A visual calibration system for roller coaxiality deviation

Abstract

This invention discloses a visual calibration system for roller coaxiality deviation, relating to the field of visual inspection and dynamic balancing calibration technology in roller manufacturing processes. The system includes a rotating support mechanism, a dynamic balancing detection unit, a visual acquisition unit, a phase synchronization unit, a microelectromechanical inertial measurement unit, and an edge computing module. The dynamic balancing detection unit acquires the roller's imbalance amount, imbalance phase, and correction surface distribution. The phase synchronization unit maps the imbalance phase to the actual circumferential position of the roller. The visual acquisition unit acquires images of the corresponding positions. The edge computing module identifies visible geometric offset features and, combined with process phase inheritance, achievable correction boundary quantities, and acquired reliable boundary quantities, generates a low-level state machine evolution boundary. It outputs calibration parameters for shaft head recalibration, outer diameter compensation machining, local grinding of the rubber coating, weight adjustment, weight removal, clamping verification, or anomaly re-checking.

Description

Technical Field

[0001] This invention relates to the field of visual inspection and dynamic balancing calibration technology in the manufacturing process of rollers, and more specifically, to a visual calibration system for roller coaxiality deviation. Background Technology

[0002] Rollers are widely used in industrial equipment such as conveying, calendering, printing, packaging, mining, and logistics. Their manufacturing quality directly affects the equipment's operational stability, conveying accuracy, surface contact uniformity, and long-term reliability. During roller manufacturing, processes such as shaft head pressing or welding, outer diameter machining, rubber coating or surface treatment, dynamic balancing testing, and dynamic balancing verification can all potentially create visible geometric offsets, including coaxiality deviations, shaft head eccentricity, outer diameter runout, rubber coating thickness areas, weld area offsets, or clamping reference drift. These geometric offsets typically manifest as abnormal imbalances with circumferential phase direction during roller rotation and dynamic balancing verification, thus affecting rework efficiency and calibration quality.

[0003] In the prior art, CN109029269A discloses a method for detecting the geometric parameters of roller products. This solution mainly focuses on detecting geometric parameters such as the coaxiality and tapered cylindricity of roller installation. The roller is supported by a detection platform, and a distance measuring device moves along the roller's axial direction to scan the cylindrical surface. Simultaneously, a position measuring device records the axial position. Key geometric parameters of the roller are then calculated based on the distance and axial position data. The core process of this solution includes: rotatably mounting the roller along its product axis on the detection platform; configuring the distance and position measuring devices; having the distance measuring device move along the roller's axial direction to scan the cylindrical surface; having the position measuring device simultaneously record the axial position; and then calculating parameters such as installation coaxiality and tapered cylindricity based on the multi-position distance data.

[0004] In the prior art, CN110426151B also discloses a workpiece dynamic balancing correction control system and a dynamic balancing correction method. This scheme calculates the position and adjustment amount of the workpiece's imbalance based on the correction datum and processing parameters, and controls the processing module to reprocess the workpiece when it meets the reprocessing range. This type of scheme can convert the dynamic balancing detection results into a basis for reprocessing control, but it still mainly focuses on correcting the position and adjustment amount of the imbalance. It does not further map the imbalance phase to the actual circumferential position of the roller, nor does it identify the specific geometric sources at this phase position, such as shaft head eccentricity, outer circle runout, thick rubber coating area, weld area offset, or clamping datum drift.

[0005] However, in the dynamic balancing verification and rework calibration scenarios during the roller manufacturing process, static geometric detection schemes struggle to directly determine the processing source corresponding to the unbalanced phase, while dynamic balancing correction control schemes struggle to distinguish the specific geometric causes of the unbalanced anomalies. In actual production, support jitter, handling vibration, and tooling repetitive positioning errors can also lead to decreased image edge sharpness and phase synchronization deviation. The aforementioned existing schemes do not integrate dynamic balancing phase, actual circumferential position of the roller, visual offset characteristics, process phase inheritance, acquisition reliability assessment, and underlying state machine boundary modulation, which still easily leads to problems such as blind grinding, repeated correction, attribution based on experience, and low verification efficiency.

[0006] Therefore, it is still necessary to provide a visual calibration system for roller coaxiality deviation, which can correspond the unbalance phase obtained by dynamic balancing detection with the circumferential position of the actual roller during the roller manufacturing process. Then, by using visual acquisition, phase synchronization, microelectromechanical inertial measurement, process phase inheritance, boundary parameter calibration and low-level state machine control, the system can perform phase attribution and reliability judgment on the roller coaxiality deviation and its related visual geometric offset, and output process-specific calibration parameters. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a visual calibration system for roller coaxiality deviation, which is used to perform phase attribution on roller coaxiality deviation and related visible geometric offset based on visual acquisition, phase synchronization and process phase inheritance, and generate corresponding processing calibration parameters to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A visual calibration system for roller coaxiality deviation includes a rotating support mechanism, a dynamic balancing detection unit, a visual acquisition unit, a phase synchronization unit, and an edge computing module. The dynamic balancing detection unit acquires the roller's imbalance amount, imbalance phase, and correction surface distribution. The phase synchronization unit maps the imbalance phase to the actual circumferential position of the roller. The visual acquisition unit acquires visual images at the actual circumferential position. The edge computing module includes a visual bias recognition module, a process phase inheritance module, a comprehensive boundary modulation module, a low-level state machine control module, and a calibration parameter output module. The visual bias recognition module is used to identify visible geometric bias features from the visual image; the process phase inheritance module is used to record the occurrence process nodes and phase continuity relationships of the visible geometric bias features on the same circumferential phase in multiple manufacturing processes; the integrated boundary modulation module is used to generate the underlying state machine evolution boundary based on the phase synchronization unbalance load, process phase bias inheritance, correction reachable boundary, and acquisition reliable boundary; the underlying state machine control module is used to perform state transition based on the underlying state machine evolution boundary; and the calibration parameter output module is used to output roller calibration parameters based on the state transition results.

[0010] In a preferred embodiment, the phase synchronization unit includes a zero-position marker, a phase trigger, and an encoder; the phase trigger is used to identify the moment when the zero-position marker passes a predetermined detection position; the encoder is used to record the angle increment from when the zero-position marker is identified to when the visual image is acquired; the phase synchronization unit converts the unbalanced phase to the uniform circumferential phase coordinates of the roller based on the zero-position marker, the phase trigger, and the encoder.

[0011] In a preferred embodiment, the phase synchronization unit is further configured to invoke a backup phase reference when the zero-position mark cannot be identified or the position deviation of the zero-position mark exceeds the zero-position identification boundary; the backup phase reference includes a shaft head keyway, end face positioning hole, weld start position, temporary calibration ring mark, fixture follow-up mark, or a reference point entered by the human-machine interaction module; when there is a fixed angle offset between the backup phase reference and the historical zero-position mark, the phase synchronization unit completes phase registration based on the fixed angle offset; when the fixed angle offset cannot be established, the process phase inheritance module suspends the phase inheritance judgment based on multiple manufacturing processes.

[0012] In a preferred embodiment, the vision acquisition unit includes an end face camera, a side camera, and a local camera; the end face camera is used to acquire images of the roller end face, zero mark, and shaft head; the side camera is used to acquire images of the roller outer circle, rubber coating layer, and weld seam area; the local camera is used to acquire locally magnified images of the area corresponding to the unbalanced phase; the vision offset recognition module is used to identify shaft head eccentricity, end face center drift, outer circle runout, rubber coating thickness area, visible morphology offset of weld seam area, and clamping reference drift based on the vision images.

[0013] In a preferred embodiment, the process phase inheritance module is used to generate process records; the process records include a unique roller number, process node, zero-position phase, visual offset type, visual offset phase, offset amplitude, acquisition confidence level, imbalance amount, imbalance phase, correction surface distribution, calibration output result, and verification result; the process phase inheritance module is also used to record zero-position identification error, camera extrinsic parameter error, encoder error, clamping repeat positioning error, and historical registration error; when the cumulative result of the error exceeds the cross-process inheritance boundary, the process phase inheritance module pauses the phase inheritance judgment based on multiple manufacturing processes and retains the visual geometric offset judgment result of the current station.

[0014] In a preferred embodiment, the edge computing module further includes a boundary calibration module; the boundary calibration module is used to generate a boundary parameter table based on qualified roller samples, shaft head eccentricity samples, outer circle runout samples, rubber coating thick area samples, weld seam area offset samples, clamping drift samples, and vibration interference samples; the boundary parameter table includes synchronization boundary, zero position identification boundary, reference drift boundary, cross-process inheritance boundary, strong coupling phase boundary, weak coupling phase boundary, retest trigger boundary, visual attribution acceptance boundary, reshoot boundary, clamping verification boundary, shaft head recalibration boundary, outer circle compensation machining boundary, rubber coating local grinding boundary, weighting correction boundary, deweighting correction boundary, correction amplitude upper limit, correction reachable level boundary, correction destructive level boundary, manual execution boundary, early warning boundary, and abnormal review boundary; the boundary records in the boundary parameter table include applicable roller type, applicable process, input data source, boundary value, trigger condition, abnormal branch, output status, and output field.

[0015] In a preferred embodiment, the system further includes a microelectromechanical inertial measurement unit (MEMS); the MEMS is used to acquire vibration data within the visual image exposure window; the acquired reliable boundary quantity is generated by optical spatial features and mechanical temporal features; the optical spatial features include image sharpness level, roller edge ghosting level, end face contour continuity level, and exposure stability level; the mechanical temporal features include vibration stability level, acceleration discreteness level, and impact disturbance level; when the acquired reliable boundary quantity is lower than the visual attribution acceptance boundary, the integrated boundary modulation module modulates the underlying state machine evolution boundary according to the unmet acceptance conditions, driving the underlying state machine control module to enter the re-shoot confirmation state, phase re-measurement state, or clamping verification state.

[0016] In a preferred embodiment, the underlying state machine control module performs state transitions according to trusted gating, reference gating, phase gating, process gating, and correction gating. When the trusted gating fails, the re-detection confirmation state takes precedence over the outer diameter compensation machining state, the local grit finishing state, the weight correction state, and the weight removal correction state. When the reference gating fails, the clamping verification state takes precedence over the weight correction state and the weight removal correction state. When the phase gating fails, the phase retest state takes precedence over the outer diameter compensation machining state and the local grit finishing state. When the re-detection confirmation state... When the phase retest state or clamping verification state reaches the upper limit of the number of times or the upper limit of time corresponding to the boundary parameter table, and the corresponding gate control conditions are still not met, the bottom state machine control module enters the abnormal review state. When multiple states among the shaft head recalibration state, the outer circle compensation machining state, the rubber coating local grinding state, the weighting correction state, and the weight removal correction state simultaneously meet the entry conditions, the bottom state machine control module prioritizes selecting the state with a low level of destructiveness, a correction amount that does not exceed the upper limit of the correction amount amplitude, and a correction attainable level that meets the current process conditions.

[0017] In a preferred embodiment, the calibration parameter output module is used to output shaft head recalibration parameters, outer diameter compensation machining parameters, rubber coating local grinding parameters, weighting correction parameters, weight removal correction parameters, clamping verification parameters, or anomaly review prompts. The shaft head recalibration parameters include the shaft head eccentricity direction, recalibration position, and retest requirements. The outer diameter compensation machining parameters include the circumferential phase of the compensation machining, the axial range, the compensation direction, and the upper limit of the compensation amount. The rubber coating local grinding parameters include the grinding circumferential range, the grinding axial range, the grinding number limit, and the post-grinding verification requirements. The weighting correction parameters and the weight removal correction parameters both include the correction surface, the correction phase, the correction quality, and the verification requirements. The clamping verification parameters include the clamping reference re-shooting requirements, the support synchronization verification results, and the re-clamping prompts. The anomaly review prompts include the anomaly type, the triggering reason, the automatic correction prohibition mark, and the manual verification prompts.

[0018] In a preferred embodiment, the edge computing module further includes a review and recording module and an early warning linkage module. The review and recording module is used to generate single-item records, batch statistics, bias frequency statistics, residual imbalance trends, and statistical summaries. The early warning linkage module is used to generate observation early warnings, maintenance early warnings, or shutdown review early warnings based on the statistical summaries. When an observation early warning is generated, the early warning linkage module outputs an increased sampling inspection command to the manufacturing execution system. When a maintenance early warning is generated, the early warning linkage module outputs a maintenance inspection command to the manufacturing execution system. When a shutdown review early warning is generated, the early warning linkage module outputs a workstation isolation command and a pause automatic calibration command to the manufacturing execution system.

[0019] The technical effects and advantages of the visual calibration system for roller coaxiality deviation of the present invention are as follows: This invention establishes a phase correspondence between the unbalance quantity, unbalance phase, and correction surface distribution obtained from dynamic balancing detection and the visible geometric offsets such as shaft head eccentricity, outer circle runout, rubber coating thickness area, weld area offset, and clamping reference drift obtained from visual acquisition. This allows the geometric deviation information in the roller manufacturing process to be transformed into the attribution basis in the dynamic balancing rework process, thereby reducing manual experience judgment and blind grinding, and improving the pertinence of rework calibration.

[0020] This invention records the offset of the same roller in processes such as shaft head pressing, outer circle machining, rubber coating or surface treatment, and dynamic balance verification under a unified circumferential phase coordinate system by using phase synchronization, process phase inheritance, and acquisition reliability assessment. It also combines image clarity, edge blur, exposure stability, and vibration stability to determine the reliability of visual data, thereby reducing the impact of industrial field vibration, support shaking, and image blur on attribution judgment and improving the reliability of visual calibration results.

[0021] This invention integrates a boundary modulation module and a low-level state machine control module, enabling the system to reasonably transition between states such as normal pass, phase retest, re-scan confirmation, clamping verification, shaft head recalibration, outer diameter compensation machining, local grinding of rubber coating, weight correction, weight removal correction, and anomaly re-inspection, and outputs calibration parameters that are specific to each process. At the same time, through verification records and three-level early warning, it traces the trends of single-piece records, batch statistics, offset frequency, and residual imbalance, improves the first-pass dynamic balancing pass rate, reduces unnecessary grinding and counterweight consumption, and enhances the quality stability of mass production. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall mechanical structure of the system of the present invention.

[0023] Figure 2 This is a diagram of the system functional modules of the present invention.

[0024] Figure 3 This is a flowchart of the boundary calibration and boundary parameter table generation process of the present invention.

[0025] Figure 4 This is a flowchart of the multi-process phase inheritance and data flow of the present invention.

[0026] Figure 5 This is a schematic diagram illustrating the correspondence between the visual bias recognition object and the phase in this invention.

[0027] Figure 6 This is a diagram showing the relationship between the four comprehensive analysis points and the boundary modulation in this invention.

[0028] Figure 7 This is a schematic diagram illustrating the evolution of the underlying state machine of this invention.

[0029] Figure 8 This is a flowchart illustrating the closed-loop process of calibration parameter output and processing verification in this invention.

[0030] Figure 9 This is a schematic diagram of the review record, three-level early warning and production line linkage of the present invention.

[0031] 100. Rotary support mechanism; 110. Left support seat; 120. Right support seat; 130. Auxiliary support roller; 140. Clamping reference; 210. Left vibration acquisition unit; 220. Right vibration acquisition unit; 230. Phase triggering unit; 240. Encoder; 310. End face camera; 320. Side camera; 330. Local camera; 340. Supplemental lighting unit; 350. Dustproof and anti-fog assembly; 400. Microelectromechanical inertial measurement unit; 500. Edge computing module; 510. Boundary calibration module Block; 520, Process Phase Inheritance Module; 530, Visual Offset Recognition Module; 540, Integrated Boundary Modulation Module; 550, Low-Level State Machine Control Module; 560, Calibration Parameter Output Module; 570, Human-Machine Interaction Module; 580, Verification Record Module; 590, Early Warning Linkage Module; 600, Roller; 610, Zero Position Marker; 620, Shaft Head; 630, Outer Circle; 640, Rubber Coating Layer; 650, Weld Area; 660, Backup Phase Reference; 700, Manufacturing Execution System. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0033] In this embodiment, the coaxiality deviation of the roller 600 mainly refers to the visible geometric offset formed by the roller 600 at the shaft head 620, end face, outer circle 630, rubber coating layer 640, weld area 650, or clamping reference 140. This visible geometric offset serves as a visual attribution feature for identifying the source of dynamic balance imbalance. The system uses the imbalance amount, imbalance phase, and left and right correction surface distribution output by the dynamic balance detection unit as the main input, and uses the shaft head 620 eccentricity, end face center drift, outer circle 630 runout, rubber coating thickness area, visible morphological offset of weld area 650, and clamping reference 140 drift obtained by the visual acquisition unit as attribution features. The system outputs processing calibration parameters such as re-clamping, re-photo confirmation, outer circle 630 compensation processing, rubber coating local grinding, weight correction, weight removal correction, or abnormal re-inspection through the underlying state machine.

[0034] Example 1, as Figure 1 and Figure 2 As shown, this embodiment provides a visual calibration system for the coaxiality deviation of the roller 600, including a rotating support mechanism 100, a dynamic balance detection unit, a visual acquisition unit, a supplementary lighting unit 340, a phase synchronization unit, a microelectromechanical inertial measurement unit 400, an edge computing module 500, a boundary calibration module 510, a process phase inheritance module 520, a visual bias recognition module 530, a comprehensive boundary modulation module 540, a low-level state machine control module 550, a calibration parameter output module 560, a human-machine interaction module 570, a verification and recording module 580, and an early warning linkage module 590.

[0035] The rotating support mechanism 100 includes a left support seat 110, a right support seat 120, an auxiliary support roller 130, and a clamping reference 140. The left support seat 110 and the right support seat 120 are respectively disposed at both ends of the rotating support mechanism 100 to support the shaft ends 620 at both ends of the roller 600. The auxiliary support roller 130 is disposed in the lower middle part of the roller 600, and the clamping reference 140 is disposed on the mounting platform of the left support seat 110 or the right support seat 120. The rotating support mechanism 100 is used to support the roller 600 to be tested and to make the roller 600 rotate according to the testing requirements. The rotating support mechanism 100 can adopt a support structure that uses the shaft ends 620 at both ends, a center support, a support roller, or a support structure that cooperates with an existing dynamic balancing machine. For a roller 600 with a shaft head 620, it is preferable to use the shaft heads 620 at both ends as the rotation positioning reference; for an integral roller 600 without an independent shaft head 620, it is preferable to use a roller support, end face positioning hole or temporary calibration ring to establish the rotation reference; for a long roller 600, the rotation support mechanism 100 can be provided with an auxiliary roller 130 or an adjustable support point along the axial direction, and the support synchronization verification is performed before testing.

[0036] The dynamic balancing detection unit includes a left vibration acquisition unit 210, a right vibration acquisition unit 220, a phase trigger unit 230, and an encoder 240. The left vibration acquisition unit 210 is mounted on the left support 110, and the right vibration acquisition unit 220 is mounted on the right support 120. The phase trigger unit 230 is used to identify the moment when the zero-position mark 610 passes through the predetermined detection position, and the encoder 240 is used to record the angle increment of the roller 600. The phase trigger unit 230 and the encoder 240 serve as shared components for dynamic balancing detection and phase synchronization. The dynamic balancing detection unit is used to acquire the current imbalance amount, imbalance phase, and left and right correction surface distribution of the roller 600. The dynamic balancing detection unit can be a vibration sensor, phase detector, and controller from an existing dynamic balancing machine, or it can be an external detection module that communicates with the dynamic balancing machine. The data output by the dynamic balancing detection unit is used for subsequent visual attribution, state machine boundary modulation, and calibration path selection.

[0037] The vision acquisition unit includes an end face camera 310, a side camera 320, and a local camera 330. The end face camera 310 is used to acquire the outer edge of the shaft head 620, the end face contour, the end face zero mark 610, and the center position of the shaft head 620. The side camera 320 is used to acquire the outer edge of the roller 600, the contour of the rubber coating layer 640, the weld area 650, and the runout trajectory of the outer circle 630 at different axial positions. The local camera 330 is used for magnified acquisition near the unbalanced phase. In different workstation arrangements, the end face camera 310, side camera 320, and local camera 330 can be selectively activated according to the object being inspected. The vision acquisition unit is equipped with a dustproof and anti-fog assembly 350, which includes a dustproof window, an air curtain blowing port, a lens anti-fog heating element, and a temperature acquisition element. When the dustproof window is contaminated or the temperature and humidity near the lens reach the preset anti-fog conditions, the dustproof and anti-fog assembly 350 performs blowing or anti-fog treatment.

[0038] The supplementary lighting unit 340 can use a ring light, strip light, coaxial light, polarized light, or strobe light source. For the black coating layer 640, it is preferable to use oblique strip light to enhance the contour boundary; for the metal reflective roller 600, it is preferable to use polarized light or coaxial light to suppress reflection; for processing environments with a lot of dust, dustproof windows and air curtain blowing ports are provided outside the camera and light source.

[0039] The microelectromechanical inertial measurement unit 400 is preferably fixed near an industrial camera bracket, inspection fixture, or roller 600 bearing seat to collect vibration data within the image exposure window. The microelectromechanical inertial measurement unit 400 is used to assist in determining whether the current image is affected by machine tool vibration, impact, or support jitter, and participates in the modulation of visual attribution acceptance boundaries, reshoot boundaries, and clamping verification boundaries.

[0040] The edge computing module 500 is used to receive dynamic balance detection data, visual image data, phase synchronization data, and vibration data, and to perform phase conversion, image recognition, process recording, comprehensive analysis, boundary modulation, and state machine control. The edge computing module 500 can be placed next to the dynamic balance detection station or integrated into the industrial control computer or vision terminal of the roller 600 processing line.

[0041] Example 2, as Figure 3 As shown, in this embodiment, a boundary calibration module 510 is set to generate various preset boundaries and thresholds. Before formal operation, the system was calibrated using qualified roller 600 samples, shaft head 620 eccentric samples, outer circle 630 runout samples, rubber coating thick area samples, weld area offset samples, clamping drift samples, and vibration interference samples. A boundary parameter table is generated based on the dynamic balance test data, visual offset data, microelectromechanical inertial measurement data, and verification results of the calibrated samples.

[0042] The boundary parameter table should include at least the following boundaries: synchronization boundary, zero-position identification boundary, reference drift boundary, cross-process inheritance boundary, strong coupling phase boundary, weak coupling phase boundary, retest trigger boundary, visual attribution acceptance boundary, reshoot boundary, clamping verification boundary, shaft head recalibration boundary, outer diameter compensation machining boundary, rubber coating local grinding boundary, weighting correction boundary, weight removal correction boundary, correction amplitude upper limit, correction achievable level boundary, correction destructive level boundary, manual execution boundary, early warning boundary, and anomaly review boundary. The boundary parameter table should be stored separately according to the diameter, shaft diameter, length, surface material, whether rubber coating is used, support method, and dynamic balance level of the roller 600.

[0043] Each boundary record in the boundary parameter table includes at least the boundary number, boundary name, applicable roller type 600, applicable process, input data source, unit, calibration sample number, calculation or calibration method, typical range, current value, validity period, trigger condition, abnormal branch, output status, output field, and update time. The boundary parameter table is written by the boundary calibration module 510 and called by the edge computing module 500 during the state machine evolution process. When any boundary record lacks a unit, current value, trigger condition, or abnormal branch, the system will not enable that boundary to participate in the underlying state machine modulation and will enter an abnormal review state or prompt for recalibration.

[0044] The output fields for different types of boundaries are fixed according to their type. Synchronization boundary outputs timestamp deviation value, unit milliseconds, available acquisition phase marker, and retake requirements; Phase boundary outputs phase difference, angle unit, strong coupling marker, weak coupling marker, and phase retest requirements; Vision boundary outputs edge width, gradient continuity rate, ghosting width, image clarity level, and retake requirements; Vibration boundary outputs root mean square acceleration value, standard deviation of acceleration, vibration stability level, and acceptance marker; Execution boundary outputs allowable correction marker, prohibition of correction marker, upper limit of correction amount, and verification requirements; Early warning boundary outputs early warning level, trigger batch, triggering device, handling action, and recovery conditions.

[0045] The following formula is preferred for generating some boundary parameters: ; in, Represents the boundary value of the x-th class of detected objects; This represents the average value of the x-th type of test object in the calibration sample; Indicates the boundary magnification factor; This represents the standard deviation of the x-th type of test object in the calibration sample.

[0046] This formula can be used to generate image sharpness boundaries, vibration stability boundaries, clamping drift boundaries, phase remeasurement discrete boundaries, visual bias amplitude boundaries, and manual execution deviation boundaries.

[0047] The settings are categorized according to roller type, working condition stability, and tolerance for misjudgment. For standard steel roller 600, typical workshop environments, and stable support conditions, The preferred value is 1.5 to 2.0; for rubber-coated rollers of 600, with reflective surfaces, or in cases of slight vibration, The preferred value is 2.0 to 2.8; for long drums (600), multi-support drums (600), and operating conditions with strong dust or vibration, The optimal value is between 2.8 and 4.0. The system can also automatically adjust based on the false alarm rate of the calibration sample; when the false alarm rate is too high, the value should be increased. Reduce when the false negative rate is high and the adjusted Write the boundary parameter table for the corresponding specification roller 600.

[0048] The clear boundary of an image is determined by both the edge width and the gradient continuity rate. Preferably, images with an edge width of no more than 2.5 pixels and a gradient continuity rate of no less than 80% are considered acceptable. Images with an edge width greater than 3.5 pixels or a gradient continuity rate less than 60% are considered to be reshot. Images in between are included in the reshot confirmation. The edge shadow boundary is determined by the grayscale expansion width of the edge of the roller 600 along the rotation direction. Preferably, images with a grayscale expansion width of no more than 3 pixels are considered acceptable, and images with a grayscale expansion width greater than 5 pixels are considered unstable.

[0049] The vibration stability boundary is determined by the root mean square (RMS) value or standard deviation of the acceleration output by the MEMS inertial measurement unit 400 within the exposure window. Under normal workshop conditions, an RMS value of acceleration not greater than 0.05g is considered stable, between 0.05g and 0.15g is considered pending confirmation, and greater than 0.15g is considered low reliability. Under conditions such as long roller 600, multiple supports, or strong vibration, the above boundaries are calibrated by vibration interference samples and written into the boundary parameter table. The timestamp synchronization boundary is determined by the maximum deviation between the camera exposure center timestamp, phase trigger timestamp, and inertial measurement sampling timestamp. Preferably, a deviation not greater than 0.5 milliseconds is considered normal synchronization, and a deviation greater than 1 millisecond indicates the need for reshoot confirmation or phase retesting.

[0050] The preferred formula for determining the phase boundary of a strongly coupled phase is: ; in, Indicates a strongly coupled phase boundary; This indicates the angular error of encoder 240 or phase trigger unit 230; This indicates visual bias phase recognition error; This indicates the phase repeatability error of the dynamic balancing detection unit; This represents the phase error derived from the time synchronization error. All the error terms mentioned above have been converted to the same angular unit before being included in the calculation. It can be calculated by multiplying the time synchronization error by the angular velocity at the time of the 600-degree acquisition of the roller; when the acquisition phase is directly calculated using the encoder's 240-degree angle increment, The equivalent angle error corresponding to the timestamp latching error.

[0051] Weakly Coupled Phase Boundary Preferred selection Two to three times. When the phase difference between the unbalanced phase and the visual bias phase is no greater than... When the phase difference is greater than 1, the system determines that the two are strongly coupled; when the phase difference is greater than 1, the system determines that the two are strongly coupled. When the phase difference is at a certain point, the system determines that the two are weakly coupled; when the phase difference is at a certain point... and During this period, the system prioritizes entering the re-scan confirmation or phase re-measurement state.

[0052] The allowable residual imbalance can be given by the company's process standards, customer acceptance specifications, or preset values ​​from the dynamic balancing machine. When a unified conversion model needs to be established, the following formula is preferred: ; in, This indicates the allowable residual imbalance; m represents the mass of the 600mm drum. This indicates the allowable residual ratio of unbalance. The value is determined by the application of the roller 600, the speed rating, the customer's acceptance standards, and the company's dynamic balance rating, and is stored in the boundary parameter table of the corresponding roller 600 specification.

[0053] Example 3, as Figure 5 As shown, in this embodiment, the system establishes a unified circumferential phase reference for the roller 600. The phase synchronization unit includes a zero-position marker 610, an encoder 240, and a phase triggering unit 230; the zero-position marker 610 is preferably disposed on the end face of the roller 600, the end face of the shaft head 620, or on a temporary calibration ring that moves with the roller 600; the encoder 240 is used to output the rotation angle of the roller 600, and the phase triggering unit 230 is used to identify the time when the zero-position marker 610 passes through a predetermined detection position.

[0054] When transferring the roller 600 between different processing steps, the system prioritizes identifying the zero-position mark 610 set on the end face of the roller 600 or the end face of the shaft head 620, and converts the visual offset phase acquired in the current process to the same circumferential coordinate system of the roller 600. After the roller 600 is re-clamped, the system first identifies the zero-position mark 610 and re-establishes the phase coordinates of the current workstation, and then registers the phase coordinates of the current workstation with the historical process records. If the zero-position mark 610 fails to be identified, the position of the zero-position mark 610 drifts abnormally, or the phase synchronization error exceeds the synchronization boundary, the process phase inheritance module 520 does not directly use the current visual offset record for attribution judgment, and the underlying state machine enters the phase retest or re-sampling confirmation state.

[0055] When the end face zero-position mark 610 is worn, contaminated, or removed by machining, the system uses a backup phase reference 660. The backup phase reference 660 includes the keyway of the shaft head 620, the end face positioning hole, the weld start position, the temporary calibration ring mark, the fixture follow-up mark, or a reference point entered by the human-machine interface module 570. If there is a calculable fixed angular offset between the backup phase reference 660 and the historical zero-position mark 610, the edge calculation module 500 completes phase registration based on the fixed angular offset; if a fixed angular relationship cannot be established, the system pauses the process phase inheritance judgment, retains only the dynamic balance phase visual attribution of the current station, and prompts to re-establish the zero-position reference. The reference point entered by the human-machine interface module 570 is only used as a temporary fallback method; the system will prioritize prompting the re-establishment of an identifiable physical zero-position reference in subsequent processes.

[0056] The fixed angle offset between the backup phase reference 660 and the historical zero mark 610 is obtained by continuously collecting the phase difference between the zero mark 610 and the backup phase reference 660 three times under the same roller 600 in a low-speed stable rotation state. If the maximum dispersion of the three phase differences does not exceed the backup reference registration boundary, the average value is taken as the fixed angle offset. If the maximum dispersion exceeds the backup reference registration boundary, the process phase inheritance module 520 pauses the process phase inheritance and only outputs the current station visual attribution result, phase retest requirements, and prompts to re-establish the zero reference.

[0057] The acquisition phase corresponding to the visual image is preferably determined using an encoder 240-degree angle increment: ; Where θ represents the acquisition phase corresponding to the visual image; This indicates the initial phase when the zero-position marker 610 is recognized by the phase triggering unit 230; This represents the angle increment recorded by encoder 240 from the zero-position trigger moment to the camera exposure center moment; mod represents the modulo operation.

[0058] For data acquisition scenarios involving 240° rotation without an encoder or low-speed uniform rotation, a conversion method using angular velocity and time difference can be used; for indexing and phase-stop data acquisition scenarios, The indexing angle is directly given after being locked by the rotary support mechanism 100. When the roller 600 is in the process of acceleration and deceleration, the system prioritizes the encoder 240 angle increment; if the encoder 240 angle increment is unavailable, it enters the phase retest state and pauses the visual attribution of the current frame.

[0059] When the drum 600 is in forward or reverse rotation, frequently starts or stops, or undergoes direction reversal detection, the phase synchronization unit records the rotation direction marker. Both forward and reverse rotation phases are converted to the same zero-position coordinate. After a direction reversal, the system first performs a zero-position confirmation before allowing visual images to participate in attribution. If the angular velocity fluctuation exceeds the speed change boundary during start-stop, the system pauses dynamic shooting and prioritizes stop-phase shooting or low-speed stable shooting.

[0060] The optimal speed change boundary is determined by the ratio of the difference between the maximum and minimum angular velocities within the acquisition window to the average angular velocity. When this ratio is no greater than 5%, the system allows the use of dynamically captured images for attribution; when this ratio is between 5% and 15%, the system enters phase retesting or low-speed stable acquisition; when this ratio is greater than 15%, the system pauses dynamic acquisition and prioritizes phase-stop acquisition. For frequent start-stop or reversing conditions, the system re-executes zero-position confirmation after each reversal.

[0061] The camera exposure center time, the zero-position marker 610 trigger time, and the microelectromechanical inertial measurement unit (MEMS) 400 sampling time are preferably latched by the same high-frequency clock source. This high-frequency clock source can be a local hardware timer within the edge computing module 500, a fieldbus distributed clock, or a synchronous timing unit in a programmable logic controller. The camera exposure trigger signal, the phase trigger unit 230 signal, and the MEMS 400 sampling signal are all generated or recorded based on this high-frequency clock source to reduce the impact of ordinary communication bus latency, software polling latency, and data upload jitter on the phase conversion results.

[0062] Each frame of visual data is bound to a data packet, which includes at least the unique number of the roller 600, the process number, the equipment number, the image frame number, the exposure center timestamp, the phase trigger timestamp, the encoder 240-degree angle, the dynamic balancing test batch number, the MEMS inertial measurement window number, and the data integrity verification mark. After receiving the data packet, the edge computing module 500 first performs timestamp consistency verification and frame number continuity verification. If there is image frame loss, vibration data missing, or abnormal phase trigger timestamp, the frame data will not be used to generate visual attribution results and will enter the re-shoot confirmation state. If multiple data loss issues occur, the underlying state machine will be in an abnormal re-check state and output a sensor link check prompt.

[0063] Process records are transmitted between different workstations via local industrial networks, Manufacturing Execution System 700 (MES), or edge servers. Data exchange protocols can employ open platform communication unified architecture protocols, master-slave based industrial Ethernet communication protocols, real-time process Ethernet protocols, industrial control automation real-time bus protocols, or enterprise internal industrial data protocols. Each process record is assigned a unique Roller 600 number, process number, equipment number, record version number, timestamp, checksum, and continuation flag.

[0064] When the network connection is lost, all data of the current workstation is cached in the local edge computing module 500 and a queue to be uploaded is established according to the unique number of the roller 600 and the current process number. If the network is restored, the image data is continued to be uploaded in the order of record version number and timestamp. If the historical data cannot be restored, the dynamic balancing phase visual attribution for the current workstation is stopped, but the phase inheritance judgment across processes is no longer performed.

[0065] Data integrity verification can be performed using CRC checksums, hash checksums, or signed data digests. If only a single frame of image is lost or the microelectromechanical inertial measurement window is missing, the system enters the re-image confirmation state. If multiple frames of data are lost consecutively, the system enters the abnormal review state. If the checksum is found to be inconsistent or conflicting with the record version, the system locks the process record and outputs a manual review prompt through the human-machine interaction module 570.

[0066] When there is a version conflict, inconsistent checksum, or the historical record is locked, the system retains the original data packet, the conflict data packet, and the latest data packet, and generates a conflict record. The unlocking process includes equipment administrator confirmation, rereading the unique number of roller 600, re-identifying the zero mark 610, re-collecting the current station data, and reviewing the historical process record. If the re-collected result can be phase-registered with the historical record, the system restores the process phase inheritance. If it cannot be registered, the system only retains the visible geometric offset judgment result of the current station and prompts the user to re-establish the phase reference through the human-machine interaction module 570.

[0067] Example 4, as Figure 5 As shown, in this embodiment, the visual bias recognition module 530 performs edge extraction, contour fitting, phase positioning, and bias amplitude recording on the image obtained by the visual acquisition unit.

[0068] The visual offset features include at least the following: shaft head 620 eccentricity, end face center drift, outer circle 630 runout, thick rubber coating area, visible shape offset of weld area 650, and clamping reference 140 drift.

[0069] Edge extraction preferably employs sub-pixel edge detection or gradient threshold edge detection. Before edge points enter contour fitting, abrupt grayscale changes, isolated points, and discontinuous edge segments are removed. Circle fitting preferably employs least squares circle fitting or random sampling consistency circle fitting. When the fitting residual exceeds the fitting residual boundary or the number of effective edge points is lower than the effective point number boundary, the current frame image is not included in visual attribution and enters the re-shooting confirmation state. The fitting iteration terminates when the change in the center between two adjacent frames is less than 0.01 mm, or when the number of iterations reaches the preset upper limit.

[0070] The boundary of the circular fitting residual can be selected from 0.03 mm to 0.08 mm. The boundary size is written into the boundary parameter table according to the camera resolution, the diameter of the 600 roller, and the repeatability of the calibration part. The boundary of the effective edge point number can be selected as 60% to 80% of the number of theoretical sampling points of the fitted contour. If the number of effective edge points is lower than this boundary, or if the fitting residual exceeds the fitting residual boundary twice in a row, the current image is blocked and no visual attribution is performed. The image is then put into a reshoot confirmation state.

[0071] For end face center drift, the visual offset recognition module 530 extracts the end face outer contour or end face reference circle contour from the end face image, obtains the end face reference center through circle fitting, and compares the center with the historical end face reference center or tooling reference center to obtain the end face center drift direction and corresponding phase.

[0072] For the eccentricity of the shaft head 620, the visual offset recognition module 530 extracts the outer circle edge of the shaft head 620 from the end face image or the outer circle image of the shaft head 620, obtains the center of the outer circle of the shaft head 620 through circle fitting, and compares the center of the outer circle of the shaft head 620 with the end face reference center, the fixture rotation center or the historical shaft head 620 center to obtain the eccentricity direction and corresponding phase of the shaft head 620.

[0073] For the runout of the outer circle 630, the visual offset recognition module 530 extracts the edge of the outer circle 630 of the roller 600 from the lateral image and records the position change of the radial edge of the roller 600 under multiple phases; the circumferential position with a larger radial change amplitude is recorded as the yaw phase of the outer circle 630. For the axially longer roller 600, the visual offset recognition module 530 selects multiple detection sections along the axial direction of the roller 600 and records the change of the edge of the outer circle 630 corresponding to each detection section.

[0074] For areas with thick coating, the visual bias recognition module 530 extracts the outer edge of the coating from the outer contour image of the coating layer 640 and compares the outer contour of the coating with the reference circular contour or the baseline contour of the same batch. Areas with large local distance differences are recorded as areas with thick coating or areas with insufficient coating grinding.

[0075] For multi-layer coated or composite coated rollers 600, the visual offset recognition module 530 records the outer layer contour offset, the visible offset of the interlayer boundary, and surface texture anomalies, respectively. If the outer layer contour offset is coupled with the unbalanced phase, the bottom state machine enters the local overcoating grinding state; if there is only a surface texture anomaly but no obvious outer layer contour offset, the system does not directly output grinding parameters, the bottom state machine enters the anomaly review state, and the calibration parameter output module 560 outputs a manual review prompt and a re-shoot confirmation requirement.

[0076] For visible morphological offsets in weld region 650, the visual offset recognition module 530 records the weld-related phase based on weld location markers, weld appearance contours, grayscale abrupt changes, texture continuity changes, or weld zone boundaries. For internal density inhomogeneity, internal welding quality distribution, or local stiffness anomalies, the system does not make a definitive judgment, but only considers them as suspected sources related to weld region 650, and further confirms them in conjunction with dynamic balance verification results.

[0077] For clamping reference 140 drift, the visual offset recognition module 530 collects the position changes of zero mark 610, fixture mark, support mark or roller reference mark in multiple images; if the above marks are displaced beyond the reference drift boundary in multiple retest images, they are recorded as clamping reference 140 drift, thereby increasing the probability of entering the clamping verification state.

[0078] For the conical roller 600, the visual offset recognition module 530 establishes multiple cross-sectional reference profiles along the axial direction. The reference radius of each cross-section is given by the design taper or calibration profile at that cross-section. The system calculates the visual offset phase of each cross-section, and then selects the phase record and correction radius of the corresponding cross-section according to the axial position of the dynamic balance correction surface.

[0079] For rollers 600 with different diameters at both ends of the shaft 620, the system establishes a first end shaft 620 reference and a second end shaft 620 reference respectively, and converts the two end shaft 620 references to a unified circumferential coordinate system through the same zero-position mark 610. The eccentricity amplitude of the two end shafts 620 is calculated separately according to their respective shaft diameters and end positioning references.

[0080] For an integrated roller 600 without an independent shaft head 620, the system preferentially uses the outer circle 630 support reference, the end face positioning hole or the temporary calibration ring as the circumferential zero reference.

[0081] When a roller 600 has multiple axial detection sections, the system first compares the visual offset phase of each section with the unbalanced phase. If the offset phases of multiple sections are all within the strong coupling range, the system prioritizes the section closest to the axial position of the dynamic balance correction surface as the main section; if the offset phases of multiple sections are different, the system marks them as axially inconsistent offsets and enters the re-sampling confirmation state, phase re-measurement state, or abnormal re-check state.

[0082] Example 5, as Figure 6 As shown, the system sets a confidence boundary quantity for acquisition, which is used to determine whether the current visual acquisition result meets the acceptance conditions for process phase inheritance and visual attribution. The confidence boundary quantity is generated by optical spatial features and mechanical temporal features. The optical spatial features include image clarity level, roller 600 edge ghosting level, end face contour continuity level, and exposure stability level. The mechanical temporal features include vibration stability level, acceleration dispersion level, and impact disturbance level acquired by the microelectromechanical inertial measurement unit 400 within the exposure window. The confidence boundary quantity is output as an acceptance field set. When any key field is lower than the corresponding acceptance condition, it is determined that it is lower than the visual attribution acceptance boundary.

[0083] Image sharpness level is determined based on the edge width and gradient continuity of the end face contour or outer circle 630 edge. When the edge width is less than the sharpness boundary and the length of the continuous edge gradient segment meets the visual acceptance requirements, the image sharpness level is acceptable; when the edge width exceeds the sharpness boundary, or the continuous edge gradient segment is interrupted, the image sharpness level is pending re-shooting. Edge motion blur level is determined based on the grayscale expansion width of the roller 600 edge along the rotation direction. When the grayscale expansion width exceeds the motion blur boundary, the edge motion blur level is unstable. Vibration stability level is determined based on the acceleration dispersion output by the MEMS / INS 400 within the exposure window. When the acceleration dispersion exceeds the vibration stability boundary, the current image enters a low acceptance level. Synchronization consistency level is determined based on the deviation between the camera exposure trigger timestamp, the zero-point trigger timestamp, and the MEMS / INS 400 sampling timestamp. When the timestamp deviation exceeds the synchronization boundary, the current image does not directly participate in visual attribution, and the system enters a re-shoot confirmation state.

[0084] When the acquired confidence boundary value is lower than the visual attribution confidence boundary, the underlying state machine control module 550 selects the corresponding state according to the unmet confidence conditions; when the image clarity level, edge blur level, or exposure stability level does not meet the confidence conditions, it enters the reshoot confirmation state; when the phase trigger timestamp, camera exposure center timestamp, or encoder 240 angle increment does not meet the synchronization conditions, it enters the phase retest state; when the vibration stability level, clamping reference 140 drift, or support synchronization index does not meet the confidence conditions, it enters the clamping verification state.

[0085] After initial use, model changeover, camera position adjustment, support mechanism maintenance, or dynamic balancing machine calibration, the system performs a calibration procedure. The system calibration follows this sequence: mechanical reference calibration, camera intrinsic parameter calibration, camera extrinsic parameter calibration, encoder 240 zero-point calibration, dynamic balancing detection unit zero-point calibration, microelectromechanical inertial measurement unit 400 zero-bias calibration, support mechanism repeatability calibration, and boundary sample calibration. The system proceeds to the next calibration step only if the result of the previous calibration meets the corresponding calibration boundary. If any calibration item fails to meet the requirements, the system will not enter automatic calibration mode and will output a maintenance prompt.

[0086] After camera intrinsic parameter calibration, the reprojection error is preferably no greater than 0.2 pixels; after camera extrinsic parameter calibration, the spatial reconstruction error of known points on the calibration component is preferably no greater than 0.05 mm; after encoder 240 zero-position calibration, the angular dispersion of three consecutive zero-position triggers is preferably no greater than 0.2 degrees; after microelectromechanical inertial measurement unit 400 static zero-bias calibration, the acceleration zero-bias fluctuation in the static state is preferably no greater than 0.02g; after support mechanism repeated positioning calibration, the end face zero-position phase deviation of the same standard roller 600 for three clamping operations is preferably no greater than 0.5 degrees. The above values ​​are parameters of preferred embodiments, and the actual values ​​can be written into the boundary parameter table based on the equipment accuracy, roller 600 specifications, and enterprise process level.

[0087] The system further performs spatial correspondence calibration between the coordinate system of the MEMS inertial measurement unit 400 and the physical pixel coordinate system of the camera. Using a stroboscopic calibration block with known edge spacing or known center position, image sequences and inertial measurement data are simultaneously acquired at a preset vibration frequency and direction to establish a transformation matrix between the acceleration axis of the MEMS inertial measurement unit 400 and the image coordinate axes of the end-face camera 310, the side camera 320, or the local camera 330. The edge calculation module 500 converts the inertial measurement data into equivalent disturbance components in the image coordinate direction based on the transformation matrix and writes them into the vibration stability level, edge ghosting level, and repeat shot requirement fields in the acquired reliable boundary quantities.

[0088] The system undergoes calibration procedures upon initial installation, production changeover, camera position adjustment, support mechanism maintenance, dynamic balancing machine calibration, replacement of the MEMS 400 inertial measurement unit, and after continuous operation reaching a preset cycle. In a typical workshop environment, the system is preferably calibrated once per shift or after 8 hours of continuous operation, and fully calibrated weekly. Under conditions of dust, high humidity, strong vibration, or frequent changeovers, the system is preferably calibrated every 4 hours. If three consecutive data points fall below the acceptance threshold, or if the same equipment triggers consecutive maintenance warnings, the system will initiate a temporary calibration.

[0089] The system compares low-speed stationary phase images, low-speed rotation images, and process speed images to distinguish between inherent machining biases and dynamic deformation biases. If a bias only appears under process speed or strong vibration conditions and disappears under low-speed stationary phase conditions, the system marks it as a dynamic deformation or support disturbance-related bias and does not directly output the 630mm outer diameter compensation machining or overmolding grinding parameters. If a bias is stable under low-speed stationary phase, low-speed rotation, and process speed conditions and remains coupled with the unbalanced phase, the system includes it in the attribution as an inherent machining bias.

[0090] When the phase difference of the same bias in the low-speed stationary phase image, low-speed rotating image, and process speed image does not exceed the strong coupling phase boundary, and the rate of change of the bias amplitude does not exceed the preset amplitude stability boundary, it is determined to exist stably. When the bias only appears in the process speed image, or when the phase difference or the rate of change of the bias amplitude exceeds the corresponding boundary, it is determined to be a dynamic deformation or support disturbance related bias. The rate of change of the bias amplitude is determined based on the ratio of the maximum amplitude difference of the same bias under different acquisition conditions to the bias amplitude in the low-speed stationary phase image, and is written into the boundary parameter table of the corresponding 600 roller specification.

[0091] The visual acquisition unit and the microelectromechanical inertial measurement unit 400 are equipped with a dustproof sealed housing, an air curtain purge port, a lens anti-fogging heating element, and a temperature acquisition element. The dustproof window of the visual acquisition unit is designated as a pollution monitoring area. When the average grayscale change in the monitoring area exceeds the window pollution boundary or the edge contrast continues to decrease, the air curtain purge is activated. If the image clarity level does not recover after purge, the system prompts for manual cleaning. The lens anti-fogging heating element is controlled based on the temperature, humidity, and dew point difference near the lens. Heating is activated when the difference between the lens temperature and the dew point temperature is less than the preset anti-fogging boundary, and the heating power is reduced when the difference returns to a safe range. The ambient temperature is preferably controlled between 0℃ and 45℃, and the relative humidity is preferably no greater than 85%. When the ambient temperature exceeds the equipment's operating boundaries, the system reduces the acquisition confidence boundary value and suspends automatic attribution.

[0092] For the long roller 600, the system performs multi-support synchronization verification before testing. Synchronization verification includes the height difference of each support roller, the contact pressure difference of the support rollers, the radial runout of the auxiliary support, and the repeatability error of the support seat. The height difference of the support rollers is preferably determined by the height difference between the center heights of adjacent support rollers. Under normal long roller 600 operating conditions, a difference of no more than 0.05 mm is considered normal synchronization; a difference greater than 0.05 mm but no more than 0.10 mm is considered requiring adjustment; and a difference greater than 0.10 mm is considered a support abnormality. The contact pressure difference of the support rollers is preferably determined by the ratio of the contact pressure difference between adjacent support rollers to the average contact pressure. A difference of no more than 15% is considered normal synchronization; a difference greater than 15% but no more than 30% is considered requiring adjustment; and a difference greater than 30% is considered a support abnormality. The radial runout of the auxiliary support is preferably no more than 0.05 mm, and the repeatability error of the support seat is preferably no more than 0.05 mm. For heavy-duty roller 600 or large-diameter roller 600, the above boundaries can be written into the boundary parameter table based on the support synchronization calibration sample.

[0093] The output data for the support synchronization verification includes at least the support point number, roller height difference, roller contact pressure difference, auxiliary support radial runout, support seat repeatability error, synchronization level, suggested adjustment direction, and retesting requirements. When any indicator is in the adjustment range, the system enters the clamping verification state and outputs a support adjustment prompt; when any indicator is in the support abnormal range, or when the same support point fails the synchronization verification twice consecutively, the system enters the abnormal review state and stops outputting parameters for 630mm outer diameter compensation machining, rubber coating local grinding, weight adjustment, and weight removal adjustment.

[0094] Example 6, as Figure 4 As shown, in this embodiment, the process phase inheritance module 520 uses the same circumferential zero-position reference to record visual offset features at multiple processing nodes. The process nodes include at least the following: after the shaft head 620 is press-fitted, after the outer diameter 630 is machined, after rubber coating or surface treatment, and before dynamic balancing verification.

[0095] Each process record includes at least the unique number of the roller 600, its specifications, process node, equipment number, zero-position mark 610 image, zero-position phase, visual offset type, visual offset phase, offset amplitude, acquisition confidence level, unbalance amount, unbalance phase, left and right correction surface distribution, calibration output result, and verification result. These records are transmitted between different workstations via a local industrial network or Manufacturing Execution System 700. If historical records are missing, the system retains the visual inspection and dynamic balancing verification functions for the current process, but does not enable process phase inheritance judgment.

[0096] In the process phase inheritance module 520, an error source and an error accumulation value are generated for each process record. The error sources include zero-position recognition error, camera extrinsic parameter error, encoder 240 error, clamping repetitive positioning error, and historical record registration error. In the edge computing module 500, the above errors are accumulated in the form of phase error and written into the process record. If the cross-process inheritance limit is exceeded, the process phase inheritance module 520 suspends the phase inheritance judgment based on multiple manufacturing processes and retains the visible geometric offset judgment result of the current station.

[0097] Cross-process inheritance boundaries are preferably determined based on cumulative phase error. For ordinary steel roller 600, direct inheritance is allowed when the cumulative phase error is no greater than 1 degree; when it is greater than 1 degree but no greater than 3 degrees, phase retesting is required; when it is greater than 3 degrees, cross-process inheritance judgment is suspended. For long roller 600, rubber-coated roller 600, or multi-support conditions, the above boundaries can be expanded to two levels of 2 degrees and 5 degrees according to the calibration sample, and written into the boundary parameter table of the corresponding specifications.

[0098] The minimum phase difference between the unbalanced phase and the visual bias phase is preferably calculated using the following formula: ; in, This represents the minimum phase difference between the unbalanced phase and the visual bias phase. This indicates the unbalanced phase output by the dynamic balancing detection unit; This indicates the visual bias phase identified by the visual bias recognition module 530.

[0099] When the same unbalanced phase is simultaneously accompanied by shaft head eccentricity (620°), outer diameter runout (630°), thick adhesive layer, or visible topographic offset in weld area (650°), the system generates a set of attribution candidates. Each attribution candidate includes process initiation level, phase coupling level, acquisition confidence level, correction achievable level, and correction destructive level.

[0100] The selection and optimization of the attribution candidate set is performed in a gating manner. The system first eliminates candidates whose acquisition confidence level is lower than the visual attribution confidence boundary, and then eliminates candidates whose correction attainability level does not meet the conditions of the current process. For the remaining candidates, priority is given to candidates that are in a strong coupling range with the unbalanced phase and appear continuously in multiple processes. If multiple candidates meet the above conditions, the system selects the candidate with the lower level of correction damage. If it is still impossible to distinguish between primary and secondary candidates, the underlying state machine enters the abnormal review state. After manual review, the calibration parameter output module 560 outputs phased correction suggestions and prioritizes correction methods with higher reversibility and lower damage.

[0101] Each attribution candidate's data format includes at least the candidate number, bias type, starting process, bias phase, bias amplitude, phase difference with the unbalanced phase, phase coupling level, acquisition confidence level, correction achievable level, correction destructive level, suggested entry state, anomaly marker, image index, and process record number. Bias types include 620° eccentricity of the shaft head, end face center drift, 630° runout of the outer circle, thick rubber coating area, 650° visible morphology bias in the weld area, and 140° drift of the clamping datum; phase coupling levels include strong coupling, transitional coupling, and weak coupling; acquisition confidence levels include acceptable, awaiting re-shooting, and low acceptance; correction achievable levels include allowable correction, restricted correction, and prohibited correction; and correction destructive levels include low, medium, and high.

[0102] The achievable correction level is determined based on whether the current process has recalibration fixtures, whether the outer diameter 630 retains compensation machining allowance, whether the rubber coating layer 640 allows for further grinding, and whether there is sufficient operating space at the weighting or weight-removal location. The destructive correction level is determined based on whether the material of the roller 600 body is changed, whether the effective thickness of the rubber coating layer 640 is affected, and whether the shaft head 620 needs to be disassembled or re-clamped. Candidates requiring only re-tapping, re-measuring, or weighting are classified as low destructive; candidates requiring minor compensation machining of the outer diameter 630 or partial grinding of the rubber coating are classified as medium destructive; candidates requiring disassembly of the shaft head 620, re-pressing, or large-scale weight removal are classified as high destructive.

[0103] The attribution candidate set is output to the integrated boundary modulation module 540 in the form of an attribution candidate table. The integrated boundary modulation module 540 first reads the acquisition confidence level and the correction achievable level, and eliminates candidates with low confidence and candidates prohibited from correction. Then it reads the phase coupling level and the starting procedure, and filters candidates that are strongly coupled and appear continuously in multiple procedures. If there are still multiple candidates, it reads the correction destructive level and selects the candidate with lower destructiveness first. If it is still impossible to distinguish between primary and secondary candidates, the underlying state machine enters the abnormal review state, and the calibration parameter output module 560 outputs phased correction suggestions and manual review prompts.

[0104] For weld area 650, if the system only detects the phase-related offset of the weld, but the low-speed geometric offset is not obvious, and the unbalanced phase stably points to weld area 650 at the process speed, the system marks the result as a suspected abnormality in the stiffness or mass distribution of weld area 650. The bottom state machine enters the abnormal review state, and the calibration parameter output module 560 outputs an abnormal review prompt. The abnormal review prompt includes the abnormality type, triggering cause, weld area 650 review phase, suggested review items, automatic correction prohibition mark, and manual review prompt, so as to avoid directly judging it as a geometric offset that can be solved by grinding the outer diameter 630.

[0105] Example 7, as Figure 6 As shown, in this embodiment, the edge computing module 500 sets four comprehensive analysis points, which enter the comprehensive boundary modulation module 540 in parallel and jointly affect the evolution boundary of the underlying state machine.

[0106] The first comprehensive analysis point is the phase synchronization unbalance load K. The phase synchronization unbalance load is used to describe the unbalance strength, phase stability, and left and right correction surface distribution of the current roller 600. It is preferred to express it in a multi-component form: ; Where U represents the imbalance obtained from the current detection; This indicates that residual imbalance is allowed; This represents the unbalanced phase dispersion obtained from multiple remeasurements. This represents the unbalanced component corresponding to the left correction plane; This represents the unbalanced component corresponding to the right correction plane; This represents a very small positive number used to avoid a denominator of zero.

[0107] Let the phase synchronization imbalance load K be denoted as ,in , , The integrated boundary modulation module 540 is incorrect. , and Instead of performing a simple weighted summation, the evolution boundary of the underlying state machine is modulated according to a conditional branching approach.

[0108] when Greater than or equal to 1 and When the phase stability condition is met, the integrated boundary modulation module 540 tightens the phase-locked window to 0.8 to 1.0 times the original phase-locked boundary and allows entry into strong coupling attribution judgment; when Greater than or equal to 1 and When the phase stability condition is not met, the integrated boundary modulation module 540 increases the priority of the retest trigger boundary, causing the underlying state machine to enter the phase retest state first; when When the calibration surface switching boundary is exceeded, the integrated boundary modulation module 540 writes the calibration surface switching mark into the underlying state machine evolution boundary, causing the calibration parameter output module 560 to prioritize outputting the left and right calibration surface verification requirements or calibration surface switching prompts; when Less than 1 and When the phase stability condition is met, the system will preferentially enter either the normal pass state or the review record state.

[0109] The three components in this formula are used to modulate the phase-locked window, the retest trigger boundary, and the correction surface switching boundary, respectively. Preferred selection 0.001 times to 0.01 times; if the dynamic balancing detection unit has already given the minimum effective resolution, then Alternatively, the minimum effective resolution can be selected.

[0110] The second comprehensive analysis point is the process phase offset inheritance amount, which is used to describe from which process the visual offset feature on the same circumferential phase begins to appear, and whether it continues to the dynamic balance verification stage.

[0111] The third comprehensive analysis point is the achievable boundary quantity for correction. This quantity is used to describe whether each correction method under the current process has the conditions for execution. The achievable boundary quantity for correction includes whether the shaft head 620 has not been finally fixed, whether the outer diameter 630 still has compensation machining allowance, whether the rubber coating layer 640 allows for further grinding, whether the weighting position allows for the installation of a balance block, whether the weight removal position allows for material removal, whether the left and right correction surfaces have operating space, and whether the current clamping datum 140 is reliable.

[0112] The fourth comprehensive analysis point is to collect a reliable boundary quantity, which is used to describe whether the current visual image and vibration environment support visual attribution.

[0113] like Figure 7 As shown, the underlying state machine checks the state entry conditions sequentially according to the trusted gating, reference gating, phase gating, process gating, and correction gating.

[0114] The reliability gate is used to determine whether the image and vibration data are reliable; the reference gate is used to determine whether the zero mark 610 and the clamping reference 140 are reliable; the phase gate is used to determine whether the unbalanced phase is stable; the process gate is used to determine whether the source of visual bias has a process inheritance relationship; and the correction gate is used to determine whether the current process has the conditions for correction.

[0115] When the trusted gate fails, the re-test confirmation state takes precedence over the outer diameter 630 compensation machining state, the rubber coating local grinding state, the weighting correction state, and the weight removal correction state; when the reference gate fails, the clamping verification state takes precedence over the weighting correction and weight removal correction states; when the phase gate fails, the phase retest state takes precedence over the outer diameter 630 compensation machining and rubber coating local grinding states; when multiple states among the shaft head 620 recalibration state, outer diameter 630 compensation machining state, rubber coating local grinding state, weighting correction state, and weight removal correction state simultaneously meet the entry conditions, the bottom state machine prioritizes the state with high reversibility of the current process, small correction amount, and low damage to the roller 600 body.

[0116] When the eccentricity and imbalance phase of the shaft head 620 are in a strong coupling range, and the process phase inheritance module 520 shows that the eccentricity has appeared since the shaft head 620 was press-fitted or has continued until the dynamic balance verification, and the calibration reachable boundary amount shows that the shaft head 620 still has the conditions for recalibration, the bottom state machine enters the shaft head 620 recalibration state; when the shaft head 620 has been finally fixed or lacks recalibration tooling, the bottom state machine enters the abnormal review state, and the calibration parameter output module 560 outputs a limited recalibration prompt.

[0117] The underlying state machine sets a state dwell counter and a state dwell time. The number of consecutive retakes in the retake confirmation state is preferably 1 to 3, and the state dwell time is preferably no more than 3 minutes; the number of consecutive retakes in the phase retest state is preferably 2 to 5, and the state dwell time is preferably no more than 5 minutes; the number of retests after clamping verification is preferably 1 to 2, and the state dwell time after clamping verification is preferably no more than 10 minutes. When the retake confirmation state, phase retest state, or clamping verification state reaches the corresponding upper limit of the number of retakes or the upper limit of the time in the boundary parameter table, and the corresponding gate control condition is still not met, the underlying state machine control module 550 enters an abnormal review state and stops outputting the outer diameter 630 compensation machining parameters, the rubber coating local grinding parameters, the weighting correction parameters, and the weight removal correction parameters.

[0118] The underlying state machine includes states such as normal pass, phase retest, re-scan confirmation, clamping verification, shaft head 620 recalibration, outer diameter 630 compensation machining, rubber coating local grinding, weight adjustment, weight removal adjustment, and abnormal review. The state machine evolves according to the tightening, loosening, and shielding of boundary parameters.

[0119] The output format for each status is uniformly generated by the calibration parameter output module 560. Normal pass status outputs the pass mark, roller 600 number, verification time, and residual imbalance; phase retest status outputs the number of retests, retest speed, and phase acquisition requirements; reshot confirmation status outputs the reshot phase, exposure parameters, and supplementary lighting parameters; clamping verification status outputs the clamping datum 140 verification requirements, support synchronization verification results, and re-clamping prompts; shaft head 620 recalibration status outputs the shaft head 620 eccentricity direction, recalibration position, and retest requirements; outer diameter 630 compensation machining status outputs the compensation machining circumferential phase, axial range, compensation direction, and upper limit of compensation amount; rubber coating local grinding status outputs the grinding circumferential range, grinding axial range, and grinding number limits; weighting correction and weight removal correction status outputs the correction surface, correction phase, correction quality, and verification requirements; abnormal review status outputs the abnormality type, triggering reason, automatic correction prohibited mark, and manual verification prompts.

[0120] Example 8, as Figure 8 As shown, in this embodiment, after the underlying state machine enters a stable state, the calibration parameter output module 560 generates corresponding machining calibration parameters. These calibration parameters can be automatic machining instructions, semi-automatic equipment parameters, or manual process prompts. For workstations that do not meet the conditions for automatic execution, the system outputs phase, position, compensation limit, and verification requirements, which are then executed by the operator or external machining equipment.

[0121] When the underlying state machine enters the normal pass state, the calibration parameter output module 560 outputs the dynamic balance verification pass result and records the roller 600 number, process node, unbalance amount, unbalance phase, correction surface distribution and image index.

[0122] When the underlying state machine enters the phase retest state, the calibration parameter output module 560 outputs the number of retests, retest speed, phase acquisition requirements, and visual acquisition requirements.

[0123] When the underlying state machine enters the repeat shooting confirmation state, the calibration parameter output module 560 outputs the repeat shooting phase, supplementary lighting mode, exposure time adjustment requirements, and the synchronous recording requirements of the microelectromechanical inertial measurement unit 400.

[0124] When the underlying state machine enters the clamping verification state, the calibration parameter output module 560 outputs a re-clamping prompt, a support seat reset prompt, a top or roller cleaning prompt, a clamping reference 140 re-photographing requirement, and a retesting requirement.

[0125] When the underlying state machine enters the shaft head 620 recalibration state, the calibration parameter output module 560 outputs the shaft head 620 eccentricity direction, recalibration position, allowed recalibration method, retest requirements after recalibration, and a prohibition on direct counterweight marking.

[0126] When the underlying state machine enters the outer diameter 630 compensation machining state, the calibration parameter output module 560 outputs the circumferential phase, axial range, compensation direction, upper limit of compensation amount, and post-machining verification requirements for the compensation machining.

[0127] When the bottom state machine enters the local grinding state of the rubber coating, the calibration parameter output module 560 outputs the grinding circumferential range, grinding axial range, grinding number limit, and post-grinding verification requirements.

[0128] When the underlying state machine enters the weighting or deweighting correction state, the calibration parameter output module 560 outputs the correction plane, correction phase, and correction mass. When the dynamic balancing detection unit has already decomposed the unbalance of the roller 600 into the unbalance components to be corrected on the corresponding correction plane, the calibration parameter output module 560 preferably uses the following formula to calculate the correction mass: ; in, Indicates the quality of the calibration; This indicates the unbalance component to be corrected on the selected correction surface; This indicates the correction radius set on the actual roller 600.

[0129] When using weighted correction Take the radial distance from the center of the balance block installation to the 600 rotation axis of the drum; when using drilling to remove weight, Take the radial distance from the geometric center of the drilling area to the rotation axis of the drum 600; when using grinding to remove weight, The radial distance from the equivalent centroid of the grinding area to the axis of rotation of the drum 600 is taken; when the drum 600 has a variable diameter or conical surface structure... Take the actual radius at the corresponding axial section of the correction; when there are multiple selectable correction positions on the same correction surface, the system prioritizes the position with small phase error, large correction radius, small impact on structural strength and high process accessibility.

[0130] For variable diameter rollers 600, tapered rollers 600, or rollers 600 that have undergone multiple grinding cycles, the calibration radius is remeasured by the vision acquisition unit or updated by feedback from the processing equipment after each grinding or weight removal. The calibration parameter output module 560 reads the latest radius record before the next calculation of calibration quality to avoid using the radius before grinding, which could lead to calibration quality deviations.

[0131] In actual execution, the calibration parameter output module 560 can also discretize the calculated correction quality according to the balance block specifications, the diameter of the weight removal tool, the width of the grinding tool, the wall thickness of the roller 600 and the process allowance, and generate the corresponding weighting mass, weight removal depth, weight removal area or grinding range.

[0132] In manual or semi-automatic operation scenarios, the human-machine interface module 570 displays the roller 600 number, correction phase, axial range, suggested action, upper limit of correction amount, verification requirements, and precautions. After completing the correction, the operator needs to confirm the actual correction position, correction method, and correction amount. If the deviation between the actual correction position and the system output phase exceeds the manual execution boundary, the system enters an abnormal review state, and the calibration parameter output module 560 outputs a manual review prompt, the actual correction position, the system output phase, the phase deviation value, and a re-verification requirement. If the manual execution deviation exceeds the limit multiple times consecutively, the system locks the automatic calibration output and prompts the team leader or process personnel for verification.

[0133] When the phase deviation between the manually corrected position and the system output exceeds the manual execution boundary after multiple consecutive manual corrections, the system locks the automatic calibration output. To unlock, process personnel need to reconfirm the calibration tool position, manual execution records, and verification data, and complete a verification using a standard sample or a roller with a known calibration position (600mm). If the verification passes, the system resumes automatic calibration output; if the verification fails, it remains locked and enters the equipment maintenance process.

[0134] After each correction, the verification recording module 580 re-acquires dynamic balance data, visual image data, and vibration data, and records the imbalance amount, imbalance phase, visual offset phase, state machine state, calibration parameters, acquisition confidence boundary quantities, and image index before and after the correction. The verification recording module 580 generates single-piece records, batch statistics, offset frequency statistics, residual imbalance trend, and statistical summaries. The single-piece records record the processing and verification results of a single roller 600; the batch statistics form quality traceability records according to roller 600 specifications, batch, process, and equipment number; the offset frequency statistics count the frequency of visual offset types; the residual imbalance trend records changes in residual imbalance after correction; and the statistical summaries output early warning judgment data to the early warning linkage module 590.

[0135] The verification record module 580 statistically analyzes the frequency of visual bias types, bias phase concentration, number of failed dynamic balance verifications, and trends in residual imbalance after correction, according to batch, specification, process, and equipment number. Early warnings are categorized into observation warnings, maintenance warnings, and shutdown verification warnings. When the frequency of the same visual bias type in the same batch reaches 10% to 20%, an observation warning is triggered, and the warning linkage module 590 outputs an increased sampling instruction to the manufacturing execution system 700. When the same bias type appears consecutively in at least 3 pieces on the same equipment or in the same process, and the reduction in residual imbalance after correction is less than 30%, a maintenance warning is triggered, and the warning linkage module 590 outputs a maintenance inspection instruction to the manufacturing execution system 700. When at least 5 pieces of excessive imbalance appear consecutively on the same equipment or in the same process, or when 3 consecutive pieces still fail to reach a normal pass state after correction, a shutdown verification warning is triggered, and the warning linkage module 590 outputs a workstation isolation instruction and a pause automatic calibration instruction to the manufacturing execution system 700. After the manufacturing execution system 700 executes the corresponding instruction, it writes the execution result back to the review record module 580.

[0136] Example 9, as Figure 9 As shown, this embodiment provides several typical operating conditions.

[0137] In the application scenario of rubber-coated rollers, roller 600 enters the dynamic balancing verification station after being pressed into the shaft head 620, machined on the outer diameter 630, and rubber-coated and ground. The dynamic balancing detection unit detects an excessive imbalance in roller 600 and outputs the imbalance phase and the distribution of the left and right correction surfaces. The phase synchronization unit converts this imbalance phase into the actual circumferential position of roller 600. The vision acquisition unit acquires images of the rubber-coated surface and the edge of the outer diameter 630 around this circumferential position, while the microelectromechanical inertial measurement unit 400 records vibration data within the exposure window. The acquired reliable boundary quantity display image is clear and the vibration is stable. The visual bias recognition module 530 identifies a thick rubber coating area at this circumferential position, and the process phase inheritance module 520 shows that this bias appears for the first time after the rubber coating process. The comprehensive boundary modulation module 540 lowers the entry boundary of the rubber coating local grinding state and raises the entry boundary of the direct weight removal state. The bottom state machine enters the rubber coating local grinding state and outputs the rubber coating grinding phase, grinding axial range, grinding number limit, and verification requirements.

[0138] In the scenario of clamping datum 140 drift, the roller 600 under test exhibits significant phase drift during dynamic balancing verification. The vision acquisition unit repeatedly captures images of the end face zero mark 610, the outer edge of the shaft head 620, and clamping datum 140, while the microelectromechanical inertial measurement unit 400 simultaneously records vibration data. After the system identifies the positional drift of clamping datum 140 and acquires reliable boundary quantities to display vibration fluctuations in the support area, the integrated boundary modulation module 540 raises the weight correction and deweight correction states to enter the boundary, lowers the clamping verification state to enter the boundary, causing the bottom state machine to enter the clamping verification state, and outputs requirements for re-clamping, support seat reset, roller cleaning, and retesting.

[0139] In the application scenario of outer diameter 630 compensation machining, no obvious eccentricity of the shaft head 620 was found after the roller 600 was press-fitted, but edge runout of the outer diameter 630 at a certain circumferential position was recorded after machining the outer diameter 630. During dynamic balancing verification, the unbalance phase was close to the phase of the runout of the outer diameter 630. The acquired reliable boundary quantity display image was reliable, and the achievable correction boundary quantity showed that the outer diameter 630 still had a compensation machining allowance. The integrated boundary modulation module 540 reduced the boundary of the outer diameter 630 compensation machining state and restricted the entry of the overmolding grinding, shaft head 620 recalibration, and direct weight removal states. The bottom state machine entered the outer diameter 630 compensation machining state and output the circumferential phase, axial range, compensation direction, and upper limit of compensation amount.

[0140] In the suspected abnormal scenario of weld area 650, the visual acquisition unit did not find obvious low-speed geometric bias, but the unbalanced verification result stably pointed to weld area 650. The system marked it as a suspected abnormality in the stiffness or mass distribution of weld area 650, and the bottom state machine entered the abnormal review state. The calibration parameter output module 560 outputs an abnormal review prompt.

[0141] The abnormality review prompts include the abnormality type, triggering cause, 650 review phase of the weld area, suggested review items, automatic correction prohibition mark, and manual review prompt. Among them, the automatic correction prohibition mark restricts the system from directly outputting the outer diameter 630 compensation machining parameters or the local grinding parameters for the rubber coating.

[0142] This system further provides various threshold and boundary generation methods through the boundary calibration module 510, solves the zero-position mark 610 failure problem through the backup phase reference 660, solves the multi-source data synchronization problem through data packet and timestamp verification, solves the multi-bias superposition problem through the unweighted attribution candidate set, solves the state machine conflict problem through gating sequence, state priority and timeout mechanism, and improves the quality traceability capability of mass production through review records, process deviation early warning and production line linkage. This solution can reduce blind grinding, repeated dynamic balancing and manual experience judgment, and improve the first pass rate and process stability in the manufacturing process of roller 600.

[0143] It should be noted that, for the sake of brevity, the foregoing method embodiments are described as a series of actions, but this does not mean that the application limits the order of the steps. Based on the ideas of this application, some steps can be executed in different orders or in parallel without affecting the functional implementation. Secondly, those skilled in the art should also understand that the specific embodiments described in the specification are preferred embodiments of the technical solutions of this application, and not limitations on the scope of protection of this application. All equivalent improvements or substitutions made within the spirit and principles of this application should be covered within the scope of protection of this application.

[0144] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A visual calibration system for roller coaxiality deviation, comprising a rotary support mechanism, a dynamic balance detection unit, a visual acquisition unit, a phase synchronization unit, and an edge computing module, characterized in that, The dynamic balancing detection unit is used to obtain the unbalance amount, unbalance phase, and correction surface distribution of the roller; The phase synchronization unit is used to map the unbalanced phase to the actual circumferential position of the roller; The visual acquisition unit is used to acquire visual images of the object at its circumferential position. The edge computing module includes a visual bias recognition module, a process phase inheritance module, a comprehensive boundary modulation module, a low-level state machine control module, and a calibration parameter output module; The visual bias recognition module is used to identify visual geometric bias features from the visual image; The process phase inheritance module is used to record the occurrence process nodes and phase continuity relationships of the visible geometric offset features on the same circumferential phase in multiple manufacturing processes. The integrated boundary modulation module is used to generate the underlying state machine evolution boundary based on the phase synchronization imbalance load, process phase offset inheritance, correction reachable boundary, and acquisition reliable boundary. The underlying state machine control module is used to perform state transitions based on the evolution boundaries of the underlying state machine. The calibration parameter output module is used to output roller calibration parameters based on the state transition results.

2. The visual calibration system for roller coaxiality deviation according to claim 1, characterized in that: The phase synchronization unit includes a zero-position marker, a phase triggering unit, and an encoder; The phase triggering unit is used to identify the moment when the zero-position marker passes through a predetermined detection position; The encoder is used to record the angular increment from when the zero-position marker is identified to when the visual image is acquired; The phase synchronization unit converts the unbalanced phase to the unified circumferential phase coordinates of the roller based on the zero-position mark, the phase triggering unit, and the encoder.

3. The visual calibration system for roller coaxiality deviation according to claim 2, characterized in that: The phase synchronization unit is also used to invoke a backup phase reference when the zero-position marker cannot be identified or the position deviation of the zero-position marker exceeds the zero-position identification boundary. The backup phase reference includes the shaft head keyway, end face positioning hole, weld start position, temporary calibration ring mark, fixture follow-up mark, or reference point entered through the human-machine interaction module; When there is a fixed angle offset between the backup phase reference and the historical zero mark, the phase synchronization unit completes phase registration according to the fixed angle offset; When the fixed angle offset cannot be established, the process phase inheritance module suspends the phase inheritance judgment based on multiple manufacturing processes.

4. The visual calibration system for roller coaxiality deviation according to claim 1, characterized in that: The visual acquisition unit includes an end-face camera, a side camera, and a local camera; The end-face camera is used to acquire images of the drum end face, zero-position mark, and shaft head; The side camera is used to acquire images of the outer circle of the drum, the rubber coating layer, and the weld area; The local camera is used to acquire magnified images of the regions corresponding to the unbalanced phase. The visual offset recognition module is used to identify shaft head eccentricity, end face center drift, outer circle runout, adhesive thickness area, visible shape offset of weld area, and clamping reference drift based on the visual image.

5. The visual calibration system for roller coaxiality deviation according to claim 1, characterized in that: The process phase inheritance module is used to generate process records; The process record includes the unique roller number, process node, zero-position phase, visual offset type, visual offset phase, offset amplitude, acquisition confidence level, imbalance amount, imbalance phase, calibration surface distribution, calibration output result, and verification result; The process phase inheritance module is also used to record zero-position identification error, camera extrinsic parameter error, encoder error, clamping repetitive positioning error, and historical registration error. When the cumulative result of the error exceeds the cross-process inheritance boundary, the process phase inheritance module pauses the phase inheritance judgment based on multiple manufacturing processes and retains the visible geometric offset judgment result of the current workstation.

6. The visual calibration system for roller coaxiality deviation according to claim 1, characterized in that: The edge computing module also includes a boundary calibration module; The boundary calibration module is used to generate a boundary parameter table based on qualified roller samples, shaft head eccentric samples, outer circle runout samples, rubber coating thick area samples, weld area offset samples, clamping drift samples, and vibration interference samples. The boundary parameter table includes the following boundaries: synchronization boundary, zero-position identification boundary, reference drift boundary, cross-process inheritance boundary, strong coupling phase boundary, weak coupling phase boundary, retest trigger boundary, visual attribution acceptance boundary, reshoot boundary, clamping verification boundary, shaft head recalibration boundary, outer circle compensation machining boundary, rubber coating local grinding boundary, weighting correction boundary, weight removal correction boundary, upper limit of correction amplitude, correction achievable level boundary, correction destructive level boundary, manual execution boundary, early warning boundary, and abnormal review boundary. The boundary records in the boundary parameter table include applicable roller type, applicable process, input data source, boundary value, trigger condition, abnormal branch, output status, and output field.

7. The visual calibration system for roller coaxiality deviation according to claim 6, characterized in that: The system also includes a microelectromechanical inertial measurement unit; The microelectromechanical inertial measurement unit is used to collect vibration data within the visual image exposure window; The acquired reliable boundary quantity is generated by optical spatial features and mechanical temporal features; The optical spatial features include image sharpness level, roller edge ghosting level, end face contour continuity level, and exposure stability level. The mechanical time characteristics include vibration stability level, acceleration dispersion level, and impact disturbance level; When the acquired confidence boundary quantity is lower than the visual attribution acceptance boundary, the integrated boundary modulation module modulates the underlying state machine evolution boundary according to the unmet acceptance conditions, and drives the underlying state machine control module to enter the re-shot confirmation state, phase re-measurement state, or clamping verification state.

8. The visual calibration system for roller coaxiality deviation according to claim 7, characterized in that: The underlying state machine control module performs state transitions according to trusted gating, reference gating, phase gating, process gating, and correction gating; When the trusted gate fails, the re-shot confirmation state takes precedence over the outer diameter compensation machining state, the local grit repair state, the weight correction state, and the weight removal correction state. When the benchmark gate fails, the clamping verification state takes precedence over the weighting correction state and the de-weighting correction state. When the phase gate fails, the phase retest state takes precedence over the outer diameter compensation machining state and the local grinding state of the rubber coating. When the repeat confirmation state, phase retest state, or clamping verification state reaches the corresponding upper limit of the number of times or the upper limit of the time in the boundary parameter table, and the corresponding gate control passage condition is still not met, the underlying state machine control module enters the abnormal review state. When multiple states among the shaft head recalibration state, the outer diameter compensation machining state, the rubber coating local grinding state, the weighting correction state, and the weight removal correction state simultaneously meet the entry conditions, the bottom-level state machine control module preferentially selects the state with a low level of correction destructiveness, a correction amount that does not exceed the upper limit of the correction amount amplitude, and a correction attainable level that meets the current process conditions.

9. A visual calibration system for roller coaxiality deviation according to claim 8, characterized in that: The calibration parameter output module is used to output shaft head recalibration parameters, outer diameter compensation machining parameters, rubber coating local grinding parameters, weighting correction parameters, weight removal correction parameters, clamping verification parameters, or abnormal re-check prompts; The shaft head recalibration parameters include the shaft head eccentricity direction, recalibration position, and retesting requirements; The outer circle compensation machining parameters include the circumferential phase of compensation machining, axial range, compensation direction, and upper limit of compensation amount; The parameters for localized grinding of the rubber coating include the circumferential grinding range, the axial grinding range, the limit on the number of grinding cycles, and the post-grinding verification requirements. Both the weighting correction parameters and the deweighting correction parameters include the correction surface, correction phase, correction quality, and verification requirements; The clamping verification parameters include clamping reference re-shooting requirements, support synchronization verification results, and re-clamping prompts; The anomaly review prompts include the anomaly type, triggering reason, a flag indicating that automatic correction is prohibited, and a prompt for manual review.

10. A visual calibration system for roller coaxiality deviation according to claim 9, characterized in that: The edge computing module also includes a review and recording module and an early warning linkage module; The review record module is used to generate single-item records, batch statistics, bias frequency statistics, residual imbalance trends, and statistical summaries. The early warning linkage module is used to generate observation early warnings, maintenance early warnings, or shutdown review early warnings based on the statistical summary. When the observation warning is generated, the warning linkage module outputs an increased sampling inspection instruction to the manufacturing execution system; When the maintenance warning is generated, the warning linkage module outputs a maintenance check command to the manufacturing execution system; When the shutdown review warning is generated, the warning linkage module outputs a workstation isolation command and a pause automatic calibration command to the manufacturing execution system.

Images