Method for manufacturing an object represented by a point cloud

Abstract

The present invention relates to a computer-aided manufacturing method based on a point cloud representation of the object to be manufactured. A voxel comprising manufacturing parameters is assigned to each point of the point cloud and transmitted to a subtractive or additive manufacturing device. The invention also relates to a method for training a machine learning model for generating manufacturing parameters of the manufacturing method. Finally, the present invention also relates to a product obtained by the claimed manufacturing method.

Description

Manufacturing process for an object represented by a point cloud technical field

[0001] The present invention relates to a method for manufacturing an object based on a point cloud and to an object obtained by this method. State of the art

[0002] Using a graphical interface, computer-aided design (CAD) software allows the creation and representation of a digital object. Computer-aided manufacturing (CAM) software then comes into play. CAM software is dedicated to controlling manufacturing machines, typically CNC machines, lasers, 3D printers, etc. Thus, for additive or subtractive manufacturing, traditional CAD and / or CAM uses mathematically defined finite surfaces. These are traditional geometries, typically triangles, squares, or NURBS when the shapes require special curvatures. Depending on the CAD and / or CAM application, these surfaces are used directly or, when closed, allow the creation of mathematical volumes. These volumes can also be used for manufacturing purposes. Examples include cubes, cylinders, etc.or, for more complex volumes, volumes delimited by NURBS surfaces or multitudes of triangles. It should be noted that these technologies use approximations; typically, a cylinder is represented using triangles, circles or NURBS shapes are converted into polygons, and so on. Thus, traditional CAD only allows the design of undefined and often approximate objects.

[0003] It is worth noting that CAD and CAM software allow users to create and then control machines. Examples include Dassault Systèmes CATIA, ALPHACAM CFAO, PTC Creo, and Solidworks. {PPESTA}-{5}-{PCT} Solidcam, etc... Because the core computing systems of CAD and CAM differ, managing the software layers to achieve dynamic CAM is complex. Dynamic CAM allows a user to automatically incorporate modifications during CAD design iterations. The design tree simply needs to be updated to integrate the new control of the manufacturing machines. It's worth noting that the capabilities of CAD and CAM software often make dynamic CAM inaccessible. In this case, the user saves their part in an international exchange format, typically *.STEP (for solid parts) or *.STL or *.IGES (for surface parts). The file is then opened, and all machining strategies must be recalculated at each iteration. These steps are particularly cumbersome if there are many design iterations.

[0004] These CAD / CAM methods are accessible and efficient for simple objects. However, they become difficult to implement for complex objects because the number of triangles connecting the points increases significantly as the number of points themselves increases. It is then necessary to use software that corrects errors during triangle creation (overlays, open surfaces, etc.). Thus, in CAM, beyond a certain number of points, it is even impossible for current computers to reasonably handle the creation of complex 3D shapes.

[0005] Another solution, used when pure 3D management is no longer feasible (requiring too much computing power), is to resort to so-called 2D management. 1 / 2 also called grayscale depth management machining. This approach involves adding textures to flat surfaces or NURBS. These projected images, generally using grayscale, utilize software from the video game or animation film industries (mesh surfaces and image textures). 2D 1 / 2, thus consists of managing intensity levels corresponding to depths orthogonal to the surface. {PPESTA}-{5}-{PCT}

[0006] However, although the 2D approach 1 / 2 allows for the management of larger point clouds, but its use is rigid because it does not allow working in a pure 3D CAD or CAM environment. The user is limited to a truncated, textured 2D visualization. This approach allows very little, if any, integration of other parts, for example, from standard CAD.

[0007] Unlike finite surfaces or volumes in CAD / CAM, real-world objects, whether natural or artificial, have no limits in internal or external definition when approaching an atomic level. Such surfaces or volumes, with their extremely detailed resolution, can be numerically approximated by point clouds containing a very large number of points, typically several hundred million or even several billion. These objects are also called "fractals" because of their characteristic structure, characterized by details within details. The definition of a fractal is as follows: "Mathematical objects whose creation or form is governed solely by irregularity or fragmentation." These characteristics can be used for decorative or technical purposes, for example.One example is a model of terrestrial topographic relief where the interest stems from the presence of countless details, high-definition textures, the surface of a lung (optimizing the gaseous interface), the surface of a butterfly wing (creating light diffractions using scales) or the surface of multi-channel electromagnetic antennas (fractal shapes optimized to handle a multitude of wavelengths), blood or nerve networks and their interweaving, a tree, as well as an internal bone structure, etc... For example, starting from a block of material of 5cm x 5cm x 5cm, knowing that 3D printing, milling or laser machining technologies make it possible to create or machine chips of 1 m. 3 The creation of a fractal object would represent the possibility of managing 1.25th 14 pm 3 .

[0008] For example, starting with a block of material measuring 5cm x 5cm x 5cm, knowing that 3D printing, milling, or laser machining technologies allow for the creation or machining of chips measuring 10 µm x 10 µm x 10 {PPESTA}-{5}-{PCT} m, or 1000 pm 3 The creation of a fractal object would represent the possibility of managing 1.25th 11 pm 3 With a 4MHz laser, this represents 8 hours of machining. A manufacturable fractal object is one that approaches the maximum resolution capabilities of the techniques used.

[0009] The fabrication of fractal objects is possible using traditional CAD and / or CAM software, but it quickly becomes limited without significant time spent on CAD / CAM management and calculations as soon as the object's definition exceeds a few million 3D surfaces or points in CAD, typically 2 million. It should be noted that no software exists that can manage fractal objects without a definition limit, typically 1 billion or more points or voxels. It should also be noted that during the fabrication process, the mixing of tool sizes and / or manufacturing technologies to optimize production time, typically through roughing phases, is also quickly limited. For example, a large tool or a large laser impact followed by small ones refining a surface becomes impossible for existing software to handle a fractal object. Brief summary of the invention

[0010] One aim of the present invention is to propose a method for manufacturing an object that is free from the limitations of known methods.

[0011] Another aim of the invention is to propose a method for manufacturing objects based on a point cloud, offering a simplification of processes, typically from 2 million points.

[0012] Another aim of the invention is to propose a method for manufacturing objects based on digital representations comprising a large number of points. {PPESTA}-{5}-{PCT}

[0013] Another objective of the invention is to propose a method for manufacturing objects that eliminates the constraints due to surface and / or volume calculations present in prior art methods.

[0014] Another objective of the invention is to enable dynamic display management, typically by managing in graphics memory only the parts currently used by the designer.

[0015] Another objective of the invention is to enable the dynamic management of CAD / CAM design iterations.

[0016] According to the invention, these goals are achieved in particular by means of an additive or subtractive manufacturing process of an object, comprising the steps of: obtaining by means of a computer a set of points forming a digital representation of the object, the set of points comprising at least 2 million points; for each point of the set of points, establishing by means of a computer a set of voxels, each voxel corresponding to a unit of material to be added by additive manufacturing or removed by subtractive manufacturing, and each voxel comprising a three-dimensional attribute (X,Y,Z) allowing the voxel to be located in space and comprising at least one parameter determining a unit of material to be deposited by additive manufacturing or removed by subtractive manufacturing; for each voxel successively, transmitting at least one parameter of the voxel to an additive or subtractive manufacturing system;For each voxel, deposit by additive manufacturing or remove by subtractive manufacturing using the additive or subtractive manufacturing system, the unit of material corresponding to the voxel as a function of at least one parameter of the voxel.

[0017] This process therefore allows the fabrication of objects by their surfaces or volumes based on point clouds comprising at least 2 {PPESTA}-{5}-{PCT} millions of points thanks to a "point-by-point" processing, particularly in the allocation of parameters that the manufacturing system will interpret to control the tool.

[0018] Advantageously, this process avoids the need to use traditional CAD / CAM techniques such as meshing or NURBS surface generation to delimit objects.

[0019] The step of obtaining the point set may include a substep of subsampling and / or oversampling by means of a computer of an initial set of points forming an initial digital representation of the object.

[0020] The points package advantageously comprises at least 15 million points, preferably at least 20 million points.

[0021] The successive points of the initial set of points may include points located on an edge of the initial digital representation of the object.

[0022] This sub-step allows, in particular, for the management of tool offsets at the edge of manufactured objects as well as increasing manufacturing precision.

[0023] According to one embodiment, the process may include a surface roughing step by additive or subtractive manufacturing prior to the surface additive or subtractive manufacturing step, the roughing step comprising the substeps of: using a computer, obtaining a subsampling of the point set and determining a digital rough based on the subsampling; manufacturing, using an additive or subtractive manufacturing system, a rough object based on the digital rough. {PPESTA}-{5}-{PCT} in which the step of depositing by additive manufacturing or removing by subtractive manufacturing the unit of material corresponding to the point on the basis of at least one voxel parameter, is carried out from the object blank.

[0024] Roughing significantly reduces manufacturing time by using coarser tools to produce a first approximation of the object. Advantageously, roughing is facilitated by the point-by-point approach, which makes it easy to obtain a roughing surface from the initial point cloud.

[0025] According to one embodiment, the angle between a surface of the object blank and a manufacturing direction determined by the orientation of a manufacturing tool is at least 40°.

[0026] According to one embodiment, the manufacturing of the object blank is carried out using a subtractive manufacturing system comprising a CNC machining center and the unit of material corresponding to the voxel as a function of at least one parameter of the voxel is removed using a subtractive manufacturing system comprising a laser ablation machining center.

[0027] According to one embodiment, the step of obtaining the point set further includes the substeps of: generating by computer-aided design (CAD) a file corresponding to a surface representation of a complementary object; converting the file into a complementary point set using a computer; wherein the point set includes at least some points from the complementary point set.

[0028] These sub-steps advantageously allow the integration of elements from traditional CAD into point clouds and the fabrication {PPESTA}-{5}-{PCT} an object based not only on a very detailed set of points (e.g. from a scan) but also on additions from traditional CAD.

[0029] The process may include a step of: using a computer, assigning to each point in the point set a complementary manufacturing attribute; at least one parameter of the voxels in the voxel set being determined on the basis of the complementary attribute of the associated point.

[0030] These additional manufacturing attributes include, for example, at least one element from: color, tint, transparency, texture, hardness, roughness, oxidation, and / or position.

[0031] According to one embodiment, the additional manufacturing attribute of each point includes a category obtained at the end of a point classification step in the set of points, preferably a point classification by a machine learning algorithm.

[0032] A voxel can also include one or more parameters selected from: - a dimension of the unit of matter, - the speed of a manufacturing tool in the manufacturing system, - the power of a manufacturing tool within the manufacturing system, - an angle between a manufacturing surface comprising the voxel and a manufacturing tool of the manufacturing system - a voxel ordering index allowing it to be positioned relative to other voxels, - a laser polarization, frequency and / or vibration when the manufacturing system includes a laser machining tool.

[0033] According to one embodiment, the manufacturing system is a subtractive manufacturing system comprising a machining center. {PPESTA}-{5}-{PCT} multi-axis, wherein at least one parameter includes a vector normal to the surface, the process further comprising the steps of: numerically determining, using a computer, an approximation surface of the surface on the basis of an average of the normal vectors assigned to each point, the approximation surface enabling mechanical orientation of the manufacturing tool and / or a manufacturing support of the surface during manufacturing.

[0034] According to one embodiment, the manufacturing system is a subtractive manufacturing system from a block of material, and in which at least one parameter of a surface voxel of the object is a manufacturing tool power, the power enabling the tool to machine a predetermined depth so as to reveal a layer of the block of a predetermined color.

[0035] Advantageously, the manufacturing tool power is selected from at least three different powers so as to allow the tool to machine at least three predetermined depths, each depth corresponding to a layer of the block of a predetermined color.

[0036] It is therefore possible to create pixels on the surface of the object in order to obtain a wide range of colors.

[0037] According to one embodiment, the set of points is obtained during an initial step of scanning an existing object and / or generating it by computer.

[0038] These goals are also achieved by means of a method for training a machine learning model to generate at least one parameter of the manufacturing process described above, comprising the steps of: {PPESTA}-{5}-{PCT} retrieve a training dataset comprising parameter measurements of an object manufactured according to the process described above, each parameter measurement being assigned to a voxel of a digital representation of the manufactured object, train a machine learning model on the basis of the training data providing as output for each voxel, an approximation of at least one parameter of the voxel.

[0039] The step of retrieving the training dataset may include machining and / or additive manufacturing according to the method described above of a sample matrix, each sample being manufactured on the basis of a set of voxels comprising at least one parameter different from an adjacent sample in the sample matrix.

[0040] During the execution of the manufacturing process, at least one parameter determining a unit of material to be deposited by additive manufacturing or removed by subtractive manufacturing can be determined by a machine learning model trained according to the process described above.

[0041] These goals are also achieved by means of a product obtained by the manufacturing process described above.

[0042] According to a particular embodiment, the product comprises a watch, jewelry and / or jewelry component, including at least one setting bezel.

[0043] Advantageously, the crimping chatons can thus be manufactured as a single piece with the part in order to limit the need for assembly. {PPESTA}-{5}-{PCT} Brief description of the figures

[0044] Examples of implementation of the invention are given in the description illustrated by the accompanying figures, in which: • Figure 1 illustrates the main steps of a manufacturing process for an object according to the present invention. • Figure 2 schematically illustrates points and associated voxels used for the manufacture of an object. • Figure 3 schematically illustrates voxels associated with a point on a surface to be manufactured, each voxel belonging to a manufacturing layer. • Figure 4a schematically illustrates the voxels associated with each of the points of a surface to be manufactured. • Figure 4b illustrates the surface to be manufactured without the voxels of figure 4a. • Figures 5a-5c schematically illustrate sets of voxels including particular size parameters. • Figures 6a and 6b illustrate an embodiment including the generation of a roughing surface. • Figure 7 illustrates a step in integrating a CAD-designed element to manufacture an object. • Figures 8a and 8b illustrate superimposed layers of different colors in the material used to make the object. {PPESTA}-{5}-{PCT} Figure 9 illustrates machining at different depths in a material comprising layers of different colors. • Figure 10a illustrates a sample matrix for training a machine learning model. • Figure 10b illustrates a detail of the sample matrix from Figure 10a. Example(s) of an embodiment of the invention

[0045] The present invention relates to a method for additive or subtractive manufacturing of an object 1 having a surface and / or a volume from an assembly comprising a very large number of points, typically at least 2 million points, forming a digital representation of the object.

[0046] Referring to Figure 1, the manufacturing process includes a first step of obtaining, using a computer, a set of points 2 (or point cloud) forming a digital representation of the object to be manufactured. Obtaining this set of points 2 can, for example, mean that a computer file containing the set of points is loaded / downloaded from a computer's internal or external memory, from a cloud, from a remote server, etc., but it can also mean that this set of points 2 can be generated by a scanning operation (e.g., lidar, radar, sonar, or other laser scanning methods), by mathematical modeling using software, or by any other method that allows obtaining a 3D model of a surface represented by a point cloud.

[0047] The set of points 2 is typically stored electronically as .stl files or any other alternative and / or equivalent format allowing the storage of 3D data. {PPESTA}-{5}-{PCT}

[0048] To overcome the limitations of prior art methods, particularly those stemming from current computer processing power constraints, this process offers a surprising point-by-point approach to managing the 3D model. For each point in the set of points 2 that form the object's digital representation, a specific number of material units to be removed by subtractive manufacturing or deposited by additive manufacturing are determined computationally and then transmitted to the manufacturing system. This drastically reduces, or even eliminates, the need for calculating triangulations, NURBS surfaces, or other surface or volumetric constructions.

[0049] Thus, the present process includes a second step during which for each point of the set of points 2, a set of voxels 3 is established by means of a computer, each voxel 3 corresponding to a unit of material to be added by additive manufacturing or removed by subtractive manufacturing, and each voxel 3 comprising on the one hand a three-dimensional attribute allowing the voxel to be located in space and on the other hand, at least one parameter determining the unit of material to be added by additive manufacturing or removed by subtractive manufacturing.

[0050] Thus, in the context of the present invention, each voxel allows the storage of spatial information, the three-dimensional attribute, as well as at least one manufacturing parameter determining a unit of material to be removed or deposited.

[0051] Additive manufacturing here refers to all 3D printing techniques such as, for example: • powder bed fusion • the binder jet • material extrusion, material projection {PPESTA}-{5}-{PCT} layering or lamination of sheets • deposition under concentrated energy • Photopolymerization allows for 3D printing of objects in materials such as metals (including steel, aluminum, titanium, platinum, etc.), polymers, ceramics, glasses, etc.

[0052] Conversely, subtractive manufacturing refers to manufacturing techniques that remove material from a block of material and includes, in particular, milling and laser machining. High-definition and ultra-high-definition laser machining, i.e., with an impact diameter between 2 µm and 35 µm and a layer thickness between 10 nm and 5 µm, are examples of subtractive manufacturing suitable for implementing the process of the present invention. Combining different additive manufacturing techniques, such as initial milling followed by laser finishing, is also a technique particularly well-suited to producing objects within the scope of the present invention.

[0053] In laser machining, the manufacturing device typically has a firing frequency of at least 400 kHz (i.e., 400,000 laser pulses per second) and can reach up to 2 MHz, or even 10 MHz. The laser firing frequency can be linked to a travel speed, resulting in more or less precise spacing between impacts. Therefore, for accuracy reasons, it can be advantageous to use pulse-on-demand lasers.

[0054] In laser machining, the generation of very large amounts of data is directly correlated with the number of impacts. To maintain economic efficiency, a high firing frequency is advantageous. This high frequency can cause heating, which can lead to accuracy problems. Therefore, using femtosecond lasers or similar technologies can be beneficial. {PPESTA}-{5}-{PCT} having the particularity of enhancing the material without allowing it time to diffuse heat.

[0055] Subtractive manufacturing can be carried out on a very wide range of materials such as metals or alloys, polymers, wood species, ceramics, mother-of-pearl, coral, glass, composite materials, etc.

[0056] The three-dimensional attribute of each voxel must allow the voxel to be located in space. This can consist of the precision of its spatial coordinates relative to a given reference frame, or the relative position of the voxel with respect to other voxels in the set. The voxel's location in space must allow a subtractive or additive manufacturing system to position a manufacturing tool so that it can remove or deposit a corresponding unit of material.

[0057] If machining or additive manufacturing is done layer by layer, i.e., manufacturing is done layer by layer, the three-dimensional attribute can include a double of planar coordinates (X,Y) placing the voxel in a given layer indicated by a third data point (Z) representing, for example, a layer number or a given thickness.

[0058] Each voxel also includes at least one parameter that determines the unit of material to be removed or deposited by the manufacturing system. This might include, for example - a dimension (diameter, thickness, etc.) of the unit of matter, - the speed of a manufacturing tool in the manufacturing system, - the power of a manufacturing tool within the manufacturing system, - an angle between a manufacturing surface comprising the voxel and a manufacturing tool of the manufacturing system - a voxel ordering index allowing it to be positioned relative to other voxels in the same layer, {PPESTA}-{5}-{PCT} - a laser polarization, frequency and / or vibration when the manufacturing system includes a laser machining tool.

[0059] In additive manufacturing, a dimension can, for example, represent the size of a laser spot (in powder bed fusion) or the amount of material to be extruded at a given point. In subtractive manufacturing, it can also refer to, for example, the size of a laser spot or the size / diameter of a milling head.

[0060] In one embodiment, a manufacturing speed parameter can be associated with each voxel. This parameter provides the manufacturing system with an indication of the manufacturing tool's speed around the voxel to which it is associated. For example, this parameter can indicate the tool's travel speed between the voxel to which the parameter is associated and the next voxel, or between that point and all adjacent points, and so on. Indeed, it can be advantageous to be able to manage the manufacturing speed dynamically, that is, based on voxels or zones (e.g., internal voxels or voxels on the object's exterior), due to thermal constraints (e.g., material cooling during manufacturing), slopes, material changes, rendering quality, manufacturing time, and so forth.

[0061] More specifically, when an area forms a significant angle with the horizontal, i.e., when an area has a steep slope, it is advantageous to be able to adjust the manufacturing speed compared to other flat areas. This can be the case, for example, when a roughing step has been performed and the surface roughing includes inclined portions. In laser machining especially, laser impacts on an inclined portion are naturally more difficult to control because the surface is not perpendicular to the laser. It is then advantageous to optimize the laser spot's movement speed based on the angle formed by the surface on which the impacts are made (e.g., the surface roughing) and the direction of the laser spot. These considerations {PPESTA}-{5}-{PCT} can also be valid in milling or additive manufacturing, where the angle of the work surface with the horizontal can affect manufacturing performance.

[0062] In some cases, manufacturing time can be affected by thermal constraints. Indeed, certain additive or subtractive manufacturing processes require cooling times to avoid altering surface quality. To limit manufacturing time losses caused by these constraints, it is advantageous for the manufacturing system to be able to assess the proximity of two voxels so that it can operate in distant zones if heat needs to be diffused to a specific area of ​​the surface.

[0063] Thus, according to one embodiment, a parameter in the form of a scheduling index is assigned to each voxel to allow the manufacturing system 4 to determine the location of one point relative to another. This parameter can allow the manufacturing system to choose to continue operating in a certain area after verifying that the voxels in that area are sufficiently far from an area requiring cooling time.

[0064] As mentioned above, the objects referred to in the context of the present invention are sometimes called "fractal" objects because of their very fine resolution, which can only be faithfully rendered digitally as very dense point clouds and is therefore difficult or even impossible to process with traditional CAD / CAM methods. In particular, a fractal object may, for example, possess similar details at arbitrarily small or large scales, be too irregular to be effectively described in traditional geometric terms, or possess intrinsic inhomogeneity.

[0065] In manufacturing terms, the fractal surface or volumetric appearance is determined by the size of the unit of material to be removed or deposited. {PPESTA}-{5}-{PCT} In effect, the manufacturable or cost-effective unit of material is defined, in the case of physical production, by the tool, or, in the case of digital creation (fractal digital objects created by computers), by the available computing power and / or resources. The size of the unit of material is linked to cost-effectiveness because it defines manufacturing or computation times. Typically, the unit of material characterizing the manufacturing process can be: a laser impact, a chip, a grain of powder, or a polymerized particle, etc.

[0066] The process includes a third step comprising the successive transmission for each voxel of its three-dimensional attribute and at least one parameter to an additive or subtractive manufacturing system.

[0067] This transmission typically depends on the type of file used to store the voxels as well as the type of manufacturing system.

[0068] In one embodiment, the management of the set of points 2 and the associated voxels 3 is performed on a computer connected via a wired or wireless network to the manufacturing system, such that the parameters are sent to the system via the Internet and / or a private network. Alternatively or complementaryly, the computer may be integrated into the manufacturing system. Alternatively or complementaryly, the voxel parameters are transmitted from the computer to the manufacturing system via an external storage device that can be connected to both the computer and the manufacturing system.

[0069] One advantage lies in the simplicity of managing the set of points 2. The latter can be grouped into packets, typically by layer, in order to easily parallelize ("multithreading") the necessary computer calculations, typically the laser trajectories calculated by layers. {PPESTA}-{5}-{PCT}

[0070] Another advantage of this method lies in its ability to dynamically display an object on the screen for the operator defining the surface / object to be manufactured. Traditional CAD software requires loading all the points that form the digital representation of the object into memory before it can be processed. Obviously, when the number of points becomes large, displaying all of them simultaneously becomes complex, if not impossible, since they are all necessary to represent the surface. Surprisingly and advantageously, this method allows loading only a portion of the points corresponding to the digital object, since it is not represented by a surface per se, but by a point cloud. It is therefore possible to dynamically display the points needed by the operator.This dynamic display allows the corresponding software to process a much larger number of points than traditional CAD software.

[0071] Referring to Figure 1, the process includes a manufacturing step of object 1 by manufacturing system 4. For each voxel, the manufacturing system removes or deposits one unit of material based on at least one parameter of that voxel. Thus, for each point in the point set 2, one or more units of material are removed / deposited until object 1 is obtained.

[0072] The manufacturing system 4 may consist simply of a machining or additive manufacturing device, but may also include an embedded computer and means of interaction for an operator. In particular, the manufacturing system may include means of performing additional data processing tasks, including file conversion and / or the interpretation of executable files.

[0073] Figure 2 illustrates a set of voxels 3 associated with a surface point of an object 1. The voxels are symbolized by cubes representing {PPESTA}-{5}-{PCT} The unit of material to be removed / deposited during manufacturing. These voxels can typically be organized into superimposed layers.

[0074] Figure 3 also illustrates an object 1 to be manufactured, where a point on the surface is associated with a set of overlapping voxels 3. A plurality of points above the object represent a theoretical surface of the material block from which the object is manufactured. The voxels are thus arranged between the object's surface and the theoretical surface of the material block. The number of voxels 3 between the object's surface and the theoretical surface determines the machining depth.

[0075] In subtractive manufacturing, the generation of voxels 3 associated with the points of the point set 2 is generally achieved by successively adding layers of voxels above and / or below the point set 2. The number of voxels is therefore dependent on the block of material in which the machining will be performed. Thus, in some embodiments, the set of voxels can form a rectangular parallelepiped, a cylinder, or more generally a prism whose base is the point set 2, formed by layers of voxels arranged above the object to be machined.

[0076] Figure 4a illustrates such an arrangement of voxels 3 above the object 1 to be machined. Each point in the point set 2 is associated with a column of voxels extending to the surface of the material block corresponding to the first machining surface. Once the material units associated with the voxels are removed, the object 1 is produced as illustrated in Figure 4b.

[0077] Figure 5a illustrates a set of 30 voxels used for the additive manufacturing of an object according to the invention. Each voxel 3 thus corresponds, for example, to a quantity of material to be deposited / fused by a 3D printing device. In particular, the size of the voxels can be variable, so as to indicate, for example, that a larger or smaller quantity of material should be deposited at the corresponding point, or that the power of an energy beam should be greater. {PPESTA}-{5}-{PCT} or less important in order to fuse more, respectively less powder. The constraints related to the sagging of certain cantilevered portions can thus be effectively addressed by varying the parameters according to the area of ​​the object.

[0078] Similarly, Figure 5b illustrates a set of voxels 30 formed from a plurality of voxels 3 arranged vertically above the object 1 to be manufactured. As in the case of additive manufacturing, the size of the voxels can be variable, for example to indicate the power or size of the machining tool (laser power / diameter / frequency, cutting tool diameter / geometry / speed, etc.).

[0079] Figure 5c also illustrates a set of voxels 30 formed from a plurality of voxels 3 arranged vertically above the object 1 to be manufactured. However, the height of each voxel 3 is less than that of the voxels 3 illustrated in Figure 5b, indicating, for example, that less power must be applied by a machining laser to achieve a shallower machining depth.

[0080] In other embodiments including a surface roughing step, the material block may have an outer surface corresponding to a surface roughing of the surface itself.

[0081] According to one embodiment, the process may include a substep of oversampling by means of a computer of an initial set of points forming an initial digital representation of the object, by adding intermediate points between successive points of the initial set of points so as to increase the density of the initial set of points to form the point set 2.

[0082] This sub-step of densifying an initial point cloud representing object 1 to be manufactured allows, in particular, the management of {PPESTA}-{5}-{PCT} manufacturing tool offsets, for example by doubling the point density of the initial set and then deactivating points outside the object to optimize the finishing of the object's contours by avoiding the use of points located on the final contour of the object during machining / deposition.

[0083] This oversampling can define a unit linked to the unit of material to be lifted / deposited, useful for example to manage tool offset as mentioned, but also to increase precision or by allowing machining by overlapping laser shots in the case of laser machining.

[0084] In one embodiment, the process includes one or more preliminary roughing steps. Roughing refers to machining or 3D printing a rough object that subsequently requires finishing to obtain the final object. Preliminary roughing significantly reduces surface manufacturing time. This is because larger quantities of material can be deposited / removed, and the tooling speed can be increased. To be efficient and fast, roughing uses simple surfaces by definition. This allows the use of traditional numerical surface modeling methods (NURBS, meshing, triangulation, etc.) for creating the surface rough. In some cases, roughing requires high precision even when performed with a large tool.Indeed, it is the calculation by subtraction between the roughing points and the finishing points that determines the amount of material to be removed before the finish. If the roughing is not clean, the finish will retain this imperfection.

[0085] The manufacturing method (additive or subtractive) used to create the rough model does not limit the manufacturing method used to create the final object. Thus, a rough model obtained by 3D printing can then be machined, a rough model obtained by machining can then be used as a 3D printing support, or the same manufacturing method can be used to create both the rough model and the final object. {PPESTA}-{5}-{PCT}

[0086] Consequently, the manufacturing system used for the roughing stage can be the same manufacturing system as that used for surface or volume manufacturing, or alternatively, it can be a different manufacturing system.

[0087] The roughing stage involves using a computer to determine a subsample of point set 2, and then to determine a digital rough model of the object based on this subsample. Subsampling point set 2 refers to removing points from the point set to reduce its number. The resulting subset forms a coarser digital representation of the object. This approximation can be used to first create a rough object, which will then be refined by more precise machining based on point set 2.

[0088] As illustrated in Figure 6a, a digital draft 5 is determined by computer. Some voxels 3 thus correspond to points of the digital draft while others, typically located between the digital draft 5 and the representation of the object to be manufactured 1, correspond to intermediate points.

[0089] Figure 6b illustrates the same situation, but without the visualization of the voxels associated with the points in the point set. Thus, only the digital draft 5 of the representation of object 1 to be manufactured is represented.

[0090] The removal of points that produce subsampling can be homogeneous, for example by removing every other point, four out of five, nine out of ten, etc., or it can be inhomogeneous and depend on the geometry of the surface to be manufactured. Inhomogeneous removal allows for consideration of factors such as local variations in height or significant slopes in certain portions of the surface to be manufactured. This can, in some cases, lead to optimization. {PPESTA}-{5}-{PCT} manufacturing time, in fact, the more the surface roughing is similar to the surface to be manufactured, the shorter the surface manufacturing time is because the manufacturing tool used for finishing has to travel less distance and fewer layers have to be deposited / machined.

[0091] Subsampling allows for the calculation, using a computer, of a digital rough model of object 1. This calculation is performed by a computer equipped with dedicated software (e.g., CATIA, SolidWorks, etc.). The calculation of the digital rough model can be carried out using traditional meshing methods (the points of the subsample are connected to form polygons, typically triangles) or using NURBS (Non-Uniform Rational B-Splines) modeling. To avoid the use of traditional CAD or CAM software, the roughing step can also be a direct trajectory tracking between the points defining this rough surface.

[0092] Although the digital roughing is obtained from points in point set 2, it must then be vertically offset to match the material block that will be used for machining or the 3D printing surface. Thus, the digital roughing can be moved layer by layer by a certain distance (or offset). An optional verification substep could be to ensure that, once the offset is applied, no points (or a small proportion of points that do not negatively impact the final quality) from point set 2 extend beyond the offset digital roughing.

[0093] The offset digital draft can then be converted by software into a set of points vertically offset from point set 2. An alignment operation in the plane perpendicular to the vertical can optionally be performed to align the points of the point set with the corresponding points of the offset point set (i.e., the corresponding points are aligned in X and Y). In this way, a point in point set 2 and its corresponding point in the set of {PPESTA}-{5}-{PCT} offset points are aligned and separated vertically by a number of layers.

[0094] Due to factors such as tool imperfections and / or thermal deformation, the roughing operation produces a blank that does not always correspond to the theoretical point cloud used to determine the digital blank. Since the generation of the finished part depends on the quality of the blank, the latter will not be precise. In some cases, it can therefore be advantageous to scan the actual blank, typically using LiDAR. The generation of the finished part is then calculated based on this scan.

[0095] One advantage of point cloud management software lies in the ability to compare two 3D point clouds. This feature allows for the creation of predictive deformations. More specifically, it is possible to create an object and then scan it. The difference between its final shape and its theoretical shape can be subtracted from the final or preliminary point cloud to obtain an extremely precise object during a second manufacturing process. This technique can be particularly useful in post-processing, such as the post-fabrication sintering of ceramics. The resulting deformations, although repetitive, are significant and non-homogeneous. Using point clouds for manufacturing, without relying on traditional CAD technologies, greatly simplifies the steps mentioned below by creating predictive deformations.

[0096] According to one embodiment, the angle between the surface of the surface roughing and the manufacturing direction is at least 40°, preferably at least 45°. The manufacturing direction is typically vertical (with respect to a standard terrestrial reference frame) and coincides with the direction orthogonal to the manufacturing layers (in both additive and subtractive manufacturing). This manufacturing direction also typically coincides with the tool orientation. {PPESTA}-{5}-{PCT} manufacturing (e.g. laser or extrusion head for 3D printing, milling head or laser in machining).

[0097] This limitation corresponds to a restriction of the angle between the horizontal (i.e., the plane orthogonal to the machining direction) and the surface of the workpiece to a maximum of 50°. In some cases, and particularly in laser machining, an excessively large angle between the workpiece surface and the horizontal can lead to defects in the final surface quality. Specifically in laser machining, if the machined surface is too steep relative to the laser direction, the laser's impact on the surface is difficult to control in terms of geometry (diameter, depth, regularity, etc.), and the surface quality can deteriorate. It is therefore advantageous to limit this angle so that it does not exceed the laser's return angle.

[0098] An adjustment of the laser power as well as the manufacturing speed can also be combined with a reduction of the angle between the horizontal and the surface of the surface roughing.

[0099] In one embodiment, two separate manufacturing systems are used: one for the roughing step and the other for the finishing step of object 1. The roughing step is typically carried out using a subtractive manufacturing system comprising a CNC milling machine, and the finishing step is, for example, carried out using a second subtractive manufacturing system comprising a laser cutting machine.

[0100] Advantageously, the process allows for combining surface fabrication from a point cloud, as described above, with surface elements derived from traditional computer-aided design (CAD). This combination of the two techniques makes it possible, for example, to insert easily CAD-generated elements into a surface created on the basis of {PPESTA}-{5}-{PCT} Tl of a set of points containing a number of points making the use of CAD impossible. These can be, for example, indexes, logos, writing elements, etc., which are inserted onto a complex surface such as a topographic model, a watch dial, or even a piece of jewelry.

[0101] According to an embodiment illustrated in Figure 7, the process includes a substep involving the CAD generation of a surface representation 6 of a complementary object (e.g., a watch dial index). In a second substep, this surface representation is converted by computer into a complementary set of points 7 such that the point set 2 on the basis of which the voxels are created also contains points from this surface representation 6.

[0102] As illustrated in Figure 7, at least one complementary object 6 is generated by CAD and then converted into a complementary point cloud 7 so that it can be incorporated into another point cloud. The resulting complementary point set can then be used for voxel generation 3 and the rest of the manufacturing process.

[0103] To facilitate the integration of CAD elements into an object, a subsample of the point set 2 that can be processed by CAD software—that is, with a correspondingly reduced number of points—can be used to create the surface representation. As with roughing, this subsample can be used to create a digital surface using standard methods (meshing, NURBS, etc.). Through a Boolean operation, the surface representation and the digital surface are combined to form a new digital surface. By applying another Boolean operation, only the elements of the surface representation that extend beyond the digital surface can be extracted. Finally, these extracted elements can {PPESTA}-{5}-{PCT} be converted into a set of points that can be integrated into the first set of points.

[0104] Continuing with the point-by-point approach, the present method also advantageously proposes to assign, using a computer, at least one manufacturing complement attribute to each point in the set of points 2. These attributes are then used to determine at least one parameter of each voxel associated with the point in question.

[0105] A complementary manufacturing attribute can include a surface or volume type, i.e., a characteristic of the surface or volume type at the point location, e.g., color, tint, oxidation, texture, roughness, hardness, a particular category (habitation, vegetation, rigid internal point, point, flexible internal point, electrically conductive internal point, or surface edge point, finish point, roughing point, etc.), but can also include technical information such as a location index, point arrangement within a layer, elevation gain, an offset determining the distance between a point in point set 2 and the corresponding point in a digital roughing, etc. These attributes can be stored in a database associated with the first point set via a suitable computer file.

[0106] Since the additional attributes are associated with the points in point set 2 and are therefore not directly communicated as such to manufacturing system 4, the storage format can be any format. Once the parameters associated with the voxels are determined, it may, in some cases, be necessary to convert the additional attributes into files interpretable by manufacturing system 4.

[0107] According to one embodiment, a surface or volume layout attribute is associated with each point of the point set 2. {PPESTA}-{5}-{PCT} This attribute allows for specific point arrangements, notably improving the surface or volume of the object. For example, the inventor observed that purely geometric arrangements, in which patterns formed by the points are repeated (aligned grid, staggered points, etc.), can lead to a less aesthetically pleasing surface finish than when the points are arranged taking into account a random factor in each layer. Thus, the points in point set 2 can possess, as a complementary attribute, a specific position, determined in particular by means of an algorithm including a random component.

[0108] In one embodiment, an additional attribute indicating whether a point is an internal or edge point of the object is associated with each point. Indeed, edge points may require a particular type of finishing, e.g., the use of a different manufacturing tool, variation in the speed or intensity / power of the tool around these points, etc.

[0109] Advantageously, complementary attributes can be used to categorize the points in point set 2. In one embodiment, the process includes a step of classifying the points in point set 2. The points are thus grouped into categories that can be associated with the points in the form of a complementary attribute. This classification step is performed by a classification algorithm, for example, based on a machine learning algorithm or other types of artificial intelligence algorithms.

[0110] It is common for some points in point set 2 to be irrelevant, or even undesirable, for the creation of the object. For example, in the case of points from a laser scan of a geographical area, some points may correspond to elements such as vegetation, buildings, construction equipment, vehicles, soil types (e.g., water, rock, concrete, grass), etc., which are not relevant. {PPESTA}-{5}-{PCT} does not wish to include this on the final surface. The classification step allows one or more categories to be associated with each point, and it is then possible to manipulate a set of points belonging to the same category, for example, to delete them, to process them separately, to change certain manufacturing characteristics such as color or texture, etc. The categories can thus be used to create filters allowing the viewing of one category or another in appropriate software. Similarly, in the case of an X-ray scan of a human arm, the creation of hard and electrically activatable structures is possible while removing fatty areas defined as such by the tomographic analysis.

[0111] The points in point set 2 can include an additional attribute corresponding to a surface texture of object 1. Depending on the selected manufacturing device, several texturing techniques are possible. The voxel parameters corresponding to the points on the surface of the final object can therefore include a parameter defined on such an additional attribute.

[0112] In additive manufacturing, the particle size of the material used, particularly powder molten on a powder bed, the layer orientation, the tool paths during deposition, etc., can determine the surface texture. The voxel parameters are determined based on these variables to obtain the texture corresponding to the complementary attribute.

[0113] In subtractive manufacturing, and particularly in laser machining, certain techniques based on modifying the frequency, polarization, and / or vibration of the laser spot allow for the modification of the surface structure to impart specific optical, mechanical, or chemical properties. This notably enables the coloring or texturing of the surface. The voxel parameters are determined based on these variables to obtain the color or texture corresponding to the desired attribute. {PPESTA}-{5}-{PCT}

[0114] For example, laser-induced periodic surface structures (LIPSS) can be created using multibeam femtolasers. The polarization and shape of these beams allow for impacts on the surface that create ripples ranging from a few micrometers to 100 nm. The arrangement of these ripples relative to each other results in specific textures.

[0115] Other ultra-high-speed pulsed laser (femtolaser) nanostructuring techniques allow for the creation of structures on the scale of a few nanometers on the surface by generating peaks and pits. The angle between the laser polarizations, as well as the laser pulse frequency, allows for variations in the structures and therefore the textures and optical effects of the final surface. These variables can be passed as parameters to the corresponding voxels.

[0116] The trajectory of the manufacturing tool (laser spot, extrusion head, milling head, etc.) during layer deposition or machining can affect the final surface finish. The inventor has observed that orthogonal trajectories from one layer to another can lead to a poor surface finish.

[0117] Thus, according to one embodiment, the trajectories of the manufacturing tool differ between two successive layers. Advantageously, the tool's trajectory on a given layer is obtained by rotating the trajectory of the previous layer. The angle of this rotation can be chosen to minimize trajectory repetition between layers; for example, an angle whose smallest common multiple of 360 is high, or a random angle whose limits are controlled to avoid statistically significant negative or repetitive effects (e.g., 37° ± X, where X is a random number between 3 and 17). This angle can be passed as a parameter to the corresponding voxels (i.e., the voxels at the end of the trajectory). {PPESTA}-{5}-{PCT}

[0118] In subtractive manufacturing using a multi-axis milling machine, typically a 4- or 5-axis machine, it is advantageous to associate each point in point set 2 with a normal vector to the surface of the digital representation of object 1, in the form of an attribute. These normal vectors allow for the determination of an approximation surface, enabling optimization of the orientation of the manufacturing tool and / or the support of the active surface during its fabrication. Indeed, if the surface represented by point set 2 includes portions where the slope and / or depth of the surface vary significantly, the trajectory and orientation of the machining tool can be complex, and the machining time lengthy.Generating a simpler approximation surface, for example by averaging normal vectors over a portion of the surface to smooth complex areas, allows for optimization of manufacturing, particularly in terms of tool and support orientation via the 4 or 5 axes of the device.

[0119] In one embodiment, it is advantageous to be able to oversample or undersample the entire set of voxels, that is, to increase or decrease the number of voxels compared to its theoretical size corresponding to the minimum unit used for manufacturing. This over / undersampling can be local, i.e., in a particular region of the object (e.g., in a given layer or set of layers, on the contour of the final surface, etc.), or global and uniform, so as to increase or decrease the resolution of the entire voxel set. This over / undersampling can, for example, depend on the desired final surface finish (smooth, rough, or the grain size required to represent certain surface details). For example, some topographic model surfaces require a high level of detail for the appearance to meet expectations (e.g.,A body of water should be as smooth as possible, while other areas can be less detailed and still have a consistent appearance (e.g., a forest is always irregular). Oversampling can also depend on the desired accuracy for a potential surface rough draft. {PPESTA}-{5}-{PCT}

[0120] In one embodiment, it is advantageous to be able to introduce modifications, linear or non-linear, random or non-random, to the set of voxels. These modifications can be local, i.e., in a particular region (e.g., in a given layer or set of layers, on the contour of the final surface, etc.), or global and uniform. These modifications can, for example, be a function of the desired final surface state, giving it specific characteristics. For instance, the bodies of water in topographic models can contain hydrophilic microstructures (pattern repetitions) allowing the model to accept a paint color adapted to that hydrophilic material in that specific location.

[0121] In laser machining, voxel oversampling also allows for the management of adaptive laser paths according to the materials or qualities sought (point jumps, generation of random zones for each point, etc.).

[0122] According to an embodiment illustrated in Figures 8a and 8b, involving a subtractive manufacturing system from a block of material, a parameter of the surface voxels of the object, i.e. the voxels of the last machining layer, consists of a manufacturing tool power enabling the machining of a predetermined depth so as to reveal a layer of the block of material of a predetermined color.

[0123] As schematically illustrated in Figures 8a and 8b, the material block can include a plurality of superimposed layers of different colors. By machining to a predetermined depth corresponding to the depth of a given color layer, it is thus possible to create areas of a given color on the surface of object 1.

[0124] Thus, a color passed as a complementary attribute to a point in the set of points 2 can be obtained by this type of variable depth machining in the last layer. {PPESTA}-{5}-{PCT}

[0125] In one embodiment, a color per voxel is obtained by machining each voxel to a given depth based on a voxel power parameter. Alternatively or complementarily, a second, finer machining operation can be performed to machine different depths for a given surface voxel.

[0126] As illustrated in Figure 8b, the material block can comprise several superimposed layers 80, 81, 82 of different colors. For example, 3 or 4 colors such as cyan, magenta, and yellow (CMY), or red, green, and blue (RGB), or cyan, magenta, yellow, and black (CMYK). Thus, each surface voxel can have a given depth or power parameter, allowing the manufacturing system to machine at the corresponding depth to reveal the layer of the corresponding color. It is therefore possible to obtain, through additive or subtractive color mixing, a very wide range of colors for the final color of the object. It is thus possible to create physical pixels to reproduce a very broad range of colors.

[0127] Figure 9 illustrates the same principle but within a single voxel, which is then machined more finely to reveal several different layers of the material block within that same voxel. A first parameter allows machining a first layer 80 to a depth corresponding to a first color, a second parameter allows machining a second layer 81 to a second depth corresponding to a second color, and a third parameter allows machining a third layer 82 to a third depth corresponding to a third color.

[0128] It is also possible to use more than three layers, for example, four, five, six, seven, eight or more than eight layers in order to have more possible colors / renderings.

[0129] According to one embodiment, the color layers are added by deposition after a first machining step so as to {PPESTA}-{5}-{PCT} The block is covered with material whose surface relief is already partially or fully formed. After deposition, a second fine machining step reveals one of the color layers by applying a corresponding laser power.

[0130] The deposition of color layers is for example carried out by PVD (from the English "physical vapor deposition") on the block of material.

[0131] In one embodiment, the manufacturing process includes an initial step of stamping and / or casting (e.g., casting of gold or other precious metals / alloys) the block of material before machining. This initial step makes it possible, for example, to obtain certain profiles or reliefs that are complex to machine, e.g., including portions with significant curvature, very steep slope variations, etc. It is thus possible to machine according to this method using parts that have already been worked by stamping and / or casting.

[0132] When the geometry of the surface represented by point set 2 is particularly complex, for example, with numerous height variations over small distances, machining time can be considerably increased due to the toolpath and / or orientation. Therefore, it is useful to be able to estimate the machining time of such complex surfaces, in order, for example, to modify the surface to reduce manufacturing time.

[0133] According to one embodiment, the machining time is estimated based on a calculation of the fractal dimension of the surface represented by the first set of points or certain profiles of this surface. A surface profile refers to the curve determined by the points contained in a vertical plane encompassing a toolpath. This profile therefore corresponds to a path that the tool must follow during manufacturing. Depending on the complexity of the surface, a profile has a fractal dimension (e.g., Hausdorff dimension or dimension of {PPESTA}-{5}-{PCT} Minkowski) between 1 and 2. The closer the fractal dimension is to 2, the longer the manufacturing time because the tool (or the surface support) must change orientation. It is therefore possible to estimate a manufacturing time based on the specific characteristics of the manufacturing device, depending on the calculated fractal dimension.

[0134] The present invention also relates to a product obtained by the process described above.

[0135] In particular, products intended for the luxury industry are advantageously produced using this process. Indeed, the level of detail, the quality of texture and surface finish made possible by this point-by-point manufacturing process, based on point clouds containing several hundred million or even several billion points, is well suited to products such as watch parts (dials, cases, crystals, etc.), jewelry, and / or fine jewelry.

[0136] This process also makes it possible to manufacture monolithic objects which are traditionally obtained by assembly (welding, brazing, screwing, etc.).

[0137] In one embodiment, the set of points 2 forms a digital representation of a watch, jewelry, and / or precious metal component comprising one or more fastening elements such as setting settings. It is particularly advantageous in terms of cost-effectiveness and repeatability to be able to manufacture such fastening elements as a single unit with the component itself.

[0138] An example of a complex part produced by this process is a watch dial with raised motifs, one or more apertures, hour markers, and a series of bezels for decorative elements. All these elements can be designed as a single point cloud using the process described above, whether they originate from {PPESTA}-{5}-{PCT} scanning and / or computer generation (pattern and kittens) or traditional CAD (indexes).

[0139] Due to the large number of parameters and voxels, the number of possibilities for adapting the manufacturing process—e.g., surface finish, colors, profiles, machining speed, etc.—is extremely significant. Optimizing all these parameters can prove very complex due to the number of experiments and trials required, as well as the subsequent comparisons between these trials.

[0140] Thus, according to one aspect, the present invention advantageously proposes a method for optimizing these parameters based on a machine learning model (i.e. algorithm).

[0141] The optimization method includes a first phase of training a machine learning model, and a second phase of using said model during which new data

[0142] This optimization method includes the following steps: • Collect training data on a part manufactured according to the process described above using at least one sensor; • Train a machine learning model based on the collected training data; • Use the trained model to automatically determine optimal parameters based on initial data for a new object to be manufactured according to the manufacturing process described above.

[0143] The training data collection stage includes, in particular, the measurement of at least one manufacturing parameter of the object using at least one sensor such as a high-precision optical sensor, a laser scanner (e.g., lidar sensor), or other types of sensors. {PPESTA}-{5}-{PCT}

[0144] In one embodiment, the sensor(s) scan the manufactured object to produce a very high-precision digital representation in the form of a point cloud. This digital representation can then be compared to the initial (theoretical) point cloud used to generate the voxel set for manufacturing.

[0145] The data collection step may include the fabrication of sample matrices 9, as illustrated in Figure 10a. A sample matrix 9 consists of a plurality of objects (i.e., samples) 90 fabricated according to the method described above, each object having its own list of predetermined parameters different from the list of other objects in the sample matrix. In this way, it is possible to generate large quantities of samples 90 (i.e., objects in the matrix) by varying the parameters.

[0146] For example, it is possible to generate sample matrices 9 by varying a single parameter, or only a small number of parameters, so that the model can extract the manufacturing characteristics associated with that parameter or those parameters. These samples 90 are typically ordered and labeled so that an operator or a recognition system can identify them. As illustrated in Figure 10b, which is a detailed view of Figure 10a, each sample 90 has a label, for example, a number, allowing it to be identified. The label can, for example, be manufactured (machined or 3D printed) at the same time as the sample matrix 9.

[0147] The model training step is essentially based on comparing the measurements taken by at least one sensor with the theoretical values ​​corresponding to the manufacturing parameters, i.e. the parameters of each voxel in the set of voxels used to manufacture the object. {PPESTA}-{5}-{PCT}

[0148] This can include a spatial comparison of voxels "Theoretical" and "measured," for example, when the sensor is a lidar sensor. This can also include a comparison of the laser power for certain voxels (corresponding, for example, to a depth in order to reach a particular color layer as explained above or to a desired surface state) with the measured result (the corresponding color measured using an optical sensor or a surface state measured with a laser scanning microscope, for example).

[0149] Machine learning is typically based on a deep learning algorithm. Many neural network architectures are suited to this type of learning. Deep convolutional neural networks such as AlexNet, VGGNet, GoogLeNet / Inception, ResNet, and DenseNet are particularly well-suited for machine learning based on data from optical sensors and / or 3D scanners.

[0150] Since the manufacturing parameters (i.e. associated with the voxels) are known, the model training is advantageously supervised because the training data (i.e. the data measured by the sensor(s)) can be labeled (i.e. tagged) by the corresponding manufacturing parameters.

[0151] Once trained, the machine learning model can then be used in a third step to automatically determine manufacturing parameters, for example based on the material, the maximum resolution / power required, the desired manufacturing speed, the type of part manufactured (e.g., very important surface finish for a watch part vs. less important surface finish for a non-visible mechanical part), etc.

[0152] Thus, according to one embodiment, one or more sample matrices are machined to train a deep learning model. Each sample differs from an adjacent sample in this {PPESTA}-{5}-{PCT} that at least one predetermined parameter varies slightly from the corresponding parameter of the adjacent sample. As a non-limiting example, a sample matrix is ​​machined to train the model to determine the best laser power for optimal color rendering. Thus, each sample in the matrix typically consists of a portion machined with power parameters varying from one sample to another, while ensuring that each sample is the desired color (i.e., that the laser machines the color layers to obtain the desired color).

[0153] Using an optical sensor, such as a confocal microscope, the sample matrix is ​​scanned to obtain digital data associated with each sample. This data can then be labeled (or not) by an operator who decides whether the sample in question is acceptable or not (e.g., whether the color rendering is good or not) and can then be used to train the machine learning model to automate the generation of optimal parameters corresponding to a desired rendering.

[0154] Once trained, the model will therefore be able, based on the input specification of new basic parameters (e.g. material, resolution, etc.), to generate as output optimal manufacturing parameters for a desired rendering. {PPESTA}-{5}-{PCT} Reference numbers used in the figures Object Set of points Voxel Set of voxels Additive or subtractive manufacturing system; Manufacturing tool of the manufacturing system; Surface of the digital blank; Surface representation Complementary set of points Material block First layer Second layer Third layer Sample matrix Sample {PPESTA}-{5}-{PCT}

Claims

Demands 1. A subtractive manufacturing process for an object (1), comprising the steps of: obtaining by means of a computer a set of points (2) forming a digital representation of the object, the set of points comprising at least 2 million points; for each point of the set of points (2), establishing by means of a computer a set of voxels (30), each voxel (3) corresponding to a unit of material to be removed by subtractive manufacturing, and each voxel comprising a three-dimensional attribute (X,Y,Z) allowing the voxel to be located in space and comprising at least one parameter determining a unit of material to be removed by subtractive manufacturing; for each voxel (3) successively, transmitting at least one parameter of the voxel to a subtractive manufacturing system; for each voxel (3), removing by means of the subtractive manufacturing system, the unit of material corresponding to the voxel as a function of at least one parameter of the voxel.

2. Method according to claim 1, the step of obtaining the point set (2) comprising a substep of upsampling by means of a computer of an initial set of points forming an initial digital representation of the object, by adding intermediate points between successive points of the initial set of points so as to increase the density of the initial set of points to form the point set (2).

3. Method according to claim 2, the successive points of the initial set of points comprising points located on an edge of the initial digital representation of the object.

4. A method according to any one of the preceding claims, the set of points (2) comprising at least 15 million points, preferably at least 20 million points. {PPESTA}-{5}-{PCT} 5. A method according to any one of the preceding claims, comprising a surface roughing step by additive or subtractive manufacturing prior to the subtractive manufacturing step of the surface, the roughing step comprising the substeps of: using a computer, obtaining a subsampling of the point set (2) and determining a digital rough based on the subsampling; manufacturing using a subtractive manufacturing system a rough object based on the digital rough, wherein the step of removing by subtractive manufacturing the unit of material corresponding to the point based on at least one voxel parameter is carried out from the rough object.

6. A method according to the preceding claim, wherein an angle between a surface of the object blank and a manufacturing direction determined by the orientation of a manufacturing tool is at least 40°.

7. A method according to any one of claims 5 to 6, wherein the fabrication of the object blank is carried out by means of a subtractive manufacturing system comprising a CNC machining center and the unit of material corresponding to the voxel as a function of at least one parameter of the voxel is removed by means of a subtractive manufacturing system comprising a laser ablation machining center.

8. A method according to any one of the preceding claims, wherein the step of obtaining the point set further comprises the substeps of: generating by computer-aided design (CAD) a file corresponding to a surface representation of a complementary object; converting the file into a complementary point set using a computer; wherein the point set comprises at least some points of the complementary point set. {PPESTA}-{5}-{PCT} 9. A method according to any one of the preceding claims, comprising a step of: using a computer, assigning to each point of the point set (2) a complementary manufacturing attribute; at least one parameter of the voxels of the voxel set being determined on the basis of the complementary attribute of the associated point.

10. Method according to the preceding claim, the additional manufacturing attribute of each point comprising at least one element among: a color, a tint, a texture, a hardness, a roughness, an oxidation, and / or a position.

11. A method according to any one of claims 9 to 10, the additional manufacturing attribute of each point comprising a category obtained at the end of a point classification step of the set of points, preferably a point classification by a machine learning algorithm.

12. A method according to any one of the preceding claims, wherein at least one voxel further comprises one or more parameters selected from: - a dimension of the unit of matter, - the speed of a manufacturing tool in the manufacturing system, - the power of a manufacturing tool within the manufacturing system, - an angle between a manufacturing surface comprising the voxel and a manufacturing tool of the manufacturing system - a voxel ordering index allowing it to be positioned relative to other voxels, - a laser polarization, frequency and / or vibration when the manufacturing system includes a laser machining tool.

13. A method according to any one of the preceding claims, wherein the manufacturing system is a subtractive manufacturing system comprising a multi-axis machining center, wherein at least one {PPESTA}-{5}-{PCT} parameter includes a normal vector to the surface, the process further comprising the steps of: numerically determining by means of a computer an approximation surface of the surface on the basis of an average of the normal vectors assigned to each point, the approximation surface allowing mechanical orientation of the manufacturing tool and / or a manufacturing support of the surface during manufacturing.

14. A method according to any one of the preceding claims, wherein the manufacturing system is a subtractive manufacturing system from a block of material, and wherein at least one parameter of a surface voxel of the object is a manufacturing tool power, the power enabling the tool to machine a predetermined depth so as to reveal a layer of the block of a predetermined color.

15. A method according to the preceding claim, wherein the manufacturing tool power is selected from at least three different powers so as to enable the tool to machine at least three predetermined depths, each depth corresponding to a layer of the block of a predetermined color.

16. A method according to any one of the preceding claims, comprising an initial step of scanning an existing object and / or computer generation so as to produce the point set.

17. A method for training a machine learning model to generate at least one parameter of the manufacturing process according to any one of claims 1 to 16, comprising the steps of: retrieving a training dataset comprising parameter measurements of an object manufactured according to the process of claims 1 to 16, each parameter measurement being assigned to a voxel of a digital representation of the manufactured object, {PPESTA}-{5}-{PCT} train a machine learning model based on training data providing as output for each voxel, an approximation of at least one parameter of the voxel.

18. A training method according to the preceding claim, the step of retrieving the training dataset comprising machining, according to the method of any one of claims 1 to 16, a sample matrix, each sample being manufactured based on a set of voxels comprising at least one parameter different from an adjacent sample in the sample matrix.

19. A method according to any one of claims 1 to 16, the at least one parameter determining a unit of material to be removed by subtractive manufacturing being automatically generated by a machine learning model trained according to the method of any one of claims 17 to 18.

20. Product obtained by the manufacturing process according to any one of claims 1 to 16 or 19.

21. Product according to the preceding claim comprising a watch, jewelry and / or jewellery component, including at least one setting bezel. {PPESTA}-{5}-{PCT}

Images