-thousandth or when the iteration reaches eight times.
[0073] The double-bolt node uses a deterministic state machine to identify the tightening stage. When both insertion channels are zero, it is in the uninserted state. When only the first insertion channel is one, it is in the single-bolt positioning state of the first bolt and the actual inserted bolt is marked as the first bolt. When only the second insertion channel is one, it is in the single-bolt positioning state of the second bolt and the actual inserted bolt is marked as the second bolt. When both channels are one and the torque has not entered the initial tightening range, it is in the two-bolt insertion state. When both channels are one and the torque-rotation data of any bolt at three consecutive points enters the initial tightening range of the calibration table, it is in the initial tightening state. When both bolts enter the final tightening range at three consecutive points and the strain change of the adjacent members and the relative posture of the components on both sides have not exceeded the allowable boundary, it is in the final tightening state. The state only moves forward in this order. When there is a signal conflict, the previous state is maintained and set to pending verification.
[0074] In the first bolt single-bolt positioning state, the model constrains the relative translation of the first bolt hole center and retains the relative rotational degree of freedom about the first bolt axis; in the second bolt single-bolt positioning state, the model constrains the relative translation of the second bolt hole center and retains the relative rotational degree of freedom about the second bolt axis; the hole center, hole axis, hole-bolt radial clearance, subsequent state transitions and tightening records all refer to the same bolt according to the actual inserted bolt identification, and it is forbidden to replace it with the hole center or axis of another bolt; in the two bolt insertion states, the relative sliding degree of freedom within the hole-bolt radial clearance range is retained; in the initial tightening and final tightening states, the equivalent preload range and nodal tangential stiffness are obtained by looking up the torque-angle interval of the bolt tightening calibration table, respectively; the calibration table uses bolt specification, connecting plate thickness, surface treatment and tightening stage as lookup keys, and the value fields are the lower and upper limits of preload, tangential stiffness and confidence interval; if no complete key is matched, it is not allowed to enter the final tightening state and the conservative lower limit value of the same specification is used;
[0075] The joint estimation output is written into the state field according to the component pose, member strain state, support displacement boundary, wind load vector, double bolt state, equivalent preload and nodal tangential stiffness. Each field records the estimated value, covariance, observation coverage and quality flag. After the estimation increment is applied, all standardized residuals are recalculated. When the absolute value of any residual is still greater than 4, the update result is retained but the corresponding component or node is marked as an object affected by abnormal observations, so that S3 can expand the set of local propagation starting points.
[0076] The measurement covariance matrix is stored in square units for each observed physical quantity. The variance units for geometric coordinates and support displacement are mm², strain variance is με², the variance unit for the tightening event time is s², the variance unit for wind speed component is (m / s)², the variance unit for torque is (N·m)², and the variance unit for rotation angle is degrees². When geometric displacement forms dimensionless residuals, it is only divided by the standard deviation in mm of the same unit. mm² is only used for covariance storage and inverse variance weighting. Torque observation is only used for tightening condition identification and preload calibration. This step does not perform torque-to-force conversion. If conversion is necessary, N·m should be multiplied by 1000 to convert to N·mm and then divided by the mm lever arm to obtain N, and it must not be converted to kg. The weight of individual residuals is as follows: Sure, Indicates the fusion moment No. The inverse variance weights of each observation, in units of the reciprocal of the square of the corresponding observation unit, and derived from the current measurement covariance version. This indicates that the standard deviation of the aforementioned observations is the same as the observed value, and the unit is the same. and The meaning of its value is consistent with the standardized residual formula;
[0077] After the state increment is solved by constraints, it is... Write the new state. Indicates the fusion moment The model state vector, after being normalized to its respective allowable range, is a dimensionless vector with each component limited to between -1 and 1. This represents the dimensionless state vector at the previous fusion time step, derived from the parent model's state version. This represents the dimensionless state increment vector obtained from this joint estimation, which originates from the results of the objective function constraint iteration. This represents the fusion time index. If the iteration matrix is ill-conditioned or has fewer than 4 valid observations, this increment is not applied. Instead, the previous model state is maintained as a safe state, and a re-sampling task is generated.
[0078] When S5 returns the command execution result, the return packet retains the pre-tightening component posture adjustment command, temporary support adjustment command, double bolt operation command and subsequent component installation sequence command formed by the decomposition of the selected action sequence. The component identifier, field coordinate system, point of action, direction, amplitude, execution stage, retest conditions, corresponding model state version, connection prerequisite relationship and sequence position in each command are used as retest data association keys. The server only receives the observations corresponding to the released commands. Command data that has not been released or has an inconsistent version enters the isolation buffer.
[0079] After each executed instruction, the server collects geometric measurement data, member strain data, temporary support displacement data, and bolt tightening data specified by the retest conditions. It uses the corresponding component interface tolerance as the geometric allowable deviation, the difference between the member stress limit and the member stress estimate in the model state version before instruction execution as the allowable stress increment, and the difference between the ridge closure margin and the lower limit of the closure margin before instruction execution as the allowable reduction of the closure margin. If any retest exceeds the corresponding allowable amount, it maintains the current adjustable state, does not move the double bolt tightening state forward, and generates an anomaly flag. Only if it does not exceed the allowable amount is the retest result allowed to be incorporated into the joint estimation.
[0080] The acceptance rules for joint estimation are as follows: when the objective function decreases, all state increments do not exceed their respective limits, and the median of the normalized residuals does not increase, the current update is submitted. If any of the conditions are not met, the current state increment is reduced to half of its original value, and the observation prediction and objective function are recalculated. If the conditions are not met for three consecutive times, the update at this fusion time is cancelled, the parent model state version is maintained, and a reason code is generated for the supplementary sampling task where the estimation does not converge. The next fusion time will only be retried after the supplementary sampling channel is restored or the total number of effective observations increases.
[0081] Anomaly suppression is performed on a continuous cycle. When the absolute value of the standardized residual of a channel is greater than 4 for three consecutive fusion times, the quality flag of that channel is set to deweighted and the measurement variance is amplified by four times. When it returns to no more than 2 for five consecutive fusion times, the original covariance version is restored. Channels in the deweighted state are still kept in the observation buffer for traceability, but the double bolt state forward movement must not be triggered alone.
[0082] Before submitting the model state version, the topology consistency rule is executed. If the relative displacement of the hole center of the final tightened connection edge exceeds the upper limit of the hole-bolt gap after the component interface pose is updated, the pose update is rejected and the corresponding node is set to the constraint conflict state. If the calculated reaction force exceeds the support bearing limit after the temporary support boundary is updated, the boundary is kept at the previous valid value and a support unloading verification task is generated. The new version is only released when all connection edges and support boundaries pass the consistency check.
[0083] When S5 executes and the retest returns to S2, the server retrieves the state before execution based on the model state version number in the instruction. It updates the retested geometry, strain, support, and fastening records according to the same time alignment and joint estimation rules. If the retested record lacks an instruction identifier or the state version is inconsistent, it will not be merged with the observations of this cycle. After completion, a new model state version that only adds and does not modify is generated. The parent version number, fusion time, parameter library version, state increment, and abnormal observation flag are saved. This model state version is then passed to S3 as the sole input for sensitivity propagation and local cohesion.
[0084] In this specific embodiment, S3 includes:
[0085] The local solver reads the new model state version output by S2 and registers changes in component assembly state, temporary support boundary increments, double bolt state switching, and wind load vector changes as state change events. Each event records the location of occurrence, change type, change amount, original state version, and new state version. When multiple events act on the same component or node, they are merged into a propagation starting point set. Events from different model state versions are not mixed in the same local solution.
[0086] The solver expands step by step along the component edges connected to the current layer in the assembly topology, with the propagation starting point as the zeroth layer. It applies a unit perturbation to the changing state variables on the linearized equilibrium equations and calculates the incremental response of the port displacement, rod stress, and future interface pose to the state variables. The displacement is recorded in mm, the stress in MPa, and the rotation angle in rad. All responses are first divided by the corresponding allowable error to form a dimensionless response, and then divided by the upper limit of the allowable change of the state variable to form a normalized sensitivity that can be compared across physical quantities.
[0087] Sensitivity cutoff threshold according to Sure, This indicates the sensitivity cutoff threshold used in this transmission, which is a dimensionless, non-negative quantity obtained from the first term of three ratios arranged in ascending order. This indicates the allowable error in stress calculation, expressed in MPa, and originating from two percent of the stress limit for the member. This represents the upper bound of the stress state quantity along the corresponding propagation path, in MPa, and is taken as the stress limit value of the member along that path. This represents a dimensionless upper bound for the allowable interface pose error after normalization to position and orientation tolerances. This represents the dimensionless upper bound of the corresponding pose state quantity, and it is set to 1. This represents the dimensionless upper bound of the allowable ridge closure error after normalization to the closure tolerance. This represents the dimensionless upper bound of the corresponding closed state quantity and is set to 1. All three ratios are dimensionless quantities.
[0088] For each candidate stopping port, the solver calculates the L2 norm of all normalized response components of that port as the sensitivity norm, and reads the port error from the condensation check record of the previous model state version. Propagation stops only when the sensitivity norm is not greater than the sensitivity cutoff threshold and the port error of the previous version is not greater than the port error cutoff threshold. In this embodiment, the port error cutoff threshold is set to 0.08 and is a dimensionless quantity. If there is no condensation check record in the previous version, it is considered that the stopping condition is not met and the expansion continues for at least one more layer.
[0089] The component-node subgraph enclosed by all the stop propagation ports is registered as a local influence domain, and the degrees of freedom within the local influence domain are retained in the subsequent local solution. The solver takes the substructure outside the domain, other than the stop propagation ports, and rearranges the degrees of freedom of the substructure outside the domain into the degrees of freedom of the stop propagation ports and the degrees of freedom of the non-ports outside the domain, and performs static condensation calculations. and , This represents the port stiffness matrix acting on the stop propagation port after the condensation of the extra-domain substructure. The unit is N / mm, and the data is derived from the assembly results of the extra-domain components and nodal tangent stiffnesses in the current model state version. This represents the stiffness submatrix corresponding to the port degrees of freedom in the external substructure, in N / mm. This represents the coupling stiffness submatrix of the port with respect to the non-port degrees of freedom outside the domain, in N / mm. This represents the stiffness submatrix of non-port degrees of freedom outside the domain, in N / mm, and the rigid body modes are eliminated through the external support boundaries before inversion. The coupling stiffness submatrix represents the coupling stiffness of the non-port degrees of freedom outside the domain to the port degrees of freedom, in N / mm. This represents the equivalent port load vector after the condensation of the extra-domain substructure, in units of N. This represents the original load vector at the port of the external substructure, in units of N. Represents the original load vector for non-port degrees of freedom outside the domain, in units of N, with subscripts... Indicates the set of degrees of freedom of the port where propagation has stopped, subscript Represents the set of non-port degrees of freedom outside the domain;
[0090] All model components, nodes, loads, and supports are assigned to local influence domains or external substructures based on unique object identifiers. The stop propagation port serves only as an assembly interface for both: the stiffness of components within the domain and the loads within the domain are only assembled into the local equations, while the stiffness of components outside the domain, the loads outside the domain, and the supports outside the domain only participate in the aforementioned condensation. The port degrees of freedom are assembled once through the same interface mapping matrix. It is forbidden to include the same component, load, or support in both the internal and external domains simultaneously. The solver uses the port stiffness and equivalent load obtained from the external condensation as boundary conditions to solve for the local mechanical response that retains all internal degrees of freedom. Then, the adjacent components on the outer layer of the stop port are temporarily incorporated into the check domain and solved independently according to the same allocation rules. The back-substitution displacement of the same port in the local solution is subtracted from the check solution displacement for each degree of freedom, and the result is divided by the corresponding allowable position deviation or allowable attitude deviation to calculate the L2 norm. As port error Indicates port The dimensionless nonnegative error, This represents a diagonal matrix consisting of the six-degree-of-freedom allowable deviations of the port. The first three terms are in mm, and the last three terms are in rad. The data is derived from the interface tolerance record. This represents the port displacement vector of the local solution, with the translation component in mm and the rotation component in rad. This represents the displacement vector at the same port in the verification solution, and its unit is . Consistent, Indicates the port number where propagation has stopped;
[0091] When the error at any port is greater than 0.08, the solver incorporates the adjacent components and their connecting nodes on the outer layer of the port into the local influence domain, re-divides the internal and port degrees of freedom, and repeats the condensation, solution, and verification. If the error decreases by less than one percent in two consecutive rounds, it incorporates two layers of adjacent components at once. The domain expansion process continues until the error at all ports is no greater than the threshold or the local influence domain covers the complete assembly topology. When the complete topology is covered, the condensation is canceled and the solution of the full model is used as the final local mechanical response.
[0092] The final local mechanical response record includes the displacement of components within the domain, axial force and bending moment of members, combined stress, stress utilization rate, predicted translation and predicted rotation of the interface, and saves the component identifier, port identifier, load scenario, model state version, local influence domain version and error upper bound for each result. The stress utilization rate is obtained by dividing the combined stress by the allowable stress of the corresponding member and is restricted to a dimensionless non-negative value. The interface pose is obtained by superimposing the local solution increment with the current measured pose.
[0093] The solver writes the final local influence domain, port stiffness, equivalent load, port error, stress envelope of adjacent members, and predicted pose of the future interface into the local response package. If the stiffness matrix is not invertible due to insufficient constraints, the relevant degrees of freedom are fixed at the current measured value and marked as unsolvable. Unusable local response packages are not included in the candidate action calculation. Normal local response packages, along with the model state version, are passed to S4 for mixed candidate action generation and subsequent intersection of feasible poses of the interface.
[0094] In this specific embodiment, S4 includes:
[0095] The candidate action generator reads the local response package of S3, discretizes three field coordinate translations in 2mm increments and three rotations in 0.1-degree increments within the allowable attitude adjustment range of the affected component, generates jacking or unloading amounts in 1mm increments within the temporary support stroke and load limit, generates two sequences for the double bolt nodes to be tightened: first bolt operation first and second bolt operation first, and selects all subsequent components that have been completed from the design pre-relationship diagram, and combines the above discrete amounts into mixed candidate actions with action identifiers according to the same model state version;
[0096] The generator first performs structural pruning, deleting combinations that exceed the component's orientation range, support stroke, support bearing capacity, or violate pre-installation relationships. For double-bolt operation sequences, it also checks the current node status. Combinations that have not reached the two-bolt insertion state must not include initial tightening or final tightening operations. Each retained action records the component translation vector, rotation vector, support adjustment vector, bolt operation sequence, subsequent installation sequence, expected execution time, and the version of the referenced local response package.
[0097] The wind load scenario generation module converts the current 2-second average wind speed and direction, the meteorological station's 10-minute forecast sequence, and their respective measurement covariance into local influence domain nodal loads. Within the expected execution period, it takes the upper and lower boundaries corresponding to the forecast mean and coverage factor to form no less than three wind load scenarios. In this implementation, the preset confidence probability is 0.95 and the coverage factor is 1.96. When the forecast sequence is unavailable, the coordinate components of the current wind speed and the safety increment corresponding to the upper boundary of the measurement uncertainty are superimposed in the unfavorable direction to form an alternative scenario. All scenarios record the m / s unit, direction angle, effective time period, and source version.
[0098] For each hybrid candidate action sequence, the generator derives a candidate state chain from the read-only parent model state version and applies component pose adjustment, temporary support adjustment, double bolt state switching, and subsequent component installation sub-actions sequentially. After each sub-action is completed, the allowed degrees of freedom of the connection edge, the tangential stiffness of the node, the support boundary, the installed component, the mass, the load, and the assembly topology version are updated. The affected domain is redefined and out-of-domain cohesion and local solution are performed. The component pose, member stress, interface predicted pose, topology version, and constraint state output in this step are used as inputs for the next sub-action. The re-linearized tangent model is allowed to be reused only when the displacement and rotation increments are both within the small increment range defined by S2. If any step solution is unavailable or the hard constraints are not satisfied, the candidate sequence is terminated and the failed sub-action and parent state version are recorded.
[0099] For each sub-action of the candidate state chain and each wind load scenario, the generator uses the updated port stiffness, equivalent load, and nodal displacement of that sub-action to solve for the stress of adjacent members and the predicted pose of future interfaces, and then uses... Forming an uncertain interval, Representing physical quantities In action Japanese wind scene The confidence interval below, unit and same, This represents the model's predicted mean, expressed in mm, rad, or MPa, and is derived from the local response calculation of the current candidate state. This represents the coverage factor corresponding to the preset confidence probability, which is a dimensionless positive number and is taken as 1.96 in this implementation. The standard deviation is obtained by linear propagation of the covariance of geometry, pose, and strain measurements, and its units are... same, This indicates the interface pose component or the stress of adjacent members. Indicates the sequence number of the mixed candidate actions. The numbers represent the wind load scene sequence. The units of the translational covariance, rotational covariance, and strain covariance used in linear propagation are mm², rad², and με², respectively. The variances after propagation are mm², rad², or MPa², depending on the output type. The standard deviation in the formula is obtained by taking the positive square root.
[0100] The generator first performs the most unfavorable boundary check on the hole position, continuous sweep collision avoidance, member stress, mid-span vertical coordinate, and ridge closure constraint for each wind load scenario, using the predicted mean and confidence uncertainty interval for that scenario. It then robustly shrinks the set of allowable poses using the corresponding uncertainty interval to form a set of feasible poses for the action in that scenario. Next, it intersects the set of feasible poses for the same action that has been robustly shrunk in all wind load scenarios for each degree of freedom. At the same time, it takes the union envelope of the upper and lower boundaries of the stress uncertainty intervals for each scenario according to the member number. If any scenario has a hard constraint out of bounds, the intersection of the feasible sets of scenarios for any degree of freedom is empty, or the upper boundary of any stress envelope exceeds the allowable stress of the member, the action is immediately deleted. It is prohibited to replace the worst unfavorable check for each scenario with the intersection between the original predicted confidence intervals. The wind load scenario, degree of freedom, constraint endpoint, or member identifier that caused the deletion is saved in the deletion record.
[0101] For actions that are not deleted, the generator uses the predicted pose of the future interface as the center of the six-degree-of-freedom set, converts the radial clearance of the hole-bolt into an insertable translational and rotational interval, converts the directional distance of the component and tool sweep body and the restricted body on site (not less than 20mm) into a spatial collision avoidance pose interval, and then intersects it with the allowable pose adjustment range of the component and the elastic displacement interval calculated from the stress limit of the rod, degree by degree of freedom. The interval boundary is recorded in mm and rad in the field coordinate system. The set simultaneously saves the constraint source, confidence boundary and corresponding action identifier.
[0102] The following are considered as hard constraints that do not allow relaxation: the ability to insert holes, the non-intersection of components and tool sweep bodies, the stress limit of rods, the allowable range of vertical coordinates at mid-span, and the lower limit of the closure margin of the ridge. The amplitude preference of translation, rotation, and support adjustment close to zero is considered as soft constraints that allow relaxation. Hard constraints are judged according to the most unfavorable boundary of feasible pose intersection and stress envelope. If any hard constraint is not satisfied, the candidate action is deleted. The relaxation amount of soft constraints is normalized according to the width of the corresponding allowable range and written into the action record for S5 to include in the comprehensive cost.
[0103] The rolling intersection module, based on the candidate state chain and the preceding relationships of subsequent components, starts from the current installation interface and maps the six-DOF feasible set of this sub-action to the updated interface coordinate system of the next sub-action through rigid body adjoint transformation. It then continues to find intersections with the feasible set of the next sub-action, which is re-formed according to its new topology, new support boundary, new node stiffness, hole-bolt gap, continuous swept body, attitude adjustment range, and stress limit. The component poses used for transformation are derived from the current candidate sub-state. Each time an intersection is completed, the sub-state version, step number, interface identifier, transformation matrix, and intersection boundary are saved until the ridge interface. If the intersection is empty or the sub-state update fails at any step, the action is marked as having no feasible pose in the future.
[0104] The sweep body constraint takes the continuous conservative sweep envelope formed between the component entity and the installation tool envelope between the start and end poses of the action as the object. The spatial index uses the field coordinate grid number as the key to retrieve adjacent forbidden bodies. For each attitude segment, the translation of adjacent poses is calculated as the sum of the maximum radius of the component or tool to the rotation axis multiplied by the rotation angle, which is used as the upper limit of the maximum displacement at any point in the segment. This is compared with the margin after deducting a 20mm safety margin from the current minimum directed clearance of the segment. Only when the upper limit of the maximum displacement is not greater than this margin is the conclusion that the segment is collision-free accepted. Otherwise, the segment is recursively bisected and the conservative envelope is recalculated until it is proven that the continuous clearance is not less than 20mm or the subdivision reaches the upper limit of the maximum point displacement of 1mm. If any subdivision segment envelope intersects with a forbidden body or the minimum continuous clearance is less than 20mm, the corresponding pose unit is deleted. The deletion record saves the collision object, attitude segment, upper limit of maximum displacement, minimum continuous clearance and subdivision termination condition. When all pose units are deleted, the mixed candidate action is directly marked as an empty set.
[0105] When determining the elastic displacement range based on the stress limit of the member, the generator applies positive and negative unit perturbations to the six interface degrees of freedom and reads the stress increment slope in the S3 local response packet. The MPa margin obtained by subtracting the upper limit of the current stress envelope from the allowable stress of the member is divided by the MPa stress increment slope corresponding to each millimeter or each radian of the same member to obtain the allowable increment in mm or rad for that degree of freedom. This calculation does not involve the conversion of torque, lever arm, force or mass. When the absolute value of the slope is less than the calibration lower limit, the member does not control that degree of freedom. When any stress margin is not greater than zero, the corresponding action is immediately eliminated. The intersection of all positive and negative allowable increments is taken as the stress constraint pose range.
[0106] Each hard constraint is first normalized to its own unit positive tolerance. The upper limit constraint is calculated by subtracting the upper limit from the comparison value and then dividing by the positive tolerance; the lower limit constraint is calculated by subtracting the comparison value from the lower limit and then dividing by the positive tolerance; for bilateral interval constraints, the upper limit violation and lower limit violation are calculated separately and the maximum of the two is taken. The comparison value comes from the most unfavorable confidence boundary of the current candidate substate under the corresponding wind load scenario. The upper limit, lower limit, and tolerance of the same unit come from the current model state version, interface tolerance records, member allowable stress records, or ridge closure calibration records, only when... Action All constraint numbers This action is only retained when the condition is met. Indicates action No. The dimensionless, signed violation value of a hard constraint is not greater than 0 when the constraint is satisfied, and its value increases with the degree of violation. Indicates the sequence number of the mixed candidate actions. This indicates the hole position, collision avoidance, stress, mid-span vertical coordinate, or closed constraint number; if any tolerance is missing or not greater than zero, the division is not performed, the action is marked as unevaluable and deleted, and the comparison value, boundary, tolerance source, and constraint number are written into the review record; if any violation is positive, the action and the corresponding constraint number are written into the deletion record.
[0107] At the ridge interface, the module normalizes the three translational degrees of freedom with allowable positional deviations and the three rotational degrees of freedom with allowable attitude deviations, and then calculates... , Indicates action The ridge closure margin is a dimensionless signed quantity. A positive value indicates that the predicted closure state is inside the feasible set, and the larger the value, the more sufficient the margin. A negative value indicates that it is outside the set. This represents a sign factor, with the innermost value being 1 and the outermost value being negative 1. This represents a six-DOF pose vector on the boundary of the feasible pose set of the roof ridge, with the first three terms in mm and the last three terms in rad. Indicates action The boundary of the feasible pose set corresponding to the roof ridge. This represents a diagonal matrix consisting of three allowable positional deviations and three allowable attitude deviations, with each diagonal term having the same unit as its corresponding degree of freedom. Indicates action The predicted closed pose vector and the unit is Consistent, Indicates the sequence number of the mixed candidate actions;
[0108] The lower limit of the ridge closure margin is jointly calibrated by the allowable position deviation of the ridge interface, the allowable attitude deviation, and the boundary of the pose uncertainty interval corresponding to a confidence probability of 0.95. In this implementation, dimensionless zero is taken as the lowest feasible boundary and 0.15 is set as the execution buffer value. Actions below zero are deleted as hard constraints, while actions between zero and 0.15 are retained but marked with a low closure margin. Finally, the candidate action record, the intersection of feasible poses of all scenarios, stress envelope, closure margin, soft constraint relaxation amount, and deletion reason are combined into a candidate evaluation package and passed to S5.
[0109] In this specific embodiment, S5 includes:
[0110] The action decision-maker reads the candidate evaluation package of S4 and first eliminates mixed candidate actions with future infeasible pose markers, hole positions that cannot be inserted, swept body intersections, stress envelope exceeding limits, mid-span vertical coordinate exceeding limits, or ridge closure margins below the lower limit. The elimination process only uses the most unfavorable values formed by S4 under all wind load scenarios and confidence boundaries. Each deletion record saves the action identifier, hard constraint type, comparison value, limit value, model state version, and corresponding scenario to avoid failed actions being repeatedly selected in subsequent retesting cycles.
[0111] For the remaining actions, the decision-maker performs minimum-maximum normalization on the negative of the ridge closure margin within the allowable closure margin range, divides the stress increment of adjacent members by the corresponding allowable stress range width, divides the L2 norm of the three-way translation vector by the allowable translation range width, divides the L2 norm of the three-way rotation vector by the allowable rotation range width, and divides the L2 norm of the temporary support adjustment vector by the allowable support adjustment range width. The resulting five terms are all truncated between 0 and 1, and the larger the value, the higher the cost. Each denominator comes from the interface tolerance, member strength, and support capacity records referenced in the current model state version.
[0112] The installation sequence change quantity is calculated by comparing the relative order of any two components in the candidate subsequent installation sequence with that in the designed installation sequence, determining the number of inversion pairs. If the number of subsequent components is not less than two, then... Normalization Indicates action The cost of changing the installation order is a dimensionless quantity between 0 and 1. This represents the number of inversions of the candidate installation order relative to the designed installation order. It is a non-negative integer derived from the results of pairwise comparisons. Indicates action The number of subsequent components involved is a non-negative integer derived from candidate action records. Indicates the sequence number of mixed candidate actions, when the number of subsequent components is less than 2. Take zero directly and do not perform the division operation;
[0113] Comprehensive cost calculate, Indicates action The overall cost is a dimensionless quantity between 0 and 1, with smaller values being better. Indicates the first The dimensionless weights of the evaluation items take values no less than 0 and the sum of the six weights is 1. They are derived from the allowable closure error allocation records of the current assembly stage. Indicates action No. The normalized cost is a dimensionless quantity between 0 and 1. The six terms correspond to the inverse of the closure margin, the stress increment of adjacent members, the amplitude of three-dimensional translation, the amplitude of three-dimensional rotation, the amplitude of temporary support adjustment, and the change in installation sequence, respectively. This indicates the evaluation item number, ranging from 1 to 6. Indicates the sequence number of the mixed candidate actions;
[0114] The closure error allocation record can be searched using the assembly stage, the number of remaining components, and the ridge closure buffer zone as the lookup key. The value fields are the six weights, the applicable model version, and the effective time. The record is calibrated by the finite element disturbance test before construction and the results of the first two rounds of field correction. In this implementation, the six weights of closure margin, stress, translation, rotation, support, and sequence are set to 0.30, 0.25, 0.10, 0.10, 0.10, and 0.15 respectively in the two component stages before ridge closure. When the model state update causes changes in the number of remaining components or the closure buffer zone, the table is looked up again. If there is no match, the most conservative record in the same stage is used and the reduction of the weights of closure margin and stress is prohibited.
[0115] The decision-maker selects the action sequence in ascending order of comprehensive cost. If the difference in comprehensive cost between two actions is no more than one ten-thousandth, they are considered the same and the action with the largest ridge closure margin is selected first. If they are still the same, the action with the smallest stress increment of adjacent members is selected. If they are still the same, the action with the fewest number of members to be adjusted is selected. If they are still the same, the action is selected in dictionary order of action identifier to ensure that the result is repeatable. The selected actions and sorting criteria are written into the decision record.
[0116] If no other actions are performed after the hard constraint is removed, the instruction manager stops releasing the final tightening instruction of the current double-bolt node and subsequent component installation instructions, keeps the component, temporary support and double-bolt node in the current adjustable state, generates a retest task containing geometric measurement points, strain channels, support displacement channels, bolt tightening channels, sampling duration and target model state version, and after the retest data is sent back to S2, the time alignment state is updated again and a new candidate evaluation package is generated.
[0117] When other mixed candidate actions exist, the instruction manager decomposes the selected action sequence into pre-tightening component posture adjustment instructions, temporary support adjustment instructions, double bolt operation instructions, and subsequent component installation sequence instructions. The control state consists of the current component posture, support displacement, double bolt tightening stage, and completed installation sequence. The control action consists of posture adjustment displacement and rotation angle, support lifting or unloading amount, bolt operation sequence, and next component identifier. The control objective is to execute the action with the lowest overall cost while keeping all hard constraints feasible. The status is checked every 200ms, and a mandatory retest is performed after each instruction is completed. Each instruction records the instruction identifier, component or node identifier, field coordinate system, point of action, direction, amplitude, execution stage, retest conditions, corresponding model status version, and previous instruction identifier. Subsequent installation sequence instructions also record the connection prerequisite relationship and sequence position. If the previous retest fails, the status version changes, or the execution equipment quality code is invalid, the instruction is frozen and enters the abnormal feedback branch. The next instruction is only released after the previous instruction retest is qualified.
[0118] After each instruction is executed, the interface coordinates, member strain, temporary support displacement and bolt tightening data specified by the retest conditions are collected. The interface geometric deviation is the six-degree-of-freedom L2 norm of the difference between the measured interface pose and the instruction target pose after interface tolerance normalization. The stress increment of adjacent members is the retest stress estimate minus the stress estimate of the model state version before execution. The reduction of the ridge closure margin is the closure margin before execution minus the closure margin calculated according to the same S4 rule after retesting. The three results are written into the instruction retest record respectively.
[0119] The retesting module first verifies the data integrity according to the instruction identifier and model status version. If there are fewer than three non-collinear valid geometric measurement points, the strain channel coverage is less than 80%, the support displacement lacks the pre-execution benchmark, or the double bolt tightening sequence is interrupted, the module does not calculate the pass conclusion but sets the instruction status to pending retesting. The component, support, and tightening stages remain unchanged, and the same retesting conditions are reissued. Only when the data is complete and the time alignment error is no more than 20ms will the module enter the allowable quantity comparison.
[0120] The allowable increase in stress and the allowable decrease in closure margin are respectively calculated according to and calculate, Indicates the object to be retested The allowable stress increment, in MPa and non-negative. This indicates the stress limit of the member in the object, in MPa, and is derived from the model state version before execution. This indicates the estimated stress value of the member before the command was executed, in MPa, and from the same version. Indicates the object to be retested The allowable reduction in the closure margin is a dimensionless, non-negative quantity. This represents the ridge closure margin before execution, a dimensionless quantity sourced from the selected action record. This represents the lower limit of the closure margin, a dimensionless quantity derived from the current ridge interface tolerance and confidence boundary. Indicates the sequence number of the object being retested in this instruction;
[0121] The digital twin assembly and calibration method for this scaffolding and roof steel frame establishes a five-state record for each instruction: release, execution, retest, judgment, and feedback. Only when the signature time, executor, and equipment quality code of the previous state are complete can the next state be allowed. When an execution error occurs, all unreleased instructions are frozen and the amplitude of the executed action, the retest result, and the reason for the error are returned to S2. When the retest is qualified, the completion flag of the current instruction is written into the same action record. This deterministic state transition prevents the final tightening instruction and subsequent installation instructions from being released out of turn.
[0122] The geometric tolerance is taken as the corresponding component interface tolerance, with the unit being mm for translation and rad for rotation. The allowable stress increment is taken as the allowable stress of the member in the model state version before execution minus the stress estimate, with the unit being MPa. The allowable reduction of the closure margin is taken as the ridge closure margin before execution minus the lower limit of the closure margin and is a dimensionless quantity. When the interface geometric deviation exceeds the normalized allowable value 1, the stress increment of any member exceeds its allowable increment, or the reduction of the closure margin exceeds the allowable reduction, the retest result is marked as an execution anomaly, subsequent instructions are stopped, and all retest data and executed instructions are returned to S2.
[0123] When all three retest results are within limits, the instruction manager marks the current instruction as complete and releases the next instruction. After all instructions are completed, the final geometry, strain, support, and fastening data are returned to S2 to form a new model state version. Decision records, instruction records, retest records, and anomaly flags are associated and saved according to the same action identifier, thereby completing a closed-loop correction of one attitude adjustment, support adjustment, double bolt operation, and installation sequence.
[0124] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
[0125] This invention combines assembly topology, node phased constraints, and on-site multi-source observation, enabling changes in component geometry, temporary support boundaries, wind load, and double-bolt fastening state to enter the same model state. This state then drives local mechanical response and future interface pose prediction, thereby transforming the chain reaction between current assembly errors and subsequent installation and ridge closure into calculable constraints and margins.
[0126] Compared to methods that adjust based solely on the current hole position or a single measurement, this invention adds node state switching, sensitivity truncation and adaptive domain expansion, reverse intersection of future feasible pose sets, and a hybrid action rolling sequence structure. This allows operations that do not meet subsequent assemblability or closure conditions to be identified before final tightening, and continuous correction is achieved through a post-execution retest and model update step.
Claims
1. A digital twin assembly and calibration method for a canopy frame and roof steel skeleton, characterized in that, include: S1. Based on the design, measured geometry and node relationships, establish an assembly topology digital twin model that characterizes the geometry and mechanics of the components, the temporary support boundary and the state of the double bolt nodes; S2. Time-align the on-site geometry, mechanics, boundary, environment, and fastening data, estimate the status of components, supports, and double-bolt nodes based on observation and prediction residuals, and update the model. S3. The sensitivity of the structural response and interface pose is propagated along the assembly topology by responding to changes in the assembly state of the components, temporary support boundaries, double bolt node states, or wind loads. Based on the sensitivity, a local influence domain is formed, and the external structure is condensed into port stiffness and equivalent load. The local influence domain is then expanded according to the port error. S4. Form a mixed candidate action that includes component orientation adjustment, support adjustment, double bolt operation and installation sequence. Calculate the stress of adjacent members and the predicted pose of future interfaces for each action based on port stiffness and equivalent load. Based on the predicted pose of future interfaces and the geometric, spatial and mechanical constraints in the model, roll to construct a set of six-degree-of-freedom feasible poses for subsequent installation and intersect them in the opposite direction of the ridge to determine the ridge closure margin. S5. Eliminate mixed candidate actions that are not feasible in the future or violate assembly accuracy, spatial collision avoidance, stress and closure constraints. Determine the comprehensive cost based on the closure margin, stress and amplitude of the remaining actions. Select and output attitude adjustment, support adjustment, double bolt operation and installation sequence instructions. After retesting, return to step S2.
2. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, S1 includes: Establish a local coordinate system for each component and record the cross-sectional parameters, material parameters, designed interface pose, and measured interface pose; The components and double-bolt nodes are respectively used as component nodes and connecting edges in the assembly topology digital twin model; a displacement boundary that can be updated by measured settlement or jacking along the direction of the support is established for the temporary support; for each double-bolt node, the measured hole center, hole axis direction, hole diameter, bolt diameter, hole-bolt gap, preload parameters and tightening stage of the two connecting holes are recorded.
3. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S2, estimating the state of the double-bolt node, includes: identifying the state of the double-bolt node as an uninserted state, a first bolt positioning state, a two-bolt inserted state, an initial tightening state, or a final tightening state based on the insertion signals of the two bolts, the torque-rotation sequence, the strain changes of the adjacent members of the node, and the relative poses of the members on both sides of the node; switching the hole-bolt contact degree of freedom and the sliding boundary according to the identified double-bolt node state, and updating the equivalent preload and the tangential stiffness of the node. The first bolt positioning state corresponds to retaining the relative rotational degree of freedom about the axis of the first bolt, the two bolts inserted state corresponds to retaining the relative sliding degree of freedom within the hole-bolt gap range, and the initial tightening state and the final tightening state correspond to the equivalent preload range determined by their respective torque-rotation intervals.
4. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S2, updating the assembly topology digital twin model, includes: registering the on-site geometric measurement data with the measurement reference points in the corresponding component's local coordinate system to form the geometric residual; forming the strain residual based on the strain data of the member and the strain calculated by the assembly topology digital twin model; forming the support residual based on the temporary support displacement data and the current displacement boundary; forming the fastening event residual based on the state transition time corresponding to the bolt fastening data and the state transition time predicted by the model; jointly estimating each residual using the inverse matrix of the measurement covariance of each residual as weights, and writing the joint estimation result into the model state version with the corresponding timestamp.
5. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S3, forming the local influence domain, includes: taking the location where the change occurs as the propagation starting point, calculating the sensitivity of the port state to the changed state quantity level by level along the component edge connected to the propagation starting point; stopping propagation at a port when the sensitivity norm at a port is not greater than the sensitivity cutoff threshold and the condensation response error of the port in the previous model state version is not greater than the port error cutoff threshold; determining the component-node subgraph enclosed by the port where propagation stops as the local influence domain, and eliminating the out-of-domain degrees of freedom into the port stiffness and equivalent load acting on the port where propagation stops through static condensation, wherein the allowable stress calculation error, allowable interface pose error, and allowable ridge closure error are divided by the upper bound of the state quantity of the corresponding sensitivity propagation path, and the three ratios obtained are arranged in ascending order of value, and the first ratio after arrangement is used as the sensitivity cutoff threshold.
6. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 5, characterized in that, Step S3, expanding the local influence domain, includes: under the same state change and the same model state version, solving the condensed local mechanical response using the port stiffness and equivalent load; taking the difference between the displacement back-substitution value of the port that stopped propagating in the local mechanical response and the corresponding port displacement value obtained after incorporating the adjacent layer of components outside the port as the port error; when any port error is greater than the port error cutoff threshold, incorporating at least one layer of adjacent components outside the port into the local influence domain and re-condensing and solving until all port errors are not greater than the port error cutoff threshold or the local influence domain is expanded to the complete assembly topology, and obtaining the displacement, internal force, stress utilization rate and future interface predicted pose of adjacent members from the final local mechanical response.
7. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S4, forming the hybrid candidate action, includes: discretizing and generating three-dimensional translation and three-dimensional rotation within the allowable attitude adjustment range of the component; generating lifting or unloading within the stroke and load limits of the temporary support; generating two sequences for the two bolts to be tightened: first bolt first operation and second bolt first operation; and generating an installation sequence for subsequent components that meet the current connection prerequisite relationship; forming a wind load scenario set based on the current wind speed and direction and the predicted wind speed and direction during the execution period of the candidate action; forming an attitude uncertainty interval and a stress uncertainty interval based on the measurement covariance of geometric measurement data, component attitude data, and member strain data, according to the coverage factor corresponding to the preset confidence probability; and applying this to each hybrid candidate action. The motion calculation involves the intersection of feasible pose sets and the stress envelope of adjacent members within the wind load scenario set. When the predicted wind speed and direction are unavailable, the current wind speed component is superimposed with the wind speed safety increment determined by the upper bound of the current wind speed measurement uncertainty to form an alternative wind load scenario. The constraints that do not allow the introduction of relaxation, such as the hole insertion constraint, the component and tool sweep body non-intersection constraint, the stress limit constraint, the allowable interval constraint of the mid-span vertical coordinate, and the lower limit constraint of the ridge closure margin, are determined as hard constraints. The motion amplitude preference that allows the introduction of relaxation and whose relaxation is included in the comprehensive cost is determined as a soft constraint. The hard constraint screening is performed according to the boundary values of the intersection of feasible pose sets and the stress envelope of adjacent members.
8. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 7, characterized in that, The construction of the six-DOF feasible pose set in step S4 includes: taking the predicted pose of the future interface as the center of the set, performing an intersection operation on the insertable translational and rotational intervals corresponding to the hole-bolt gap, the pose intervals when the component and tool sweep bodies do not intersect, the allowable attitude adjustment range, and the elastic displacement interval determined by the stress limit of the rod; transforming the intersection obtained in the current installation step to the interface coordinate system of the next installation step according to the connection relationship of the subsequent components, and continuing to find the intersection with the intersection corresponding to the next installation step until the ridge interface; normalizing the three translational degrees of freedom with the allowable position deviation of the ridge interface, and normalizing the three rotational degrees of freedom with the allowable attitude deviation of the ridge interface, and within the normalized six-DOF pose space, assigning a positive or negative sign to the shortest distance from the set boundary to the predicted closed state located inside or outside the set obtained by the intersection of the ridge interface, respectively, to obtain the ridge closure margin. The lower limit of the closure margin is jointly determined by the allowable position deviation of the ridge interface, the allowable attitude deviation, and the confidence interval boundary determined by the measurement uncertainty according to the preset confidence probability and coverage factor.
9. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S5 determines the comprehensive cost by: normalizing and then weighting the negative of the ridge closure margin, the stress increment of adjacent members, the three-dimensional translation, the three-dimensional rotation, the temporary support adjustment, and the installation sequence change, respectively. The installation sequence change is the number of inverse pairs of the candidate installation sequence relative to the designed installation sequence. When the number of subsequent components is not less than 2, the installation sequence change is normalized to half the product of the number of subsequent components and the number minus one. When the number of subsequent components is less than 2, the installation sequence change is zero and this normalization is not performed. The remaining normalization scales are taken as the allowable range width of the corresponding quantity, and each weight is determined according to the distribution ratio of the allowable closure error in the current assembly stage among the evaluation quantities. When the comprehensive costs of two action sequences are the same, the one with the largest ridge closure margin is selected first; if still the same, the one with the smallest stress increment of adjacent members is selected; if still the same, the one requiring the fewest adjustments is selected.
10. The digital twin assembly and correction method for a canopy frame and roof steel skeleton according to claim 1, characterized in that, Step S5, which outputs and executes the attitude adjustment, support adjustment, double bolt operation, and installation sequence instructions, includes: when no other mixed candidate actions exist after eliminating mixed candidate actions, stopping the release of the final tightening instruction and subsequent installation instructions for the current double bolt node, maintaining the current adjustable state, collecting retest data, and returning to step S2 to update the model and regenerate mixed candidate actions; when other mixed candidate actions exist, decomposing the selected action sequence into component attitude adjustment instructions before final tightening, temporary support adjustment instructions, double bolt operation instructions, and subsequent component installation sequence instructions, and recording the component identifier, field coordinate system, point of action, direction, amplitude, execution stage, retest conditions, and corresponding model state version in each instruction; and recording the connection prerequisite relationship and sequence position in the subsequent component installation sequence instructions; in each instruction After execution, the system collects the geometric measurement data, member strain data, temporary support displacement data, and bolt tightening data specified under the retest conditions. It then determines the interface geometric deviation, stress increment of adjacent members, and reduction in ridge closure margin, respectively. The corresponding component interface tolerance is used as the geometric allowable deviation. The difference between the member stress limit and the member stress estimate in the model state version before command execution is used as the stress allowable increment. The difference between the ridge closure margin and the lower limit of the closure margin before command execution is used as the closure margin reduction allowable decrease. If the interface geometric deviation exceeds the geometric allowable deviation, the stress increment of adjacent members exceeds the stress allowable increment, or the ridge closure margin reduction exceeds the closure margin reduction allowable decrease, the retest result is marked as an execution anomaly. Subsequent commands are stopped and the system returns to S2; otherwise, the next command is released.
