Subdivision of a large part and CAD model using computer-aided engineering (CAE) and computed tomography (CT)

Abstract

The invention relates to a computer-based method for gradual subdivision of large computer-aided graphic design (CAD) and part models using different principles to analyze production defects and their criticality levels, and to perform quality and life prediction of the final product.

Description

[0001] SUBDIVISION OF A LARGE PART AND CAD MODEL USING COMPUTER-AIDED ENGINEERING (CAE) AND COMPUTED TOMOGRAPHY (CT)

[0002] Technical Field

[0003] The invention relates to a computer-based method for gradual subdivision of large computer-aided graphic design (CAD) and part models using different principles to analyze production defects and criticality levels, and to perform quality and life prediction of the final product.

[0004] Prior Art

[0005] The methods of correlating CT analysis with non-linear analysis CAE and establish automatization methods are not used in common practice. These are done as discrete / separate processes.

[0006] In addition, there is no preliminary evaluation of CT scan results and method for slicing of CAD models into sub-models based on the measurements and evaluations from production, instead the processes are performed randomly and manually.

[0007] In addition, there is no decision-making mechanism for determine which region / zone of the part should be quickly put into non-linear analysis by preevaluating the initial CT images with image processing.

[0008] There is no holistic approach that combines machine learning, CAE analysis and CT scanning information to subdivide large part models.

[0009] CT, CAE and Machine learning methods used as discrete systems have different applications. They cannot be used together for integrated quality control and performance control on cast or press-formed parts.

[0010] When a large part model is to be fully analysed by non-linear methods, there is a serious waste of time and resources. A full CT defect metric analysis of a large part (defect ratio, defect sizes) without subdivision by prior visual processing takes a huge amount of computation power, RAM power and time.

[0011] The well-known state of the art document "Machine Learning in Computer Aided Engineering" discusses the use of machine learning in computer aided engineering.

[0012] When the existing works in the art are examined, there is a need to develop a computer-based method that enables the realization of the quality level and life estimation of the final product by examining the production defects and their criticality levels by gradually subdividing the large computer-aided graphic design (CAD) and part models with different principles in order to solve the above- mentioned problems.

[0013] Aim of the Invention

[0014] The aim of the present invention is to to develop a computer-based method that enables the gradual subdivision of large computer-aided graphic design (CAD) and part models using different principles to examine production defects and their criticality levels and to realize the quality and life prediction of the final product.

[0015] It is a further aim of the present invention to provide a computer-based method for increasing the automation of CT analysis times and analysis methods and enabling product control at larger quantity in the serail production

[0016] It is another aim of the present invention to provide a computer-based method for accelerating non-linear analysis and increasing efficiency in the analysis of a large part model.

[0017] Another aim of the present invention is to realize a computer-based method that enables multi-parameter and multi-functional dependent / including evaluations by increasing the use of artificial neural network types and in particular the use of large learning models (LLM). Another aim of the present invention is to realize a computer-based method to solve the problem of high demand of processor power, RAM power and time.

[0018] Detailed Description of the Invention

[0019] The invention relates to a computer-based method for gradual subdivision or segmentation of large computer-aided graphic design (CAD) and part models with different principles, to examine production defects and their criticality levels and to establish the quality and life estimation of the final product, comprising the steps;

[0020] - Selection of material and computer-aided graphic design (CAD) model and determination of computed tomography (CT) parameters according to the selected material and model,

[0021] - Determination of computed tomography (CT) parameters (power, voltage, focus type, filter, number of slices),

[0022] - Determination of material and manufacturing defects in computed tomography (CT) images by X-ray tomography,

[0023] - Examination / Evaluation of 3D merged X-ray tomography slices in computed tomography (CT) analysis by sweeping along different coordinate axes, and comparison of two-dimensional sequential images along those axis with sample data within datasets to establish correlations using machine learning and visual analysis algorithms based on deviations from the ideal state,

[0024] - Recording of sensor and similar measurement results from the production step as parameters or boundary conditions for the comparison step with process monitoring methods,

[0025] - Comparison of computed tomography (CT) image dataset using a large learning model (LLM) with the results of multi-parameter experiments, analytical calculations and simulation of different database results obtained by simulation in the database established with images of products produced with different parameters and qualities in the available physical scans,

[0026] - Comparison of the data from the previous comparison step with the measurement and sensor data from the production step and reference values established by similar measurements, - Where necessary, calculation of CT metrics of selected 2D sections and determination of the worst defects and sections by analytical method,

[0027] - Defect / void representation in finite element or point cloud modeling by placing the actual defect in the finite element model where the defect is mathematically defined with the XFEM method, numerically embedding the element or point in the element modeling or point cloud modeling into the identification equation,

[0028] - Then, the analysis of the two-dimensional computer-aided engineering (CAE) model with finite element analysis method for the sections,

[0029] - In parallel, during the comparison, the CT cross-sectional images produced by the current production will be matched against the measurement-sensor data / experiment results / CAE analysis data already stored in the data sets,

[0030] - Optimization of time and resources by canceling certain analyses in the computer-aided engineering (CAE) step by correlating X-ray tomography images of cross-sections in parallel with images in the database, or by validating as the reference analysis for the said section,

[0031] - Computer-aided engineering (CAE) and computed tomography (CT) correlation of both X-ray cross-sectional images and two-dimensional computer-aided engineering (CAE) analysis of image correlations,

[0032] - Analysis, correlation and comparison of cross-sections, geometric point coordinates, defect geometries, nodal reaction forces, stress and strain levels in computer-aided engineering (CAE) and computed tomography (CT) correlation, reaction forces acting on this point cloud when the 2D crosssection is converted to point cloud data, and production process data in computer-aided engineering (CAE) analysis,

[0033] - Computer-aided engineering (CAE) and computed tomography (CT) correlation checked with a large learning model (LLM), followed by subdivision,

[0034] - Parallelization of the computational unit by checking whether there is a need for repeating computed tomography (CT) scan and parameter adjustment,

[0035] - If no new analysis is needed, i.e. the data set contains the appropriate data, the data in the LLM is projected onto the DDSR in the CAE point cloud or CAE finite element model, - 2-dimensional CAE analysis results are superimposed and the point eigenvalues of the intersecting cloud points or nodes are transferred to 3 dimensions by minimum - maximum - arithmetic mean and, if necessary, regression and interpolations,

[0036] - Obtaining a homogeneous structure and continuum distribution of features by the said transition and determining the weak and strong regions of the part by projecting the value space / data space created by these data to subregions or the entire model,

[0037] - Comparison of computed tomography (CT) defect metric calculation results with the information from a three-dimensional computer-aided engineering (CAE) model dataset,

[0038] - If three-dimensional computer-aided engineering (CAE) analysis is required, computer-aided engineering (CAE) analysis to include regional defect metrics,

[0039] - Displaying the results on the large model, giving the strength of the integrated model point by point and providing production control,

[0040] - At each step, the results of the performance metrics and defect correlations calculated for the master part are fed back to optimize the parameters in production.

[0041] In the method according to the invention, the material and CAD model are first selected and CT parameters are determined according to the selected material and model. In the CT image, defects from the material and production are determined by X-ray tomography method. The solid model, which is rotated 360o axially in the machine, is superimposed with the images of its individual slices and the structural image of the whole structure created from the X-ray results is determined as defective and defect-free regions within the sensitivity of the equipment. Since X- ray transmission rates are different in the material and cavity, defects are determined from the image contrasts. The intensity and geometric form factors of each defect type are different.

[0042] In CT analysis, the 3-dimensional combined master data is sorted and checked in different Cartesian or cylindrical or cubic coordinate directions with unitary progression steps. These two-dimensional sequential images, which are staggered planarly and controlled in 2D by the sweeping process, are compared with sample data in the datasets and correlated with machine learning according to deviations from the ideal state. Sensor and similar measurement results from the production step are recorded as parameters or boundary conditions for the comparison step with process monitoring methods.

[0043] These above mentioned process operations include measurement of time-dependent temperature values of the molten metal in the crucible or chamber, composition analysis data, temperature, filling signal and pressure data at different points in the casting or pressure forming mold, mold filling time, cooling and heating circuit data and solidification zone data, roller speeds and forces in extrusion method, product rolling speed, temperature distributions, stresses in the rolling direction with optical stress analysis, temperatures and forces applied in the forming and shaping process, press speeds and times, laser power and generated temperatures, laser focal length, laser spot diameter in 3D laser printing process: The process duration includes the magnetic field applied at a level that will allow the metal to be shaped from the molten state in the aforementioned methods.

[0044] The results of multi-parameter experiments, analytical calculations and simulation of the CT image dataset, which is a database of images of products manufactured with different parameters and qualities in the pre-available physical scans, are compared with the results of different databases using the Large Learning Model (LLM). When necessary, the CT metrics of selected 2D sections are calculated and an analytical method is used to identify the worst defects and sections. The two- dimensional (2D) CAE model is then analysed for each cross-sections using finite element analysis.

[0045] The process of dividing the cross-sections into elements or point clouds can be as follows. Firstly, the visual cross-section can be directly divided into elements and a standard finite element model can be used. The second is to define the defects in the cross-section during image analysis for the extended finite element method (XFEM) by running a program subroutine that may be using an analytical / numerical model such as Beremin, etc., by determining the coordinate, geometric properties (roundness, opening, folding) and dimensions of the defect- gap-irregularity, or thirdly, by embedding the defects into the element equations in the finite element formulation.

[0046] In parallel with the segmentation process, correlation of the X-ray images of the cross-sections with the images in the database is used to optimize time and resources in the CAE step by cancelling certain analyses or validating the reference analysis in database. In CAE and CT correlation, image correlations of both X-ray cross-sectional images and 2D CAE analyses are performed and a large learning model is used when necessary.

[0047] The analysis, correlation and comparison of the geometric point coordinates, defect geometries, nodal point reaction forces, stress and strain levels in the CAE analysis, reaction forces acting on this point cloud when the 2D cross section is converted to point cloud data, and manufacturing process data are then analysed, correlated and compared in the CAE and CT correlation.

[0048] After the CAE and CT correlation is checked with a large learning model (LLM), subdivision / segmentation is performed. Computational unit parallelization is performed by checking whether there is a need for CT scanning and parameter changes again.

[0049] If there is no need for re-scanning, if there is no need for new analysis, i.e. if the data set contains the appropriate data, the existing CT scan data within the large learning model is taken, the intersection and regression of the heat maps of the desired features (forces, crack etc. dimensions, acting stress and strain amounts) in the 2D data in different directions determined / calculated / correlated in the previous steps are made, and the part performance prediction is made by projecting it on the sub-region to be analyzed in detail or on the entire 3D model in the point cloud or CAE model.

[0050] When re-scanning or 3D analysis is required, the target regions are selected in three dimensions and 3D CT scans are performed for each region and metrics are calculated. The information from the 3D CAE model dataset is compared with the calculation results of CT defect metrics. If a three-dimensional CAE analysis is required, the CAE analysis is performed to include the regional defect metrics. By displaying the results on the large model, the strength of the integrated model is given point by point and production control is ensured.

[0051] At each step, the results of the performance metrics and defect correlations calculated for the master part are fed back to optimize the parameters in production.

[0052] An exemplary embodiment of the method according to the invention is the X- ray / CT analysis of a large aluminum machine part. The strength values, connection points, loads, temperature effects and boundary conditions that the part must meet are determined according to CAD data. Depending on each operating condition and each test condition, the critical areas of the part are determined and these data are entered into one of the data sets in advance / pre-values in the program. These regions can be referred to as part’s specification regions.

[0053] The desired power values, working focal lengths, sensitivity and number of cross- sectional images are selected according to the material and part type. This optimizes the degree of detail of the image and the minimum measurable values of error metrics.

[0054] Since the molded part formed in part production has a complex geometry, the X- ray projection cross-sectional image is taken, then the axially rotated part is rotated 360 degrees, creating as many projection images as the number of cross-sections. These images are combined to create CT data. This data is the point cloud data of the material or structure according to the X-ray absorption coefficient, which can be analyzed in 3D and 2D. The cross-sectional image of the part can be examined by moving unit lengths in Cartesian / cylindrical or cubic coordinates on this data. Defects and cross-sectional thickness variations are visible. The part can be divided into any number of regions on this data.

[0055] The use of these images in 2D image processing and correlation is as follows. According to the coordinate type (example is Cartesian), a cross-sectional image is obtained in x, y, z directions at each unit advance. By sweeping sequentially in each direction, the cross-sectional quality of the manufactured part is analyzed. That is, based on the parameter evaluation from the production and the image analysis of this cross section, the ideal cross section in the data set, the sample image information obtained in different parametric experimental design studies, theoretically correlated with the data of defective cross sections of different ratios / morphologies / types, deviations from the ideal state and the criticality of the defects are analyzed.

[0056] As a result of the correlation, the cross-sectional analysis in each direction and the sensor and similar measurement results from the production step are embedded into the active analysis data set. The 2D cross-sectional analyses are now obtained in all directions by LLM and correlation / machine learning. The quality of each subregion and the whole part is matched. Here, certain regions can be given quality conformity by preliminary evaluation.

[0057] For the rest of the sections, however, in critical regions, 2D CAE analyses for quality correlation of the swept sections are computed planarly in the x-y-z directions. Here the cross-sectional image is meshed with a point cloud or a finite element method. Defects are identified in one of the three methods described above. This analysis network and its results can be converged with the closest analysis result in the predefined data set, or the calculation can be performed and new input can be inserted into the data set. Time and resource optimization is achieved by performing the analysis when necessary and correlating with the closest dataset result in other cases. (An example of a method that can be used here is the regression algorithm).

[0058] CAE results and CT - X-ray tomography results are correlated with the results in the datasets and the processing speed and accuracy are increased by using the LLM method. Thus, the need for subdivision and 3D metric calculation steps is accelerated as a result of rapid 2D analysis and correlation.

[0059] Defect containing sub-regions were identified until this step. With CAE and CT correlations, 2D cross-sectional analysis-CAE correlations, the theoretical stress, strain and maximum force levels that each point can bear were selected and determined. These values can be called point eigenvalues. With this data, the subregions to be analyzed in detail are identified

[0060] By combining this information with data from the production and process phase, production parameter confirmation or adjustment can be recommended.

[0061] When no new scan is required, the 2D images are stacked one after the other in the x-y-z axes directions. 2-D CAE analysis results are superimposed. In these overlays, the point eigenvalues of the intersecting cloud points or nodes are minimum - maximum - arithmetic mean and, if necessary, regression and interpolations are used to obtain a homogeneous structure and feature transition while upgrading to 3 dimensions. The value space / data space created by these data is projected to sub-regions or the whole model to determine the weak and strong regions of the part.

[0062] When re-scanning or 3D CT analysis is required, the metrics of the sub-regions to be analyzed in detail are calculated. Projection of these metrics on 3D CAE models is carried out with one of the 3 methods mentioned above and the analysis is performed and the criticality of the regions, part strength is calculated over the nodal points or cloud points. If necessary, process regulation is carried out by intervening in the production parameters.

[0063] Thanks to the method of the invention, the CT analysis of the structure of a product whose CAD data and material properties are selected, together with the process parameters, can be analyzed faster by dividing it into sub-regions and finite element analysis can be performed faster by using a large learning model (LLM) and a data set.

[0064] Currently, CT analysis and CAE analysis for a single part is performed in 8 + 24 hours, 32 hours in total. Thanks to the method according to the invention, this time is reduced to 3 + (12-15), 15-18 hours in total.

Claims

CLAIMS1. A computer-based method for gradual subdivision / segmentation of large computer-aided graphic design (CAD) and part models with different principles, to examine production errors and criticality levels and to realize the quality and life estimation of the final product, characterized by comprising the stepsSelection of material and computer-aided graphic design (CAD) model and determination of computed tomography (CT) parameters according to the selected material and model,- Determination of computed tomography (CT) parameters (power, voltage, focus type, filter, number of slices),- Determination of material and manufacturing defects in computed tomography (CT) images by X-ray tomography,- Examination / Evaluation of 3D merged X-ray tomography slices in computed tomography (CT) analysis by sweeping along different coordinate axes, and comparison of two-dimensional sequential images along those axis with sample data within datasets to establish correlations using machine learning and visual analysis algorithms based on deviations from the ideal state,- Recording of sensor and similar measurement results from the production step as parameters or boundary conditions for the comparison step with process monitoring methods,Comparison of computed tomography (CT) image dataset using a large learning model (LLM) with the results of multi-parameter experiments, analytical calculations and simulation of different database results obtained by simulation in the database established with images of products produced with different parameters and qualities in the available physical scans,Comparison of the data from the previous comparison step with the measurement and sensor data from the production step and reference values established by similar measurements,- Where necessary, calculation of CT metrics of selected 2D sections and determination of the worst defects and sections by analytical method,- Defect / void representation in finite element or point cloud modeling by placing the actual defect in the finite element model where the defect is mathematically defined with the XFEM method, numerically embedding the element or point in the element modeling or point cloud modeling into the identification equation, Then, the analysis of the two-dimensional computer-aided engineering (CAE) model with finite element analysis method for the sections,In parallel, during the comparison, the CT cross-sectional images produced by the current production will be matched against the measurement-sensor data / experiment results / CAE analysis data already stored in the data sets,Optimization of time and resources by canceling certain analyses in the computer-aided engineering (CAE) step by correlating X-ray tomography images of cross-sections in parallel with images in the database, or by validating as the reference analysis for the said section,Computer-aided engineering (CAE) and computed tomography (CT) correlation of both X-ray cross-sectional images and two- dimensional computer-aided engineering (CAE) analysis of image correlations,Analysis, correlation and comparison of cross-sections, geometric point coordinates, defect geometries, nodal reaction forces, stress and strain levels in computer-aided engineering (CAE) and computed tomography (CT) correlation, reaction forces acting on this point cloud when the 2D cross-section is converted to point cloud data, and production process data in computer-aided engineering (CAE) analysis,Computer-aided engineering (CAE) and computed tomography (CT) correlation checked with a large learning model (LLM), followed by subdivision,- Parallelization of the computational unit by checking whether there is a need for repeating computed tomography (CT) scan and parameter adjustment,If no new analysis is needed, i.e. the data set contains the appropriate data, the data in the LLM is projected onto the DDSR in the CAE point cloud or CAE finite element model,2-dimensional CAE analysis results are superimposed and the point eigenvalues of the intersecting cloud points or nodes are transferred to 3 dimensions by minimum - maximum - arithmetic mean and, if necessary, regression and interpolations,Obtaining a homogeneous structure and continuum distribution of features by the said transition and determining the weak and strong regions of the part by projecting the value space / data space created by these data to sub-regions or the entire model,Comparison of computed tomography (CT) defect metric calculation results with the information from a three-dimensional computer-aided engineering (CAE) model dataset,If three-dimensional computer-aided engineering (CAE) analysis is required, computer-aided engineering (CAE) analysis to include regional defect metrics,- Displaying the results on the large model, giving the strength of the integrated model point by point and providing production control, At each step, the results of the performance metrics and defect correlations calculated for the master part are fed back to optimize the parameters in production.

2. A computer-based method according to claim 1, characterized in that the extended finite element method (XFEM) method can be used instead of the finite element analysis method for the analysis of cross-sections of a two- dimensional computer-aided engineering (CAE) model.

3. A computer-based method according to claim 1, characterized in that the computer-aided engineering (CAE) and computed tomography (CT) correlation in both X-ray cross-sectional images and two-dimensional computer-aided engineering (CAE) analyses in image correlation, wherein a large learning model can be used as needed.