Industrial ct calibration method, device, system and medium based on regularized tolerance band

Abstract

The application discloses an industrial CT calibration method and device based on a regularization tolerance band, a system and a medium. The industrial CT calibration method based on the regularization tolerance band comprises the following steps: performing a ball center pixel observation on a to-be-detected object under multiple rotation angle postures to obtain a geometric parameter set; constructing a re-projection residual error, a specification residual error and a tolerance band regularization residual error based on each effective observation value in the geometric parameter set; concatenating all the re-projection residual errors, the specification residual error and the tolerance band regularization residual error to obtain a total residual error vector; and performing a minimization solving on the total residual error vector to obtain a decoupled geometric parameter and a three-dimensional structure point. The tolerance band regularization residual error is determined by a nominal prior value, a tolerance parameter and an effective observation value. The application can effectively improve the geometric fidelity of measurement.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of industrial CT calibration technology, and in particular to an industrial CT calibration method, apparatus, system, and medium based on regularization tolerance bands. Background Technology

[0002] Industrial CT is widely used in nondestructive testing, defect analysis, and dimensional metrology. The accuracy of its 3D reconstruction and measurement results highly depends on the precise calibration of projection geometric parameters. Geometric calibration based on a standard spherical phantom is usually modeled as a problem of minimizing reprojection residuals: given pixel observations of the sphere center at each rotation angle, the geometric parameter vector and the 3D structural point set are jointly estimated to minimize the reprojection error output by the projection operator. However, in engineering practice, it often exhibits ill-posed and ill-conditioned characteristics. For example, the Gram matrix of its normal equation is approximately rank-deficient along several directions, and there exists a near-zero space. As a result, the estimated values ​​of individual geometric parameters deviate significantly from their true physical values. If the estimated values ​​of such parameters are directly used, it is easy to mistake the fitting compensation for the true geometric deviation, thereby affecting the reliability of dimensional measurement and defect location. Summary of the Invention

[0003] This application provides an industrial CT calibration method, apparatus, system, and medium based on regularization tolerance bands, which can effectively improve the geometric fidelity of measurements.

[0004] In a first aspect, embodiments of this application provide an industrial CT calibration method based on regularized tolerance bands, including: The geometric parameter set is obtained by observing the center pixel of the object under multiple rotational poses. Based on each valid observation in the set of geometric parameters, construct reprojection residuals, normalized residuals, and tolerance band normalized residuals; The total residual vector is obtained by concatenating all the reprojection residuals, the normalized residuals, and the tolerance band regularized residuals. Minimize the total residual vector to obtain the geometric parameters and three-dimensional structural points of the decoupled object to be detected; The tolerance band regularized residual is determined by the nominal prior value, the tolerance parameter, and the effective observation value.

[0005] The industrial CT calibration method based on regularized tolerance band according to the first aspect of this application has at least the following beneficial effects: by introducing the regularized residual of the tolerance band to constrain the geometric parameters, without introducing a systematic bias into the identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, as well as the reproducibility across initial values ​​and data subsets, improving the stability of geometric parameter estimation, and effectively improving the geometric fidelity of measurement.

[0006] According to some embodiments of the first aspect of this application, the tolerance band regularized residual is determined by the following formula:

[0007] in, The number of valid observations in the set of geometric parameters. The tolerance parameter is... The nominal prior value, The valid observation value , The tolerance band canonical residual, This is the strength coefficient.

[0008] According to some embodiments of the first aspect of this application, the step of minimizing the total residual vector to obtain the decoupled geometric parameters and three-dimensional structure points includes: The total residual vector is minimized using a damped Gauss-Newton algorithm or a trust region reflection algorithm to obtain the decoupled geometric parameters and three-dimensional structural points.

[0009] According to some embodiments of the first aspect of this application, the intensity coefficient has the same pixel dimension as the reprojection residual.

[0010] According to some embodiments of the first aspect of this application, the tolerance parameter is determined by the mechanical manufacturing or assembly tolerance of the effective observation.

[0011] According to some embodiments of the first aspect of this application, the tolerance band regularized residual is determined by the following formula:

[0012] in, For the Softplus function, The number of valid observations in the set of geometric parameters. The strength coefficient, The tolerance parameter is... The nominal prior value, For the valid observations, The tolerance band canonical residual, This is the sharpness coefficient.

[0013] According to some embodiments of the first aspect of this application, the step of minimizing the total residual vector to obtain the decoupled geometric parameters and three-dimensional structural points of the object to be detected includes: The total residual vector is minimized using a damped Gauss-Newton algorithm or a trust region reflection algorithm to obtain the decoupled geometric parameters and three-dimensional structural points of the object to be detected.

[0014] According to some embodiments of the first aspect of this application, the set of geometric parameters includes at least one of the following parameters: source-rotation center distance, source-detector distance, focal lateral offset value, focal longitudinal offset value, detector principal point lateral offset value, detector principal point longitudinal offset value, in-plane tilt angle, out-of-plane tilt angle, and turntable angle step value.

[0015] Secondly, embodiments of this application provide an operation control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the industrial CT calibration method based on regularization tolerance band as described in the first aspect.

[0016] The operation control device according to the second aspect of this application has at least the following beneficial effects: by introducing tolerance band regularized residuals to constrain geometric parameters, without introducing systematic biases into identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, as well as the reproducibility across initial values ​​and data subsets, improving the stability of geometric parameter estimation, and effectively improving the geometric fidelity of measurement.

[0017] Thirdly, embodiments of this application provide an industrial CT system, which is used to implement the industrial CT calibration method based on regularization tolerance bands described in the first aspect. The industrial CT system includes: The observation and acquisition module is used to obtain a set of geometric parameters by observing the center pixel of the object under multiple rotational poses. The residual assembly module is used to construct reprojection residuals, normalized residuals, and tolerance band regularized residuals based on each valid observation value in the geometric parameter set, and to concatenate all the reprojection residuals, normalized residuals, and tolerance band regularized residuals to obtain the total residual vector. The optimization solution module is used to minimize the total residual vector to obtain the decoupled geometric parameters and three-dimensional structural points.

[0018] The industrial CT system according to the third aspect of this application has at least the following beneficial effects: by introducing tolerance band regularized residuals to constrain geometric parameters, without introducing systematic biases into identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, as well as the reproducibility across initial values ​​and data subsets, improving the stability of geometric parameter estimation, and effectively improving the geometric fidelity of measurement.

[0019] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the industrial CT calibration method based on regularization tolerance bands as described in the first aspect embodiment above.

[0020] The computer-readable storage medium according to the fourth aspect of this application has at least the following beneficial effects: by introducing tolerance band regularized residuals to constrain geometric parameters, without introducing systematic biases into identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, as well as the reproducibility across initial values ​​and data subsets, improving the stability of geometric parameter estimation, and effectively improving the geometric fidelity of measurement.

[0021] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0022] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0023] Figure 1 A detailed flowchart of the industrial CT calibration method based on regularization tolerance band provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the operation control device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of an industrial CT system provided in an embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] It is understandable that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0026] Industrial CT is widely used in nondestructive testing, defect analysis, and dimensional metrology. The accuracy of its 3D reconstruction and measurement results highly depends on the precise calibration of projection geometric parameters. Geometric calibration based on a standard spherical phantom is usually modeled as a problem of minimizing reprojection residuals: given pixel observations of the sphere's center at each rotation angle, the geometric parameter vector and the 3D structural point set are jointly estimated to minimize the reprojection error output by the projection operator. However, in engineering practice, it often exhibits ill-posed and ill-conditioned characteristics. For example, the Gram matrix of its normal equation is approximately rank-deficient along several directions, and there exists a near-zero space. As a result, the estimated values ​​of individual geometric parameters deviate significantly from their true physical values. If the estimated values ​​of such parameters are directly used, the fitting compensation error is easily mistaken for the true geometric deviation, thus affecting the reliability of dimensional measurement and defect location.

[0027] Based on this, this application provides an industrial CT calibration method, device, system, and medium based on a regularized tolerance band. By introducing a regularized residual of the tolerance band to constrain geometric parameters, without introducing a systematic bias into the identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, as well as the reproducibility across initial values ​​and data subsets, improving the stability of geometric parameter estimation, and effectively improving the geometric fidelity of the measurement.

[0028] Firstly, referring to Figure 1 , Figure 1 A detailed flowchart of the industrial CT calibration method based on regularization tolerance band provided in the embodiments of this application includes, but is not limited to, the following steps: Step S100: Perform sphere center pixel observation on the object to be detected under multiple rotation angle poses to obtain a set of geometric parameters; Step S200: Construct reprojection residuals, normalized residuals, and tolerance band normalized residuals based on each valid observation in the geometric parameter set; Step S300: Concatenate all reprojection residuals, normalized residuals, and tolerance band regularized residuals to obtain the total residual vector; Step S400: Minimize the total residual vector to obtain the geometric parameters and three-dimensional structural points of the decoupled object to be detected.

[0029] Understandably, an industrial CT system includes an observation and acquisition module, which comprises a rotary stage, an X-ray emitter, and a detector. The object to be inspected is a structure containing a standard sphere. The process of observing the sphere's center pixel under multiple rotational orientations can be described as follows: the object is placed on the rotary stage; the X-ray emitter emits X-rays that penetrate the object; and the detector receives the penetrated X-rays to determine the geometric parameter set of the object. This geometric parameter set includes at least one parameter among the following: source-rotation center distance, source-detector distance, focal lateral offset, focal longitudinal offset, detector principal point lateral offset, detector principal point longitudinal offset, in-plane tilt angle, out-of-plane tilt angle, and rotary stage angle step. During sphere center pixel observation, reliable observation points with sphere center pixel coordinates may not be obtained due to image blurring, extraction failure, or exceeding the detector's field of view. Therefore, after acquiring the geometric parameter set, valid observation values ​​can be determined from it. Each valid observation value includes the observed pixel coordinates and the corresponding rotary stage rotation angle. The number of valid observations is determined by the number of rotation angles and the number of standard spheres in the object being tested. For example, if 12 spheres are used to make observations at 720 angles, the number of valid observations is 8640.

[0030] Specifically, by acquiring projected images at different angles (such as uniformly distributed 360°), image processing algorithms (such as Hough transform or centroid method) are used to extract the observed pixel coordinates of the sphere's center on the detector, while the corresponding rotation angle of the rotating stage is recorded to obtain the geometric parameter set.

[0031] It should be noted that after determining all valid observations in the geometric parameter set, reprojection residuals, normed residuals, and tolerance band regularized residuals can be constructed based on these valid observations. The reprojection residual is the difference between the theoretical projected pixel coordinates obtained from the projection of the current geometric parameters and 3D structure points at each rotation stage angle and the observed pixel coordinates. The normed residual is used to eliminate the baseline deficit in free network adjustment. The normed residual can be a linear combination of all observed pixel coordinates to ensure that the objective function has a unique solution in the normed degrees of freedom direction. The tolerance band regularized residual is determined by the nominal prior value, tolerance parameter, and valid observations. The nominal prior value provides the initial point or design reference point for the geometric parameters, the tolerance parameter defines the allowable physical deviation range of the parameters, and the valid observations provide the actual measurement data. The tolerance band regularized residual compares the estimated values ​​of the geometric parameters with the nominal prior value and combines the tolerance parameter to determine whether a penalty needs to be applied, thereby ensuring that the final geometric parameter estimates have physical meaning and reliability in engineering practice.

[0032] Specifically, the tolerance band canonical residual is determined by the following formula:

[0033] in, The number of valid observations in the geometric parameter set. The strength coefficient, For tolerance parameters, The nominal prior value, For valid observations, For tolerance zone regularized residuals, slope operator Furthermore, the tolerance parameters corresponding to different valid observations are also different. The tolerance parameters can be set according to the mechanical tolerance configuration or assembly tolerance to ensure the accuracy of the tolerance band setting. For example, the tolerance parameter for lateral offset can be 0.5mm, the tolerance parameter for tilt angle can be 0.2°, and the tolerance parameter for angle step can be 0.001°.

[0034] Specifically, if the valid observations satisfy This indicates that no data offset has occurred; if the valid observations satisfy... This indicates a data offset, meaning the valid observations are outside the tolerance band. Therefore, if the valid observations fall within the tolerance band, the tolerance band regularization residual can be output as zero, in which case the valid observations will not introduce any bias. If the valid observations are outside the tolerance band, curvature is selectively superimposed in the relaxation direction of the likelihood surface, thereby suppressing parameter drift in a targeted manner without sacrificing the accuracy of identifiable parameter estimation. By maintaining zero bias within the tolerance band and selectively suppressing drift outside the band, the compensation problem caused by parameter unidentifiability is solved while preserving data-driven accuracy, improving the stability of geometric parameter estimation and effectively enhancing the geometric fidelity of the measurement.

[0035] It should be noted that, according to design specifications and quality control standards, industrial CT systems are allowed a tolerance of ±0.1 mm. Based on this, the tolerance parameter for the source-detector distance can be set to 0.1 mm. During calibration, the effective observed value of the source-detector distance is obtained through observation of the sphere's center pixel, and then compared with the nominal prior value. If the deviation between the two is within 0.1 mm, the tolerance band normalized residual is 0. If the deviation exceeds 0.1 mm, the tolerance band normalized residual of this parameter will be calculated according to the above formula, imposing a corresponding penalty on deviations exceeding the tolerance range, improving the physical interpretability and robustness of the calibration results, thereby making the industrial CT calibration results more accurate and reliable.

[0036] Specifically, the industrial CT system is used for non-destructive testing of battery cells. The observation and acquisition module of the industrial CT system includes a rotary table, an X-ray emitter, a detector, and a robotic arm for moving the battery cells to the rotary table. During the assembly of the industrial CT system, in order to avoid parts jamming or being damaged due to excessive tightness, an assembly tolerance of ±0.1 mm is allowed between the rotary table and the X-ray emitter, and an assembly tolerance of ±0.1 mm is also allowed between the center of the rotary table and the detector. Based on this, the tolerance parameter for the source-detector distance can be set to 0.1 mm, and the tolerance parameter for the source-rotation center distance can be set to 0.1 mm.

[0037] Furthermore, since the basic unit of the reprojection residual is a pixel, and the root mean square of the reprojection is a core indicator for measuring calibration accuracy, setting the intensity coefficient to have the same pixel dimension as the reprojection residual ensures that the regularization intensity of the tolerance band regularized residual is equivalent to the root mean square of the reprojection. By setting the same dimension, it is possible to effectively avoid overemphasizing or ignoring a certain residual term during optimization due to dimension mismatch, thus enabling the regularization constraint to act more accurately on the decoupling and correction of geometric parameters, improving the robustness and convergence consistency of the calibration algorithm.

[0038] It should be noted that the tolerance band canonical residual can also be determined by the following formula:

[0039] in, Sharpness coefficient The Softplus function is used to avoid discontinuities in the first derivative at inflection points and to improve the convergence of second-order optimization. In some embodiments, β can be between 10 and 100 to avoid the tolerance band boundary becoming blurred due to an excessively small sharpness coefficient and the function becoming too close to the original ramp function due to an excessively large sharpness coefficient. Furthermore, due to the continuous differentiability of the total residual vector, various gradient-based optimization algorithms can be effectively utilized when minimizing the total residual vector, thereby improving the accuracy and reliability of the calibration results.

[0040] It is understandable that reprojection residuals can only constrain projection consistency, not the physical rationality of parameters—when there is strong coupling between parameters, different parameter combinations can produce almost the same reprojection error. Normalized residuals address the mathematical solvability of the geometric baseline and are unrelated to control parameter drift. Tolerance band regularized residuals address parameter compensation problems, not interfering with data fitting within the band and selectively suppressing drift outside the band. Therefore, after constructing reprojection residuals, normalized residuals, and tolerance band regularized residuals based on each valid observation in the geometric parameter set, all reprojection residuals, normalized residuals, and tolerance band regularized residuals can be concatenated to obtain a total residual vector. This allows for simultaneous perception of the gradients of all residuals with respect to all parameters during subsequent optimization, thus balancing data fitting, geometric solvability, and parameter physical rationality in each iteration. The total residual vector is determined by the following formula:

[0041] in, The reprojection residual term is obtained by summing all the reprojection residuals. To standardize residuals, This is the tolerance band regularized residual term obtained by summing all the tolerance band regularized residuals.

[0042] Understandably, after calculating the total residual vector, a damped Gauss-Newton algorithm or a trust-region reflection algorithm can be used to minimize the total residual vector. By iteratively adjusting the values ​​of geometric parameters and 3D structural points, the norm of the total residual vector is gradually reduced until the convergence condition is met. During this process, due to the existence of tolerance-band canonical residuals, the optimization algorithm is guided to search for the optimal solution within a physically reasonable parameter space, thus effectively avoiding significant deviations of parameter estimates from the true physical values. Therefore, various geometric parameters in an industrial CT system can be accurately determined, and the precise 3D spatial coordinates of a standard spherical phantom can be obtained simultaneously, thereby achieving effective decoupling between geometric parameters and 3D structural points.

[0043] Understandably, calibration methods in related technologies, when processing the geometric parameters of industrial CT systems, may suffer from an approximate rank deficiency in the Gram matrix of the normal equations due to the complexity of the system's internal structure. This can lead to significant fluctuations in the estimated values ​​of certain geometric parameters during optimization, even deviating from their true physical values ​​and affecting the final measurement accuracy. For example, when minimizing the total residual vector, industrial CT calibration methods in related technologies often suffer from an approximate rank deficiency in the Gram matrix of the normal equations along several directions, resulting in near-zero space and causing the estimated values ​​of individual geometric parameters to deviate significantly from their true physical values. In some embodiments, a standard ceramic sphere phantom of known diameter is first placed on the rotary table of the industrial CT system. During operation of the industrial CT system, the standard ceramic sphere phantom is penetrated by X-rays at multiple rotational angles, and the detector captures a series of projected images. For example, a projected image is acquired at each rotation angle from 0 degrees, 1 degree, 2 degrees up to 359 degrees. Subsequently, image processing is performed on each projected image, for example, by using edge detection and circle fitting algorithms to accurately identify the observed pixel coordinates of the standard sphere phantom on the detector plane, while simultaneously recording the corresponding rotation angle of the rotary table. These pixel coordinates constitute the effective observations used for calibration, thus obtaining the initial set of geometric parameters. Then, based on these effective observations, reprojection residuals, normalized residuals, and tolerance band regularized residuals are constructed. Specifically, for the reprojection residuals, the theoretical projected pixel coordinates are calculated based on the currently estimated geometric parameters (e.g., source-detector distance, focal lateral offset, etc.) and the three-dimensional position of the standard phantom, and compared with the actually observed pixel coordinates to obtain the pixel-level deviation. For the normalized residuals, an L2 norm penalty can be applied to certain geometric parameters (e.g., detector principal point offset) to prevent them from becoming excessive during optimization. For the tolerance band regularized residuals, for example, the nominal prior value of the source-detector distance is 1000 mm, and its mechanical manufacturing tolerance (tolerance parameter) is ±2 mm. When constructing the residuals, if the currently estimated source-detector distance deviates from 1000 mm by more than 2 mm, the tolerance band regularized residual formula will be used to calculate and apply a corresponding penalty to deviations exceeding the tolerance range. Subsequently, all reprojection residuals, normalized residuals, and tolerance band regularized residuals are concatenated to form a total residual vector. This total residual vector comprehensively reflects the inconsistencies between observed data and model predictions, the smoothness requirements of parameters, and the physical constraints on parameter rationality. Finally, the total residual vector is minimized. For example, a damped Gauss-Newton algorithm or a trust region reflection algorithm can be used to iteratively adjust geometric parameters (such as source-detector distance, focal offset, detector tilt angle, etc.) and the three-dimensional spatial coordinates of the standard phantom to minimize the norm of the total residual vector.During the optimization process, when the optimization algorithm attempts to estimate the source-detector distance to a physically unreasonable value (e.g., 900 mm or 1100 mm), the tolerance band regularized residual incurs a significant penalty to suppress parameter deviation and ensure the optimal solution is found within a physical tolerance range of 1000 mm ± 2 mm. This yields the precise decoupled geometric parameters and the three-dimensional structural points of the standard spherical phantom. Compared to calibration methods in related technologies, when faced with complex geometric parameter estimation, the inherent mathematical ill-conditioning often leads to instability in the optimization process, resulting in estimated geometric parameters that may deviate significantly from the actual physical values, or even produce non-physical solutions. In the above embodiment, without the constraint of the tolerance band regularized residual, the optimization algorithm might estimate the source-detector distance to a value far exceeding its physical tolerance range, such as 950 mm or 1050 mm, rendering the calibration results useless for practical application. By introducing tolerance band regularized residuals to constrain geometric parameters, without introducing systematic biases into identifiable parameters, parameter compensation and drift caused by actual unidentifiability are selectively suppressed, thereby improving the physical interpretability and stability of geometric parameter estimation and thus improving the geometric fidelity of the measurement.

[0044] It should be noted that during the process of solving for the total residual vector, a Jacobian matrix can also be constructed. By calculating the parametric covariance matrix, the correlation matrix and condition number can be obtained. Specifically, the regularization activation state of the regularized residuals for each tolerance band can be checked. If the valid observations satisfy... If the effective observation is marked as a weakly identifiable direction, it indicates that its calibration result is strongly constrained by prior knowledge and should be compared with the design value; if the effective observation satisfies This indicates that the effective observation is data-driven and the estimate has high reliability. By constructing the Jacobian matrix, we can distinguish whether the tolerance band regularized residual is the true geometric bias or a result of fitting compensation, providing a reference for subsequent parameter processing.

[0045] Understandably, in the process of CT reconstruction projection, the geometric parameters of the object to be inspected are first calibrated, then the actual workpiece is scanned, and finally, an image reconstruction algorithm is used. During image reconstruction, geometric parameters obtained from the optimization solution (including source-rotation center distance, source-detector distance, focal offset, detector principal point offset, in-plane / out-of-plane tilt angle, etc.) and three-dimensional structural points are used. Specifically, in the back-projection reconstruction formula, these parameters are used to establish rotation and translation matrices between the reconstruction coordinate system and the X-ray source-detector coordinate system, thereby transforming the reconstruction points from the reconstruction coordinate system to the detector coordinate system. Then, their projection positions on the detector plane are calculated, and finally, the back-projection assignment is completed. The accuracy of the back-projection directly depends on the precision of the geometric parameters. If the parameters are biased, the reconstructed image will produce geometric artifacts and reduce spatial resolution. By introducing a tolerance band regularized residual to constrain the geometric parameters, parameter compensation is suppressed while preserving data-driven accuracy, resulting in higher physical fidelity and reproducibility of the calibration results, thereby improving the quality of the reconstructed image.

[0046] Specifically, after the battery cell is assembled, if lithium plating occurs inside the cell or there are holes in the electrode, it is difficult to judge from the outside of the cell. In this case, an industrial CT system can be used to perform non-destructive testing on the battery cell. During the testing process, the object to be tested can be placed on a rotating stage, and then the rotating stage can be started to rotate. The object to be tested is penetrated by X-rays at multiple rotation angles. The detector captures a series of projection images. Each projection image is processed to obtain the observed pixel coordinates of the object to be tested on the detector plane. At the same time, the corresponding rotation angle of the rotating stage is recorded to determine the geometric parameter set of the object to be tested. Then, the geometric parameters (including source-rotation center distance, source-detector distance, focal offset value, detector principal point offset value, in-plane / out-of-plane tilt angle, etc.) and three-dimensional structural points are obtained by optimizing the geometric parameter set. The battery cell is then placed on a rotating stage, which is then started to rotate. The detector acquires the cell's parameters, and image reconstruction is performed based on these parameters using a back-projection algorithm. The back-projection reconstruction formula requires establishing rotation and translation matrices between the reconstruction coordinate system and the X-ray source-detector coordinate system using geometric parameters and 3D structural points. This transforms the reconstructed points of the battery cell from the reconstruction coordinate system to the detector coordinate system. The projected position on the detector plane is then calculated, and the back-projection assignment is completed. The accuracy of the back-projection directly depends on the precision of the geometric parameters. If the geometric parameters deviate, the reconstructed image will produce geometric artifacts and reduce spatial resolution, leading to misjudgments or omissions of minute defects (such as electrode pores, lithium plating, and interlayer anomalies). By introducing a tolerance band regularized residual to constrain the geometric parameters, parameter compensation is suppressed while preserving data-driven accuracy. This results in higher physical fidelity and reproducibility of the calibration results, thereby improving the quality of the reconstructed image and ultimately enhancing the accuracy of battery cell detection.

[0047] Specifically, during the production of multilayer circuit boards, due to reasons such as production equipment failure and mechanical fatigue, problems such as open circuits, short circuits, or missing solder joints may occur on the circuit boards. However, these defects are difficult to distinguish with the naked eye. In this case, an industrial CT system can be used to perform non-destructive testing on the multilayer circuit boards. During the testing process, the object to be tested can be placed on a rotary table, and then the rotary table can be started to rotate. The object to be tested is penetrated by X-rays at multiple rotation angles. The detector captures a series of projection images. Each projection image is processed to obtain the observed pixel coordinates of the object to be tested on the detector plane. At the same time, the corresponding rotation angle of the rotary table is recorded to determine the geometric parameter set of the object to be tested. Then, the geometric parameters (including source-rotation center distance, source-detector distance, focal offset value, detector principal point offset value, in-plane / out-of-plane tilt angle, etc.) and three-dimensional structural points are obtained by optimizing the geometric parameter set. The multilayer circuit board is then placed on a rotating platform, which is then rotated. The detector acquires the circuit board parameters, and image reconstruction is performed based on these parameters using a back-projection algorithm. The back-projection reconstruction formula requires establishing rotation and translation matrices between the reconstruction coordinate system and the X-ray source-detector coordinate system using geometric parameters and 3D structural points. This transforms the reconstructed points of the multilayer circuit board from the reconstruction coordinate system to the detector coordinate system. The projected position on the detector plane is then calculated, and the back-projection assignment is completed. The accuracy of the back-projection directly depends on the precision of the geometric parameters. If the geometric parameters deviate, the reconstructed image will produce geometric artifacts and reduce spatial resolution, leading to misjudgments or omissions of minor defects (such as open circuits, short circuits, voids, and missing solder joints). By introducing a tolerance band regularized residual to constrain the geometric parameters, parameter compensation is suppressed while preserving data-driven accuracy. This results in higher physical fidelity and reproducibility of the calibration results, thereby improving the quality of the reconstructed image and ultimately enhancing the accuracy of multilayer circuit board detection.

[0048] Specifically, liquid cooling plates are key components of thermal management systems in new energy vehicles and data centers. Their internal flow channels are complex, with numerous welding interfaces. Problems such as flow channel blockage, welding defects, and uneven wall thickness can lead to poor performance or reduced lifespan in actual use. Therefore, industrial CT systems can be used for non-destructive testing of liquid cooling plates. During the testing process, the object to be inspected is placed on a rotating stage, which is then rotated. The object is penetrated by X-rays at multiple rotational angles, and the detector captures a series of projected images. Each projected image is processed to obtain the observed pixel coordinates of the object on the detector plane, while the corresponding rotation angle is recorded. This determines the geometric parameter set of the object. The geometric parameters (including source-rotation center distance, source-detector distance, focal offset, detector principal point offset, in-plane / out-of-plane tilt angle, etc.) and three-dimensional structural points are then obtained through optimization of the geometric parameter set. The liquid-cooled plate is then placed on a rotating stage, which is then started to rotate. The detector acquires the plate's parameters, and image reconstruction is performed based on these parameters using a back-projection algorithm. The back-projection reconstruction formula requires establishing rotation and translation matrices between the reconstruction coordinate system and the X-ray source-detector coordinate system using geometric parameters and 3D structural points. This transforms the reconstructed points of the liquid-cooled plate from the reconstruction coordinate system to the detector coordinate system. The projected position on the detector plane is then calculated, and the back-projection assignment is completed. The accuracy of the back-projection directly depends on the precision of the geometric parameters. If the geometric parameters deviate, the reconstructed image will produce geometric artifacts and reduce spatial resolution, leading to misjudgments or omissions of minor defects (such as internal flow channel blockage, welding defects, and uneven wall thickness). By introducing a tolerance band regularized residual to constrain the geometric parameters, parameter compensation is suppressed while preserving data-driven accuracy. This results in higher physical fidelity and reproducibility of the calibration results, thereby improving the quality of the reconstructed image and ultimately enhancing the accuracy of plate parameter detection.

[0049] Secondly, referring to Figure 2 This application provides an operation control device 200, including a memory 210, a processor 220, and a computer program stored in the memory 210 and executable on the processor 220. The processor 220 executes the program to implement the industrial CT calibration method based on regularization tolerance band as described in the first aspect embodiment above, for example, executing... Figure 1 Method steps S100 to S400.

[0050] The memory 210, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs, such as the industrial CT calibration method based on regularization tolerance band in the above embodiments of this application. The processor 220 implements the industrial CT calibration method based on regularization tolerance band in the above embodiments of this application by running the non-transitory software program and instructions stored in the memory 210.

[0051] The memory 210 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function. The data storage area may store data required for executing the industrial CT calibration method based on regularization tolerance bands as described in the above embodiments. Furthermore, the memory 210 may include high-speed random access memory 210, and may also include non-transitory memory 210, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. It should be noted that the memory 210 may optionally include memory 210 remotely located relative to the processor 220, and these remote memories 210 can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0052] Thirdly, referring to Figure 3 This application provides an industrial CT system, which includes an observation acquisition module, a residual assembly module, and an optimization solution module. The observation acquisition module performs sphere-centered pixel observations of the object under inspection at multiple rotational poses to obtain a set of geometric parameters. The residual assembly module constructs reprojection residuals, normalized residuals, and tolerance-band regularized residuals based on the effective observations in the geometric parameter set, and concatenates all reprojection residuals, normalized residuals, and tolerance-band regularized residuals to obtain a total residual vector. The optimization solution module minimizes the total residual vector to obtain the decoupled geometric parameters and three-dimensional structural points. By introducing tolerance-band regularized residuals to constrain the geometric parameters, without introducing systematic biases into identifiable parameters, it selectively suppresses parameter compensation and drift caused by actual indiscernibility, thereby improving the physical interpretability of geometric parameter estimation, reproducibility across initial values ​​and data subsets, and stability of geometric parameter estimation, effectively improving the geometric fidelity of the measurement.

[0053] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. These instructions are used to cause a computer to execute the industrial CT calibration method based on regularized tolerance bands as described in the first aspect embodiment above, for example, executing... Figure 1Method steps S100 to S400 are described above. Those skilled in the art will understand that all or some of the steps in the methods disclosed above, and the system, can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all physical components can be implemented as processors, such as central processing units, digital signal processors, or microprocessors executing software, or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs). Such software can be distributed on a computer-readable medium, which can include computer storage media or non-transitory media and communication media or transient media. As is known to those skilled in the art, computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc DVD or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information transmission medium.

[0054] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0055] The above is a detailed description of the preferred embodiments of this application. However, this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. An industrial CT calibration method based on regularized tolerance bands, characterized in that, include: The geometric parameter set is obtained by observing the center pixel of the object under multiple rotational poses. Based on each valid observation in the set of geometric parameters, construct reprojection residuals, normalized residuals, and tolerance band normalized residuals; The total residual vector is obtained by concatenating all the reprojection residuals, the normalized residuals, and the tolerance band regularized residuals. The total residual vector is minimized to obtain the geometric parameters and three-dimensional structural points of the decoupled object to be detected. The tolerance band regularized residual is determined by the nominal prior value, the tolerance parameter, and the effective observation value.

2. The industrial CT calibration method based on regularized tolerance bands according to claim 1, characterized in that, The tolerance band regularized residual is determined by the following formula: in, The number of valid observations in the set of geometric parameters. The tolerance parameter is... The nominal prior value, For the valid observations, , The tolerance band regularized residual, This is the strength coefficient.

3. The industrial CT calibration method based on regularized tolerance bands according to claim 2, characterized in that, The intensity coefficient has the same pixel dimension as the reprojection residual.

4. The industrial CT calibration method based on regularized tolerance bands according to claim 2, characterized in that, The tolerance parameter is determined by the mechanical manufacturing or assembly tolerance of the effective observation.

5. The industrial CT calibration method based on regularized tolerance bands according to claim 1, characterized in that, The tolerance band regularized residual is determined by the following formula: in, For the Softplus function, The number of valid observations in the set of geometric parameters. The strength coefficient, The tolerance parameter is... The nominal prior value, For the valid observations, The tolerance band regularized residual, This is the sharpness coefficient.

6. The industrial CT calibration method based on regularized tolerance bands according to claim 1, characterized in that, The process of minimizing the total residual vector to obtain the decoupled geometric parameters and three-dimensional structural points of the object to be detected includes: The total residual vector is minimized using a damped Gauss-Newton algorithm or a trust region reflection algorithm to obtain the decoupled geometric parameters and three-dimensional structural points of the object to be detected.

7. The industrial CT calibration method based on regularized tolerance bands according to claim 1, characterized in that, The set of geometric parameters includes at least one of the following parameters: source-rotation center distance, source-detector distance, focal lateral offset value, focal longitudinal offset value, detector principal point lateral offset value, detector principal point longitudinal offset value, in-plane tilt angle, out-of-plane tilt angle, and turntable angle step value.

8. An operation control device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the industrial CT calibration method based on regularization tolerance bands as described in any one of claims 1 to 7.

9. An industrial CT system, characterized in that, The industrial CT system is used to implement the industrial CT calibration method based on regularized tolerance bands as described in any one of claims 1 to 7, and the industrial CT system includes: The observation and acquisition module is used to obtain a set of geometric parameters by observing the center pixel of the object under multiple rotational poses. The residual assembly module is used to construct reprojection residuals, normalized residuals, and tolerance band regularized residuals based on each valid observation value in the geometric parameter set, and to concatenate all the reprojection residuals, normalized residuals, and tolerance band regularized residuals to obtain the total residual vector. The optimization solution module is used to minimize the total residual vector to obtain the decoupled geometric parameters and three-dimensional structural points of the object to be detected.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the industrial CT calibration method based on regularization tolerance bands as described in any one of claims 1 to 7.

Images