WO2019087032A1

WO2019087032A1 - Method for the reconstruction of cad models through parametric adaptation

Info

Publication number: WO2019087032A1
Application number: PCT/IB2018/058429
Authority: WO
Inventors: Lapo Governi; Yary Volpe; Rocco Furferi; Monica Carfagni; Francesco BUONAMICI; Alessandro Lapini
Original assignee: Universita' Degli Studi Di Firenze
Priority date: 2017-10-30
Filing date: 2018-10-29
Publication date: 2019-05-09
Also published as: IT201700123187A1

Abstract

Method for the semi-automatic reconstruction of CAD models of physical objects from a 3D scan through adaptation of an associative parametric model defined a priori (1) on a scanned reference datum (2) through an optimization algorithm minimizing, with an target function, the global error, defined by a suitable metric, between the associative parametric CAD model defined a priori and the scanned model (2) by modifying a vector of parameters extracted from the function tree of the associative parametric model (1).

Description

"METHOD FOR RECONSTRUCTION OF CAD MODELS THROUGH

PARAMETRIC ADAPTATION"

******

Field of the Invention

The invention relates to a method for reconstructing associative parametric CAD models (hereinafter, briefly referred to as "CAD models") through parametric adaptation, in particular a method for reconstructing CAD models of physical objects coming from a 3D scan such as to reduce the errors in the reconstruction and the discretion during the reconstruction by the operator. State of the art

In the state of the art, methods are known for reconstructing parametric and associative digital geometric models from 3D scans are known, i.e. CAD models provided with a logic function tree, or "features" and modelling history of physical objects.

This objective is achieved starting from the acquisition of coordinates on the surfaces of the object and through the subsequent processing of the data acquired within a software for reconstructing the scanned data, i.e. a reverse engineering software.

Typically, the process, the final object of which is the generation of a CAD model that is characterized by a good dimensional accuracy with respect to the physical object, accuracy that can be assessed through an error map evaluated on the reconstructed surface by calculating the distances separating the points detected by the CAD model, and which is characterized by topological correctness, i.e. which is constituted by functions and geometric constraints, such as parallelism, perpendicularity, continuity between surfaces and other, which reflect the original intent of the designer, develops through various steps.

There exist also other reconstruction strategies, which are however less used, such as the automatic reconstruction of free-form surfaces.

Generally, the current reconstruction methods are reasonably effective for reconstructing CAD models by experienced users and are provided with suitable software, however there are some important aspects that can be improved and that, at the state of the art, constitute important limits. Referring to the described steps of reconstructing, it is also known that the order according to which the operator sets the reconstruction of the object is a relevant aspect of the whole process. First of all, it is necessary to note that a given degree of uncertainty and error always affects the reconstruction process, unlike the direct modelling wherein the operator is in full control of the model.

The reconstruction process, in fact, involves the generation of mathematical surfaces adapting to the scanned data in an approximate way. The human factor, moreover, always introduces a possibility of error in the choices related to the reconstruction strategy (for example, but not exclusively, in the compromise between accuracy of adaptation and satisfaction of ideal geometric constraints).

The chain of steps of the modelling phase generates an error which accumulates during the process (typically, depending on the order according to which the modelling steps are performed), leading to a progressive separation of the reconstructed model from the actual one.

This effect is burdened by the introduction of geometric constraints on the surfaces generated by the user, which lead the model to separate from the reference datum.

In the modelling step, the order of reconstruction of surfaces also influences the type of constraints and relationships that the user has the possibility to impose, for example, it is very often necessary to create geometric references, such as points, axes, planes, curves, to allow the imposition of symmetries and geometric relationships of which the operator is aware.

In the modelling step, the reconstructing strategy chosen by the operator is therefore fundamental for obtaining a model affected by acceptable errors and at the same time representative of the original intent of the designer (a requirement which is typically imposed through the perfect satisfaction of geometric constraints), this makes the whole process largely discretionary and depending on the capabilities of the individual operator.

Furthermore, an error in the reconstruction process can only emerge at the end of the process itself.

The software tools currently available, while allowing the generation of an associative parametric CAD model, do not allow to update the model reconstruction taking into account the scanned data. In other words, a possible modification of the CAD feature tree produces a change in the model which, however, is no longer linked to the scanned data which originated the initial CAD model.

This implies that, if it is necessary to modify a parameter / constraint being used in the modelling step, the reconstructed geometry is no longer adapted to the scanned data. In practice, this means that the process must be started from the beginning, this being very time-consuming.

The operator can try to limit the negative effects imposed by the disclosed method, by performing a large number of operations and attempts, with a little efficient trial and error approach.

To try to keep controlled the problem of loss of adherence to the scanned data caused by the changes to the CAD model, US7821513 proposes a solution for the evaluation and visualization of a local dimensional error related to the operations of reconstruction, defined as "Accuracy Analyizer" which provides for the visualization of a deviation map between a scanned model and a parametric CAD model to be used to communicate the information relating to the loss of accuracy connected to the operations performed on the CAD model or on the "mesh" to the user.

The loss of accuracy is calculated from the point-to-point distances between the two models. However, the visualization of the map implies a simple aid to the operator and does not determine a real unique solution for the minimization of the error, and certainly not a process of complete reconstruction starting from a point cloud, thus not preventing the generation of unsatisfactory reconstructed models and not eliminating the need for reconstructing the model from the beginning in case of errors.

A further problem is represented by the limited allowed operations in terms of geometric constraints which can be imposed to guide the adaptation step of the surfaces, the most advanced software allows to specify a single geometric constraint during the reconstruction of a surface, such as an axis of revolution in the reconstruction of an axial-symmetrical surface.

However, very often, the needs of reconstruction would lead to the imposition of several reports simultaneously so as to be able to exploit the information coming from various regions of the acquired data. The imposition of a large number of interconnected geometrical relationships (which is extremely common for example in the reconstruction of mechanical parts) separates the reconstructed CAD model from the scanned data during the execution of the subsequent modelling steps.

A further negative aspect is given by the fact that the disclosed method of reconstruction is based on the independent and sequential reconstruction of the regions of the scanned data identified at the beginning of the reconstruction process.

This approach leads to a partial exploitation of the information content that can be extracted from the scanned data. In a truly effective process, the information and the relationships among all the regions of the scanned data should be exploited simultaneously and automatically from the beginning of the modelling process. For example, by imagining having to reconstruct the geometry of a cylindrical solid starting from a scanned mesh, the direction of the axis of revolution of the cylinder itself should not be deduced from the only analysis of the lateral surface, but also considering the normal calculations by analysing the flat superior and inferior faces.

In general, therefore, the available tools at the state of the art only allow a partial exploitation of the information contained in the scanned data.

The currently-used approach is oriented to the minimization of local errors, recorded in the reconstruction of a single surface and not from the overall reconstruction error, herein referred to as an error involving simultaneously all the geometric features of the reconstructed associative parametric model, even if not necessarily directly recalled in the definition of the metric used for the calculation of the error.

Operators are given very reduced possibilities with regard to the insertion of geometrical constraints of various types during modelling (especially when these refer to the existing relationships among the parts of the model generated by different regions of the scanned data) and finally there are considerable limitations in the "flexibility" of the modelling process, which does not guarantee a history of associative modelling with respect to the scanned data.

Object of the invention

It is therefore needed a method for reconstructing associative parametric CAD models through the parametric adaptation, in particular a method for reconstructing CAD models of physical objects from a 3D scan, such as to reduce the reconstruction errors and, at the same time, to limit the errors caused by the discretion of the operator when reconstructing.

Summary of the Invention

These objects have been achieved by developing a method according to one or more of the appended claims.

A first advantage consists in the fact that the method allows to remove the need for an iterative process as a solution following wrong choices made during the reverse modelling procedure, proposing an auxiliary tool in the exploration of the viable solutions in order to find the best possible configuration.

A further advantage is given by the fact that with the method of the invention, it is possible to exploit a template which can be available a priori, for example, from a library (and therefore does not require additional efforts for the generation of a model), and that, alternatively, the CAD template can be created without referring directly to the scanned data but using the conventional techniques already known by a CAD designer.

A further advantage is given by the fact that the method allows the reconstruction of an associative parametric CAD model, which is easily modifiable and exploitable downstream of the reconstruction (for example, but not exclusively, for a re-engineering of the object).

An even further advantage consists in the fact that, according to the method, all the necessary geometric constraints (in any number) are considered simultaneously during the reconstruction process, thus allowing the integration of complex geometric relationships of a wide range and in any number during the step of reconstruction.

An even further advantage consists in the fact that the optimization algorithm aims at achieving an excellent "global" on the scanned model thus making it possible a greater exploitation of the information content of the scanned data and the achievement of a globally more accurate result from the dimensional point of view. An even further advantage consists in the fact that the whole process becomes, generally, easier to manage and requires fewer specific skills for the operator responsible for the reconstruction.

An even further advantage consists in the fact that the method allows to implement a reverse engineering process based on the fitting of a parametric template selected / defined at the beginning of the reconstruction and therefore characterized by a topology which is consistent with that of the object to be reconstructed.

These and other advantages will be better understood by anyone skilled in the art from the description below and the accompanying drawings, given as a non-limiting example, wherein:

- Figure 1 shows the modelling steps according to a traditional approach;

- Figures 2a-2c show schematically a point cloud of the reference data of the subsequent triangulation and the creation of a mesh in a conventional data reconstruction process;

- Figure 3a shows a reconstruction of a quadrilateral with independent evaluation of the regions (sides);

- Figure 3b shows a reconstruction of a quadrilateral with independent evaluation of the regions and subsequent introduction of geometrical constraints to impose that the quadrilateral is a square;

- Figure 3c shows a direct reconstruction of a square with evaluation of the regions and simultaneous introduction of the geometric constraints;

- Figures 4a-4c show a schematization of the adaptation process of the parametric CAD template (4a) to the scanned data in the form of a polygonal mesh (4b) until the adapted parametric model (4c) is obtained;

- Figure 5 shows the diagram of the entire reconstruction process with adaptation of the parametric model;

- Figure 6 shows an example of a parametric feature tree;

Detailed description

With reference to the accompanying drawings, it is disclosed a preferred embodiment of a method for semi-automatic or automatic reconstruction of associative parametric CAD models. With reference to Figure 2, it is shown an example of generation of a polygon mesh starting from the point cloud acquired by a 3D scanner.

With reference to Figure 1 , it is shown a traditional example of the modelling step in the 3D reconstruction of a physical object starting from its polygonal model obtained through 3D scanning and subsequent triangulation of the point cloud.

The initial data 16, typically a point cloud in the three-dimensional space, can be obtained through the use of various 3D acquisition technologies, such as a laser scanner or other equivalent systems capable of generating a cloud of three-dimensional points measured on the surfaces of a solid object of any size.

The initial data 16, that is the point cloud, is then processed following a procedure that typically includes a first processing step of the acquired initial data, the processing step, wherein the operations of data decimation, reconstruction through the triangulation of a surface which is tessellated between the points, meshing 17 and finally a possible operation of correction and repair of the mesh are carried out. In a second processing step, the mesh is inserted into the reconstruction environment of the CAD model 6. Subsequently, it can be envisaged a segmentation step producing the separation of the data processed in distinct regions 7; this operation is carried out through an analysis of the geometric properties of the mesh surface. The obtained data are then aligned 8 according to a convenient Cartesian triplet to start the generation step of the CAD model.

The third processing step (classification) involves the association of geometrical functions, such as revolution surfaces, primitive surfaces and other, to the regions separated in the segmentation step.

The fourth processing step involves the generation of the desired analytical CAD surfaces using, as a reference, the regions previously identified and classified, for example with the realization of revolutions 9, cuts 10, holes 1 1 , trimming of free surfaces 12, extrusion of parts 13 14 and connections 15.

Typically, these steps are performed within commercial software, in the best cases provided with associative parametric modelling engines, with processes that lead to optimal results only in case of use by expert users. According to the invention, and with reference to Figures 3-6, it is proposed a 3D reconstruction method through the adaptation of an associative parametric CAD model defined a priori 1 and subsequently adapted on the scanned reference datum 2 through a procedure aiming to reduce the global reconstruction error, in particular a method for reconstructing CAD models of physical objects from a 3D scan in order to reduce the reconstruction errors and to limit the errors caused by the discretion of the operator when reconstructing.

The method involves the use of a whole associative parametric CAD model defined a priori with feature-based CAD modelling tools (based on 2D and 3D parametric functions, such as for example, but not exclusively, the generation of sketches, extrusions, revolutions , extruded cuts, revolved cuts, connections, chamfers, free-form surfaces, hole generation, sweep, loft) that the operator considers to represent the "design intent" of the part to be rebuilt and therefore wants to find in the reconstructed model at the end of the process.

In the present disclosure, the whole associative parametric CAD model refers to an associative parametric CAD model consisting of at least two geometric features such as revolutions, extrusions, sweeps, lofts, etc. and topologically consistent with the whole scanned 3D object, i.e. respecting the correspondence among the geometric features between the CAD model and the physical object, the set of geometric entities and how these are connected to each other to achieve the definition of the geometry of the object.

In different embodiments of the invention, the associative parametric CAD model can be generated by the user or be available from a library.

The method also provides for the adaptation of the associative parametric model defined a priori 1 , guided by an optimization algorithm which minimizes an target function, i.e. a function typically adopted in a choice problem and characterized by one or more variables and which expresses the purpose on the basis of which it is intended to make the choice, i.e. the minimization of a global error, defined by an appropriate metric, between the associative parametric CAD model defined a priori and the scanned model 2.

The optimization algorithm performs the adaptation by modifying a vector of parameters extracted from the function tree of the associative parametric model 1 . The parameters, which may be the whole or a part of those defining the associative parametric model 1 (possibly selected by the operator at the beginning of the reconstruction), are responsible for the geometry and position and orientation of the associative parametric model defined a priori 1 and can be for example, but not exclusively, parameters such as dimensions (linear but not only), angles, number of occurrences of a geometric feature (for example, but not exclusively, the number of a repetition of holes), coordinates of points or vectors. The optimization process is interrupted when a predefined condition is reached, the function tree of initial modelling of the associative parametric model defined a priori 1 is then updated using the best set of parameters identified by the optimization algorithm to generate the final reconstructed CAD model 3.

As an example (Figure 3), consider the reconstruction of a square approximating the reference datum: depending on the reconstruction sequence chosen, the position of the segments (sides) 5 could be evaluated in a completely independent manner, leading to the reconstruction of a generic quadrilateral (Figure 3a), and the final model obtained by further adding parallelism, equality and orthogonality constraints to obtain an actual square (Figure 3b), with an unsatisfactory result as it is not well overlapped to the reference datum; or it could be reconstructed, according to the method of the present invention, considering from the beginning the geometrical relationships insisting between the sides of a square, such as parallelism, equality and orthogonality, to generate a topological^ correct geometry and at the same time well overlapped to the reference datum (Figure 3c), thus reducing the global error.

Figure 4 shows the inputs of the proposed procedure, i.e. the parametric CAD model defined a priori 1 and the scanned data 2.

The first, defined in the parametric and associative modelling environment, carries the information on the topology of the object to be reconstructed, defined thanks to its feature tree and, in particular, of the geometric relationships defined among the features, whereas the second allows to have real measured data on the shape and dimensions of the surfaces to be reconstructed.

The data of the two elements are processed in the optimization step to achieve the generation of a result 3 (reconstructed CAD model) which is able to contain both the desired geometric structure from the point of view of the geometric features and the dimensions which best reproduce the scanned data.

This result is also defined in the associative parametric format as deriving from the associative parametric model defined a priori, being specifically a modified version of it.

The entire reconstruction process, with the use of the described method, is shown in Figure 5 and consists of the steps of acquisition of the reference data on the physical object through the application of a 3D acquisition technology, selection or generation of a parametric CAD model 1 consistent with the body being reconstructed, definition of a set of parameters, derived by the parametric model defined a priori 1 , responsible for the position and orientation of the model, and which dimensions (linear but not only), angles, number of occurrences of a geometric feature (for example, but not exclusively, the number of a repetition of holes), coordinates of points or vectors, initial recording (interactive or automatic using known methods) between the scanned model 2 and the model parametric defined a priori, segmentation according to the known methods of the scanned model 2, matching of the segmented entities with the corresponding surfaces of the parametric model defined a priori 1 (interactive or automatic) and definition through the adaptation procedure of the parameters minimizing the global error between the associative parametric model defined a priori 1 and the scanned model 2. The method of the invention therefore provides the following steps:

3D acquisition by scanning a physical object to be reconstructed, generation or retrieval from a whole associative parametric CAD model associated with the topology of the physical object,

selection of a set of parameters (which can coincide with the whole set of parameters characterizing the parametric model), derived directly from the parametric model, associated at least with the position and orientation of the parametric model and with the dimensions of its geometric characteristics. The dimensions may be linear dimensions, but also angles, number of occurrences of a geometric feature, for example, but not exclusively, the number of a repetition of holes, coordinates of points or vectors;

(possible) 3D alignment between the acquired 3D model and the associative parametric CAD model,

(possible) segmentation of the acquired 3D polygonal model (interactive or automatic using known methods) in order to group into regions, the triangular faces constituting the model according to their belonging to primitive surfaces and / or geometrical features such as (but not exclusively) planes, spheres, cylinders, revolution surfaces, loft surfaces, free-form surfaces,

(possible) matching of one or more of said regions of the acquired 3D model to one or more corresponding CAD surfaces of the parametric model, iterative adaptation "fitting" of the parametric model to the acquired 3D model through an optimization step that minimizes with a target function an error defined as a global error between the CAD model and the acquired 3D model, wherein the variables of the target function are called selected parameters.

Advantageously, the fitting step is thus guided by a metric capable of describing how "distant" is the shape of the CAD template generated from time to time with respect to the reference data. The whole optimization step is guided by suitable optimization algorithms; possible implementations of the optimization step can foresee the use, single or joint, of global optimization algorithms (for example, but not exclusively, heuristic techniques) or local optimization (for example, but not exclusively, techniques such as the descent along the gradient).

In various possible embodiments of the method, in order to achieve this result, several choices are possible to evaluate the global error between the two models.

A first choice, to be preferred for reasons of ease of application and effectiveness, is to use as an estimate of the global error an appropriate mathematical combination of the Euclidean distances, evaluated between the corresponding surfaces belonging to the acquired 3D model and the CAD template. These distances can be evaluated exploiting known techniques such as, but not exclusively, point-to-point distance or point-to-plane distance. The mathematical combination used as an estimate of the global error can be, for example but not exclusively, the average of the average distances evaluated among corresponding surfaces or the average of the maximum distances or even the maximum of the maximum distances.

However, in the choice of the global error function, it can be possible to use metrics which are different and not directly linked to the Euclidean distance, such as the comparison of the volumes of the two models, the areas of the surfaces of the two models or the corresponding surfaces, the position of the axes of inertia of the whole model or some parts of it and the value of the corresponding moments of inertia, the dimension of the bounding boxes, the coincidence of the centers of gravity.

Further global error indexes that can be obtained from the two models could also be the position of the symmetry axes and planes, of the revolution axes and of other geometric entities such as the correspondence of the silhouettes and profiles obtained by projections on the main planes of the object (e.g. two cylinders could be compared by evaluating the differences between the rectangles obtained by projecting them onto a plane containing the revolution axis).

In each iteration, the method involves the generation of new values of the selected parameters and the evaluation of the global error associated to them, until obtaining an associative parametric CAD model reconstructed by updating the CAD template according to the best values of the found selected parameters which, therefore, minimize the overall error.

As evident from the foregoing, the method of the invention achieves important advantages, in particular because it allows to fully exploit, and in any case far more with respect to the traditional systems, the information contained in the data extracted from the scanned physical object not minimizing local errors, but a global reconstruction error. A further advantage is the achievement, as a result, of an easily modifiable associative parametric CAD model; this result is, in fact, easily usable for future applications of the model. Furthermore, the obtained result is characterized by a correct topology that is consistent with the interpretation of the geometry as intended by the author of the CAD template (and therefore in terms of model topology and geometric constraints) and with the actual dimensions of the scanned object.

Furthermore, the proposed method allows the user to consider geometric constraints of various types and to perform the reconstruction process with greater flexibility (the parameter selection step allows, in fact, to establish which ones should be kept constant and which ones should be kept variable during the optimization). In the specific case wherein the user draws the CAD template at the beginning of the reconstruction, it is in fact possible to choose the most convenient modelling strategy according to the considered application, and therefore not only which geometrical features are

present, but also in which order and with which relationships these determine the shape of the final object.

The present invention has been described according to preferred embodiments, however equivalent variants can be conceived without departing from the scope of the present invention.

Claims

1 . Method for the reconstruction of an associative parametric CAD model of a physical object comprising the steps of

3D acquisition of a 3D model of a physical object to be reconstructed, retrieval from a library or generation of a whole associative parametric

CAD model characterized by a topology which is consistent with that of the 3D model of the physical object to be reconstructed,

selection of a set of a part or of the whole parameters that determine the shape and position of the associative parametric CAD model, directly derived from the associative parametric model and extracted from the feature tree of the model, related at least to the position and to the orientation of the associative parametric model and to the dimensions of the geometric features that constitute it,

iterative adaptation of the associative parametric CAD model to the 3D model acquired through an optimization step which minimizes a global error function between said acquired 3D model and said associative parametric CAD model,

in each iteration, generation of new values of the selected parameters and evaluation of the global error associated with them,

achievement of a reconstructed associative parametric CAD model by updating the whole associative parametric CAD model according to the best values of the found selected parameters minimizing the global error.

2. Method according to claim 1 , wherein said global error is calculated according to one or more metrics which are selected preferably from a metric based on an Euclidean distance, on a comparison among the volumes of the two models, on a comparison among the areas of the surfaces of the two models or of the corresponding surfaces, on a comparison between the position of the inertia axes of the whole model or of some parts of it and the value of the corresponding moments of inertia, on a comparison among the dimensions of the bounding boxes, on the coincidence of the centers of gravity.

3. Method according to claim 1 or 2, comprising a segmentation step of the 3D model acquired in isolated surface regions constituting the topology of the 3D model and a matching step of one or more of the regions of the acquired 3D model to one or more corresponding CAD regions of the associative parametric model.

4. Method according to claim 3, wherein said global error function is a combination of Euclidean distance function among corresponding regions.

5. Method according to claim 4, wherein said Euclidean distance function is a point-to-point distance or a point-to-plane distance.

6. Method according to claim 5, wherein said global error function is an average distance function of the average distances or the average of the maximum distances or the maximum of the maximum distances, said distances being evaluated among corresponding surfaces between the acquired 3D model and the associative parametric CAD model.

7. Method according to one of the preceding claims, wherein said adaptation step of the values of the optimization parameters is iterated until a certain error threshold value is reached.

8. Method according to one of the preceding claims, wherein said adaptation step of the values of the optimization parameters is iterated until a maximum number of prefixed iterations.

9. Method according to claim 3, wherein said matching step is an automatic matching step of the surfaces between the acquired 3D model and the associative parametric CAD model.

10. Method according to one of the preceding claims, wherein said iterative adaptation step performs a heuristic optimization algorithm.

1 1 . Method according to one of the preceding claims, comprising a 3D alignment step between the acquired 3D model and the associative parametric

CAD model.

12. Computer product comprising a memory wherein it is stored a computer program, which, if executed in a computer device, implements the method of one or more of the preceding claims.