Method for monitoring the weaving of a workpiece

Abstract

One aspect of the invention relates to a method (100) for monitoring the weaving of a workpiece by an electronic weaving machine, the method comprising: - obtaining (110) configuration and weaving setpoint data delivered to the electronic weaving machine in order to weave the workpiece; - generating (120) a digital model of the electronic weaving machine by aggregating the obtained configuration and setpoint data; - delivering (140) the generated digital model of the electronic weaving machine (120) to the machine learning model in order to obtain a prediction of the tension exerted at the individual motors during the weaving of the workpiece; and - monitoring (150) the weaving of the workpiece by comparing the predicted tension exerted on an individual motor with the tension measured by the tension sensor associated with the individual motor.

Description

DESCRIPTION TITLE: Method for monitoring the weaving of a piece TECHNICAL FIELD OF THE INVENTION

[0001] The technical field of the invention is that of monitoring and managing the weaving of parts.

[0002] The present invention relates to a method for monitoring a weaving of a piece, the weaving of the piece being carried out by an electronic weaving machine. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0003] Weaving is a process of joining threads together using a weaving machine to create a fabric. It involves interlacing threads stretched lengthwise along the weaving machine, called warp threads, with threads stretched widthwise along the weaving machine, called weft threads.

[0004] Current weaving machines, which we will call in this application "electronic weaving machines", are mechanized and computer-controlled. Thanks to the use of these machines, it is now possible to obtain, in a limited time, complex woven composite material parts to be woven having specific technical properties, in particular mechanical properties. Woven composite material parts are used in particular in various fields such as aeronautics, for example for the manufacture of engine blades.

[0005] An electronic weaving machine uses a harness to hold guide bars in place and return springs to maintain yarn tension. The guide bars and return springs make up the underframe of the weaving machine. In addition, heddles are attached to the harness. The heddle harness helps hold the yarns in place to create intricate designs. The heddles attached to the harness are driven by individual motors equipped with tension sensors. The tension sensors measure the tension exerted on each motor. When the warp yarns are lifted through the harness, a space, called the shed, is created between the warp yarns. The shed is the result of the warp yarns being separated into two layers: upper and lower.

[0006] The main anomalies impacting the weaving of a piece are breakages or jamming of the warp threads, breakages of heddles or collision between a lance, leading the weft threads, and some warp threads. These weaving anomalies, which can negatively impact the quality of the woven piece or reveal a machine malfunction, are difficult to spot, given the complexity of the process. Controls such as X-ray tomography are certainly applied to the piece after its manufacture, but they are unreliable and time-consuming to perform, and cannot be applied in real time.

[0007] Some electronic weaving machines have a very large number of motors that are equipped with tension sensors. For example, the Jacquard Staubli Unival 100 machine is equipped with approximately 9,000 motors, each motor being equipped with a sensor that quantifies its force on a scale of 0 to 100%. Monitoring a weave made by such a machine is therefore complex to achieve.

[0008] It is known to monitor and manage the weaving of a piece using an alarm system directly implemented on the electronic weaving machine. This alarm system stops weaving when the value returned by one of the motor sensors is outside a predetermined range of values. This alarm system has many disadvantages. For example, this system does not take into account the specificities of the piece and / or the assembly. In addition, this system may generate a large number of false alarms and / or will not detect a large number of anomalies.

[0009] There is therefore a need to improve the monitoring and management of a piece of weaving carried out by an electronic weaving machine. SUMMARY OF THE INVENTION

[0010] The invention provides a solution to the problems mentioned above, in particular by enabling the monitoring of a piece of weaving carried out by an electronic weaving machine which takes into account the specificities of the piece and the assembly. Thus, the monitoring method according to the invention processes the data acquired during the weaving of a piece and uses a model of the loom capable of predicting the response of the machine sensors to the weaving instructions. The response predicted by the model is successively compared to the new data from the weaving in progress. The differences between the model's prediction and the new data can be reported as defects and located on a map of the piece.

[0011] Thus, an electronic weaving machine compatible with the method according to the invention has an encoder and a motor for each of the eyelets. The role of the motor is to position the eyelets in accordance with the instruction provided by the user. The data captured comes from each of the motors. The method according to the invention therefore consists of identifying the signals from the motors that are not consistent with the instruction given to the machine. In order for this operation to be carried out efficiently, it was necessary to identify the physical quantities most correlated with the signal recorded by the motor. The quantities identified are in particular the position of the eyelets for each of the insertions and the adjustment of the lower loom. Concerning the position of the eyelets for each of the insertions, when the eyelet is in a position that does not comply with the instruction, the motor reacts by trying to bring it back to the position that complies with the instruction.This reaction will induce a rapid change in the motor torque that is normally captured by the acquisition system. The position of the eyelets can be recovered, for example, from the cross-processing of the weaving board and the shed opening instructions. The adjustment of the lower loom concerns in particular the parameters concerning the fabric call, commonly called "feed", and the parameters of the guide bars. The advance of the preform can generate an overtension of the warps that can prevent the motor from correctly calling the yarn. The adjustment of the guide bars has a direct impact on the geometry of the shed openings and therefore on the condition of balance of the warp threads. Incorrect positioning can indeed lead to warps being hit with the rapier. This type of event could potentially introduce overtensions captured by the system. These two pieces of information can be stored in the weaving board.

[0012] A first aspect of the invention relates to a computer-implemented method of monitoring weaving of a piece by an electronic weaving machine, the weaving machine comprising individual motors, each individual motor among the individual motors being equipped with a tension sensor measuring the tension exerted at said individual motor, the method comprising: Obtaining weaving setup and instruction data provided to the electronic weaving machine for weaving the piece, the weaving setup and instruction data provided to the electronic weaving machine comprising: data from a weaving board, the weaving board providing a positioning of each of the threads during an insertion during the weaving of the piece, and Gathering data describing diameters of warp threads and a distribution of said warp threads between rows and columns of the harness of the weaving machine, and Data from a shed opening instruction describing high and low positions for the rails as a function of an insertion number and a position in the harness, Generate a digital model of the electronic weaving machine by aggregating the configuration and setpoint data obtained, Obtaining a trained machine learning model taking as input a digital model of the electronic weaving machine and being configured to provide as output a prediction of a tension exerted at the individual motor among the individual motors of the electronic weaving machine during weaving of the piece, the digital model of the weaving machine being a digital data structure aggregating the configuration and weaving setpoint data of the electronic weaving machine, Providing the generated digital model of the electronic weaving machine to the machine learning model in order to obtain a prediction of the tension exerted on the individual motors of the electronic weaving machine during the weaving of the piece, and Monitor the weaving of the part by comparing the prediction of the tension exerted at an individual motor among the individual motors and the tension measured by the tension sensor associated with that individual motor.

[0013] Thanks to the invention, the monitoring method takes into account the specificities of the weaving machine by integrating, for example, data describing the diameters of the different warp threads as well as their distribution between the rows and columns of the harness of the weaving machine, the definition of the harness, the type(s) of thread used, the stiffness of the springs and / or the geometry of the shed opening. The monitoring process also takes into account the specific characteristics of the woven piece by integrating, for example, data concerning the thickness and density of the weft yarn used for weaving the piece.

[0014] In addition to the characteristics just mentioned in the preceding paragraph, the method for monitoring the weaving of a piece by an electronic weaving machine may have one or more additional characteristics from among the following, considered individually or according to all technically possible combinations: the method further comprises the modification of the conditions of use of the weaving machine when a difference in tension between the prediction of the tension exerted at the individual motor among the individual motors and the tension measured by the tension sensor associated with said individual motor is greater than a predetermined threshold voltage value, the modification of the conditions of use of the weaving machine comprises at least one action from among: Identify a degradation of a component of the weaving machine, Stop the weaving machine, and Saving the position of the weaving defect on the woven piece, the electronic weaving machine takes as input different warp yarns and comprises guide bars, return springs and a harness to which heddles are attached, the heddles being driven by the individual motors and in which the obtained weaving configuration and setpoint data, the method further comprising data relating to the lower loom of the electronic weaving machine describing displacements of the guide bars and stiffnesses of the return springs located under the heddles, the piece woven by the electronic weaving machine is an aeronautical piece, obtaining the trained machine learning model comprises the sub-steps: Obtaining a history of data sets from piece weavings performed by the electronic weaving machine, each data set in the history of data sets corresponding to data collected during the weaving of the piece and including weaving setup and setpoint data provided to the electronic weaving machine and tension measured by the tension sensor at said individual motor among the individual motors of the electronic weaving machine, Generate, for each data set in the data set history, a digital model of the electronic weaving machine, said digital model being a digital data structure aggregating the configuration and weaving setpoint data of the electronic weaving machine, Obtaining a machine learning model taking as input the digital model of the electronic weaving machine and being configured to provide as output a prediction of a tension exerted at an individual motor of the electronic weaving machine during the weaving of the piece, and Train the machine learning model using a regression algorithm that penalizes, for each dataset in the dataset history, a difference between the measured voltage and the voltage prediction generated as output by the machine learning model. The regression algorithm is a gradient boosting algorithm.

[0015] The substeps that can be included in obtaining the machine learning model constitute a training process. The training process is designed to enable the machine learning model to generalize the weave monitoring to new parts. In other words, the trained machine learning model is configured to be usable in a process for monitoring weaves of parts different from the woven parts of the training data. Similarly, the training model The trained automatic machine is also configured to be used in a process for monitoring piece weavings made with new configurations of the electronic weaving machine, i.e. with configurations of the electronic weaving machine that were not in the training data.

[0016] A second aspect of the invention relates to a system configured to implement the steps of the method of monitoring a weaving of a piece by an electronic weaving machine.

[0017] A third 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 of monitoring the weaving of a piece by an electronic weaving machine.

[0018] A fourth aspect of the invention relates to a computer-readable recording medium comprising instructions which, when executed by a computer, cause the computer to implement the steps of the method of monitoring a weaving of a piece by an electronic weaving machine.

[0019] 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

[0020] The figures are presented for information purposes only and in no way limit the invention. Figure 1 shows a flowchart of an example method for monitoring a weaving of a piece according to the invention. Figure 2 shows a flowchart of an example of a step of the method for monitoring a weaving of a piece according to the invention. Figure 3 shows a graphical representation of an example of moulting instructions that can be provided to an electronic weaving machine. Figure 4 shows an example of the evolution of the height of the guide bars during the weaving of a piece by an electronic weaving machine. Figure 5 shows a graph including an example of prediction of the tension exerted at an individual motor of the electronic weaving machine during weaving of a piece and an example of tension measured at an individual motor tension among the individual motors of the electronic weaving machine during weaving of the piece. Figure 6 shows a graphical representation of data describing the diameters of the different warp threads and their distribution between the rows and columns of the weaving machine harness.

[0021] Unless otherwise specified, the same element appearing in different figures has a single reference. DETAILED DESCRIPTION

[0022] The figures are presented for information purposes only and in no way limit the invention.

[0023] Figure 1 shows a flowchart of an example of an exemplary method 100 according to the invention. The essential steps of this exemplary method are indicated by a rectangle with solid lines and the optional step is indicated by a rectangle with linked dotted lines.

[0024] The invention relates to a method 100 for monitoring the weaving of a piece, the weaving of the piece being carried out by an electronic weaving machine.

[0025] The method 100 is implemented by a computer or a processor. By "computer-implemented" is meant that the steps, or substantially all of the steps, are executed by at least one computer or processor or any other similar system. Thus, steps are performed by the computer, possibly fully automatically, or semi-automatically. In examples, the triggering of at least some of the steps of the method may be performed by user-computer interaction. The level of user-computer interaction required may depend on the intended level of automation and balanced against the need to implement the user's wishes. In examples, this level may be user-defined and / or predefined.

[0026] A typical example of a computer implementation of a method is to execute the method with a system adapted for this purpose. The system may include a processor coupled to a memory and a user interface graphical user interface (GUI), the memory having recorded on it a computer program including instructions for implementing the method. The memory can also store a database. Memory is any hardware suitable for such storage, possibly comprising several distinct physical parts.

[0027] A first step 110 of the method 100 consists of obtaining configuration and weaving instruction data provided to the electronic weaving machine for weaving the piece. The term “obtain” can be defined in the present application as “receive” or “calculate”.

[0028] The weaving setup and instruction data obtained in step 110 includes data from a weaving board. The weaving board provides the positioning of each of the yarns, including the high shed and the low shed, during each of the insertions when weaving the piece.

[0029] The weaving configuration and instruction data obtained in step 110 also include spooling data describing the diameters of the different warp threads as well as their distribution between the rows and columns of the harness of the weaving machine. Figure 6 shows a graphical representation of data describing the diameters of the different warp threads as well as their distribution between the rows and columns of the harness of the weaving machine. This spooling data makes it possible in particular to identify the motors of the weaving machine whose warp threads are most likely to interact with each other, and therefore to jam each other. In other words, this spooling data thus makes it possible to group the different motors into “columns” whose warp threads are most likely to interact with each other, and in particular to jam each other.The method 100 can thus take into account this neighborhood data between the motors to refine its predictions. The information concerning the diameters of the different warp threads is also important since it makes it possible to identify the threads having the greatest tendency to rub the neighboring threads and exerting additional weight on the motors that drive them. Indeed, the larger a thread diameter, the more the thread tends to rub the neighboring threads and exert additional weight on the motors that drive them. This snagging data can be obtained in the form of Excel files which can be converted into objects, for example with the python language, containing the relationships between each motor of the loom and its position in the harness, the position being described. by the row and column numbers, as well as the diameter of the warp threads passing through each collar.

[0030] The weaving configuration and instruction data obtained in step 110 also include shed opening instruction data. The shed opening instruction data describes the high and low positions of the heddles as a function of an insertion number and a position in the harness. These positions can be calculated from one or more machine instruction files. The weaving machine instruction files can for example be at least one of: a “*.tif” type file containing the information on the weaving weave, a “*.gof” type file containing the information on the geometry of the shed opening for each program, a “*.bmp” type file containing the programs used for each motor and each insertion over time, and a “*.shd” type file making it possible to make the link between the “*.gof” type files and the “*.bmp” type files.

[0031] A representation of the data contained in an example file containing the programs used for each motor and each insertion over time is illustrated in Figure 3 for a Staübli Unival 100 type weaving machine. Thus, Figure 3 illustrates shedding instructions that can be provided by a Staübli Unival 100 type weaving machine. The data concerning the shed opening allows for better accuracy in predicting the tension of one or more motors because it determines the amplitude of the movement of the heddles, and therefore indirectly the force applied by the motors.

[0032] In a variant, compatible with the preceding variants, the weaving configuration and instruction data obtained in step 110 also include data relating to the lower loom, i.e. the guide bars and the comb of the electronic weaving machine. The data relating to the lower loom include data describing the movements of the guide bars and data describing the rigidities of the return springs located under the heddles. Figure 4 illustrates an example of the evolution of the height of the guide bars during the 6000 insertions during the weaving of a blade of an aircraft engine. The horizontal axis of the graph corresponds to the number of insertions when weaving an aircraft engine blade and the vertical axis corresponds to the height of the guide bars. Curve 410 corresponds to the amplitude between the minimum and maximum height of the guide bars during an insertion and curve 420 corresponds to the minimum and maximum heights of the guide bars during an insertion. These data relating to the lower loom make it possible to improve the accuracy of the prediction of the tension applied to one or more motors because the movements of the reed and the guide bars apply traction to the fabric which translates into tension in the heddles. Indeed, the height of the guide bars is useful for determining the torques of the motors, because by raising or lowering the fabric, they indirectly exert traction on the warp threads which is reflected on the heddles.Similarly, the higher the loom's forward speed, the greater the traction induced on the heddles. The movements of the reed and guide bars therefore apply traction to the fabric, which translates into tension in the motors.

[0033] In an example, consistent with the previous examples, the spring stiffnesses are obtained from torque measurement data on a set of parts. Thus, an example of a procedure for obtaining spring stiffnesses consists of empirically calculating the stiffness and the empty length of each of the springs of the loom from torque measurement data on a set of parts. A first step of this example procedure consists of calculating, for each motor and each insertion, the median of the torque measurements on all the parts, which makes it possible to eliminate statistical anomalies. A second step of this example procedure consists of filtering the points, i.e. the training data, for which at least three successive insertions are associated with the same heddle height, which makes it possible to place oneself in a case of static equilibrium of the forces between the motor and the spring.A third step in this example procedure is to calculate the slope and the intercept of a linear regression between the torque and the beam height for all of these points. These two values ​​are used to characterize the stiffness and the unloaded length of the spring and can therefore be used in the remainder of the method 100.

[0034] In one example, consistent with the previous examples, the height of each heddle is obtained before, during and after each insertion. Indeed, when the electronic weaving machine does not allow continuous measurement of the value of the motor torque, this value may be dependent on the future movement of the beam. For example, when a beam moves from a low position to a high position, the motor will anticipate the upward movement from the first cycle, and as a result, the tension measured during the first insertion will be higher than when the beam was also in the low position during the second cycle.

[0035] In a variant, compatible with the previous variants, data concerning the behavior of neighboring wires, for example the wires of the same column of the harness, can also be obtained at step 110. Indeed, neighboring wires can interact with each other by friction, get stuck and therefore influence the tension of the other motors of the same column. For example, it is possible for each motor of a column to obtain the height of the beam before, during and after each insertion.

[0036] In a variant, compatible with the preceding variants, measurement data specific to each part can also be obtained at step 110. The measurement data correspond to the torque values ​​of each motor of the machine for each of the insertions. The measurement data are for example contained in files of type “*.Ret”.

[0037] In a second step 120, a digital model of the electronic weaving machine is generated by aggregating the configuration and setpoint data obtained.

[0038] In one example, consistent with the previous examples, the digital model of the electronic weaving machine can be generated 120 using different steps. For example, the file of an image, which may be of type “*.tif”, containing the information of the weaving weave, includes, for each motor and each insertion, a binary number indicating when the motor is in the low or high position. In order to retrieve this data, it is thus possible to construct a parser of this type of file which stores this heddle lifting data in the form of a table, for example a table of type “NumPy”. Each of the files of type “*.gof”, containing the information of the geometry of the shed opening for each program, can also be analyzed using a parser in order to retrieve the value in millimeters of the elevation of each heddle in the high and low positions, for each insertion, and for a set value specific. Each “*.gof” file represents a different value of the setpoint. The “*.shd” file, which makes it possible to link the “*.gof” and “*.bmp” files, can be converted by scripts, for example python scripts, into a table of correspondence between a setpoint value and a color code. Finally, the “*.bmp” file, containing the programs used for each engine and each insertion over time, can also contain, for each insertion and each engine, a color code. This color code can be stored in the form of a table, for example a “NumPy” table, using a dedicated python script. When all the information has been retrieved from the files, the first step is to combine the data from the “*.shd” file and the “*.bmp >> to match each “motor, insertion” pair not a color code, but a setpoint value. Then the data present in the “*.gof” type files are added to match each “motor, insertion” pair not a setpoint value, but the elevation values ​​in the high and low positions, coming from the “*.gof” type file, with the corresponding setpoint value, for the given motor and insertion. Concretely, the result of this operation can be stored in the form of two tables, for example “NumPy” type tables, one containing the low positions of the beams for each “motor, insertion” pair, the other the high positions for each “motor, insertion” pair. Finally, it is possible to combine this information with the “*.tif", containing the information of the weave pattern, in order to know, for each "motor, insertion" pair, whether it is necessary to choose the elevation value of the heddle in the high position or in the low position. The final result of this series of operations can for example be an array, for example a "NumPy" type array, containing for each motor and each insertion the height, for example the height in millimeters, of the corresponding heddle. This data is useful for predicting the tension of the motors. Indeed, each heddle is actuated by both a motor and a return spring. Thus, the more the heddle is pulled upwards by the motor, the more the force applied by the spring increases, the torque applied by the motor therefore also increases in reaction.

[0039] In one example, consistent with the previous examples, the 120 generation of the digital model of the electronic weaving machine may further understand data aggregation in a dedicated class of the Python language, called Loom, allowing data to be manipulated efficiently. In addition, the use of the occupied memory space has been optimized in particular by a choice of numeric types adapted to the different variables, for example integers stored in memory on 8 or 16 bits.

[0040] In one example, consistent with the preceding examples, when data relating to the lower loom of the electronic weaving machine is obtained in step 110, the generation 120 of the digital model of the electronic weaving machine may further comprise the storage of data relating to the lower loom of the electronic weaving machine. The data relating to the lower loom of the electronic weaving machine is present in a file of type “*.tif”, and indicates for each insertion the thickness of the weft yarn used, the speed of the conveyor belt moving the fabric, the information whether the fabric is moving forward or is stopped, and the height of the guide bars. This data may be retrieved and stored using a parser. The storage of this data may be carried out in different arrays, for example in an object constructed in python language comprising different arrays of type “NumPy >>.

[0041] In one example, consistent with the preceding examples, when data regarding the behavior of neighboring yarns is obtained at step 110, the generation 120 of the digital model of the electronic weaving machine may further comprise reducing the obtained data, for example, using principal component analysis. For example, when 32 motors are present per column, 96 data may be obtained for each insertion. With principal component analysis, this data may be reduced to 10 synthetic variables that may be used for the generation 120 of the digital model of the electronic weaving machine.

[0042] In one example, consistent with the preceding examples, when measurement data is obtained in step 110, the generation 120 of the digital model of the electronic weaving machine may further comprise the storage of this measurement data. This data may be retrieved and stored using a parser. The storage of this data may be carried out in different tables, for example tables of type "NumPy". The information concerning the torque of each motor of the machine for each of the insertions can be encoded for example as a 5-bit integer representing the torque at the time of insertion as a percentage of the maximum torque that the motor can apply.

[0043] The method 100 comprises a third step 130 of obtaining a trained machine learning model taking as input a digital model of the electronic weaving machine and being configured to provide as output a prediction of a tension exerted at an individual motor of the electronic weaving machine during the weaving of the piece. The digital model of the weaving machine is a digital data structure aggregating the configuration and weaving setpoint data of the electronic weaving machine.

[0044] In one example, consistent with the preceding examples, step 130 of method 100 comprises four sub-steps illustrated in Figure 2. Figure 2 shows a flowchart of an example of step 130 comprising four sub-steps 131 to 134.

[0045] A first sub-step 131 of step 130 consists of obtaining training data. The training data obtained is a history of data sets originating from a weaving of one or more parts carried out by an electronic weaving machine. The electronic weaving machine from which training data was obtained is preferably of the same type as the electronic weaving machine monitored using the method 100. For example, the training data consists of data collected on several thousand motors of an electronic weaving machine during several thousand insertions during the weaving of a part. In another example, compatible with the previous example, the data was collected during the weaving of one or more aeronautical parts such as an aircraft engine blade.Each data set in the history corresponds to the data collected during the weaving of a piece by the electronic weaving machine. Each data set in the data set history includes the configuration and weaving setpoint data of the electronic weaving machine and one or more tensions measured by tension sensors during a weaving of a piece by the electronic weaving machine. Among the measured tensions, it is possible to identify which tension measurement corresponds to a particular individual motor. Furthermore, it should be noted that within each data set, the correspondence. between the configuration and weaving setpoint data of the electronic weaving machine and the tension(s) measured by one or more tension sensors during weaving of a piece by the electronic weaving machine is possible. In other words, it is possible to know which configuration and weaving setpoint data of the electronic weaving machine corresponds to the tension measured by tension sensors during weaving of a piece by the electronic weaving machine.

[0046] Preferably, these configuration and instruction data are of the same type and composition as or similar to the configuration and instruction data used during the method 100. Thus, all of the data that can be obtained in step 110 of the method 100 can also be obtained in sub-step 131 in order to be used in sub-steps 132 to 134.

[0047] In a second sub-step 132 of step 130, a digital model of the electronic weaving machine is generated for each data set of the data set history. The digital model of the electronic weaving machine is a digital data structure that aggregates the configuration and weaving setpoint data of the electronic weaving machine of the data set considered. The digital model can also aggregate data relating to the geometry of the woven piece. The digital model allows easy and efficient manipulation of this data. It can for example be a vector of numbers. Furthermore, it should be noted that the correspondence between each generated digital model of the electronic weaving machine and each tension measured by the tension sensors at the individual motors of the machine is possible.Thus, in step 134, it is possible to find the voltage(s) measured by the voltage sensor(s) at the individual motors of the machine corresponding to the digital model of the electronic weaving machine provided as input to the machine learning model. In addition, the aggregation of the data and the generation of the digital model may be preferentially similar to steps 120 and 132. For example, if the digital model generated in step 120 contains data of three types aggregated in a specific order, the digital model generated in step 132 must also contain data of these three types aggregated in the same order. Preferably, the digital model is generated in step 132, using the same method as the method used to generate the digital model in step 120.

[0048] A third sub-step 133 of step 130 consists of obtaining a machine learning model. A machine learning model can be defined as a mathematical model having the ability to learn, for example to predict a result, from training data. In the invention, the machine learning model is a model taking as input a digital model of an electronic weaving machine as generated in step 120 and step 132. The machine learning model is configured to provide as output a prediction of one or more tensions exerted at one or more individual motors of an electronic weaving machine during the weaving of the piece.

[0049] In one example, consistent with the previous examples, the machine learning model is a combination of a large number of individual models, such as decision trees. These are hundreds or thousands of individual models that are highly dependent on each other, with a common goal.

[0050] In a fourth sub-step 134 of step 130, the machine learning model obtained in sub-step 133 is trained. The training of the machine learning model comprises a first part, during which the digital model of the electronic weaving machine generated in step 132 is provided as input in order to obtain as output a prediction of one or more tensions exerted at one or more individual motors of the electronic weaving machine during the weaving of the piece. In a second part, this prediction, as well as the tension or tensions measured during the weaving of the piece, are used by a regression algorithm. The objective of the regression algorithm is to penalize, for each set of digital data in the history of digital data sets, a difference between the tension measured by the tension sensor and the prediction of the tension generated as output by the machine learning model.Thus, the machine learning model is trained on the basis of a regression algorithm, using a history of piece weaving data produced by the electronic weaving machine, to predict the “normal” signal of the sensors based on the setpoint. For example, it is possible to use extracted measured torque values ​​stored in “*.Ret” type files. The term “normal” here means the signal measured by the tension sensors in the absence of an anomaly. This algorithm is completely agnostic of the actual operation of the electronic weaving machine.

[0051] In one example, consistent with the preceding examples, the data collected during the monitoring method 100 may further be used as training data for the example step 130 comprising the four sub-steps 131 to 134. Thus, each monitoring of a weave of a new piece may participate in the training and refinement of the machine learning model obtained using the four sub-steps 131 to 134.

[0052] In one example, consistent with the preceding examples, the training of the machine learning model is based on a gradient boosting algorithm. A gradient boosting algorithm consistent with the invention trains a large number of decision trees, in order to obtain a model capable of predicting tensions exerted on individual motors of the electronic weaving machine during the weaving of the piece, by iteratively improving the predictions through gradient descent optimization. For example, it is possible to use the LightGBM library with a model comprising several hundred or even several thousand decision trees in order to perform step 134.

[0053] Thus, step 130 makes it possible to obtain a machine learning model capable of predicting tensions exerted at the level of individual motors of the electronic weaving machine during the weaving of the piece. The trained machine learning model can therefore be used in the remainder of the method 100 for monitoring the weaving of a piece.

[0054] A fourth step 140 of the method 100 consists of providing the generated digital model of the electronic weaving machine 130 to the machine learning model in order to obtain a prediction of the tension exerted at one or more individual motors of the electronic weaving machine during the weaving of the piece.

[0055] In a fifth step 150 of the method 100, the weaving of the piece is monitored using the prediction, obtained in step 140, of the tension exerted at one or more individual motors of the electronic weaving machine during the weaving of the piece and the tension measured by one or more tension sensors. The monitoring is for example carried out by calculating the difference between the prediction, obtained in step 140, of the tension exerted at one or more individual motors of the electronic weaving machine during the weaving of the piece and the measured tension. by one or more tension sensors. In another example, the tension prediction is performed for a period comprising a predetermined number, for example between 5 and 20, of insertions and a quadratic deviation between the predicted tension and the tension measured over each period is calculated to monitor the weaving of the part.

[0056] A sixth optional step 160 of the method 100 consists of modifying operating conditions of the weaving machine when a tension difference between the prediction of the tension exerted at one or more individual motors of the electronic weaving machine and the tension measured by the tension sensor(s) is greater than a predetermined threshold tension value. Figure 5 shows a graph comprising two curves 510 and 520. The horizontal axis of the graph corresponds to the number of insertions performed and the vertical axis of the graph corresponds to the tension at one or more individual motors of the electronic weaving machine during weaving of a piece. Curve 510 is an example of a prediction of the tension exerted at one or more individual motors of the electronic weaving machine during weaving of the piece that can be obtained in step 140.Curve 520 is the tension measured at one or more individual motors of the electronic loom during the weaving of the piece. For this example, the tension prediction and the measured tension are very close for the first 100 insertions. From the 100th. ème insertion, a significant gap exists between the voltage prediction and the measured voltage. Thus, for this example, an anomaly is detected from the 100 ème insertion if this deviation is greater than the predetermined threshold voltage value. An example of a predetermined threshold voltage value is between 1 and 10 newtons.

[0057] The monitoring method 100 may be used during the weaving of a piece or after the weaving of a piece. When an anomaly is detected during the weaving of the piece, the modification of the operating conditions of the weaving machine may consist of stopping the electronic weaving machine. The stopping of the electronic weaving machine may be done automatically, semi-automatically or manually. Alternatively, when an anomaly is detected during the weaving of the piece, the position on the woven piece of the weaving defect may be saved. When an anomaly is detected after the weaving of the piece, the position on the woven piece of the anomaly may also be identified. The monitoring method 100 may also be used to identify degradation of a weaving machine component or the life of weaving machine components that drive the yarns.

[0058] In a preferred embodiment, compatible with the previous variants, the weaving monitored by the method 100 is the weaving of an aeronautical part such as an aircraft engine blade.

Claims

CLAIMS

1. A computer-implemented method (100) of monitoring weaving of a piece by an electronic weaving machine, the weaving machine comprising individual motors, each individual motor of the individual motors being equipped with a tension sensor measuring the tension exerted at said individual motor and the method comprising: - Obtaining (110) weaving configuration and instruction data provided to the electronic weaving machine for weaving the piece, the weaving configuration and instruction data provided to the electronic weaving machine comprising: o Data from a weaving carton, the weaving carton providing a positioning of each of the threads during an insertion when weaving the piece, and o Gathering data describing diameters of the warp threads and a distribution of said warp threads between rows and columns of the harness of the weaving machine, and o Data from a shed opening instruction describing high and low positions for the heddles as a function of an insertion number and a position in the harness, - Generate (120) a digital model of the electronic weaving machine by aggregating the configuration and setpoint data obtained, - Obtaining (130) a trained machine learning model taking as input a digital model of the electronic weaving machine and being configured to provide as output a prediction of a tension exerted at the individual motor among the individual motors of the electronic weaving machine during weaving of the piece, the digital model of the weaving machine being a digital data structure aggregating the configuration and weaving setpoint data of the electronic weaving machine, - Providing (140) the generated digital model of the electronic weaving machine (120) to the machine learning model in order to obtain a prediction of the tension exerted on the individual motors of the electronic weaving machine during the weaving of the piece, and - Monitor (150) the weaving of the part by comparing the prediction of the tension exerted at an individual motor among the individual motors and the tension measured by the tension sensor associated with said individual motor.

2. The method (100) of claim 1 wherein the method further comprises modifying the operating conditions of the weaving machine (160) when a tension difference between the prediction of the tension exerted at the individual one of the individual motors and the tension measured by the tension sensor associated with said individual motor is greater than a predetermined threshold tension value. [Claim s] Method (100) according to claim 1 or 2 wherein the modification of the conditions of use of the weaving machine (160) comprises at least one action among: - Identify a degradation of a component of the weaving machine, - Stop the weaving machine, and - Save the position of the weaving defect on the woven piece.

4. A method (100) according to any preceding claim wherein the electronic weaving machine takes as input different warp yarns and comprises guide bars, return springs and a harness to which heddles are attached, the heddles being driven by the individual motors and wherein the obtained weaving configuration and setpoint data (110), the method further comprising data relating to the lower loom of the electronic weaving machine describing displacements of the guide bars and stiffnesses of the return springs located under the heddles. [Claim s] Method (100) according to any one of the preceding claims in which the part woven by the electronic weaving machine is an aeronautical part. [Claim s] A method (100) according to any preceding claim wherein obtaining (130) the trained machine learning model comprises the substeps: - Obtaining (131) a history of data sets from piece weavings performed by the electronic weaving machine, each data set of the history of data sets corresponding to data collected during the weaving of the piece and including weaving setup and setpoint data provided to the electronic weaving machine and a tension measured by the tension sensor at said individual motor among the individual motors of the electronic weaving machine, - Generate (132), for each data set of the data set history, a digital model of the electronic weaving machine, said digital model being a digital data structure aggregating the configuration and weaving instruction data of the electronic weaving machine, - Obtaining (133) a machine learning model taking as input the digital model of the electronic weaving machine and being configured to provide as output a prediction of a tension exerted at an individual motor of the electronic weaving machine during the weaving of the piece, and - Train (134) the machine learning model using a regression algorithm penalizing, for each data set in the data set history, a difference between the measured voltage and the prediction of the voltage generated as output by the machine learning model.

7. Method (100) according to the preceding claim in which the regression algorithm is a gradient boosting algorithm.

8. System configured to implement the steps of the method (100) according to any one of the preceding claims.

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 any one of claims 1 to 7.

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 any one of claims 1 to 7.

Images