SYSTEM AND METHOD FOR PRECISION DIMENSIONAL CONTROL OF A MECHANICAL STRUCTURE

Abstract

SYSTEM AND METHOD FOR PRECISION DIMENSIONAL CONTROL OF A MECHANICAL STRUCTURE The invention relates to a method for precision dimensional control of a mechanical structure (2) characterized in that it comprises: a step (E1) of receiving an image (IM1) of said mechanical structure (2), called the initial image; a step (E2) of determining, from said initial image, a high-resolution binary mask (M), called the structural mask; a step (E3) of identifying in said structural mask (M) at least one region of interest (R) from a predetermined set of regions of interest; a step (E4) of determining at least one geometric measurement from each identified region of interest; a step (E5) of validating the conformity of the dimensions of said mechanical structure from the determined geometric measurements of the regions of interest and a predetermined set of measurements of regions of interest (D; ML1; ML2). Figure for the abstract: Figure 1

Description

Title of the invention: SYSTEM AND METHOD FOR PRECISION DIMENSIONAL CONTROL OF A MECHANICAL STRUCTURE Technical field of the invention

[0001] The present invention relates to the field of quality assessment in the mass production of objects, directly during production. More particularly, it relates to a method and a system for the automatic and highly accurate verification of predetermined geometric specifications and the temporal monitoring of measurements of said geometric specifications. Technological background

[0002] Quality inspection in industrial manufacturing is a major challenge for companies seeking to maintain high production standards. Traditional methods of product conformity assessment have significant limitations that compromise the effectiveness and accuracy of quality control.

[0003] Currently, manufacturers rely primarily on statistical sampling methods, where only a limited fraction of production is examined manually or via partially automated processes. This approach has several critical drawbacks. First, it does not guarantee exhaustive verification of all manufactured items, potentially allowing undetected defects to go unnoticed. Second, manual analysis is time-consuming, costly, and prone to human error. Finally, in the time between the sampling of the part to be inspected and the inspection results, production continues, and if a defect is detected, all parts produced in that interval potentially have the same defect.

[0004] Furthermore, existing techniques lack the ability to accurately monitor geometric parameters over time. It therefore becomes difficult to detect early signs of degradation in production quality and other problems that can lead to such degradation, such as equipment wear or fluctuations in manufacturing processes.

[0005] There is therefore a need for a method to automatically verify the design dimensions of all objects manufactured by an industrial production line. The method must also allow for the analysis of temporal trends in said design dimensions and the detection of any degradation in production quality. Objectives of the invention

[0006] The invention aims to provide an automatic method for evaluating the geometric specifications of a series of objects manufactured by an industrial production line.

[0007] The invention aims in particular to provide, in at least one embodiment, a method for analyzing temporal variations in the measured values ​​of geometric specifications and for detecting statistical trends associated with a degradation in the production quality of said manufactured objects.

[0008] The invention also aims to provide, in at least one embodiment of the invention, a method for detecting outlier measurement values. Description of the invention

[0009] To this end, the invention relates to a method for the precise dimensional control of a mechanical structure comprising:

[0010] - a step of receiving an image of said mechanical structure, called image initial,

[0011] - a step of determining, from said initial image, a binary mask high resolution, also known as structural mask,

[0012] - an identification step in said structural mask of at least one region of interest from a predetermined set of regions of interest,

[0013] - a step of determining at least one geometric measure from each Area of ​​interest identified,

[0014] - a step of validating the conformity of the dimensions of said mechanical structure based on determined geometric measurements of regions of interest and a predetermined set of measurements of regions of interest.

[0015] By proceeding in this way, the invention proposes to overcome the limitations of traditional quality control methods by offering a fully automated and precise process for evaluating the geometric specifications of manufactured objects based on a predetermined set of measurements that must be respected during the manufacturing process of mechanical structures. Thus, the invention guarantees a comprehensive and reliable inspection, while reducing human error and the costs associated with manual approaches.

[0016] The first step of receiving an initial image of the mechanical structure provides the basic data necessary for analyzing the manufacturing quality of the mechanical structure. This image constitutes the starting point of the process and allows the visual characteristics of the object to be inspected to be captured. By automating this step, the invention eliminates reliance on human observation, thereby improving the reproducibility and accuracy of the results.

[0017] The second step, which determines a high-resolution binary mask from the initial image, allows the object to be segmented precisely. This structural mask It highlights the contours and different zones of the mechanical structure, thus facilitating the identification and measurement of regions of interest. This segmentation ensures that the relevant parts of the object can be analyzed with the desired precision.

[0018] The next step, which identifies at least one region of interest in the structural mask, makes it possible to target the specific areas where geometric measurements need to be taken. By relying on a predetermined set of regions of interest, this step ensures consistency in the analysis and allows the focus to be placed on the key shapes of the mechanical structure.

[0019] Determining geometric measurements from the identified regions of interest allows for the precise quantification of the dimensions of the targeted areas, thus providing objective and measurable data for assessing the conformity of the object. By automating this task, the invention reduces measurement errors and ensures a uniform evaluation for each object produced.

[0020] Advantageously and according to the invention, the method comprises a step, consecutive to the step of determining at least one geometric measurement from each identified region of interest, of validating the conformity of the dimensions of said mechanical structure from the determined geometric measurements of the regions of interest and a predetermined set of measurements of regions of interest.

[0021] Advantageously, the step of validating the conformity of the dimensions of the mechanical structure, based on the geometric measurements obtained and a predetermined set of references, makes it possible to determine whether the object meets the required specifications. This automated validation ensures rapid and reliable decision-making, while eliminating human bias. Furthermore, by integrating this step into a continuous process, the invention makes it possible to monitor production quality in real time.

[0022] Thus, the process helps to solve the known problems of traditional methods. Automation, measurement accuracy, targeted segmentation, and analysis of temporal trends combine to offer a complete and robust solution to the challenge of quality control in industrial manufacturing.

[0023] Advantageously and according to the invention, the method includes a preliminary step of acquiring said initial image by a camera comprising a telecentric lens so that the initial image is an orthogonal view of the mechanical structure.

[0024] According to this aspect of the invention, acquiring the initial image with a camera equipped with a telecentric lens makes it possible to obtain an orthographic image, that is, one without perspective distortion, which guarantees increased accuracy in the dimensional analysis of the mechanical structure. The telecentric lens ensures that the apparent dimensions of the objects remain constant, regardless of their position. within the camera's field of view. This eliminates perspective-related errors, which could distort geometric measurements and compromise the reliability of precision dimensional control.

[0025] The use of a camera with a telecentric lens also facilitates the capture of high-resolution images, which is essential for generating an accurate structural binary mask. A superior initial image improves the detection of contours and details of the mechanical structure, thereby optimizing subsequent process steps, including the determination of regions of interest and geometric measurements. This configuration ensures a solid foundation for the entire inspection process.

[0026] Furthermore, this image acquisition method allows for standardization of the process by minimizing variations due to acquisition conditions, such as the viewing angle or the distance between the camera and the structure. This contributes to increased reproducibility of the results, which is particularly important in industrial settings where consistency in controls is paramount. Consequently, this initial step enhances the overall reliability of the precision dimensional control process.

[0027] Advantageously and according to the invention, the determined structural mask has a resolution on the order of a micrometer.

[0028] According to this aspect of the invention, the structural mask determined with a resolution on the order of a micrometer makes it possible to obtain extremely high precision in the analysis of the dimensions of the mechanical structure. This high resolution ensures detailed detection of contours and geometric features, which improves the quality of the measurements performed. Consequently, even minimal dimensional deviations are identified with high reliability, thus guaranteeing better control of tolerances.

[0029] The micrometric resolution of the structural mask also facilitates the identification of regions of interest with increased accuracy. This makes it possible to target specific areas of the mechanical structure with exceptional clarity, thus reducing the risk of errors or omissions in the analysis. This ability to precisely isolate relevant regions optimizes the inspection process by making the results more consistent and reproducible.

[0030] Furthermore, this feature improves the ability to detect defects or dimensional variations that might not be visible with lower resolutions. This contributes to strengthening the overall reliability of the process by enabling a more rigorous assessment of the conformity of the mechanical structure. Consequently, this approach meets the most stringent quality control requirements in fields where accuracy is paramount.

[0031] Advantageously and according to the invention, the step of determining said structural mask comprises the following sub-steps:

[0032] - a substep for adjusting the image brightness,

[0033] - a substep of image color space transformation,

[0034] - a substep for detecting the contours of said image,

[0035] - a substep of contour filtering based on the contour measurement and / or the distance to the centroid of the longest contour and / or the principal axis of said image,

[0036] - a sub-step of creating a mask, called an intermediate mask, from said filtered outlines.

[0037] According to this aspect of the invention, the image brightness adjustment step improves the visual quality of the initial image by standardizing variations in lighting. This operation facilitates the extraction of relevant information by reducing undesirable effects related to non-uniform lighting conditions. It thus contributes to greater accuracy in subsequent image processing steps.

[0038] Transforming the image's color space makes it possible to convert visual data into a space more suitable for analysis. This transformation optimizes the separation of elements of interest by highlighting the specific characteristics of the mechanical structure. It plays a crucial role in preparing the data for subsequent steps, particularly edge detection.

[0039] Image edge detection identifies the boundaries and shapes present in the mechanical structure. This step extracts essential geometric information by isolating significant contours, thus enabling the precise definition of areas of interest. It provides a solid foundation for filtering and intermediate mask creation.

[0040] Edge filtering, based on criteria such as edge measurement, distance to the centroid of the longest edge, or the principal axis of the image, refines the extracted data by eliminating irrelevant elements. This process ensures that only the most significant edges are retained, thereby improving the reliability and relevance of the results obtained in subsequent steps.

[0041] Creating an intermediate mask from the filtered contours synthesizes the extracted information into a usable binary representation. This intermediate mask is a key step in generating the final structural mask, ensuring a precise match between the relevant contours and the areas of interest. This operation facilitates the identification and analysis of regions of interest for precision dimensional control.

[0042] Alternatively, the structural mask determination step may include different approaches to adapt to specific constraints or varied configurations of the mechanical structure. For example, the image brightness adjustment substep may be performed using adaptive algorithms that locally modify the brightness according to variations in lighting within the image, rather than a global adjustment.

[0043] If desired, edge detection can be performed using advanced techniques, such as convolutional neural networks trained to identify specific edges, instead of using conventional methods like Canny or Sobel. Furthermore, edge filtering can incorporate additional criteria, such as texture or spatial frequency analysis, to improve the accuracy of selecting relevant edges. Finally, the creation of the intermediate mask can include a fusion step with data from other sources, such as CAD models or laser measurements, to enrich the representation and ensure a better match with the actual structure. These variations allow the process to be adapted to specific needs while maintaining the objective of producing an accurate and usable structural mask.

[0044] Advantageously and according to the invention, the process includes a substep of refining said intermediate mask by a sub-pixel rendering method.

[0045] According to this aspect of the invention, integrating a substep of refining the intermediate mask using a sub-pixel rendering method significantly improves the accuracy of the data extracted from the initial image. Using this method, the contours and details of the mechanical structure are represented with increased precision, thus reducing approximations related to pixel resolution. This improvement in the quality of the structural mask ensures a better match between the processed image and the physical reality of the structure, thereby optimizing subsequent steps in the process.

[0046] The sub-pixel rendering method also contributes to better identification of regions of interest in the structural mask. By refining the boundaries and shapes of the relevant areas, it facilitates more precise and reliable segmentation. This results in more consistent extraction of geometric features, even in cases where the details are particularly complex or subtle. This increased accuracy in identifying regions of interest enhances the overall robustness of the process.

[0047] Furthermore, the use of this method in refining the intermediate mask improves the quality of the geometric measurements determined from the regions of interest. The calculated dimensions more accurately reflect the actual characteristics of the mechanical structure, thereby reducing margins of error and increasing reliability. results. This is particularly beneficial in contexts where strict dimensional tolerances must be respected.

[0048] Finally, this refinement substep contributes to a more rigorous validation of the conformity of the mechanical structure's dimensions. By relying on more precise data and more reliable measurements, the process makes it possible to detect deviations from predefined specifications with increased sensitivity. This strengthens the system's ability to ensure that the controlled mechanical structures meet the expected quality and performance requirements.

[0049] Alternatively, the precision dimensional control method for a mechanical structure can incorporate alternative embodiments for the intermediate mask refinement substep using a sub-pixel rendering method. For example, this method can rely on advanced interpolation algorithms, such as bilinear or bicubic interpolation, to improve the accuracy of contours in the structural mask. Another approach involves using specially trained convolutional neural networks to detect and refine sub-pixel details in binary images, thereby achieving better definition of regions of interest. Furthermore, adaptive filtering techniques, such as Wiener filtering or methods based on statistical models, can be applied to reduce noise and optimize the quality of the intermediate mask before its refinement.These variants allow for the exploitation of different strategies to maximize the reliability of geometric measurements extracted from regions of interest, while adapting to the specificities of the mechanical structures analyzed and the constraints of the imaging systems used.

[0050] Advantageously and according to the invention, the method includes a step of adjusting the orientation of said intermediate mask by a centering algorithm.

[0051] According to this aspect of the invention, the step of adjusting the orientation of the intermediate mask using a centering algorithm optimizes the accuracy of the precision dimensional control process. By adjusting the orientation of the intermediate mask, the algorithm ensures that the regions of interest are correctly aligned with respect to a defined reference frame, thereby improving the reliability of the extracted geometric measurements. This step also helps to reduce potential errors related to misalignments or shifts in the initial image, thereby strengthening the consistency of the results obtained.

[0052] The use of a centering algorithm in this step ensures better utilization of the data from the initial image. By recentering the intermediate mask, the process adapts to any variations in the position or orientation of the captured mechanical structure. This allows for efficient processing. structures exhibiting complex configurations or asymmetries, while maintaining high robustness in dimensional analysis.

[0053] This orientation adjustment step also promotes increased automation of the process. By integrating an algorithm capable of automatically managing the centering of the intermediate mask, the process becomes less dependent on manual intervention, which improves repeatability and reduces the risk of human error. This results in better reproducibility of the results, even in demanding industrial environments.

[0054] Finally, this technical feature contributes to improved compatibility of the process with machine vision systems. By adjusting the orientation of the intermediate mask, the process integrates more easily into automated production lines, where speed and accuracy of dimensional checks are key requirements. This ensures reliable and rapid analysis of mechanical structures, even in large-scale production environments.

[0055] Advantageously and according to the invention, the method includes a step of generating a time series from each determined geometric measurement.

[0056] According to this aspect of the invention, the step of generating a time series from each determined geometric measurement makes it possible to ensure continuous and precise monitoring of the evolution of the dimensions of the mechanical structure over time. This approach provides a dynamic view of the potential variations in the measurements, which facilitates the identification of trends or anomalies that could indicate degradation, deformation, or deviations from the initial specifications. By integrating this temporal dimension, the method significantly enriches the analysis by providing contextualized and evolving data, thereby improving the overall understanding of the behavior of the mechanical structure.

[0057] Generating time series from the determined geometric measurements also contributes to strengthening the reliability of the diagnoses performed. By comparing current data with past records, it becomes possible to detect subtle variations that might go unnoticed in a static analysis. This ability to trace the history of measurements makes it possible to establish correlations between dimensional changes and specific operating conditions or events, thereby enhancing the predictive capabilities of the process.

[0058] Furthermore, this step promotes better traceability of the checks performed on the mechanical structure. The time series constitute a usable database for documenting changes in dimensions and for justifying the conformity or non-conformity of the structure over time. This documentation can prove invaluable in industrial contexts where transparency and the justification of decisions are important requirements.

[0059] Finally, integrating this step into the process opens the way to the use of advanced analytical tools, such as machine learning algorithms or predictive models, which can leverage time series data to anticipate failures or optimize maintenance processes. This gives the process a proactive dimension, allowing action to be taken before major problems arise, and thus helping to extend the lifespan of mechanical structures while reducing the costs associated with repairs or replacements.

[0060] Advantageously and according to the invention, the method includes a step of comparing each element of each time series to at least one predetermined tolerance threshold, and in that an anomaly detection algorithm classifies said element according to one of the classes among: "outlier", "quality degradation", "desired quality" according to the evolution of the time series of each geometric measurement with respect to each predetermined threshold.

[0061] According to this aspect of the invention, the step of comparing each element of each time series to a predetermined tolerance threshold ensures a precise and rigorous analysis of the collected data. This comparison guarantees that each measurement is evaluated against defined criteria, which contributes to better control of potential deviations in the dimensions of the mechanical structure. By proceeding in this way, the process quickly identifies variations that could compromise the quality or conformity of the structure.

[0062] The integration of an anomaly detection algorithm that classifies each element according to specific categories, such as "outlier," "quality degradation," and "desired quality," provides an automated and intelligent analysis capability. This classification makes it possible to distinguish critical problems from acceptable variations, thus facilitating rapid and informed decision-making. By systematically categorizing the elements, the process optimizes the inspection process and reduces the risk of human error in interpreting the results.

[0063] The combination of these two steps, namely comparison to tolerance thresholds and classification by the algorithm, enhances the reliability and efficiency of precision dimensional control. This mechanism ensures proactive anomaly detection and continuous quality assessment, thereby improving the overall robustness of the process. Consequently, the process meets modern industrial requirements for precision and performance.

[0064] This process not only allows verification of the dimensional conformity of each object produced, but also allows analysis of temporal variations in geometric measurements in order to detect trends indicating a progressive or sudden degradation of production quality.

[0065] Based on statistical analyses, the method can detect statistical trends that could indicate a progressive degradation of quality, such as equipment wear or fluctuations in manufacturing processes. By identifying these problems at an early stage, the invention helps to prevent major defects and maintain high production standards.

[0066] The invention also relates to a precision dimensional control system for a mechanical structure, comprising:

[0067] - a camera equipped with a telecentric lens, configured to capture an image of said mechanical structure, called initial image,

[0068] - a computing unit, capable of:

[0069] - receive said initial image captured by the camera,

[0070] - to determine, from said initial image, a high-resolution binary mask, called structural mask,

[0071] - identify, in said structural mask, at least one region of interest among the predetermined set of regions of interest stored in memory,

[0072] - determine at least one geometric measure from each region of interest identified

[0073] According to this aspect of the invention, the camera equipped with a telecentric lens makes it possible to capture an initial image of the mechanical structure with high precision and without perspective distortion. This configuration ensures that the dimensions measured on the image faithfully reflect the actual dimensions of the object, thereby improving the reliability of the geometric measurements performed by the system. Furthermore, the telecentric lens ensures uniformity in data capture, regardless of the structure's position in the field of view, which facilitates analysis and reduces potential errors.

[0074] Advantageously and according to the invention, the system includes a storage memory comprising a predetermined set of regions of interest and a predetermined set of measurements associated with these regions of interest, moreover, the computing unit (8) is capable of validating the conformity of the dimensions of said mechanical structure (2) by comparing the determined geometric measurements of the regions of interest with the associated predetermined measurements stored in the memory (7).

[0075] The integrated storage memory contains a predefined set of regions of interest along with the measurements associated with those regions. This feature allows the system to have a directly accessible reference database for performing quick and accurate comparisons. By storing this information, the system can efficiently identify relevant areas of the mechanical structure and assess their conformity without requiring recalculations or additional analyses, thereby optimizing processing time and improving the consistency of results.

[0076] The processing unit receives the initial image captured by the camera and generates a high-resolution binary mask, called a structural mask. This process segments the image by isolating the contours and shapes of the mechanical structure, thus facilitating the identification of regions of interest. The high resolution of the mask ensures accurate detection of details, even for complex or small structures, contributing to a more detailed and reliable analysis.

[0077] The identification of regions of interest in the structural mask relies on a predefined set of data stored in memory. This step allows for the rapid and precise localization of specific areas of the structure that require geometric analysis. By using predefined information, the system reduces the risk of interpretation errors and ensures a consistent match between the measured data and the expected references.

[0078] Determining geometric measurements from the identified regions of interest is a key step in the process. By extracting specific dimensions from these areas, the system obtains precise quantitative data that serve as the basis for conformity assessment. This ability to directly measure the geometric characteristics of the regions of interest ensures an objective and reproducible analysis.

[0079] Validating the conformity of the mechanical structure's dimensions relies on comparing the determined geometric measurements with predefined measurements stored in memory. This step verifies whether the structure meets the expected specifications, which is essential to guarantee the quality and reliability of the final product. By automating this comparison, the system reduces human intervention and improves the overall efficiency of the inspection process.

[0080] Advantageously and according to the invention, the precision dimensional control system of a mechanical structure includes a mechanical structure lighting device emitting light of white spectral composition.

[0081] According to this aspect of the invention, the precision dimensional control system incorporates a lighting device designed to emit light with a white spectral composition. This feature makes it possible to obtain homogeneous and neutral illumination, which improves the quality of the images captured of the mechanical structure. White light, due to its balanced spectral composition, ensures faithful reproduction of details and contrasts, thus facilitating the precise identification of the contours and geometric features of the structure. This directly contributes to the reliability of subsequent processing steps, in particular the generation of the high-resolution binary mask and the extraction of regions of interest.

[0082] The use of white spectrum lighting also ensures optimal compatibility with a wide range of materials and surfaces, whether reflective, matte, or textured. This versatility reduces the risk of errors related to to variations in lighting or visual artifacts, which enhances the system's robustness under diverse conditions. Furthermore, white light minimizes chromatic distortions, which is particularly important for precise geometric measurements, as it avoids biases introduced by color casts.

[0083] By incorporating such a lighting device, the system ensures better repeatability of results, as lighting conditions remain constant and controlled. This is particularly relevant for industrial applications where measurement accuracy and reproducibility are key requirements. Moreover, this configuration simplifies the image sensor calibration process, as it eliminates the need to compensate for complex spectral variations, thereby optimizing implementation time and reducing operational costs.

[0084] In other words, the invention also relates to a precision dimensional control system for a mechanical structure according to the invention, characterized in that it includes a mechanical structure lighting device emitting light of white spectral composition.

[0085] The invention also relates to a precision dimensional control system for a mechanical structure according to the invention, characterized in that it comprises a support for mechanical structures having a convex profile suitable for positioning a mechanical structure on said convex profile.

[0086] The invention also relates to a precision dimensional control system for a mechanical structure according to the invention characterized in that the support for mechanical structures has a substantially "V" shaped profile.

[0087] The invention also relates to a precision dimensional control system for a mechanical structure, according to the invention, characterized in that it comprises a conveyor belt including at least one support for mechanical structures and configured to position a mechanical structure in the axis of the camera. List of figures

[0088] Other objects, features and advantages of the invention will become apparent from the following description, given by way of non-limiting example only, and which refers to the accompanying figures in which:

[0089] [Fig.l] is a schematic view of a method for precision dimensional control of a mechanical structure according to an embodiment of the invention.

[0090] [Fig.2] is a schematic view of a precision dimensional control system of a mechanical structure according to an embodiment of the invention.

[0091] [Fig.3] is a schematic view of a predetermined set of region measures of interest of a mechanical structure, according to an embodiment of the invention.

[0092] [Fig.4] is a schematic view of a mechanical structure, according to a mode of realization of the invention.

[0093] [Fig.5] is a schematic view of a high-resolution binary mask, according to a mode of the realization of the invention.

[0094] Detailed description of an embodiment of the invention

[0095] In the figures, the scales and proportions are not strictly respected for the purposes of illustration and clarity.

[0096] In addition, identical, similar or analogous elements are designated by the same references in all figures.

[0097] Fig. 1 represents, schematically and partially, a method for precision dimensional control of a mechanical structure according to an embodiment of the invention.

[0098] The method begins with a preliminary step E0 of acquiring the initial image IM1 of a face of a mechanical structure 2 to be analyzed. This acquisition is carried out using a camera 6 equipped with a telecentric lens, ensuring precise and distortion-free capture of the dimensions of the mechanical structure 2. This type of lens is particularly suitable for applications requiring highly accurate measurements of the dimensions of an object.

[0099] The process continues with a step 1E1 of receiving the initial image IML. This image is transmitted to a computing unit 8, where it will be used as a basis for the following steps. The image IM1 can be received in different digital formats depending on the needs of the analysis, for example, in high resolution to inspect the details of said mechanical structure.

[0100] Next, the process continues with a step E2 of determining a binary structural mask M from the initial IML image. This step involves transforming the image into a simplified, but highly accurate, representation of the contours of the mechanical structure. The mask has a resolution on the order of a micrometer, making it possible to capture the smallest details, for example, micro-imperfections on the edges of a metal joint. This step E2 comprises several substeps:

[0101] First, a substep E2.1 of brightness adjustment is carried out to optimize the contrasts of the IML image. This adjustment is essential to differentiate areas of interest, for example, in an image showing shadows on the edges of a metallic component.

[0102] Next, a substep E2.2 of transformation of the color space into greyscale is carried out.

[0103] The process continues with a substep E2.3 of contour detection, allowing the main shapes of the structure to be extracted.

[0104] These contours are then filtered E2.4 according to criteria such as their length or their distance from the centroid of the longest contour. Only the filtered contours will be retained for subsequent measurements.

[0105] Finally, an intermediate mask is created from the filtered contours E2.5.

[0106] The process includes a substep E2.6 of refining the intermediate mask by a sub-pixel rendering method. This method improves accuracy by adjusting the contours at a sub-micrometer scale, for example, to better define the irregular edges of the mechanical structure.

[0107] The process continues with a substep E2.7 of adjusting the orientation of the mask by a centering algorithm, ensuring that the mask is perfectly aligned with the main axis of the mechanical structure.

[0108] The process continues with a step E3 of identifying the regions of interest D, L1, L2 in the structural mask. These regions are predefined according to the critical areas of the mechanical structure to be analyzed.

[0109] The process continues with a step E4 of determining the geometric measurements from the identified regions of interest. These measurements include dimensions such as diameters, lengths or angles, for example, to verify their conformity to a technical standard, as illustrated in [Fig. 5] described below.

[0110] The process continues with a step E5 of validation of the conformity of the dimensions of the mechanical structure. This validation compares the measurements obtained to predefined references.

[0111] The process continues with a step E6 of generating a time series from the geometric measurements, thus allowing the evolution of the dimensions to be followed over time.

[0112] Finally, the process concludes with a step E7 comparing the elements of each time series with a predefined tolerance threshold. This step uses an anomaly detection algorithm to classify the elements according to categories such as "outlier," "quality degradation," or "desired quality." For example, if a measurement repeatedly exceeds the tolerance, this could indicate accelerated wear or a defect in the manufacturing of the mechanical structures. Within the framework of the precision dimensional control process for a mechanical structure, the classification of geometric measurements into different categories allows for the evaluation of the conformity of the analyzed structure. This classification is based on the evolution of the measurements in a time series and their comparison with predetermined tolerance thresholds. Three main classes are defined: "outlier," "quality degradation," and "desired quality."

[0113] An "outlier" is a measurement that deviates significantly from the expected values.

[0114] The "quality degradation" class includes measurements that, while not outliers, show an unfavorable trend over time. This degradation may be due to wear of manufacturing tools, gradual changes in production parameters, or material aging. Early identification of this trend makes it possible to anticipate the necessary adjustments to maintain optimal quality and prevent the occurrence of critical defects.

[0115] Finally, the "desired quality" class corresponds to measurements that conform to dimensional requirements and fall within the defined tolerance thresholds. A measurement classified in this category means that the mechanical structure meets the expected specifications and shows no significant signs of degradation. The objective of dimensional control is to maximize the number of measurements belonging to this category while identifying and correcting deviations before they compromise the overall quality of the product.

[0116] The automatic classification of measurements into these three categories, based on the analysis of time series and tolerance thresholds, thus makes it possible to ensure precise monitoring of dimensional quality and to detect early deviations in the manufacturing process. This contributes to continuous improvement and better control of industrial processes.

[0117] Fig. 2 represents, schematically and partially, a precision dimensional control system dedicated to the inspection of mechanical structures 2. It is based on a combination of optical and electronic devices enabling precise validation of the dimensions of a mechanical structure by comparing measurements obtained to predefined data.

[0118] Said system comprises a camera 6 equipped with a telecentric lens, configured to capture an initial image IM1 of the mechanical structure to be inspected. This image is then processed by a computing unit 8 which implements the process of [Fig. 1], namely: the generation of a high-resolution binary mask M from the image, called the structural mask, the identification in this mask of predetermined regions of interest R stored in the storage memory 7, and the calculation of geometric measurements associated with these regions. These measurements are then compared to predefined values ​​to validate the dimensional conformity of the structure.

[0119] The system also includes a lighting device 3, positioned on either side of the mechanical structure 4 in a plane perpendicular to the focal axis of the camera, emitting white light to ensure homogeneous and optimal illumination of the mechanical structures during image acquisition. Furthermore, the system includes a specific support 4, having a convex profile, which allows for the correct positioning of the mechanical structures 2, ensuring ideal stability and orientation for photographing the mechanical structure 2.

[0120] The system also includes a conveyor belt comprising a plurality of supports for mechanical structures 4. Said belt being configured to position a mechanical structure 2 in the axis of the camera 6.

[0121] The system also includes an extraction device 9 for removing mechanical structures from the production line whose measurements of the regions of interest do not conform to predetermined values. The extraction device pushes the defective mechanical structure 2 off the conveyor belt.

[0122] Fig. 3 represents schematically and partially a predetermined set of regions of interest on a mechanical structure 2 used to generate a high-resolution binary mask in the process of Fig. 1.

[0123] Fig. 4 represents, schematically and partially, a mechanical structure 2.

[0124] Figure 5 schematically and partially represents a high-resolution binary mask resolution obtained by the process of [Fig.1] applied to a mechanical structure 2 illustrated in [Fig.4]. This structural mask highlights the precise contours of said mechanical structure, thus allowing the extraction of dimensional measurements essential for its quality control.

[0125] In this figure, several measurements are annotated: - MD represents the maximum diameter of a circular section located in the upper part of the mechanical structure 2. This measurement is essential to verify the conformity of the machining or molding of this area, which may be a fixing axis, a pivot point or a functional element subjected to specific mechanical stresses. - ML1 corresponds to the total height of the mechanical structure, measured from the base to the top of the circular section defined by MD. This measurement is used to ensure the consistency of the overall dimensions of the part and to detect any variations in height resulting from manufacturing defects, such as poor cutting or excessive material shrinkage. - ML2 designates the maximum width of the base of the mechanical structure. This measurement is crucial to ensure the stability and correct assembly of the part within a larger mechanical system.

[0126] These measures are directly associated with the regions of interest defined in [Fig.3], where: - D corresponds to the area where MD is measured. - L1 corresponds to the total height, where ML1 is calculated. - L2 corresponds to the base of the structure, where ML2 is evaluated.

[0127] The acquisition and analysis of these dimensions in the binary mask M guarantees sub-micrometer precision, essential for detecting possible non- conformities. Thanks to the approach of the process detailed in [Fig.1], each extracted measurement can be compared to predefined tolerance thresholds and integrated into a time series allowing to follow the dimensional evolution of mechanical structures in production.

Claims

Demands

1. A method for the precise dimensional control of a mechanical structure (2) characterized in that it comprises: - a step (E1) of receiving an image (IM1) of said mechanical structure (2), called the initial image, - a step (E2) of determining, from said initial image, a high-resolution binary mask (M), called the structural mask, - a step (E3) of identifying in said structural mask (M) at least one region of interest (R) from a predetermined set of regions of interest, - a step (E4) of determining at least one geometric measurement from each identified region of interest.

2. A method for precision dimensional control of a mechanical structure according to claim 1, characterized in that it comprises a step (E5), consecutive to step (E4) of determining at least one geometric measurement from each identified region of interest, of validating the conformity of the dimensions of said mechanical structure from the determined geometric measurements of the regions of interest and a predetermined set of measurements of regions of interest (D; ML1; ML2).

3. A method for precision dimensional control of a mechanical structure according to claim 1, characterized in that it comprises a preliminary step (EO) of acquiring said initial image (IM1) by a camera comprising a telecentric lens such that the initial image (IM1) is an orthogonal view of the mechanical structure.

4. A method for precision dimensional control of a mechanical structure according to any one of claims 1 to 3, characterized in that the step (E2) of determining said structural mask (M) comprises the following substeps: - a substep (E2.1) of adjusting the brightness of the image, - a substep (E2.2) of transforming the color space of the image, - a substep (E2.3) of detecting the contours of said image, - a substep (E2.4) of filtering the contours according to the measurement of the contour and / or the distance to the centroid of the longest contour and / or the principal axis of said image, - a substep (E2.5) of creating a mask, called an intermediate mask, from said filtered contours.

5. A method for precision dimensional control of a mechanical structure according to claim 4, characterized in that it comprises a substep (E2.6) of refining said intermediate mask by a sub-pixel rendering method.

6. A method for precision dimensional control of a mechanical structure according to any one of claims 4 to 5, characterized in that it comprises a step (E2.7) of adjusting the orientation of said intermediate mask by a centering algorithm.

7. A method for precision dimensional control of a mechanical structure according to any one of claims 1 to 6, characterized in that it comprises a step (E6) of generating a time series from each determined geometric measurement.

8. A method for precision dimensional control of a mechanical structure according to claim 7, characterized in that it comprises a step (E7) of comparing each element of each time series to at least one predetermined tolerance threshold, and in that an anomaly detection algorithm classifies said element according to one of the classes among: "outlier", "quality degradation", "desired quality" as a function of the evolution of the time series of each geometric measurement with respect to each predetermined threshold.

9. A precision dimensional control system for a mechanical structure (2), comprising: - a camera (6) equipped with a telecentric lens, configured to capture an orthographic image (IM1) of said mechanical structure (2), referred to as the initial image, - a computing unit (8), capable of: • receiving said initial image (IM1) captured by the camera (6), • determine, from said initial image (IM1), a high-resolution binary mask (M), called structural mask, • identify, in said structural mask (M), at least one region of interest (R) from the predetermined set of regions of interest stored in memory (7), • determine at least one geometric measure from each region of interest (R) identified.

10. Precision dimensional control system for a mechanical structure (2), according to claim 9, characterized in that it comprises a storage memory (7) comprising a predetermined set of regions of interest and a predetermined set of measurements associated with these regions of interest, and in that the computing unit (8) is capable of validating the conformity of the dimensions of said mechanical structure (2) by comparing the determined geometric measurements of the regions of interest with the associated predetermined measurements stored in the memory (7).

11. Precision dimensional control system for a mechanical structure (2), according to any one of claims 9 to 10, characterized in that it comprises a lighting device (3) for mechanical structures emitting light of white spectral composition.

12. Precision dimensional control system for a mechanical structure (2) according to any one of claims 9 to 11, characterized in that it comprises a support for mechanical structures (4) having a convex profile suitable for positioning a mechanical structure on said convex profile.

13. Precision dimensional control system for a mechanical structure (2) according to claim 12, characterized in that the support for mechanical structures (4) has a substantially "V" shaped profile.

14. Precision dimensional control system for a mechanical structure (2), according to any one of claims 9 to 13, characterized in that it comprises a conveyor belt including at least one support for mechanical structures (4) and configured to position a mechanical structure (2) in the axis of the camera (6).

Images