Manufacturing of a 3D woven part with regularization of the textile structure

Abstract

One aspect of the invention relates to a computer-implemented process for manufacturing a 3D woven part from a composite material comprising the definition of a textile preform of the part.

Description

DESCRIPTION TITLE: MANUFACTURE OF A 3D WOVEN PIECE WITH REGULARIZATION OF THE TEXTILE STRUCTURE TECHNICAL FIELD OF THE INVENTION

[0001] The technical field of the invention is that of parts made from woven composite material, in particular aeronautical parts such as aircraft engine blades.

[0002] In particular, the invention relates to the manufacture of a preform of a woven composite part by defining manufacturing parameters according to the desired properties of said part. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0003] A woven composite material is an assembly comprising at least one woven textile framework called reinforcement and a binder called matrix. The reinforcement comprises strands, also called "threads", woven using a loom following a theoretical weaving topology by layer, defined according to at least two orientations also called reinforcement axes. Each layer comprises different types of threads woven according to the two reinforcement axes and forms the elementary structure of the reinforcement. The reinforcement axes are conventionally called "warp" and "weft". In general, the warp corresponds to the main weaving direction, and the weft corresponds to the transverse direction, orthogonal to the warp. The way in which the strands are interwoven, i.e. the pattern according to which said strands are woven, is conventionally called "weave" and is composed of one or more layers.

[0004] The design of a 3D woven composite part requires precise control of its properties, which arise from the structure of the weave as well as the type of material, the type of reinforcement and the manufacturing process. The development of such a part is therefore complex and begins with the design and manufacturing of a preform. The design of the preform, i.e. the 3D woven reinforcement, is then a tedious process involving numerous back-and-forths in order to find the best compromise between the imposed mechanical characteristics (warp-weft ratio, volume fraction of fibers, etc.) and the manufacturing constraints (weaves, available yarn count, position in the harness, etc.) of the department responsible for manufacturing.

[0005] Convergence towards the optimal compromise is typically performed by an operator specialized in composite weaving and responsible for ensuring the regularity of the designed preform. In particular, the operator seeks to minimize the number of warp or weft yarns that are woven (i.e., woven at the preform columns) and floated (i.e., not woven at the preform columns) over a small number of insertions, generally less than ten insertions. For example, a floated yarn over fewer than ten columns and a woven yarn over fewer than 10 columns are considered defects. In this case, the search for the compromise is made with the support of a numerical weaving model and by visualizing the programmed preform.

[0006] Defining a textile part is time-consuming, with a processing time of several months, depending on the complexity of the part. This long duration therefore has a major impact on the iterations between the specifier and the manufacturing department and, consequently, on the development times of the parts.

[0007] Several digital textile programming weaving models are known from the prior art, such as: General tools, for example NedGraphics™, do not allow for taking into account the specified parameters or constraints related to the weaving of a 3D preform. In addition, these weaving models are not suitable for the definition of large parts and / or parts of high complexity; Dedicated tools for programming 3D woven preforms, for example 3D Composite Structures™, which take into account the specifications and properties of 3D textile preforms such as the number of layers, the spacing of wefts between two columns of weft yarns, the weave characteristics, etc. These weaving models include a textile layer optimization algorithm but do not take into account manufacturing constraints or weave optimization. Tools dedicated to the programming of 3D woven preforms and allowing to optimize locally, i.e. cell by cell, the definition of the textile preform, by means of heuristic techniques, in order to respect the constraints imposed by the specifier, for example by adding or removing layers and / or choice of weaves. However, these weaving models do not allow to regularize the digital definition of the preform during optimization, thus producing discontinuities between cells and increasing the number of anomalies.

[0008] So, even if there are digital tools to automatically determine the compromise, these tools are not able to regularize the solution found.

[0009] There is therefore a need for an automatic means of generating an optimized and regular digital definition of a preform of a 3D woven composite part. SUMMARY OF THE INVENTION

[0010] The invention provides a solution to the problems mentioned above, by allowing the automatic definition of the properties of a preform while guaranteeing the regularity of these properties.

[0011] A first aspect of the invention relates to a computer-implemented method for manufacturing a 3D woven part in a composite material comprising the definition of a textile preform of the part, the method comprising: Obtain a target value of a property of the textile preform; Obtain a regularization term; Adjusting a parameter of the textile preform to conform the textile preform to the target value of the property of the textile preform, the adjustment comprising a modification of the parameter of the textile preform and being implemented from: From a comparison between the target value of the property of the textile preform and a modeled value of the property of the textile preform obtained using a numerical model of the textile preform; and From the regularization term applied to the modification of the parameter of the textile preform; Issue a command including the definition of the textile preform for the manufacture of the textile preform by a manufacturing machine, the definition of the textile preform including the set parameter.

[0012] A "preform" is a three-dimensional woven reinforcement blank of the part to be manufactured. The preform is characterized by one or more properties, which may be geometric and / or mechanical properties, such as a thickness, a volume fraction, a surface density, etc. The terminology "target value" of the preform property therefore designates the value of one of the properties of the preform which is predetermined, for example by the specifier. This target value meets a specific need and application for which the composite part is intended. The target value can therefore be defined according to the specifications of the part.

[0013] The term "numerical definition" of the preform refers to the set of properties of the preform, each of which has a value associated with it. The numerical definition therefore describes a specific draft of the woven reinforcement. In other words, it is a categorical variable, for example defined by an elementary weaving pattern such as a plain weave, a twill weave, etc. The numerical definition is in a digital format allowing it to be read by a computer and transmitted, for example via a hardware medium such as an external hard drive or via a wired or wireless connection device. The digital format is, for example, a table including the properties associated with their respective values ​​and / or a plan of the preform. The plan includes one or more schematic representations in one dimension, two dimensions and / or three dimensions of the preform.

[0014] A regularization term is a weighting coefficient that imposes regularity on the numerical definition of the preform when adjusting the preform parameter. In other words, the regularization term is used to control the incidence of anomalies in the numerical definition of the preform arising from the trade-off between the target value of the numerical definition property of the preform and the manufacturing constraints.

[0015] A "preform parameter" means a physical characteristic of said preform, the adjustment of which makes it possible to conform the definition of the preform to the target value of the preform property. For example, the preform parameter is a number of threads.

[0016] A "numerical model of the preform" is a model that allows the calculation of a property of the preform, of the same magnitude as the target value of the property of the preform, in order to compare the target value of the property of the preform and the modeled property value with each other. This model calculates the property from the manufacturing constraints of the preform, for example those of the manufacturing machine and / or related to its environment, and from the parameter of the preform. The numerical model of the preform can also take into account other predetermined parameters and / or properties, in particular imposed by the specifications, such as a preform weave, a preform thickness, a preform surface density, etc. The digital model of the preform is, for example, a digital model dedicated to the programming of 3D woven preforms and making it possible to locally optimize, i.e. cell by cell, the definition of the textile preform, subsequently called a “digital model with local optimizer”, or a meta-model of this model.

[0017] A "modification" of the preform parameter means an adjustment of the value of the parameter to be set in order to achieve the optimal compromise. For example, the modification of the preform parameter is a modification of the number of wires.

[0018] The claimed invention therefore makes it possible to determine a digital definition of the preform allowing the manufacture of the preform when the digital definition is supplied to a preform manufacturing machine, the characteristics of which correspond to the properties supplied by the specifier, and which is regular, thus minimizing the number of anomalies in the digital definition of the preform. It is therefore a regulated digital definition. The balance between compliance with the theoretical criteria, taking into account the manufacturing constraints, and minimizing the number of anomalies is obtained by taking into account the regularization term when adjusting the parameter of the preform.

[0019] Thus, the adjustment of the digital definition of the preform is fully automated and does not require operator intervention to be carried out, nor to correct the regularity of said digital definition. The proposed approach therefore has better robustness than the approaches of the art which require operator intervention.

[0020] The claimed invention also makes it possible to obtain this adjusted digital definition of the preform, by adjusting the parameter of the preform, in a short time which makes it possible to accelerate the development of new parts. This is particularly advantageous in the case of preforms of high complexity and / or large size.

[0021] Advantageously, the approach proposed in this first aspect can be applied to the entirety or a portion of the part to be manufactured; the preform will then only concern the part concerned.

[0022] Advantageously, the adjustment of the preform parameter is entirely controlled by the regularization term.

[0023] Advantageously, the proposed approach can be applied to any woven composite part, regardless of its particularities which are taken into account by the numerical model of the preform.

[0024] In addition to the characteristics which have just been mentioned in the preceding paragraph, the method according to the first aspect of the invention may have one or more additional characteristics among the following, considered individually or according to all technically possible combinations.

[0025] In one embodiment, adjusting the parameter of the textile preform comprises: Initialize the textile preform parameter; Determining, by means of a first iterative search algorithm, a set value of the parameter of the textile preform, said set value of the parameter being determined by canceling a difference between the target value of the property of the textile preform and the modeled value of the property of the textile preform taking into account the regularization term applied to the modification of the parameter of the textile preform, the first iterative search algorithm searching for the modification of the parameter of the textile preform which allows said cancellation from the initialized parameter of the textile preform; Comparing a final value of the textile preform property or another property of the textile preform to a tolerance, the final value of the textile preform property or another property of the textile preform being obtained by using the digital model of the textile preform with the set value of the parameter, and: When the final value of the textile preform property or the other textile preform property meets the tolerance, issue the command including the textile preform definition; Otherwise, modify the regularization term and re-implement the textile preform parameter adjustment with the modified regularization term.

[0026] In one embodiment, the method according to the first aspect further comprises modifying the target value of the property and implementing the method again when modifying the regularization term and implementing the adjustment of the parameter of the textile preform again does not allow the final value of the property of the textile preform to meet the tolerance.

[0027] In one embodiment, the property of the textile preform is a volume fraction of the textile preform and the first iterative search algorithm is configured to solve VF cibte - f k (JT, a) + / ?Aa = 0, with: VF target the target value of the volume fraction of the textile preform; f k the digital model of the textile preform; 1T a target value of a thickness of the textile preform; a the modification of the parameter of the textile preform; / ? the regularization term; and Aa a Laplacian of the modification of the parameter of the textile preform.

[0028] Thanks to these characteristics, the preform parameter is easy to adjust since it is sufficient to solve the defined equation to determine it. This equation is, moreover, a simple cancellation, it can be easily solved using any algorithm for finding the poles of a function. These algorithms allow, in particular, to solve this type of equation very quickly and with great precision. Preferably, the Krylov-Newton algorithm is used to solve this equation.

[0029] Furthermore, the proposed setting allows to automatically regularize the numerical definition of the preform in its entirety during its execution and allows to minimize the number of anomalies of the preform, even if it is divided into cells. In the latter case, the properties of the cells are therefore smoothed from one cell to another.

[0030] Furthermore, the regularization term makes it possible to select the optimal solution on the Pareto front of the adjustment problem to be solved, i.e. to weight the error on the volume fraction by the regularity of the solution.

[0031] The regularity of the solution is ensured if the second-order derivatives of the modification field of the parameter to be adjusted are cancelled. The Laplacian of the modification of the parameter to be adjusted serves this purpose and makes it possible to detect anomalies of the preform, in particular discontinuities and regions with strong degressivity of the active layers. A "high-degression region" refers to regions of the part where the thickness decreases rapidly from one cell to the next by shrinking several layers in these cells. A layer is active when a thread is woven into it; otherwise, the layer is inactive.

[0032] In one embodiment, the textile preform comprises one or more yarns of said composite material, the parameter of the textile preform is a number of yarns, and the modification of the parameter of the textile preform is an increase or a decrease in the number of yarns.

[0033] In one embodiment, the parameter of the textile preform is initialized to a maximum number of threads.

[0034] In one embodiment, the digital model of the textile preform takes into account a thickness of the textile preform, a weave of the textile preform and / or a surface density of the textile preform.

[0035] In one embodiment, the textile preform comprises a plurality of cells, each cell being associated with a value of the parameter of the textile preform, said parameter of the textile preform being adjusted for each cell of the plurality of cells.

[0036] In one embodiment, the digital model of the textile preform is a meta-model modeling a digital weaving model, the meta-model taking as input the modification of the parameter of the textile preform and producing as output the modeled value of the property of the textile preform.

[0037] In one embodiment, the digital model is a third-order polynomial regression applied to a database created by a plurality of different modeling runs performed with the digital weaving model.

[0038] By using this meta-model, the adjustment of the preform parameter is almost instantaneous or very fast since the meta-model is an analytical function that replaces the numerical model of the preform. In addition, the use of the meta-model makes it possible to make the adjustment problem continuous. That is to say, if the modification of the preform parameter is of discrete values, for example the number of threads to remove, then the meta-model allows these values ​​to be made continuous, for example to achieve additions / removals of a positive real and not an integer of threads, thus simplifying and accelerating the adjustment.

[0039] Furthermore, when the regularization term is zero, the optimal solution coincides with the solution obtained by the weaving model optimization algorithm, but with much shorter computation times.

[0040] A second aspect of the invention relates to a computer program product comprising instructions which, when the program is executed on a computer, cause the latter to implement the steps of the method according to the first aspect.

[0041] A third aspect of the invention relates to a computer-readable recording medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to the first aspect.

[0042] The invention and its various applications will be better understood by reading the following description and examining the accompanying figures. BRIEF DESCRIPTION OF THE FIGURES

[0043] The figures are presented for information purposes only and in no way limit the invention. Figure 1 is a block diagram illustrating the sequence of steps of a method according to the invention. Figure 2 is a set of images representing a property of a blade preform by applying the method of Figure 1 for four different situations. Figure 3 is a set of images representing the preform anomalies for each situation in Figure 2. Figure 4 is a set of images representing a preform error for each situation in Figures 2 and 3. Figure 5 is a set of images of an enlargement of a portion of the blade and associated characteristics for a given situation. Figure 6 is an image of the anomalies of one of the situations in Figure 5. Figure 7 is a set of images of an enlargement of a portion of the blade and associated characteristics for a given situation. Figure 8 is a set of images of an enlargement of a portion of the blade and associated characteristics for a given situation. DETAILED DESCRIPTION

[0044] Unless otherwise specified, the same element appearing in different figures has a single reference.

[0045] The invention relates to a method for manufacturing a 3D woven part in a composite material comprising the definition of a textile preform of the part. This part is to be manufactured for a specific application and must meet a certain number of requirements of a specification. These specifications specify in particular the properties that the manufactured part must have to be suitable for its use. Tolerances can be defined in the specifications, the characteristics of the part thus being able to support a margin of approximation with respect to the imposed requirements. The characteristics of the textile preform must therefore also meet these requirements. In the following, the term "preform" means "textile preform".

[0046] Unlike the known approaches of the prior art, the method described herein makes it possible to find a regulated digital definition of the preform by taking into account the regularity of said digital definition. The search for the regulated digital definition thus takes into account the requirements of the specifications, the manufacturing constraints and the regularity of the digital definition. The method according to the invention thus makes it possible to minimize the number of anomalies in the preform, such as discontinuities, non-monotony or a strong degressivity of the active layers. The regularity is taken into account by introducing a regularization term into the adjustment problem to be solved.

[0047] The definition of the preform can be in the format of a digital document, for example a plan, a table, a text, etc.

[0048] As illustrated in Figure 1, the method 100 for manufacturing a 3D woven part in a composite material comprising the definition of a textile preform of the part comprises four steps numbered from 110 to 140. The method 100 is implemented by a computer (not shown). The computer comprises a processor and volatile and / or non-volatile memory. The computer includes in its memory instructions which, when executed by the processor, allow the implementation of the steps of the method 100.

[0049] The first step 110 is a step of obtaining a target value of the preform property. The target value of the preform property is obtained by the computer. The target value of the preform property is, for example, provided by the specifications.

[0050] For this purpose, the computer comprises a device for obtaining the target value of the property of the preform. This device is, for example, a port for connecting an external device to the computer, a communication network connection device through which the data can pass or a human-machine interface allowing the interaction of an operator such as a GUI graphical interface.

[0051] The obtained property may include, but is not limited to, one or more of the following properties: a volume fraction of the preform, a thickness, a fiber volume, a weave, a areal density. The areal density is defined as a ratio of the areal mass in the weft direction to the areal mass in the warp direction. The target value of the preform property may be the same over the entire preform or may vary over all or part of the definition of the preform.

[0052] The adjusted digital definition of the preform is therefore sought so as to respect the target value of the preform property.

[0053] The second step 120 is then a step of obtaining the regularization term. The regularization term is preferably a scalar but can also be a vector whose size depends on a number of cells of the preform.

[0054] The regularization term is produced by a dedicated algorithm or is provided by an operator. This involves initializing the regularization term so that the computer can implement the adjustment. For example, the regularization term is initialized to a positive or zero value.

[0055] The regularization term may be modified later, for example by incrementation, to gradually reduce the number of irregularities.

[0056] The third step 130 is then a step of adjusting a parameter of the preform to conform the preform, in particular its numerical definition, to the target value of the property of the preform. The adjustment takes into account the target value of the property of the preform, the manufacturing constraints and the regularization term to determine the adjusted numerical definition. The adjusted numerical definition is determined by adjusting the parameter of the preform. The adjustment therefore comprises a modification of the value of the parameter of the preform.

[0057] Manufacturing constraints are constraints imposed by the conditions under which the part is manufactured. These are typically constraints of the machine used to manufacture the part or the environment in which the machine is located. Manufacturing constraints are, for example, a space limit, a load limit, a machine force limit, a woven / floated length restriction, a maximum layer limit related to the harness depth, a maximum / minimum weft spacing, a homogeneous and gradual layer output, etc.

[0058] The tuning problem is based on a comparison between the target value of the preform property and a modeled value of the preform property obtained using a numerical model of the preform. The comparison also takes into account the regularization term applied to the modification of the preform parameter.

[0059] Preferably, the parameter of the preform is a number of threads of said preform. In this case, the modification of the parameter of the preform is a decrease or an increase in the number of threads of the preform. Alternatively, the parameter of the preform is a number of layers and the modification of the parameter is, for example, the removal of a layer, the division of a layer into several layers, the addition of a layer, etc. The possible modifications are dependent on the weave of the preform.

[0060] Setting step 130 consists of three substeps numbered 131 through 133. These three substeps can be performed multiple times, if necessary, to find the set numerical definition.

[0061] The first sub-step 131 is a step of initializing the parameter of the preform. The initialization can be done randomly, for example by following a normal distribution over an interval of values ​​including a predefined maximum value and a predefined minimum value. These two values ​​can be defined automatically by the computer according to the requirements of the specifications and / or manufacturing constraints. These two values ​​can also be defined by an operator.

[0062] Preferably, the preform parameter is the number of threads and, at initialization, the number of threads is a maximum number of threads. The maximum number of threads corresponds to the maximum quantity of threads that can be woven according to the requirements of the specifications and taking into account manufacturing constraints. In particular, the maximum number of threads is a function of the thickness of the part and the cross-section of a thread. The maximum number of threads also depends on the weave of the preform.

[0063] Substep 132 is then a step for determining an adjusted value of the preform parameter. The adjusted value of the parameter is obtained by canceling a difference between the target value of the preform property and the modeled value of the preform property taking into account the regularization term applied to the modification of the preform parameter.

[0064] Preferably, this search for the set value of the parameter is carried out by means of a first iterative search algorithm. This algorithm has the function of searching for the modification of the parameter of the preform which allows said cancellation from the initialized parameter of the preform, that is to say from the initial value of the parameter determined in the previous sub-step 131. In other words, when the parameter of the preform is a number of threads, the algorithm searches for the number of threads to be added or removed so that the defined preform complies with the requirements of the specifications.

[0065] The first iterative search algorithm is preferably an optimization algorithm such as gradient descent, simulated annealing or a Krylov-Newton method.

[0066] Preferably, the equation to be canceled is VF cibie - f k (IT, a) + a = 0, with: VF ciblethe target value of the volume fraction of the preform; f k the digital model of the preform; 1T a target value of the preform thickness; has the modification of the preform parameter; the regularization term; and has a Laplacian of the modification of the preform parameter.

[0067] The target value of the preform volume fraction and the value modeled by the numerical model of the preform are dimensionless. The regularization term has a dimension equal to the inverse of the Laplacian dimension. Initializing the regularization term to a positive or zero value favors the adjustment of the volume fraction and does not take the Laplacian into account during the first iteration.

[0068] Solving the equation allows finding a balance between the error of the modeled value of the preform volume fraction with respect to the target value of the volume fraction, and the regularity of the solution. The uniqueness of the solution of the equation, with respect to the modification of the preform, is guaranteed by the monotonicity of the error on the target volume fraction. The monotonicity of the error is guaranteed by minimizing the Laplacian when solving the equation.

[0069] The adjustment problem, i.e. the previous equation, can be solved on the entire preform or on one or more portions of the preform.

[0070] Furthermore, the adjustment problem can be applied globally to the preform or only in several cells. In the latter case, the preform is divided into a plurality of small cells, and the adjustment problem amounts to solving the previous equation in each cell of the plurality of cells. For this purpose, the regularization term can be a scalar or a vector whose size is equal to the total number of cells. The adjustment can optionally be carried out in a part of the cells of the plurality of cells, or even in a single cell; the size of the vector of the regularization term is then adapted to the number of cells in which the adjustment is carried out.

[0071] A cell has dimensions between 0.1 mm and 20 mm in width and 0.1 mm and 20 mm in height. For example, each cell has a width of 2 mm and a height of 10 mm.

[0072] In the case where the preform is subdivided into the plurality of cells, the preform parameter is assigned to each cell, independently of each other. In other words, each cell is associated with a value of the preform parameter. The preform parameter is then set individually for each cell of the plurality of cells and the regularization term ensures the regularity of the preform definition at the global level of the preform, between adjoining cells.

[0073] Also, each cell is assigned the property or, if applicable, properties. The target value of the preform property can be constant over the entire part or can vary from one cell to another. The cell can be assigned one or more types of armor, in the form of a main armor and one or more secondary armors. The other parameters / properties of the preform can also be different from one cell to another.

[0074] When the adjustment is made globally on the preform, the preform parameter as well as its modification apply to the preform as a whole.

[0075] The regularity of a cell is defined by the Laplacian of the modification of the parameter to be adjusted. The Laplacian is preferably determined by a numerical solver from the numerical model of the preform. For example, the solver is a finite difference model of order 2. When the Laplacian is zero, the optimal solution of the equation is monotonic.

[0076] The numerical model of the preform allows the calculation of the modeled value of the preform property from the manufacturing constraints, the preform parameter and, possibly, other properties of the preform imposed by the specifications. In particular, the numerical model of the preform determines a value of the volume fraction from the modification of the parameter, for example the modification of the number of wires, taking into account the manufacturing constraints and possibly properties other than the volume fraction of the preform, for example the thickness of the preform, the armor of the preform, etc.

[0077] The numerical model of the preform can be the numerical model with local optimizer. Preferably, the numerical model of the preform is a meta-model, also called a substitution model, constructed from the numerical model with local optimizer. The meta-model replaces the numerical model with local optimizer to produce the value of the volume fraction of the preform to be compared to the modeled value of the volume fraction of the preform. The meta-model therefore models the solver included in the numerical model with local optimizer. The interest is to drastically reduce the calculation times since once the meta-model is constructed, it behaves like an analytical function and the estimation by its means is almost instantaneous.

[0078] Another advantage is that the meta-model allows, through its extrapolation capacity, to make the adjustment problem continuous. Indeed, the numerical model can only take into consideration properties and parameters of the preforms that are physically realistic. In particular, the numerical model can only consider an integer number of wires and / or layers, forcing the tuning to a search among discrete values ​​for these parameters. The meta-model, acting as an interpolation or regression function, solves this drawback by allowing continuous extrapolation to intermediate solutions with discrete values. This reduces the complexity of the tuning problem related to the discretization of the preform parameter.

[0079] The meta-model is built from a database using a linear, non-linear, kernel, etc. regression or interpolation mechanism. The meta-model can alternatively use a more sophisticated learning mechanism, for example that of a neural network.

[0080] The database used by the meta-model comprises a plurality of samples determined by the numerical model to be substituted. Each sample associates a volume fraction with one or more properties of the preform, one or more parameters of the preform and / or one or more constraints, which made it possible to determine said volume fraction via the numerical model to be substituted. The plurality of samples therefore forms a search space for the volume fraction, the different dimensions of which are the information used as input to calculate the volume fraction.

[0081] The meta-model requires as input data, at a minimum, the modification of the preform parameter. Each sample therefore associates a modeled value of volume fraction with a modification of the preform parameter.

[0082] Samples are, for example, defined randomly or pseudorandomly, such as by Latin hypercube sampling. Alternatively, samples can be defined as being regularly distributed according to ranges of values ​​defined for each input quantity of the meta-model, i.e., the one or more properties, the one or more parameters and / or the one or more constraints.

[0083] Preferably, the meta-model is a third-order polynomial regression of the database that associates a volume fraction with a modification of the parameter to be adjusted. Preferably, the samples of the database associate a volume fraction VF with a thickness IT, the inverse of the thickness, a modification of the preform parameter a and the inverse of a modification of the parameter of the preform a. The meta-model is then such that f(JT, a): -> VF. The database includes at least 100 samples, preferably at least 1000 samples.

[0084] Since the effect of modifying the parameter to be adjusted on the volume fraction is highly dependent on the armor used, it is alternatively possible to build several meta-models, one for each armor of interest. Armors of interest are a plurality of armors that can be used to manufacture the preform. These armors of interest are, for example, defined according to the requirements of the specifications and / or manufacturing constraints. Armors of interest can be different between two different parts and / or preforms to be manufactured. The modification of the preform parameter can also be different for one, several or each armor of interest. For each armor, the modification of the preform parameter is then defined such that a E Ufc=i[Lfc-i, L fc [ where L k is a predetermined set of parameter modifications for each key signature, L o= 0 and A is the total number of weaves of interest. The predetermined set is defined from the specifications requirements and / or manufacturing constraints. When the parameter is the number of wires, then L k ER + and is a range of values ​​indicating the number of wires that can be added or removed from the preform.

[0085] In an embodiment where the preform parameter is the number of threads, each sample in the database associates a modeled value of the volume fraction with a decrease in the number of threads from the maximum number of threads. One advantage is to simplify the construction of the meta-model since it is then not necessary to have samples for an increase in the number of threads but only for a decrease in the number of threads.

[0086] After the first iterative search algorithm has set the parameter of the preform, the computer implements sub-step 133 which is a step of comparing the definition of the preform, or a final value of one of the properties of said definition, with one or more tolerances. The final value of one of the properties corresponds to the value of the property of the preform after setting the parameter. The final value of each property is therefore associated with the set parameter of the preform. The final value of each property is obtained by using the digital model of the preform with the set value of the parameter, i.e. the preform is modeled using the model with the set value of the parameter. These Tolerances can be those defined in the specifications. These tolerances indicate whether the preform defined with the set parameter complies with the requirements of the specifications. These tolerances are, for example, defined in the form of thresholds or ranges of values.

[0087] Preferably, two tolerances are used: the first is a tolerance on a deviation between the final value of the preform property, i.e. the modeled value of the preform property taking into account the set value of the preform parameter, and the target value of the preform property; the second is a tolerance with respect to a maximum number of anomalies. A maximum number of anomalies is then another property of the preform.

[0088] The first tolerance is, for example, a range of values ​​defined by a minimum value lower than the target value of the preform property, and a maximum value higher than the target value of the preform property so as to include the target value of the preform property in the value range. When the set value of the preform property is within the value range, then the resulting preform complies with the specification, otherwise it does not.

[0089] The second tolerance is, for example, a predetermined threshold that indicates the maximum number of anomalies beyond which the preform does not conform to the specifications. When the number of anomalies of the preform is less than the predetermined threshold, the preform after adjustment conforms to the specifications. The number of anomalies is calculated from the distribution of the modifications of the parameter of the preform in the definition of the preform. In particular, an anomaly is a non-monotonicity of the solution of the equation on a plurality of consecutive cells; the plurality of consecutive cells comprises between 2 and 20 consecutive cells, for example 10 cells.

[0090] When all tolerances are simultaneously met, the numerical definition of the preform complies with all the requirements of the specification. This is then the adjusted numerical definition of the preform. The preform can then be manufactured in accordance with the adjusted numerical definition. The adjusted numerical definition of the preform is generated using the digital model of the preform, or another digital weaving model, with the adjusted value of the preform parameter. If the meta-model is used, the adjusted value of the preform parameter in the adjusted numerical definition can be rounded to the nearest integer. This approximation preserves the monotony and regularity of the solution. The computer puts then implements the fourth step 140 of issuing a command to the manufacturing machine to manufacture said preform from the set digital definition. The command comprises the digital definition of the preform, itself comprising the set parameter of the preform, i.e. the set value of the parameter of the preform. For this purpose, the computer comprises a device for issuing the command to the manufacturing machine. When the machine receives the command, it manufactures the preform according to the specifications of the digital definition of the preform included in the command.

[0091] When at least one of the tolerances is not met, the definition of the preform obtained by the adjustment is not compliant and is not manufactured. The regularization term must then be adjusted to iterate steps 120 and 130 of the method 100 again, until the definition of the preform after adjustment complies with the requirements. The adjustment 130 is therefore implemented again with the modified regularization term, i.e. the modified value of the regularization term.

[0092] When the first tolerance is not respected, the regularization term is reduced by a predetermined decrease.

[0093] When the second tolerance is not respected, the regularization term is increased by a predetermined increase.

[0094] The predetermined increase and predetermined decrease are set by the operator depending on the application case.

[0095] When neither the first tolerance nor the second tolerance are met, no adjusted numerical definition can be found to conform the manufacture of the preform to the specifications. In this case, the target value of the preform property is modified, i.e., the specifications of the specifications concerning this property are adjusted. The method 100 is then implemented again.

[0096] The adjustment of the regularization term can be done by means of a second iterative search algorithm. This second algorithm can be the same as the first algorithm or another algorithm. Preferably, the second algorithm is an algorithm that does not impose a priori on the nature of the problem, such as a genetic algorithm, a particle swarm, a Nelder-Mead method, a Broyden-Fletcher-Goldfarb-Shanno (BFGS) method, etc. Preferably, the second algorithm does not require the calculation of a derivative. This second algorithm makes it possible to iteratively adjust the regularization term, from of an initialization of the regularization term, by calling at each of its iterations the first algorithm which then searches for the adjusted numerical definition of the preform using the regularization term produced by the second algorithm. At the end of the first algorithm, the second algorithm compares the definition of the preform obtained with the tolerances and determines whether the value of the regularization term must be modified and repeat the setting of the first algorithm with an adjustment of the regularization term, or whether the preform complies with the requirements, in which case the manufacturing command is issued.

[0097] In other words, the second algorithm has the function of finding the minimal regularization term such that all anomalies are resolved and / or the number of anomalies respects the tolerance.

[0098] The second algorithm can be applied following two approaches: a global approach and a local approach. The global approach consists of using the regularization term as a scalar to solve the problem quickly while regularizing the preform definition also on cells without anomalies. The local approach consists of adjusting the regularization term only in cells with anomalies. In the local approach, the regularization term is progressively increased to regularize cells with anomalies in a way that preserves the spatial correlation between cells induced in the equation by the Laplacian.

[0099] When the regularization term is zero, the set value of the preform parameter is such that the final value of the preform property is equal to the target value of the preform property. If the regularization term is very large, the solution becomes monotonic.

[0100] The adjusted regularization term, obtained by the second iterative search algorithm, is case-dependent because an anomaly can be more or less critical depending on the part and the position of the anomaly on the part. The regularization term can take this criticality into account, especially when it is in vector form including a component per cell of the preform.

[0101] Furthermore, when the regularization term is in vector form, it is possible to cancel the components for cells without anomalies and to make non-zero only the components of the regularization term for cells with anomalies, allowing a sparse regularization of the preform definition.

[0102] At the end of the search for the optimal regularization term by the second iterative search algorithm, it is possible that the definition of the preform obtained via the adjustment does not respect the tolerances. In other words, no adjusted numerical definition is found to conform the manufacture of the preform to the specifications. In this case, the target value of the property of the preform is modified, that is to say that the specifications of the specifications concerning this property are adjusted. Optionally, the values ​​of other properties and / or parameters can be modified, such as that of the thickness of the preform, the surface density of the preform, the armor of the preform, etc. The method 100 is then implemented again in its entirety until the adjusted numerical definition of the preform is determined.The modification of the target value(s) of the preform property(ies) can be implemented automatically, by predefined increment or decrement, or can be implemented by an operator.

[0103] Illustrations of the definition of a blade preform after adjustment using the method 100 are provided in Figures 2 to 8.

[0104] In Figure 2, the number of active layers of the preform is indicated by the gray scale for several values ​​of the regularization term (Figure 2(a) / ? = 0, Figure 2(b) p = 20, Figure 2(c) / ? = 40 and Figure 2(d) / ? = 80). It is observed that increasing the value of the regularization term changes little the “topology” of the solution, i.e. the value of the modeled volume fraction is close to the target value. On the contrary, the regularization has a non-negligible impact on the regions presenting an anomaly, as presented in Figure 3.

[0105] In Figure 3, for this same part, the number of remaining anomalies is indicated for each value of the regularization term (Figure 3(a) / ? = 0, Figure 3(b) p = 20, Figure 3(c) p = 40 and Figure 3(d) p = 80). The following are counted: 3608 anomalies for p = 0, 1094 anomalies for p = 20, 603 anomalies for p = 40 and 382 anomalies for p = 80. It is thus observed that, for this part, the number of anomalies is inversely proportional to the increase in the regularization term. The proposed method 100 therefore makes it possible to reduce this number of anomalies by adjusting the value of the regularization term.

[0106] In Figure 4, for this same part, the error on the estimation of the volume fraction is indicated, compared to the target volume fraction, for each value of the regularization term (Figure 4(a) p = 0, Figure 4(b) p = 20, Figure 4(c) p = 40 and Figure 4(d) p = 80). Each cell of the preform is associated with a value of the grayscale, which indicates the percentage error of the volume fraction after adjustment compared to the target value of the volume fraction. The average error over all cells is then estimated at 4.3% for p = 0, 4.6% for P = 20, 4.9% for p = 40 and 5.3% for p = 80. Thus, the average error on the volume fraction of the preform cells increases little with a large increase in the value of the regularization term. The proposed approach is therefore quite robust with respect to the value of the volume fraction despite the introduction of the regularization term.

[0107] In Figures 5, 7 and 8, an enlargement of the preform on a part of its cells is presented for which several properties are evaluated. The value of the regularization term is here the same for each cell. The following are thus measured, for this enlargement and for each of Figures 5, 7 and 8: the number of active layers per cell (Figures 5(a), 7(a) and 8(a)), the error on the volume fraction with respect to the target value (Figures 5(b), 7(b) and 8(b)) and the value of the regularization term (Figures 5(c), 7(c) and 8(c)). The value of the regularization term, for Figures 5, 7 and 8 is zero for each cell, equal to 1.7 for each cell and different for each cell, respectively. It is observed that the number of active layers and the error on the volume fraction are substantially the same for the three cases presented.The use of a regularization term whose value is different for each cell therefore does not degrade the quality of the definition of the preform, compared to the case where the regularization term has a constant value for all cells.

[0108] Figure 6 shows the cells in the enlargement of Figure 5 for which an anomaly is present, after adjustment, in the case where the regularization term is zero. These cells are in black; the cells without anomaly are in gray. It is noted that Figure 6 is strongly correlated with Figure 8(c): the cells showing an anomaly in Figure 6 are those where the value of the regularization term has been modified by the method 100.

Claims

CLAIMS

1. A computer-implemented method (100) for manufacturing a 3D woven part in a composite material comprising defining a textile preform of the part, the textile preform of the part comprising one or more threads of said composite material, the method comprising: - Obtaining (110) a target value of a property of the textile preform, the property of the textile preform being a volume fraction of the textile preform; - Obtain (120) a regularization term; - Adjusting (130) a parameter of the textile preform to conform the textile preform to the target value of the property of the textile preform, the adjustment comprising a modification of the parameter of the textile preform, the parameter of the textile preform being a number of threads and the modification of the parameter of the textile preform being an increase or a decrease in the number of threads, the adjustment comprising and being implemented from: o Initializing (131) the parameter of the textile preform; o Determining (132) a set value of the parameter of the textile preform by a first iterative search algorithm configured to solve VF cible - f k (JT, a) + ?Aa = 0, with: • VP'dbie the target value of the volume fraction of the textile preform; • f k a digital model of the textile preform calculating a modeled value of the property of the textile preform; • IT a target value of a thickness of the textile preform; • modification of the parameter of the textile preform; • p the regularization term, the regularization term being a weighting coefficient which imposes a regularity on the numerical definition of the preform when adjusting the preform parameter; and • The Laplacian of the modification of the parameter of the textile preform; - Issue (140) a command comprising the definition of the textile preform for the manufacture of the textile preform by a manufacturing machine, the definition of the textile preform comprising the set parameter.

2. The method (100) of claim 1, wherein adjusting (130) the parameter of the textile preform comprises: - Comparing (133) a final value of the property of the textile preform or another property of the textile preform to a tolerance, the final value of the property of the textile preform or another property of the textile preform being obtained by using the digital model of the textile preform with the adjusted value of the parameter, and: o When the final value of the property of the textile preform or the other property of the textile preform meets the tolerance, issuing (140) the command comprising the definition of the textile preform; o Otherwise, modifying the regularization term and implementing again the setting (130) of the parameter of the textile preform with the modified regularization term.

3. The method (100) of claim 2, further comprising modifying the target value of the property and re-implementing the method (100) when modifying the regularization term and re-implementing the adjustment (130) of the parameter of the textile preform does not allow the final value of the property of the textile preform to meet the tolerance.

4. Method (100) according to one of the preceding claims, in which the parameter of the textile preform is initialized to a maximum number of threads.

5. Method (100) according to one of the preceding claims, in which the digital model of the textile preform takes into account a thickness of the textile preform, an armor of the textile preform and / or a surface density of the textile preform.

6. Method (100) according to one of the preceding claims, wherein the textile preform comprises a plurality of cells, each cell being associated with a value of the parameter of the textile preform, said parameter of the textile preform being adjusted for each cell of the plurality of cells.

7. Method (100) according to one of the preceding claims, in which the digital model of the textile preform is a meta-model modeling a digital weaving model, the meta-model taking as input the modification of the parameter of the textile preform and producing as output the modeled value of the property of the textile preform.

8. Method (100) according to claim 7 in which the digital model is a polynomial regression of order 3 applied to a database created by a plurality of different models carried out with the digital weaving model.

9. A computer program product comprising instructions which, when the program is executed on a computer, cause the latter to implement the steps of the method (100) according to one of claims 1 to 8.

10. A computer-readable recording medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method (100) according to one of claims 1 to 8.

Images