A high-precision drilling method and device for a circuit board

Abstract

The application discloses a kind of high-precision drilling processing method and device of circuit board.Belongs to multilayer circuit board processing technical field, comprising: in the inner layer conductive pattern of multilayer circuit board, multiple self-checking composite target units are arranged, each unit includes at least two physical independent and corresponding same theoretical virtual target center sub-mark group.X-ray image is collected after laminating, the registration relationship between design, image and drilling equipment workpiece coordinate system is established, and the position consistency of candidate virtual target center is verified according to the measured coordinates of each sub-mark group;Local coordinate compensation model is established using the qualified target unit, and the model is verified outside using the qualified target unit which does not participate in the determination of model parameters and is located in the applicable area of model.Only when the verification is passed, the compensation coordinates are outputted and drilling is executed, to reduce the hole position deviation caused by abnormal target or failed model, improve the drilling processing precision and reliability.

Description

Technical Field

[0001] This invention relates to the field of multilayer circuit board manufacturing technology, and in particular to a method and apparatus for high-precision drilling of circuit boards. Background Technology

[0002] As circuit boards (PCBs) evolve towards higher density, miniaturization, and multilayering, the allowable positional deviation between drilled holes and conductive patterns such as inner layer pads and interconnects is gradually decreasing, placing higher demands on the positional accuracy and stability of PCB drilling. Multilayer PCBs are typically formed by laminating multiple conductive pattern layers and insulating dielectric layers. During lamination, overall expansion and contraction, local deformation, or interlayer positional shifts may occur, causing deviations between the designed hole positions before lamination and the target positions of the inner layers after lamination. If drilling is performed directly according to the designed hole positions, the hole positions may shift relative to the inner layer conductive patterns.

[0003] Existing circuit board drilling compensation methods typically utilize X-ray imaging to obtain the position of inner-layer targets. A coordinate transformation or position compensation relationship is established based on the theoretical and measured positions of one or more targets, and the drilling coordinates are then adjusted accordingly. However, the integrity of the target's pattern, imaging quality, position identification errors, and localized non-uniform deformation of the circuit board can all affect the target position measurement results. When a target with abnormal patterns, imaging abnormalities, or unreliable position measurement exists, if that target is directly involved in establishing the compensation model, its measurement deviation may be transmitted to the drilling coordinates along with the model.

[0004] Furthermore, while the compensation model meets the fitting requirements at the target location involved in the modeling, it cannot be guaranteed that the model has sufficient predictive reliability in the local area where the borehole is to be drilled. In some existing solutions, the consistency of the target's own measurement results and the applicability of the compensation model in non-modeling locations are not adequately verified, or the applicable area of ​​the compensation model is not clearly constrained. Therefore, when there are target anomalies, local deformation characteristics do not match the adopted model, or the location of the borehole to be drilled exceeds the suitable compensation area, inappropriate compensation coordinates may still be generated, affecting the accuracy of the borehole location.

[0005] Therefore, it is necessary to provide a high-precision drilling method for circuit boards to improve the reliability of target measurement data and local coordinate compensation models, reduce hole position deviations caused by abnormal targets or inapplicable compensation models, and thus improve the accuracy and stability of circuit board drilling. Summary of the Invention

[0006] Based on this, the present invention provides a high-precision drilling method for circuit boards, comprising: setting multiple self-calibrating composite target units in at least one inner conductive pattern of a multilayer circuit board, each self-calibrating composite target unit including at least two physically independent sub-marker groups, each sub-marker group having a preset theoretical geometric relationship relative to the same theoretical virtual target center, and being independently identifiable in an X-ray image; after laminating the multilayer circuit board, acquiring an X-ray image containing the self-calibrating composite target units, establishing a coordinate registration relationship between the design coordinate system, the X-ray image coordinate system, and the workpiece coordinate system of the drilling equipment, and connecting the theoretical virtual target center, The measured coordinate information of the sub-marker groups and the theoretical hole position to be drilled are unified into the same coordinate system; candidate virtual targets are determined according to the measured coordinate information of each sub-marker group in the same self-calibrating composite target unit, and the positional consistency among multiple candidate virtual targets is used to determine whether the self-calibrating composite target unit meets the preset internal calibration conditions; for self-calibrating composite target units that meet the preset internal calibration conditions, the measured target is determined according to the candidate virtual targets judged by positional consistency; for self-calibrating composite target units that do not meet the preset internal calibration conditions, they are not used as modeling target units or external verification target units; multiple A self-calibrating composite target unit that meets the preset internal verification conditions is used as a modeling target unit. A local coordinate compensation model is established based on the correspondence between the theoretical virtual bullseye and the measured bullseye of the modeling target unit, and the applicable area of ​​the local coordinate compensation model is determined. The number and spatial distribution of the modeling target units ensure that the parameters of the local coordinate compensation model are uniquely determined, and the theoretical hole location to be drilled is located within the applicable area of ​​the model. At least one self-calibrating composite target unit that meets the preset internal verification conditions, does not participate in the parameter determination of the local coordinate compensation model, and whose theoretical virtual bullseye is located within the applicable area of ​​the model is selected as... As an external verification target unit, the theoretical virtual bullseye of each external verification target unit is input into the local coordinate compensation model to obtain the corresponding predicted bullseye. When the positional deviation between the predicted bullseye and the corresponding measured bullseye of each external verification target unit meets the preset model verification conditions, the local coordinate compensation model is determined as a valid local coordinate compensation model, and the compensation coordinates of the hole to be drilled are determined and drilling is performed. When any positional deviation does not meet the preset model verification conditions, the compensation coordinates of the hole to be drilled are not output using the corresponding local coordinate compensation model, and drilling is not performed on the hole to be drilled based on the local coordinate compensation model.

[0007] Furthermore, each of the self-verifying composite target units includes at least three mutually distinguishable sub-marker pairs, each sub-marker pair constitutes a sub-marker group and includes two conductive sub-markers, the theoretical centers of the two conductive sub-markers being symmetrical about the theoretical virtual target center; different sub-marker pairs are distinguished from each other by spacing and / or arrangement direction, and all conductive sub-markers in the same self-verifying composite target unit are formed in the same inner conductive pattern.

[0008] Furthermore, the midpoint of the measured coordinates of the two conductive sub-markers in each sub-marker pair is determined as the corresponding candidate virtual bullseye; a first consistency threshold and a second anomaly threshold greater than the first consistency threshold are preset; when the distance between any two candidate virtual bullseyes is not greater than the first consistency threshold, the average coordinates of all candidate virtual bullseyes are determined as the measured bullseye; when there is one and only one abnormal candidate virtual bullseye, and the distance between any two candidate virtual bullseyes in the remaining candidate virtual bullseyes is not greater than the first consistency threshold, and the distance between the abnormal candidate virtual bullseye and each of the remaining candidate virtual bullseyes is greater than the second anomaly threshold, the abnormal candidate virtual bullseye is excluded, and the average coordinates of the remaining candidate virtual bullseyes are determined as the measured bullseye; when none of the above conditions are met, the corresponding self-verification composite target unit is determined to not meet the preset internal verification conditions.

[0009] Furthermore, the X-ray attenuation of the two conductive sub-markers in the same sub-marker pair after background correction is obtained under background matching conditions, wherein the background matching condition is that the X-ray projection background of the two conductive sub-markers meets a preset background consistency requirement; when the sum of the two X-ray attenuation values ​​is not less than a preset positive lower limit, the attenuation difference index of the corresponding sub-marker pair is determined according to the ratio of the absolute value of their difference to their sum; when the distance between any two candidate virtual bullseyes is not greater than the first consistency threshold and the attenuation difference index of each sub-marker pair is not greater than the preset attenuation difference threshold, or when the sub-marker to which the abnormal candidate virtual bullseye belongs... When the attenuation difference index of a target pair is greater than the preset attenuation difference threshold, while the attenuation difference indexes of all other sub-target pairs are not greater than the preset attenuation difference threshold, the corresponding self-calibrating composite target unit is determined to meet the dual-channel internal calibration condition. When the position consistency judgment and attenuation consistency judgment are inconsistent, any of the X-ray attenuation values ​​is invalid, or the sum of the two X-ray attenuation values ​​is less than the preset positive lower limit, X-ray images are reacquired according to the preset imaging parameter sequence and the dual-channel internal calibration is re-executed. If the dual-channel internal calibration condition is still not met after reaching the preset maximum number of reacquisitions, the corresponding self-calibrating composite target unit is deactivated.

[0010] Further, three self-calibration composite target units that satisfy the dual-channel internal calibration conditions and whose theoretical virtual targets are not collinear are selected as modeling target units, and a local affine transformation model is established according to the three modeling target units; the theoretical hole positions of the holes to be drilled and the theoretical virtual targets of at least one off-model verification target unit that satisfies the dual-channel internal calibration conditions, does not participate in the determination of model parameters and is different from the three modeling target units are all located within the area enclosed by the three modeling target units; an off-model residual is determined according to the difference between the predicted target center of each off-model verification target unit and the corresponding measured target center, and a verification uncertainty control amount is determined according to the dispersion degree of candidate virtual target centers of the respective calibration composite target units participating in modeling and off-model verification, the calibrated coordinate recognition uncertainty and the preset local non-affine model allowance; only when the modulus of each off-model residual and the verification uncertainty control amount are both not greater than a preset model verification threshold, the local affine transformation model is determined as the current effective local affine transformation model, otherwise the corresponding local affine transformation model is not used to output compensation coordinates.

[0011] Further, before performing drilling, a plurality of board surface reference calibration marks that do not participate in the parameter determination of the coordinate registration relationship and whose number and spatial distribution are sufficient to check each degree of freedom of the coordinate registration relationship are used to independently verify the coordinate registration relationship, and a registration residual is determined according to the distance between the theoretical position of each board surface reference calibration mark and the measured position converted by the coordinate registration relationship; when each registration residual is not greater than a preset registration verification threshold, the independent verification is determined to be qualified, and when any registration residual is greater than the preset registration verification threshold, the independent verification is determined to be unqualified; when the independent verification is unqualified, outputting the compensation coordinates of corresponding holes to be drilled is stopped, and X-ray images are re-acquired within the preset maximum number of re-registrations, coordinate registration, dual-channel internal calibration, establishment of the local affine transformation model and off-model verification are re-performed, and independent verification is re-performed on the re-established coordinate registration relationship; the output of the compensation coordinates of the corresponding holes to be drilled is resumed only after both the re-performed independent verification and off-model verification are qualified; when the independent verification and off-model verification are not simultaneously qualified after reaching the preset maximum number of re-registrations, drilling of the corresponding area is stopped.

[0012] Furthermore, multiple currently effective local affine transformation models are established for multiple borehole regions that are spatially adjacent or partially overlapping. For each currently effective local affine transformation model, according to the selection rule predetermined before processing and yielding a unique result under the same input data, a diagnostic model-outside verification target unit is determined from the model-outside verification target units used in the last effective model-outside verification of that model. The vector pointing from the predicted target center of the diagnostic model-outside verification target unit to its actual target center is determined as the model-outside residual vector of the corresponding model. A residual lower limit and a direction angle threshold are preset. The residual lower limit is greater than zero and not greater than the preset model verification threshold, and the direction angle threshold is greater than or equal to zero and less than 90 degrees. When the magnitudes of the model-outside residual vectors of at least two currently effective local affine transformation models reach the residual lower limit, and the angle between any two model-outside residual vectors is not greater than the direction angle threshold, the at least two currently effective local affine transformation models are determined as a diagnostic model group, and the output of compensation coordinates using the diagnostic model group is suspended.

[0013] Furthermore, the union of the modeling target units and the diagnostic model external verification target units used by each local affine transformation model in the diagnostic model group is determined as the candidate exclusion target set, and the original model state of the diagnostic model group is saved before performing the single target exclusion test; a single target exclusion test is independently performed on each self-verifying composite target unit in the candidate exclusion target set, and each single target exclusion test starts from the original model state and excludes only one self-verifying composite target unit. The local affine transformation model in the original model state that uses the self-verifying composite target unit excluded in this test as the modeling target unit or the diagnostic model external verification target unit is determined as the relevant model; for each relevant model, a preset substitution is used to... The modeling target units and diagnostic model external verification target units selected from the pool that are not included in the original model state of the relevant model, and the alternative target units that satisfy the dual-channel internal verification conditions and corresponding role space conditions, are used to re-establish and verify the relevant model. Specifically, when the alternative target unit is used to replace the modeling target unit, the corresponding role space conditions include that the parameters of the replaced local affine transformation model can be uniquely determined, and that the theoretical hole location to be drilled and the theoretical virtual bullseye of the target unit used for external verification are both located within the applicable area of ​​the replaced model. When the alternative target unit is used to replace the diagnostic model external verification target unit, the corresponding role space conditions include that the alternative target unit did not participate in the determination of model parameters and that its theoretical... The virtual bullseye is located within the applicable area of ​​the model; the external residual vectors of unrelated models and their original models remain unchanged, and the target replacement results, model parameters, and model states formed in the previous single-target exclusion experiment are not carried over to the subsequent single-target exclusion experiment; the residual persistence condition is determined when the magnitudes of the external residual vectors of each model in the diagnostic model group after single-target exclusion all reach the lower limit of the residuals and the angle between any two external residual vectors is not greater than the direction angle threshold; a recovery threshold and a minimum relative improvement ratio are preset, wherein the recovery threshold is greater than zero and less than the lower limit of the residuals; when the self-validating composite target unit excluded this time participates in at least two related models in the diagnostic model group, the magnitude of the new external residual vector of each related model is not greater than the lower limit of the residuals. When the recovery threshold is reached and the relative reduction ratio of the magnitude of the out-of-model residual vector before exclusion is not less than the minimum relative improvement ratio, the corresponding single-target exclusion test is determined to meet the preset improvement condition. When all single-target exclusion tests can be completed and all meet the residual persistence condition, the union of the applicable regions of each local affine transformation model in the diagnostic model group is determined as the suspected local deformation region. When exactly one single-target exclusion test meets the preset improvement condition, and the self-verifying composite target unit excluded in this test participates in all local affine transformation models in the diagnostic model group, and all other single-target exclusion tests meet the residual persistence condition, the excluded self-verifying composite target unit is determined as a suspected abnormal target unit.If any single-target exclusion test cannot be completed, or if the results of all single-target exclusion tests do not fall under any of the above situations, the use of the output compensation coordinates of the diagnostic model group should remain suspended; the output compensation coordinates should not be used until the local affine transformation model, which needs to be reconstructed or re-verified due to the single-target exclusion test, has passed out-of-model verification again.

[0014] Furthermore, for one or more local regions covering the same borehole to be drilled, one or more currently effective local affine transformation models verified outside the model are established respectively. The corresponding candidate compensation coordinates are determined using each of the currently effective local affine transformation models. The candidate compensation coordinates are then fused using an arithmetic mean or a pre-determined non-negative normalized weight to determine the final compensation coordinates. The internal consistency error control value of the target is determined based on the discreteness of the candidate virtual bullseye of the self-verifying composite target unit participating in the modeling and the calibrated coordinate identification uncertainty. The out-of-model verification error control value is determined based on the out-of-model residual obtained from the last effective out-of-model verification of the currently effective local affine transformation model, the discreteness of the candidate virtual bullseye of the out-of-model verification target unit, the coordinate identification uncertainty, and the unobserved margin of space determined based on the preset upper limit of the residual space change rate control and the distance between the verification position and the borehole position. The candidate compensation coordinate discrete error control value is determined based on the maximum distance between each candidate compensation coordinate and the final compensation coordinate. When there is only one candidate compensation coordinate, the candidate compensation coordinate... The discrete error control value is set to zero; multiple parameters not involved in the coordinate registration relationship are used to determine plate surface reference verification marks, whose quantity and spatial distribution are sufficient to verify each degree of freedom of the coordinate registration relationship, to independently verify the coordinate registration relationship. The coordinate registration error control value is determined based on the maximum distance between the theoretical position of each plate surface reference verification mark and the measured position after the coordinate registration relationship transformation, the registration uncertainty, and the margin for repeated clamping and coordinate transfer state changes; the maximum distance between the center of the command hole and the center of the measured hole in the calibration test hole is used as the reference value. The deviation and the measurement uncertainty of the hole center determine the positioning error control value of the drilling equipment; the target internal consistency error control value, model external verification error control value, candidate compensation coordinate discrete error control value, coordinate registration error control value and drilling equipment positioning error control value are linearly added to obtain the hole position comprehensive error control value, wherein each error control value is a non-negative quantity using the same length unit; when the hole position comprehensive error control value is not greater than the preset hole position tolerance of the hole to be drilled, drilling is performed according to the final compensation coordinate; otherwise, drilling is not performed on the hole to be drilled.

[0015] On the other hand, the present invention also provides a high-precision drilling processing device for circuit boards, including an X-ray imaging mechanism, a drilling mechanism, a processor and a memory, wherein the processor is communicatively connected to the X-ray imaging mechanism, the drilling mechanism and the memory, and the memory stores a computer program that can be executed by the processor.

[0016] When the processor executes the computer program, it controls the X-ray imaging mechanism and the drilling mechanism to perform the steps in the high-precision drilling method for the circuit board after the multilayer circuit board is laminated, on the multilayer circuit board equipped with a self-calibrating composite target unit.

[0017] The principles and effects of the present invention will be further explained below in conjunction with the above technical solutions:

[0018] (1) Each self-calibrating composite target unit includes at least two physically independent sub-marker groups that can be independently identified in X-ray images and have a preset theoretical geometric relationship with the same theoretical virtual target center. By determining candidate virtual targets based on the measured coordinates of each sub-marker group and performing internal verification based on the positional consistency among multiple candidate virtual targets, insufficient consistency of target measurement results can be detected before modeling and external verification, reducing the risk that imaging or recognition deviations of a single sub-marker group will directly affect the measured target center.

[0019] (2) Using only self-calibrating composite target units that meet the preset internal verification conditions for the establishment of local coordinate compensation models or external verification helps to block unreliable target data from being transmitted to the compensation model and borehole coordinates, thereby improving the credibility of modeling input data and verification data.

[0020] (3) By establishing the coordinate registration relationship between the design coordinate system, the X-ray image coordinate system and the workpiece coordinate system of the drilling equipment, and unifying the target information and the theoretical hole position to be drilled into the same coordinate system, the target detection, model calculation and drilling execution can adopt the same coordinate reference, reducing the influence of coordinate transformation error on the drilling position.

[0021] (4) By making the number and spatial distribution of the modeling target units uniquely determine the parameters of the local coordinate compensation model and limiting the theoretical hole position to be drilled within the applicable area of ​​the model, it helps to avoid model parameter uncertainty and use of the compensation model in unsuitable areas, and reduces hole position compensation deviation caused by unconstrained extrapolation or model mismatch.

[0022] (5) By using qualified target units that are not involved in the determination of local coordinate compensation model parameters and are located within the applicable area of ​​the model for external model verification, and only using the model to determine compensation coordinates and perform drilling when all external model verification results meet the preset model verification conditions, the predictive performance of the compensation model for non-modeled locations can be checked before drilling. When any verification result does not meet the conditions, the corresponding model output compensation coordinates are stopped, thereby reducing the risk of unsuitable models being used for drilling.

[0023] Therefore, this invention forms a step-by-step control process from internal target verification, coordinate unification, local model establishment, external model verification to drilling execution, which helps to reduce hole position deviations caused by abnormal targets, model mismatch, and inappropriate coordinate compensation, and improves the positional accuracy, reliability, and stability of circuit board drilling. Detailed Implementation

[0024] The following embodiments are used to illustrate the technical principles and possible implementations of the present invention, and are not intended to limit the scope of protection of the present invention. All coordinates in this document are two-dimensional coordinates within the circuit board surface, and the magnitude of a vector and the distance between two points are Euclidean distances in a unified coordinate system; length quantities in the same formula use the same unit. The parameter ranges and non-limiting values ​​listed herein are used to illustrate possible implementations of the present invention; the specific processing accuracy depends on equipment calibration, circuit board structure, processing conditions, and error budget allocation.

[0025] The basic process of this method is as follows: Multiple self-calibrating composite target units are set in at least one inner layer conductive pattern of a multilayer circuit board; X-ray images containing the target units are acquired after lamination; coordinates, image coordinates, board surface coordinates, and drilling workpiece coordinates are uniformly designed; internal verification is performed based on candidate virtual bullsiers obtained from multiple sub-marker groups within the same target; a local coordinate compensation model is established using qualified targets, and external verification is performed using qualified targets that did not participate in modeling. After external verification is passed, the compensation coordinates can be determined using the current effective model to control drilling; in the preferred embodiment using the following multi-model residual diagnosis and hole position comprehensive error control, the sources of abnormality are also excluded based on the residuals of multiple related models and the controlled single target, and the compensation coordinates are calculated only using the current effective model that has completed diagnosis and re-passed external verification. Drilling is performed when the hole position comprehensive error control value meets the hole position tolerance.

[0026] The "local coordinate compensation model" referred to in this specification is a coordinate transformation model that maps theoretical coordinates within the applicable area of ​​the model to measured coordinates after lamination. Depending on the complexity of the circuit board deformation within the applicable area and the number of available targets, rigid body transformation models, similarity transformation models, affine transformation models, piecewise affine transformation models, local low-order polynomial models, or interpolation models based on local control points can be used. Rigid body transformation models are used for local deformations mainly involving translation and rotation; similarity transformation models also allow uniform scaling; affine transformation models also allow anisotropic scaling and shearing; piecewise affine, local low-order polynomial, or local interpolation models are used in cases where there is spatial non-uniform deformation and sufficient control points are available. When establishing any local coordinate compensation model, the number of modeling targets is not less than the number required to solve the independent parameters of the model used, and the parameter design matrix formed by the theoretical virtual bullseye of the modeling targets has full column rank; when using the same number of modeling observations as the number of unknown parameters, the model parameters have a unique solution; when using more modeling observations than the number of unknown parameters and solving in a least-squares manner, the least-squares solution is unique. When using constraint or regularization for solving, the constraint or regularization rules are predetermined to ensure that the same input data yields a unique solution. If the number of modeling targets is insufficient, the parameter design matrix is ​​not of full rank, or the number of model conditions exceeds the preset allowable value, the corresponding model is not determined as a usable model. Instead, a new modeling target is selected, the applicable area of ​​the model is narrowed, or drilling in the corresponding area is stopped. Each model undergoes external validation using targets that were not involved in parameter determination. The applicable area of ​​the model can be determined based on the convex hull of the modeling target, a preset neighborhood, or a local area covered by external validation.

[0027] The "positional consistency" can be determined by the maximum pairwise distance between candidate virtual bullseyes, the maximum deviation of a candidate virtual bullseye from its coordinate average or geometric median, or by a preset robust clustering rule. The "two-level threshold and single anomaly exclusion" described below is a preferred deterministic implementation. The "preset model validation conditions" can be achieved by ensuring that the external residual does not exceed the validation threshold, the sum of the external residual and the corresponding coordinate uncertainty control does not exceed the validation threshold, or that the external residual and the coordinate uncertainty control do not exceed their respective error budgets. The thresholds and error budgets are determined before processing based on hole position tolerance, imaging resolution, coordinate calibration results, and equipment capabilities, and are not adjusted afterward based on the validation results.

[0028] When the same borehole is covered by multiple currently valid models, the final compensation coordinates can be the arithmetic mean of the candidate compensation coordinates or a pre-determined weighted average. The weighting coefficients can be determined based on the model's external validation residuals, the geometric quality of the modeled target, the distance from the borehole to the validation target, or a combination of the above factors. Each weighting coefficient is non-negative and their sum is 1. Regardless of the fusion method used, all models participating in the fusion must be in a currently valid state.

[0029] For ease of explanation, relevant terms and coordinate relationships are first defined. A "conductive electronic marker" is a conductive pattern formed within an inner conductive pattern that can be identified by X-rays. A "self-calibrating composite target unit" includes at least two sub-marker groups that do not physically share conductive electronic markers. Each sub-marker group has a preset theoretical geometric relationship relative to the same theoretical virtual target center and can be independently identified in an X-ray image. Each group can independently determine a candidate virtual target center corresponding to the theoretical virtual target center based solely on the measured coordinates of its conductive electronic markers. A "candidate virtual target center" is the target center independently calculated by a sub-marker group based on measured coordinates; a "measured target center" is the coordinates of multiple candidate virtual targets used for modeling or verification after internal verification, screening, and combination; a "predicted target center" is the coordinates obtained by inputting the theoretical virtual target center of the verification target into the model. In the following text, the "self-verifying composite target unit" will be referred to as the "target unit", the "modeling target unit" as the "modeling target", the "external verification target unit" as the "verification target", and the "current effective local affine transformation model" as the "current effective model".

[0030] In an implementation with only two independent sub-label groups, when the distance between the candidate virtual bullseyes obtained from the two sub-label groups is no greater than a preset first consistency threshold, the average coordinates of the two are determined as the measured bullseye, and the corresponding target unit is determined to meet the geometric internal verification condition. When using dual-channel internal verification, the corresponding attenuation consistency condition must also be met. When the distance between the two candidate virtual bullseyes is greater than the first consistency threshold, since there is a lack of a third independent sub-label group to provide majority consistency evidence, a candidate virtual bullseye is not selected from the two to recover the measured bullseye. Instead, the corresponding target unit is determined to not meet the internal verification condition in the current image, and it is not used to establish or verify the local coordinate compensation model. It is then processed according to the following limited re-acquisition rules. At least three sub-label groups are used to further identify and exclude unique abnormal sub-label groups.

[0031] Define the design coordinate system D, the X-ray image coordinate system X, the circuit board surface coordinate system B, and the drilling equipment workpiece coordinate system M. The theoretical conductive electronic markers, theoretical virtual bullseye, and theoretical hole positions are derived from the same design or CAM data, using the same origin, units, board surface orientation, and mirroring rules. After distortion correction of the X-ray image, the pixel coordinates are set using calibration points. Convert to panel coordinates:

[0032]

[0033] When using a two-dimensional affine transformation, at least three non-collinear calibration points should be used; if more than three, the least squares method can be used. When cone-beam distortion or multi-field stitching exists, first use a calibrated distortion correction model, then unify to coordinate system B. After the circuit board is clamped onto the drilling rig, establish the following using positioning holes, outer layer reference marks, or a dedicated board surface reference:

[0034]

[0035] in, This is the transformation matrix from image coordinates to board coordinates. For the corresponding translation vector, For the coordinates of the board surface; This is the transformation matrix from the plate surface coordinates to the workpiece coordinates. For the corresponding translation vector, These are the coordinates of the equipment and workpiece. and Both represent translation, rotation, scaling, and necessary shearing. When imaging and drilling share the calibrated workpiece coordinate system, the two-stage transformations can be combined. Modeling can be done in coordinate system B and then the final coordinates can be transformed to M, or the theoretical and measured coordinates can be unified to M before calculation. Regardless of the path used, the inputs for the same distance, threshold, model, and error calculations are all in the same coordinate system, avoiding redundant compensation.

[0036] After the registration relationship is established, independent plate surface datum verification marks that were not involved in the registration parameter solution are used for verification. When using two-dimensional affine registration, at least three non-collinear independent verification marks covering the processing area are used; for other registration models, verification marks sufficient to verify each degree of freedom and covering the processing area are used. Let the theoretical position and the measured position after transformation of the s-th verification mark be . , Then the registration residual of the s-th check mark is:

[0037]

[0038] All The maximum value in is denoted as The coordinate registration error control value is set as follows:

[0039]

[0040] in To calibrate and verify the uncertainty, Allowance for repeated clamping, This represents the margin for state changes between imaging and drilling within a preset maximum transport time and temperature / humidity range. Preset registration verification threshold. When each independent plate surface reference verification mark meets When the registration relationship is determined by independent registration verification; when any independent plate surface reference verification mark satisfies If the registration relationship fails the independent registration check, it will be determined that the registration relationship has not passed the check. A re-check will be performed if the preset time, environmental conditions, or clamping conditions are exceeded. If the independent check fails, the output of the corresponding borehole's compensation coordinates will be immediately stopped, and the current valid state of all local coordinate compensation models in the corresponding area will be paused. This is done within the preset maximum number of re-registration attempts. Internally, X-ray images are reacquired, coordinate registration relationships are re-established, and the measured coordinate information of each sub-marker group is re-identified. Internal verification, identical to that used in the original model, is re-executed. If the original model used dual-channel internal verification, dual-channel internal verification is re-executed. Subsequently, a local coordinate compensation model is re-established, and external model verification is re-executed. Before restoring the output compensation coordinates, the re-established coordinate registration relationships are independently verified again using the plate surface reference verification marks that were not involved in the registration parameter solution. Only when both the re-performed independent registration verification and external model verification are qualified are the corresponding compensation coordinates for the borehole to be drilled restored. If the above conditions are still not met simultaneously, drilling in the corresponding area shall be stopped. The candidate virtual bullseye, measured bullseye, model parameters, compensation coordinates, and external residuals of the model before re-registration can be saved as historical records, but shall not be used for model building, diagnosis, or borehole release after re-registration.

[0041] In a preferred composite target structure, each target unit includes m distinguishable sub-marker pairs. Each sub-tag pair constitutes a sub-tag group. The theoretical virtual bullseye is O, and the theoretical centers of the two conductive sub-tags in the k-th sub-tag pair are:

[0042]

[0043]

[0044] in Different sub-marker pairs are encoded using different spacing, orientations, or combinations thereof; each sub-marker pair does not share conductive sub-markers. All conductive sub-markers in the same self-calibrating composite target unit are formed in the same inner conductive pattern. The two members of the same sub-marker pair preferably have the same nominal material, layer thickness, shape, and size, and are formed in the same pattern transfer and etching process; "identical" allows for tolerances within manufacturing specifications. Inconsistencies can be detected between two sub-marker groups, but at least three sub-marker groups are used for unique anomaly identification and rejection; more sub-marker pairs can also be set to increase redundancy.

[0045] Let the pixel size converted to the board surface be... The minimum feature size w of the conductive electronic marker satisfies:

[0046]

[0047] in The minimum number of pixels required for segmentation, centroid removal, and edge culling can be 4 to 10. For example, the conductive sub-marker is a circular copper graphic with a diameter of 0.20 mm, and the three sub-marker pairs are oriented at 0 degrees, 60 degrees, and 120 degrees, respectively, with center distances of 1.20 mm, 1.60 mm, and 2.00 mm, respectively; the actual values ​​are determined based on pixel size, board space, etching capability, and design rules.

[0048] Let the original, dark field, and flat field pixel values ​​at pixel u be respectively , and The corrected transmission intensity is:

[0049]

[0050] Pixels with excessively small denominators, bad pixels, saturation, or exceeding the effective dynamic range are excluded. When the device grayscale and transmission intensity are inversely related, they are first converted according to the calibration relationship. Conductive electronic markers can be located by threshold segmentation, template matching, edge fitting, connected component filtering, or a combination thereof, and misidentification is eliminated according to preset shape, size, orientation, and spacing. The measured coordinates can be taken as the geometric centroid or the attenuation contrast-weighted centroid.

[0051]

[0052] Where Ω represents the set of valid pixels involved in coordinate recognition. Let be the two-dimensional coordinates of pixel u on the board, and ω(u) be a preset non-negative weight, satisfying... It can erode edge pixels inward or perform sub-pixel geometric fitting. Each imaging process uses preset rules or a fixed parameter sequence, and results are not selected afterward according to the desired outcome.

[0053] After completing the image recognition of conductive sub-labels, candidate virtual bullseyes are determined based on the affine midpoint preservation property of the centrally symmetric sub-label pairs. For ,have:

[0054]

[0055] Therefore, when the deformation within the composite target range can be approximated by the same affine transformation and the members are correctly identified, each sub-marker pair should obtain candidate bullseyes that are close to each other. The measured coordinates of the k-th sub-marker pair are... , The candidate virtual bullseye is:

[0056]

[0057] Let the sum of the coordinates of all candidate virtual bullseye be:

[0058]

[0059] Preset When any two The distance between them is no greater than At that time, it was determined that:

[0060]

[0061] and with As the measured bullseye, the target satisfies the geometric internal verification. If there exists and only one anomalous index q, such that... The distance between any of the remaining candidate bullseyes is no greater than [missing value]. ,and The distance to each of the other candidate bullseyes is greater than Then it is determined that:

[0062]

[0063] Geometric verification implementation methods can As the measured bullseye; when using dual-channel calibration. The attenuation condition must also be met. All other cases are geometrically irrecoverable under the current image, including anomalies with more than two candidate bullseyes, multiple central clusters of undetermined principal clusters, and deviations within a certain range. and The abnormal index cannot be uniquely identified or the number of valid candidate bullseyes is insufficient; this state is not equivalent to the physical target being permanently invalid.

[0064] If the coordinate error control value of the two members of the kth sub-mark pair is , Then, the error control value for the corresponding candidate target center coordinates of this sub-marker can be taken as:

[0065]

[0066]

[0067] in, ; All The maximum value in is determined as .

[0068]

[0069] in Covers manufacturing symmetry tolerances, identification errors, and normal non-affine variations within the target span. And it should not be less than a stable and resolvable plate displacement increment. Alternatively, it can be determined using a preset quantile from repeated imaging of a normally calibrated target. Then add resolvable intervals to form .

[0070] The term "background matching condition" as used in this specification refers to the fact that two conductive sub-markers in the same sub-marker pair adopt the same target area and background sampling area design rules, and the X-ray projection paths corresponding to the two conductive sub-markers are identical in terms of background conductive material other than the target conductive sub-marker, or their differences do not exceed a preset background matching tolerance. The identical or tolerance-range differences include at least the number of layers, material type, nominal thickness, and projected area of ​​the background conductive material. To form the background matching condition, in the target inner layer, a local copper-avoiding background area is formed around each conductive sub-marker, excluding the mark itself and containing no conductive material of the same layer. The two members of the same sub-marker pair adopt the same design shape and size of the background area. In other conductive pattern layers, projection copper-avoiding windows are set at corresponding positions along the preset X-ray projection direction, or a background matching pattern is set such that the distribution of background conductive material in the two projection paths satisfies the background matching tolerance.

[0071] The projection window is determined during the graphic design phase, with a minimum margin. satisfy:

[0072]

[0073] in For the maximum interlayer alignment tolerance, The distance from the target layer to the farthest related layer. The maximum X-ray beam tilt angle, The upper limit of registration error is pre-defined during the design phase. Margin is provided for the edges of the target and background areas. (Batch testing) Exceed In such cases, the background matching is not directly deemed valid; instead, the target is re-registered, re-imaged, or deactivated.

[0074] Define the target region for the j-th member of the k-th sub-tag pair. and background area For the same sub-marker pair, the same region shape, area, erosion width, and statistical rules are used. Only when both the target and background regions have sufficient effective pixels, the intensity is within the effective dynamic range, the image is not saturated, the signal-to-noise ratio meets the requirements, and the current registration error does not cause the target or background region to exceed the preset projection window, is the background-corrected X-ray attenuation calculated based on the corresponding intensity. If the background matching condition cannot be confirmed, the corresponding attenuation information is determined to be invalid. The average transmission intensity is:

[0075]

[0076]

[0077] When the strength is greater than the effective lower limit of the equipment, and it is not saturated, and When, calculate:

[0078]

[0079] like ,but:

[0080]

[0081] Preset Insufficient pixels, low signal-to-noise ratio, ineffective intensity, saturation, non-positive attenuation, or denominator less than [value missing]. At that time, the attenuation data is invalid. This can be determined based on repeated imaging of normal calibration sub-marker pairs. When the average transmission intensity errors between the target and background regions are approximately uncorrelated, for... The uncertainty of the attenuation can be estimated using the following formula:

[0082]

[0083] in, The uncertainty is the average transmission intensity of the target area. The uncertainty is the average transmission intensity of the background region. The uncertainty of the attenuation is given; it is propagated to the attenuation difference index to obtain... And can be taken ,in The attenuation difference index for normal calibration sub-marker pairs is the center value or upper limit benchmark, where z is the preset coverage factor. This is a non-negative additional margin. Alternatively, it can be determined by adding a preset multiple to the median absolute deviation. The threshold is fixed before detection. Multiple imaging levels are used to calibrate the effective intensity. and Unless it has been proven that the same threshold applies to all gears.

[0084] Dual-channel judgment includes three categories: First, all Satisfying the first positional consistency condition and all ,reserve Secondly, the unique geometric anomaly index q simultaneously satisfies And the rest ,reserve Thirdly, if the attenuation of a geometrically anomalous sub-marker pair is normal while the attenuation of other sub-marker pairs is abnormal, or if the attenuation of any sub-marker pair is abnormal despite being geometrically consistent, it indicates inconsistency between the two channels, and the current measured target center qualification will be revoked. Attenuation anomalies only indicate abnormalities in the material or imaging response and do not uniquely correspond to a certain defect.

[0085] If geometric recovery is impossible, any attenuation data is invalid, the two channels are inconsistent, or image recognition fails, the measured target center obtained based on the current image becomes invalid, and a limited reacquisition process begins. The initial X-ray image acquisition is not counted in the reacquisition count; each reacquisition reacquires an X-ray image according to a preset imaging parameter sequence and re-executes the coordinate and intensity recognition of the conductive electron markers, candidate virtual target center calculation, background correction attenuation calculation, and positional and attenuation consistency checks. Maximum number of reacquisitions. The preset finite positive integer can be 1 to 3; if the corresponding internal verification condition is met after any re-collection, the re-collection is terminated and the result of that re-collection is adopted. If the corresponding internal verification conditions are still not met, the corresponding target unit is deactivated and will not be used to establish or verify the local coordinate compensation model. When the projected background structure itself does not meet the background matching conditions, the target area is missing, the target exceeds the imaging range, or there are no imaging parameters in the preset imaging parameter sequence that can bring the target area and the background area into the effective dynamic range, the corresponding target unit can be deactivated directly.

[0086] After obtaining multiple target units that passed the dual-channel internal verification, three targets were selected as modeling targets, with their theoretical and measured bullseye as follows: , , The model is:

[0087]

[0088] Construct a 3x3 matrix H from three sets of theoretical coordinates, where the i-th row is a matrix H. and demand To avoid near-collinearity, the area of ​​the triangle is:

[0089]

[0090] in, , , You can also request... and satisfy . and Preset according to coordinate error, area size, model condition number, and hole position tolerance. Hole to be drilled. Located inside or on the boundary of a closed modeling triangle, satisfying:

[0091]

[0092] Instead of making unconstrained extrapolation to the outer hole position of the triangle, the target or area is adjusted.

[0093] A fourth, different, dual-channel qualified target that was not involved in parameter solving was selected for out-of-model validation, and its theoretical bullseye was determined. Located inside the triangle, preferably satisfying , , as well as The predicted bullseye and residual are:

[0094]

[0095] The set of retained candidate bullseyes for modeling target i is The measured target center was Each candidate bullseye will be retained. and The maximum distance between them is denoted as and define:

[0096]

[0097] For the validation targets, each candidate bullseye will be retained. and The maximum distance between them is denoted as and define . , Determined by repeated imaging, repeated localization, centroid calibration, or subpixel fitting residuals; discarded candidate points are not considered when using target recovery. u can only be zero if the calibration data conservatively covers the common identification uncertainty; the overall image common registration error is determined by... Additional charges apply.

[0098] The verification uncertainty control quantity is:

[0099]

[0100] Let V be the set of valid validation target indices for the current model; when there are multiple external validation targets, all of them will be... The maximum value in is determined as If there is only one, take The aforementioned , , and All are non-negative engineering control quantities using the same length unit. and Used to conservatively characterize the discreteness of candidate virtual bullseyes within the corresponding target and the repeatability of target-level coordinate recognition. Used to characterize the propagation of these control variables to the corresponding verification locations and the preset local non-affine model margin. This represents the maximum validation uncertainty control value across all validation locations of the current model.

[0101] in The preset non-negative local non-affine model margin can be determined by adding a conservative margin based on repeated measurements of normal plates, historical calibration results, or numerical simulation results covering the preset local non-affine variation range. The hole position error budget allocated to the model verification stage is obtained by subtracting other error budgets from the total hole position tolerance. Predetermine the model validation threshold And satisfy:

[0102]

[0103] Only when And each external validation target satisfies Only then is the model determined to be the currently valid model. If If any external validation residual does not meet the above conditions, then the candidate compensation coordinates for the corresponding borehole will not be output, and the modeling target, validation target, model applicable area will be reduced, or model applicable area will be redefined in a preset order within the maximum number of model adjustments. When there are multiple validation targets, each one must pass.

[0104] If verification fails, the maximum number of adjustments can be made. The modeling target, validation target, or region reduction are performed in a preset order, ensuring that the model remains solvable, triangle quality is maintained, region coverage is achieved, and the validation target's internal position remains valid. Evidence of previous validation failures must not be ignored due to subsequent target changes: if the original validation target remains internally valid, the data is effective, and it falls within the adjusted applicable region after re-imaging, it continues to be a mandatory validation target; it is only removed if it is internally invalid, identified as a suspected anomaly through controlled exclusion, or no longer belongs to the region due to region rezoning. When a target is removed due to region rezoning, the original failure location remains suspended until a new model is built and independently validated. The initial modeling does not count the number of adjustments. A value of 1 to 5 can be selected; drilling in the corresponding area should be stopped when the upper limit is reached or no suitable combination is found.

[0105] In other implementations, the local coordinate compensation model can also employ local similarity transformation, piecewise affine transformation, local bilinear interpolation, or a local polynomial model. Similarity transformation uses at least two qualified targets with different theoretical positions; affine transformation uses at least three non-collinear qualified targets; bilinear models use at least four qualified targets that make the parametric equations full rank; polynomial models use qualified targets with a number of coefficients that make the design matrix full rank. Each model also includes at least one qualified verification target that does not participate in parameter solving, and the applicable region of the model is determined based on the convex hull, interpolation unit, or predefined local region of the modeling target. Both the borehole to be drilled and the verification target are located within this region. The following residual diagnosis method uses the three-point affine model as the preferred diagnostic object; other models only employ corresponding diagnoses when the role preservation, substitution rules, and re-solution conditions after single target exclusion are predefined, and the three-point affine substitution rules are not directly applied.

[0106] In the preferred embodiment employing independent registration verification, before proceeding to residual diagnosis and before sending the physical drilling command, the independent registration verification is completed according to the aforementioned coordinate registration and independent verification methods. Affine registration uses at least three non-collinear independent verification markers covering the processing area. If the verification fails, the aforementioned process of re-acquiring X-ray images, re-establishing coordinate registration relationships, dual-channel internal verification, model reconstruction, external model verification, and re-independent registration verification is fully executed; the old target center, old model parameters, and old external model residuals must not be used.

[0107] For spatially adjacent or partially overlapping regions, multiple currently valid local affine models are established. Partial overlap occurs when the intersection area of ​​the model regions is greater than zero; partial overlap occurs when there is no intersection but the minimum boundary distance is not greater than a preset value. The time frame is spatially adjacent. The three modeling target sets of the i-th model are: The selection rules for external validation targets for diagnostic models are fixed before the start of this batch of circuit board manufacturing: if a model has only one validation target that has passed external validation, that validation target is determined as... When there are multiple validation targets that have passed out-of-model validation, first select the validation target with the largest residual magnitude from the last valid out-of-model validation. If the residual magnitudes are the same, select the validation target with the smallest fixed target number to ensure that the same input data yields a unique result. Define the set of participating targets:

[0108]

[0109] remove Other verification targets are not included. It is also not considered a candidate for exclusion in this diagnosis. Selection rules, , and The residual from the last effective out-of-model validation is fixed throughout the single-target exclusion and recursive diagnosis process and cannot be changed afterwards. If data is invalid, internally unqualified, or exceeds the applicable range of the model during the diagnosis, another validation target should not be directly used. Instead, the corresponding test should be identified as a diagnostic uncertainty test or the diagnosis should be terminated. After re-imaging and modeling, a new diagnosis can be started according to the same rules, but the old comparison results should not be used.

[0110] Diagnostic validation targets The theoretical virtual bullseye and the measured bullseye are denoted as follows: and The diagnostic residual of the i-th model is:

[0111]

[0112] The direction is unified from the predicted target center to the measured target center. At least two spatially correlated models are selected from the currently effective models to form a candidate model group. Preset:

[0113]

[0114] The angle between any two non-zero residuals is:

[0115]

[0116] When calculating the included angle, each residual vector is a non-zero vector, and the cosine value generated by floating-point operations is truncated to [-1, 1]. When each candidate model group... And any When this happens, the candidate model group is identified as the diagnostic model group M. Above the range where direction is dominated by noise; The preset control upper limit for the residual vector measurement disturbance mode, and ; The maximum azimuth error control amount caused by this disturbance can be determined according to... Determined by combining the allowable variation in the direction of the residual field If the residual direction condition is not met, the candidate model group is determined but the target anomaly diagnosis is not initiated, and the model remains in its current effective state. If the residual direction condition is met, the diagnostic model group is placed in the pending diagnosis state. Before all single target exclusion tests, model state updates, and re-model external verification are completed, the output of candidate compensation coordinates is paused and no actual drilling commands are sent.

[0117] The candidate exclusion target set is as follows:

[0118]

[0119] For each diagnostic model, target cells located on the same multilayer circuit board, whose effective measured target centers have been obtained based on the X-ray images and coordinate registration relationships used at the start of the diagnosis, meet the dual-channel internal verification conditions, are not deactivated, and are not included in the original target set of the model, are determined as the preset replacement pool for that model. Target cells serving as replacement modeling targets also satisfy the conditions of model solvability, modeling area, triangle quality, and coverage of the area to be drilled; target cells serving as replacement verification targets also satisfy the condition of being located within the applicable area of ​​the corresponding model and not participating in the solution of model parameters.

[0120] Before starting all single-target exclusion tests, a snapshot of the model state at the start of diagnosis is saved. This snapshot includes at least the three modeling targets for each model, the diagnostic external validation targets, the set of participating targets, model parameters, the model's applicable region, the set of mandatory external validation targets, the original external residuals, the effective state of each participating target, and the preset substitution pool for each model, the measured bullseye of the substitution targets, the dual-channel internal verification state, the target-level coordinate uncertainty control quantity, spatial location, and fixed target number. All single-target exclusion tests use the preset substitution pool and sorting parameters frozen at the start of diagnosis. A single-target exclusion test is performed independently for each target t in U; ​​only one t is excluded at a time, and each test is performed independently after restoring the aforementioned model state snapshot. The substitution targets, model parameters, residuals, verification results, or target states formed in the previous exclusion are not carried over to the next test. If images are reacquired during the experiment, or if the measured data, internal validation status, or validity status of any participating target or substitute target changes, the current diagnosis is terminated, the model is rebuilt with the updated data, and a new diagnosis is initiated. The results are not reused with the original experimental results. Define the relevant model:

[0121]

[0122] For this exclusion of target t, it will satisfy The model was determined to be the relevant model for this experiment, and will meet the requirements. The model was determined to be an irrelevant model for this experiment; the relevant or irrelevant state was only determined for the excluded targets in this experiment and was not considered a permanent attribute of the model. When t is the modeling target, the other two original modeling targets and the original Select a target replacement t from the pre-defined replacement pool of the corresponding model that satisfies the conditions of triangle solvability, area, and coverage; then remodel and validate; t is... When three original modeling targets are retained, a qualified dual-channel alternative verification target located within the triangle is selected from the pre-defined alternative pool of the corresponding model. When a t plays different roles in different models, it is processed according to its original role for each model. Unrelated models and their original residuals remain unchanged.

[0123] The substitution rules are fixed before processing. First, discard any backup targets that do not meet the dual-channel internal verification conditions or the corresponding role space conditions. Then, proceed according to the target-level coordinate uncertainty control quantities. A unique sequence of replacement targets is formed by ranking them in ascending order of size, spatial distance from the replaced target or the applicable area of ​​the corresponding model in ascending order of size, quality of the triangle formed after replacement in descending order of quality, and fixed target number in ascending order of size. When the target-level coordinate uncertainty control quantities are the same, they can also be sorted in descending order of the margin between the attenuation difference index and the attenuation difference threshold. The same input data yields a unique replacement order according to this rule. The new residuals after the reconstruction of the relevant model and the original residuals of the unrelated model constitute the complete residual set for this experiment. The diagnostic test model does not automatically regain validity simply because residuals can be calculated; it must still meet certain requirements after the diagnosis is completed. And it was re-validated through all out-of-model validations.

[0124] If the modulus of each residual in the complete residual set is not less than And the angle between any two residuals is no greater than 1. This experiment is a "residual persistence test". Additional assumption:

[0125] ,

[0126] The residual of the i-th related model after excluding t is When t participates in at least two related models, and each related model satisfies:

[0127]

[0128]

[0129] Since the original residuals entering the diagnostic model group satisfy The denominator of the relative improvement ratio is not zero. Experiments that meet the above recovery threshold and minimum relative improvement ratio conditions are designated as "single-target exclusion improvement experiments." If the same-direction condition disappears but the amplitude does not simultaneously meet both of the above conditions, it is not considered evidence of single-target anomaly. If any relevant model cannot obtain a qualified substitute target, cannot recover the three-stage model and one-test, cannot maintain regional coverage, cannot obtain effective out-of-model residuals, or if the experimental results do not meet either the residual persistence condition or the single-target exclusion improvement condition, the experiment is designated as a "diagnostic uncertainty experiment." Such diagnostic uncertainty experiments cannot be considered evidence of single-target anomalies and enter the diagnostic uncertainty branch.

[0130] because The single-target exclusion improvement experiment requires that the new residual modulus of the relevant model not exceed [a certain value]. The residual persistence test requires that the modulus of each residual in the complete residual set is not less than 1. Therefore, a single test cannot belong to both of the above categories simultaneously; all other results are classified as diagnostic indeterminate tests. Thus, each single-target exclusion test has a unique classification, and the subsequent diagnostic status is determined based on the mutually exclusive combinations of classifications from all tests.

[0131] After all tests are completed, the following mutual exclusion rules shall apply:

[0132] (1) Suspected local deformation: All tests can be completed and are all residual continuous tests. The union of all applicable regions of the diagnostic model group is determined as the initial influence region; the initial influence region can be divided into multiple grid cells according to the preset grid size. The grid cell containing the starting position of the model's external residual vector, and the grid cell that is spatially adjacent to the grid cell and whose corresponding external residual vector direction difference is not greater than the preset direction threshold, are determined as continuous or adjacent regions that need to be re-divided and re-modeled. When a continuous region that meets the above conditions cannot be obtained according to the residual spatial distribution, the union of the applicable regions of the model is determined as the suspected local deformation region. The registration and imaging calibration are checked again. After passing the check, the original model is invalidated. The influence region is divided into sub-regions where each region can form three modeling targets that meet the dual-channel internal verification conditions and at least one independent verification target that meets the dual-channel internal verification conditions, does not participate in the model parameter solution, and is located within the applicable region of the model. Drilling is resumed after the new model passes independent verification.

[0133] (2) Suspected single-target anomaly: The experiment with exactly target t is the improvement experiment, and all other experiments are continuous, and If t is deactivated; if t belongs to the set of mandatory verification targets for the relevant model, remove it from the corresponding mandatory set and save the reason and diagnosis record; invalidate the original model containing t, and reconstruct and verify it using the remaining qualified targets. Each reconstructed model retains at least one external verification target that meets the dual-channel internal verification conditions, is not involved in solving the model parameters, and is located within the applicable region of the reconstructed model; the reconstructed model can only be restored to the current valid model after being verified by all existing mandatory verification targets and at least one external verification target.

[0134] (3) Mixed state: exactly t in one experiment satisfies the improvement conditions, For a proper subset of M, there must be at least two irrelevant models whose residuals are still in the same direction, and all exclusion experiments except for the improvement experiment corresponding to t are continued. Temporarily discontinue t for relevant branches; if t belongs to the mandatory verification target set, remove it and save the reason and record. Then, reconstruct and verify the relevant model using the remaining qualified targets, ensuring that each reconstructed model retains at least one external verification target that meets the dual-channel internal verification conditions, does not participate in model parameter solving, and is located within the applicable region of the reconstructed model. Recursively diagnose a new diagnostic group consisting of at least two irrelevant models. If there are fewer than two irrelevant models, t is not considered a suspected anomaly and is directly diagnosed as uncertain. If any recursive branch is incomplete or uncertain, the entire original model group does not output compensation coordinates; the mixed state is only completed after all branches are completed and the models used for hole positions have been re-verified by all existing mandatory verification targets and at least one independent external verification target.

[0135] (4) Diagnostic uncertainty: There are more than two improvement tests, at least one diagnostic uncertainty test, non-continuous tests do not meet the amplitude improvement, only one relevant model improves, there are fewer than two unrelated models, recursion cannot continue, or one of the first three categories is not met. Pause all relevant models, and the target range can be expanded by the preset step size for re-diagnosis; stop drilling in the region when the maximum number of times is still uncertain.

[0136] The recursion depth, the number of regional partitioning layers, and the number of times the target range is expanded are all preset finite positive integers. Each recursion must reduce the number of models or shrink the region to be processed, and the count is not reset when switching paths; drilling is not allowed if any diagnosis is not completed.

[0137] In the implementation using multi-model residual diagnosis, after completing the diagnostic model group determination and, if a diagnostic model group exists, completing all single-target exclusion tests, model state updates, and re-external model verification, only models covering the borehole to be drilled and currently in a valid state are selected. If no diagnostic model group satisfies the residual unidirectional condition, the normal release calculation path is entered after completing the diagnostic model group determination. Models that are failed, pending diagnosis, suspended, incomplete diagnosis, or failed the last external model verification are not allowed to participate. Each currently valid local affine transformation model covering the same borehole is established according to the aforementioned three-point affine modeling and external model verification method. The applicable region of each model includes the theoretical borehole position, and each model independently satisfies the verification uncertainty control quantity and external model residual requirements. A model is not exempt from its own external model verification due to spatial overlap with another model. Let the theoretical borehole position be covered. The current number of valid models is n; No drilling is performed. The i-th model... The three modeling targets are , The centroid coefficient of the borehole to be drilled satisfies:

[0138]

[0139] The final compensated coordinates for the candidate and arithmetic mean are:

[0140]

[0141]

[0142] hour .

[0143] The set of retained candidate bull centers for the i-th model and the j-th modeling target is: The measured target center was Each candidate bullseye will be retained. and The maximum distance between them is denoted as and define:

[0144]

[0145] in To address the target-level identification uncertainty, the recovered bullseye ignores discarded points. The internal consistency error propagation value corresponding to the i-th model is:

[0146]

[0147] All The maximum value in is determined as .

[0148] The last valid validation target set for the i-th model is Let the theoretical virtual bullseye, the measured bullseye, and the final residual of the verification target v be denoted as follows: , and The set of retained candidate virtual bullseyes that participated in determining the measured bullseye is denoted as . Each candidate virtual bullseye will be retained. and The maximum distance between them is denoted as And determine the control quantity for the uncertainty of the verification target coordinates:

[0149]

[0150] The unobserved margin for the space between the verification location and the borehole position can be taken as follows:

[0151]

[0152] During the process calibration phase, at least two independent verification locations with different theoretical positions and not involved in the solution of corresponding model parameters are set for the applicable region of the i-th model. The theoretical coordinates of the a-th and b-th independent verification locations are denoted as follows: and The corresponding external residual vectors of the model are denoted as follows: and And determine the rate of change of the residual space:

[0153]

[0154] The maximum residual space change rate for all valid verification location pairs is denoted as... and compare it with the preset non-negative margin. The sum is determined as follows:

[0155]

[0156] It is a dimensionless, non-negative control quantity; the independent verification locations, stack-up structures, imaging parameters, registration methods, and model area scales used are the same as those of the circuit board to be processed, or more conservative conditions are used. When there are insufficient verification locations on the current board, pre-calibrated ones under similar stack-up and process conditions are used. Alternatively, the maximum value obtained from numerical simulation covering the preset maximum local non-affine field gradient can be added to a non-negative margin; if there is no calibration, similar historical data, or numerical simulation basis, it should not be used. If the value is set to zero, the unobserved margin of the corresponding model is determined as unreliable, and the model is not used in the compensation coordinate fusion and error release for the corresponding borehole location. If there is no other effective model with a valid residual space rate of change control that covers the current effective model for the borehole location, then no compensation coordinates are output and no borehole is drilled. The out-of-model verification error component corresponding to the v-th verification target of the i-th model is:

[0157]

[0158] All The maximum value in is determined as .

[0159] The discrete error components of the candidate coordinates corresponding to the i-th model are:

[0160]

[0161] At that time, all The maximum value in is determined as ; hour . The determination of the error control value is based on independent registration verification data corresponding to the current plate, current imaging state, and current clamping state. Before determining this error control value, independent registration verification is performed using multiple independent plate surface reference verification marks that are sufficient in number and spatial distribution to check each degree of freedom of the current coordinate registration relationship and have not participated in the calculation of registration parameters. The value is then determined according to the aforementioned calculation relationship. If no valid independent registration verification data is obtained or the independent registration verification fails, the previous coordinate registration error control value is not used, and the corresponding model is not used to release the hole position. Determined through calibration test drilling covering the entire machining range: Drill according to actual machining conditions. There are 1 calibration hole, and the command center and the actual hole center obtained by independent measurement of the l-th calibration test hole are respectively denoted as _l_. and Where l is 1 to The integer is used to determine the equipment positioning deviation of the l-th calibration test borehole:

[0162]

[0163] All The maximum value in is denoted as The measurement uncertainty of the calibration test borehole is denoted as . And determine:

[0164]

[0165] The positioning error control values ​​for this equipment cover equipment positioning, spindle runout, drill bit or beam deflection, tool wear, clamping, and hole forming errors.

[0166] The overall error control value for hole position is:

[0167]

[0168] Characterizes the propagation of the discretization of candidate target centers within the modeled target and the uncertainty in target-level coordinate identification to the location to be drilled; Characterizes the residual of the last effective out-of-model validation, the uncertainty of validation target identification, and the unobserved margin in space between the validation location and the borehole location; This characterizes the degree of inconsistency among multiple currently effective models for the candidate compensation coordinates of the same borehole. Taking zero at time only indicates that there is no discretization between models, not that the model error is zero; Characterizes coordinate registration, repeated clamping, and coordinate transfer and plate state changes during the detection to drilling process; It characterizes the comprehensive deviation of the drilling equipment positioning and the actual hole-forming process under the current calibration conditions.

[0169] All five items are non-negative control quantities using the same length unit in the same coordinate system, and are determined based on the same workpiece, the same coordinate registration state, and the current effective model state corresponding to the actual drilling. After re-imaging, re-registration, changing the clamping state, changing the tool state, or re-establishing the model, the error control quantities that no longer correspond to the current state become invalid and are recalculated. Linear addition is a deterministic conservative machining release rule, which does not require that the individual items be statistically independent, nor is it interpreted according to the sum of squared variances. The radial tolerance of the hole position is... When the true positional tolerance diameter ΦT is given in the design, . Output Q and drill; Sometimes Q is not output and drilling is not performed.

[0170] Regarding target placement, self-calibrating targets are set synchronously with the inner layer conductive patterns. They can be located at process edges, non-functional areas between unit boards, non-functional areas reserved on finished boards, or dedicated areas corresponding to high-precision holes. Preferably, at least four qualified targets are used for the same drilling area, with three forming a triangle covering the hole area and the fourth located within it. When model adjustment and controlled exclusion are required, a fifth or more redundant targets are preferred. Adjacent areas can share targets to form adjacent or overlapping models. When different inner layer targets may project overlapping, they are distinguished by different spacing, direction, shape, overall offset, or by separate field of view and angle imaging.

[0171] X-ray parameters include at least tube voltage, tube current, exposure or frame average, pixel size, magnification, beam direction, and dynamic range, determined according to board thickness, laminated copper thickness, mark size, and equipment curves to ensure unsaturated target and background. For mechanical drilling, the drill bit, rotation speed, feed rate, cover plate, backing plate, and number of laminates; or for laser drilling, the wavelength, pulse, focus, and scanning parameters are preset according to material and equipment specifications; calibration and trial drilling are performed under conditions consistent with or more conservatively than the actual processing conditions.

[0172] The above , , , , , , , , , The maximum number of tests is determined before testing based on equipment calibration, circuit board tolerances, and error propagation relationships. Record the equipment, pixel ratio, imaging level, calibration sample or theoretical error, calculation rules, applicable stack-ups, hole position tolerance, and effective time. Do not temporarily change the thresholds based on abnormalities in the board under test.

[0173] Limited resampling is possible when target or background data is invalid, geometry is unrecoverable, or channel conflicts occur; modeling degradation, or holes or validation targets are not in the applicable area. Or the external residuals of the model exceed Adjust the model in a timely manner and stop drilling in the corresponding area after reaching the adjustment limit; when registration verification fails, the old residuals are discarded and the entire process is redone; when diagnostic uncertainty occurs, multiple improvement candidates, mixed branches are not completed, or recursion cannot continue, diagnostic uncertainty is diagnosed; when there is no effective model or Drilling is not performed at any time. The maximum number of resampling, re-registration, model adjustment, diagnostic recursion, regional partitioning, and expanded target selection are all preset finite positive integers, accumulated for the same plate and region, and are not cleared when switching paths; if any budget is exhausted, the process terminates by disabling the target, model failure, or stopping hole / region drilling, without forming an infinite loop.

[0174] Each group can also independently obtain a candidate virtual target corresponding to the same theoretical virtual target by three or more conductive electron markers according to a preset affine centroid relationship, but each group remains physically independent. When no attenuation channel is needed, only internal geometric verification can be performed; the highly robust implementation uses dual channels and all models and verifications ensure the target passes both channels. Multiple currently valid models of the same hole can be fused using an arithmetic average or by fusion using non-negative normalized weights fixed before processing, and the results are calculated using the fused coordinates. In the preferred embodiment employing multi-model residual diagnosis and five-item error release control, regardless of whether mechanical drilling or laser drilling is used, the actual drilling command is only sent when internal verification, model validation, and diagnostic model group determination have all been completed, the relevant diagnosis and re-model external validation have also been completed when the target anomaly diagnosis is triggered, and the five error controls meet the release conditions.

[0175] In other embodiments, a high-precision drilling apparatus for circuit boards is also provided, including an X-ray imaging mechanism, a drilling mechanism, a processor, and a memory. The processor is communicatively connected to the X-ray imaging mechanism, the drilling mechanism, and the memory, respectively. The memory stores a computer program executable by the processor. When the processor executes the computer program, it controls the X-ray imaging mechanism to acquire X-ray images of a multilayer circuit board equipped with a self-calibrating composite target unit after lamination, and executes the corresponding steps implemented after circuit board lamination in the high-precision drilling method for circuit boards described in any of the foregoing embodiments. When the high-precision drilling method for circuit boards determines that the corresponding drilling execution conditions are met, the processor controls the drilling mechanism to perform drilling according to the determined compensation coordinates; when the corresponding drilling execution conditions are not met, the processor does not send the corresponding drilling command to the drilling mechanism.

Claims

1. A method for high-precision drilling of circuit boards, characterized in that, include: Multiple self-verifying composite target units are set in at least one inner conductive pattern of a multilayer circuit board. Each self-verifying composite target unit includes at least two physically independent sub-marker groups. Each sub-marker group has a preset theoretical geometric relationship relative to the same theoretical virtual bullseye and can be independently identified in an X-ray image. After laminating the multilayer circuit board, an X-ray image containing the self-calibrating composite target unit is acquired. The coordinate registration relationship between the design coordinate system, the X-ray image coordinate system and the workpiece coordinate system of the drilling equipment is established. The theoretical virtual target, the measured coordinate information of the sub-marker group and the theoretical hole position to be drilled are unified into the same coordinate system. Candidate virtual bullseyes are determined based on the measured coordinate information of each sub-marker group in the same self-verifying composite target unit. The positional consistency among the multiple candidate virtual bullseyes is used to determine whether the self-verifying composite target unit meets the preset internal verification conditions. For self-verifying composite target units that meet the preset internal verification conditions, their measured bullseyes are determined based on the candidate virtual bullseyes that pass the positional consistency judgment. For self-verifying composite target units that do not meet the preset internal verification conditions, they are not used as modeling target units or external verification target units. Multiple self-calibrating composite target units that meet the preset internal verification conditions are selected as modeling target units. A local coordinate compensation model is established based on the correspondence between the theoretical virtual bullseye and the measured bullseye of the modeling target units, and the applicable area of ​​the local coordinate compensation model is determined. The number and spatial distribution of the modeling target units enable the parameters of the local coordinate compensation model to be uniquely determined, and the theoretical hole position of the hole to be drilled is located within the applicable area of ​​the model. At least one self-verifying composite target unit that meets the preset internal verification conditions, does not participate in the parameter determination of the local coordinate compensation model, and whose theoretical virtual bullseye is located within the applicable area of ​​the model is selected as the model external verification target unit. The theoretical virtual bullseye of each of the model external verification target units is input into the local coordinate compensation model to obtain the corresponding predicted bullseye. When the positional deviation between the predicted target center and its corresponding measured target center of each of the model external verification target units meets the preset model verification conditions, the local coordinate compensation model is determined as an effective local coordinate compensation model, and the compensation coordinates of the hole to be drilled are determined and drilling is performed using it; when any of the positional deviations does not meet the preset model verification conditions, the compensation coordinates of the hole to be drilled are not output using the corresponding local coordinate compensation model, and drilling is not performed on the hole to be drilled based on the local coordinate compensation model.

2. The high-precision drilling method for circuit boards according to claim 1, characterized in that, Each of the self-calibrating composite target units includes at least three mutually distinguishable sub-marker pairs. Each sub-marker pair constitutes a sub-marker group and includes two conductive sub-markers. The theoretical centers of the two conductive sub-markers are symmetrical about the theoretical virtual target center. Different sub-marker pairs are distinguished from each other by spacing and / or arrangement direction. All conductive sub-markers in the same self-calibrating composite target unit are formed in the same inner conductive pattern.

3. The high-precision drilling method for circuit boards according to claim 2, characterized in that, The midpoint of the measured coordinates of the two conductive sub-markers in each sub-marker pair is determined as the corresponding candidate virtual bullseye; a first consistency threshold and a second anomaly threshold greater than the first consistency threshold are preset; when the distance between any two candidate virtual bullseyes is not greater than the first consistency threshold, the average coordinates of all candidate virtual bullseyes are determined as the measured bullseye; when there is one and only one abnormal candidate virtual bullseye, and the distance between any two candidate virtual bullseyes in the remaining candidate virtual bullseyes is not greater than the first consistency threshold, and the distance between the abnormal candidate virtual bullseye and each of the remaining candidate virtual bullseyes is greater than the second anomaly threshold, the abnormal candidate virtual bullseye is excluded, and the average coordinates of the remaining candidate virtual bullseyes are determined as the measured bullseye; when none of the above conditions are met, the corresponding self-verification composite target unit is determined to not meet the preset internal verification conditions.

4. The high-precision drilling method for circuit boards according to claim 3, characterized in that, The X-ray attenuation of the two conductive sub-markers in the same sub-marker pair after background correction is obtained under background matching conditions, wherein the background matching conditions are that the X-ray projection background of the two conductive sub-markers meets a preset background consistency requirement; when the sum of the two X-ray attenuation amounts is not less than a preset positive lower limit, the attenuation difference index of the corresponding sub-marker pair is determined according to the ratio of the absolute value of the difference between the two to the sum of the two; when the distance between any two candidate virtual bullseyes is not greater than the first consistency threshold and the attenuation difference index of each sub-marker pair is not greater than the preset attenuation difference threshold, or when the attenuation difference index of the sub-marker pair to which the abnormal candidate virtual bullseye belongs is greater than the preset attenuation difference threshold and the attenuation difference index of the other sub-marker pairs is not greater than the preset attenuation difference threshold, the corresponding self-verification composite target unit is determined to meet the dual-channel internal verification conditions; When the position consistency judgment and the attenuation consistency judgment are inconsistent, any of the X-ray attenuation values ​​is invalid, or the sum of the two X-ray attenuation values ​​is less than the preset positive lower limit, X-ray images are re-acquired according to the preset imaging parameter sequence and the dual-channel internal verification is re-executed; if the dual-channel internal verification condition is still not met after reaching the preset maximum number of re-acquisitions, the corresponding self-verification composite target unit is deactivated.

5. The high-precision drilling method for circuit boards according to claim 4, characterized in that, Three self-calibrating composite target units that satisfy the dual-channel internal verification conditions and whose theoretical virtual target centers are not collinear are selected as modeling target units. A local affine transformation model is established based on the three modeling target units. The theoretical hole position to be drilled and the theoretical virtual target center of at least one model external verification target unit that satisfies the dual-channel internal verification conditions, is not involved in the model parameter determination, and is different from the three modeling target units are all located within the area enclosed by the three modeling target units. The model external residual is determined based on the difference between the predicted target center of each model external verification target unit and its corresponding measured target center. The verification uncertainty control quantity is determined based on the discreteness of the candidate virtual target centers of the respective verification composite target units involved in modeling and model external verification, the calibrated coordinate identification uncertainty, and the preset local non-affine model margin. The local affine transformation model is determined as the currently effective local affine transformation model only when the magnitude of each model external residual and the verification uncertainty control quantity are not greater than the preset model verification threshold. Otherwise, the corresponding local affine transformation model is not used to output compensation coordinates.

6. The high-precision drilling method for circuit boards according to claim 5, characterized in that, Before drilling, the coordinate registration relationship is independently verified using multiple plate surface reference verification marks, which are determined by parameters not involved in the coordinate registration relationship and whose quantity and spatial distribution are sufficient to verify each degree of freedom of the coordinate registration relationship. The registration residual is determined based on the distance between the theoretical position of each plate surface reference verification mark and the measured position after transformation by the coordinate registration relationship. When all the registration residuals are not greater than the preset registration verification threshold, the independent verification is deemed qualified. When any of the registration residuals is greater than the preset registration verification threshold, the independent verification is deemed unqualified. When the independent verification is unqualified, the output of the compensation coordinates of the corresponding borehole is stopped, and X-ray images are reacquired within the preset maximum number of re-registrations. The coordinate registration, dual-channel internal verification, local affine transformation model establishment, and model external verification are re-executed, and the re-established coordinate registration relationship is independently verified again. The compensation coordinates of the corresponding borehole to be drilled will only be output after both the independent verification and the external verification have passed. If the independent verification and the external verification have not passed simultaneously after the preset maximum number of re-registrations has been reached, the drilling of the corresponding area will be stopped.

7. The high-precision drilling method for circuit boards according to claim 5, characterized in that, Multiple currently effective local affine transformation models are established for multiple borehole regions that are spatially adjacent or partially overlapping. For each of the currently effective local affine transformation models, a diagnostic model-outside verification target unit is selected from the model-outside verification target units used in the last effective model-outside verification of the model, according to a selection rule predetermined before processing and yielding a unique result under the same input data. The vector pointing from the predicted target center of the diagnostic model-outside verification target unit to its actual target center is determined as the model-outside residual vector of the corresponding model. A residual lower limit and a direction angle threshold are preset. The residual lower limit is greater than zero and not greater than the preset model verification threshold, and the direction angle threshold is greater than or equal to zero and less than 90 degrees. When the magnitudes of the model-outside residual vectors of at least two of the currently effective local affine transformation models reach the residual lower limit, and the angle between any two of the model-outside residual vectors is not greater than the direction angle threshold, the at least two currently effective local affine transformation models are determined as a diagnostic model group, and the output of compensation coordinates using the diagnostic model group is suspended.

8. The high-precision drilling method for circuit boards according to claim 7, characterized in that, The union of the modeling target units and diagnostic model-external validation target units used by each local affine transformation model in the diagnostic model group is determined as the candidate exclusion target set, and the original model state of the diagnostic model group is saved before performing the single target exclusion test. A single target exclusion test is independently performed on each self-verifying composite target unit in the candidate exclusion target set. Each single target exclusion test starts from the original model state and excludes only one self-verifying composite target unit. The local affine transformation model in the original model state that uses the excluded self-verifying composite target unit as the modeling target unit or the diagnostic model-external validation target unit is determined as the relevant model. For each relevant model, a selection from a preset replacement pool is used... The modeling target unit and diagnostic model external verification target unit not included in the original model state of the relevant model, and the alternative target unit that satisfies the dual-channel internal verification conditions and the corresponding role space conditions, are selected to re-establish and verify the relevant model. Specifically, when the alternative target unit is used to replace the modeling target unit, the corresponding role space conditions include that the parameters of the replaced local affine transformation model can be uniquely determined, and that the theoretical hole location to be drilled and the theoretical virtual bullseye of the target unit used for external verification are both located within the applicable area of ​​the replaced model. When the alternative target unit is used to replace the diagnostic model external verification target unit, the corresponding role space conditions include that the alternative target unit did not participate in the determination of model parameters and that its theoretical virtual bullseye... The bullseye is located within the applicable area of ​​the model; the external residual vectors of unrelated models and their original models remain unchanged, and the target replacement results, model parameters, and model states formed in the previous single-target exclusion test are not carried over to the subsequent single-target exclusion test; the residual persistence condition is determined when the magnitude of the external residual vectors of each model in the diagnostic model group after single-target exclusion reaches the lower limit of the residual and the angle between any two external residual vectors is not greater than the direction angle threshold; a recovery threshold and a minimum relative improvement ratio are preset, the recovery threshold is greater than zero and less than the lower limit of the residual; when the self-validating composite target unit excluded this time participates in at least two related models in the diagnostic model group, the magnitude of the new external residual vector of each related model is not greater than the recovery threshold. When the relative reduction ratio of the magnitude of the out-of-model residual vector before exclusion is not less than the minimum relative improvement ratio, the corresponding single-target exclusion test is determined to meet the preset improvement condition. When all single-target exclusion tests can be completed and all meet the residual persistence condition, the union of the applicable regions of each local affine transformation model in the diagnostic model group is determined as the suspected local deformation region. When exactly one single-target exclusion test meets the preset improvement condition, the self-verifying composite target unit excluded in this test participates in all local affine transformation models in the diagnostic model group, and all other single-target exclusion tests meet the residual persistence condition, the excluded self-verifying composite target unit is determined as a suspected abnormal target unit.If any single-target exclusion test cannot be completed, or if the results of all single-target exclusion tests do not fall under any of the above situations, the use of the output compensation coordinates of the diagnostic model group should remain suspended; the output compensation coordinates should not be used until the local affine transformation model, which needs to be reconstructed or re-verified due to the single-target exclusion test, has passed out-of-model verification again.

9. The high-precision drilling method for circuit boards according to claim 5, characterized in that, For one or more local regions covering the same borehole, one or more currently valid local affine transformation models verified outside the model are established. Each of these models is used to determine corresponding candidate compensation coordinates. These candidate coordinates are then fused using an arithmetic mean or a pre-determined non-negative normalized weight to determine the final compensation coordinates. The internal consistency error control value for the target is determined based on the discreteness of the candidate virtual bullseye of the self-verifying composite target unit participating in the modeling and the calibrated coordinate identification uncertainty. The out-of-model verification error control value is determined based on the out-of-model residual obtained from the last valid out-of-model verification of the currently valid local affine transformation model, the discreteness of the candidate virtual bullseye of the out-of-model verification target unit, the coordinate identification uncertainty, and the unobserved margin in space determined by the preset upper limit of the residual space change rate control and the distance between the verification position and the borehole position. Finally, the candidate compensation coordinate discrete error control value is determined based on the maximum distance between each candidate compensation coordinate and the final compensation coordinate. When there is only one candidate compensation coordinate, the candidate compensation coordinate discrete error control value is zero. Using multiple parameters not involved in the coordinate registration relationship, and whose quantity and spatial distribution are sufficient to verify each degree of freedom of the coordinate registration relationship, the coordinate registration relationship is independently registered and verified. The coordinate registration error control value is determined based on the maximum distance between the theoretical position of each of the plate surface reference verification marks and the measured position after transformation by the coordinate registration relationship, the registration uncertainty, and the margin for repeated clamping and coordinate transfer state changes. The drilling equipment positioning error control value is determined based on the maximum deviation between the command hole center and the measured hole center of the calibration test hole, and the hole center measurement uncertainty. The target internal consistency error control value, model external verification error control value, candidate compensation coordinate discrete error control value, coordinate registration error control value, and drilling equipment positioning error control value are linearly added together to obtain the hole position comprehensive error control value. Each error control value is a non-negative quantity using the same length unit. When the comprehensive error control value of the hole position is not greater than the preset hole position tolerance of the hole to be drilled, drilling is performed according to the final compensation coordinates; otherwise, drilling is not performed on the hole to be drilled.

10. A high-precision drilling device for circuit boards, characterized in that, The method includes an X-ray imaging mechanism, a drilling mechanism, a processor, and a memory. The processor is communicatively connected to the X-ray imaging mechanism, the drilling mechanism, and the memory. The memory stores a computer program that can be executed by the processor. When the processor executes the computer program, it controls the X-ray imaging mechanism and the drilling mechanism to perform the steps in the high-precision drilling method for circuit boards according to any one of claims 1 to 9 after the multilayer circuit board is laminated, on a multilayer circuit board equipped with a self-calibrating composite target unit.

Images