Optimizing the manufacturing of composite parts

Abstract

To provide a system and a method for facilitating fabrication of a composite part.SOLUTION: An illustrative method includes: loading a numerical control (NC) program (115) that directs lay-up of a tow by a fiber placement machine (130) to create a laminate (140) for curing into a composite part; extracting tow information written into the NC program; applying inputs based on the tow information to a neural network (117) that has been trained with a measurement value (119) describing tow placement within other laminates that have been laid-up; and reporting likelihood of a fabrication error that is out of tolerance, based on an output of the neural network.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to the field of manufacturing, and in particular to the manufacturing of fiber reinforced composite parts. [Background technology]

[0002] Multi-layer stacks of build materials (e.g., carbon fiber reinforced polymer (CFRP)) can be formed into various shapes and cured to produce a composite part. To facilitate the production of composite parts, fiber placement machines (e.g., advanced fiber placement (AFP) machines and automated tape layup (ATL) machines) can be utilized. For example, a fiber placement machine can lay up one or more layers of multiple tows of build materials that form a stack that is cured.

[0003] The operation of the fiber placement machine may be controlled by a numerically controlled (NC) program that directs the operation of the fiber placement machine as the tow layup progresses. The fiber placement machine may feed multiple tows at one time to the layup in a single course. The course is fed in one "run" of the fiber placement machine head across the layup. The fiber placement machine head contains multiple lanes, each capable of placing a tow, so that individual tows can be paid out or terminated mid-strip as commanded by the NC program.

[0004] The specifications for a composite part may determine the number of layers in each region of the part and may also determine the fiber orientation of each layer of the part. Because the NC programs are based on, but not governed by, the specifications, many different NC programs may be used to perform the layup for a given specification. Each NC program may apply different combinations of strip and / or machine parameters to reach the parameters of the given specification. Some NC programs may be more favorable (e.g., faster to manufacture) and some less favorable (e.g., creating out-of-tolerance conditions). It is difficult to determine in advance which NC program will be more beneficial from a manufacturing perspective compared to other NC programs. Also, for designing a simple composite part, there are an infinite number of NC programs that can be used for that design, so it is not practical to simply lay up a laminate based on each of many NC programs and compare the results. These issues are exacerbated as specifications change or evolve over time and as composite parts (e.g., aircraft wings) become very large, complex, and / or take longer to manufacture.

[0005] It would therefore be desirable to have a method and apparatus that takes into account at least some of the above-mentioned problems, as well as other possible problems. Summary of the Invention

[0006] The embodiments described herein predictively identify similarities between a new NC program that has not been used to manufacture a composite part and a prior NC program that has already been used to manufacture a composite part. Measurements taken during operation according to the prior NC program can be used to locally identify the likelihood of manufacturing errors for the proposed NC program. The techniques described herein can utilize machine learning models, such as trained neural networks, to perform such operations.

[0007] One embodiment is a method for facilitating the manufacture of a composite part, the method comprising: laminating tows with a fiber placement machine to produce a layup for curing into a composite part. The method includes loading a numerical control (NC) program that directs a layup, extracting tow information written into the NC program, applying inputs based on the tow information to a neural network trained with measurements indicative of tow placement in other laminates in the layup, and reporting the likelihood of out-of-tolerance manufacturing errors based on the output of the neural network for the NC program.

[0008] A further embodiment is a non-transitory computer readable medium embodied with instructions programmed to, when executed by a processor, cause a method for facilitating manufacturing of a composite part, the method including loading a numerically controlled (NC) program directing the layup of tows by a fiber placement machine to produce a layup for curing into a composite part, extracting tow information written into the NC program, applying inputs based on the tow information to a neural network trained with measurements indicative of tow placement in other laid-up laminates, and reporting a likelihood of out-of-tolerance manufacturing errors based on an output of the neural network for the NC program.

[0009] A further embodiment is an apparatus for facilitating manufacturing of a composite part, the apparatus including: a memory storing a numerically controlled (NC) program that directs a fiber placement machine to lay up tows to produce a layup for curing into a composite part; and a controller that identifies tow information written into the NC program, applies inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other laid-up laminates, and reports a likelihood of out-of-tolerance manufacturing errors based on an output of the neural network for the NC program.

[0010] Yet a further embodiment is a method for facilitating manufacturing of a composite part, the method including laying up a layup using a fiber placement machine, obtaining measurements of the layup indicative of locations that are out of tolerance, training a neural network based on the measurements, analyzing a numerical control (NC) program for laying up a new layup through the neural network, and generating a heat map indicative of the likelihood of locations on the layup exhibiting an out of tolerance condition based on output from the neural network.

[0011] Other exemplary embodiments (e.g., methods and computer-readable media relating to the above embodiments) are described below. The above-described features, functions, and advantages may be achieved individually in various embodiments or may be combined in yet further embodiments. The details of which will become apparent with reference to the following description and drawings. [Brief description of the drawings]

[0012] Some embodiments of the present disclosure will now be described, by way of example only, with reference to the accompanying drawings, in which the same reference signs indicate the same elements or the same types of elements, and in which:

[0013] [Figure 1] FIG. 1 illustrates a manufacturing environment in an illustrative embodiment. [Figure 2A-2B] 1 is a flowchart illustrating a method for predicting manufacturing errors for NC programs used in a manufacturing environment in an illustrative embodiment. [Figure 2C] 1 is a flowchart illustrating training of a neural network in an exemplary embodiment. [Diagram 3] FIG. 2 is a perspective view of an exemplary embodiment of a fiber placement machine. [Figure 4] FIG. 2 is a top view illustrating multiple strips being laid up by a fiber placement machine in an exemplary embodiment. [Figure 5-6] FIG. 1 illustrates two different laminates laid up according to two different NC programs in an exemplary embodiment. [Figure 7] FIG. 13 illustrates a heat map showing the likelihood of manufacturing errors at different locations in an example embodiment. [Figure 8] FIG. 1 illustrates conversion of a stripe into a binary image in an exemplary embodiment. [Figure 9] 1 is a flowchart illustrating a method for making predictions for NC programs used in a manufacturing environment in an illustrative embodiment. [Figure 10] FIG. 1 is a flowchart illustrating an aircraft manufacturing and service method in an illustrative embodiment. [Figure 11] FIG. 1 is a block diagram illustrating an aircraft in an illustrative embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] The drawings and the following description are intended to describe specific exemplary embodiments of the present disclosure. Therefore, even if not explicitly described or illustrated herein, those skilled in the art will understand that various modifications embodying the principles of the present disclosure are possible within the scope of the present disclosure. Furthermore, the examples described herein are intended to help understand the principles of the present disclosure, and should not be construed as being limited to the specifically described examples and conditions. Therefore, the present disclosure is not limited to the specific embodiments or examples described below, but is limited by the claims and their equivalents.

[0015] Composite parts, such as carbon fiber reinforced polymer (CFRP) parts, are initially laid up in layers, collectively referred to as a preform or laminate. The individual fibers within each layer of the preform are arranged parallel to one another, but the different layers may have different fiber orientations to increase the strength of the resulting composite part along different dimensions. The preform may contain a viscous resin that hardens to cure the preform and form a composite part (e.g., for an aircraft). Carbon fibers impregnated with uncured thermosetting or thermoplastic resins are called "prepregs." Other types of carbon fibers include "dry fibers" that are not impregnated with thermosetting resins, but may contain a tackifier or binder. Dry fibers are infused with resin before curing. For thermosetting resins, the process of solidifying the resin is an irreversible process called curing, while for thermoplastic resins, the resin changes to a viscous state when reheated.

[0016] 1 is a block diagram illustrating a manufacturing system 100 in an exemplary embodiment. The manufacturing system 100 includes any system, device, or component operable to generate and modify NC programs for directing the layup operations of a fiber placement machine. The manufacturing system 100 is modified to detect and report the likelihood of manufacturing errors outside of the tolerances contained in the NC program.

[0017] In this embodiment, the manufacturing system 100 includes a manufacturing control system 110 that communicates with a fiber placement machine 130 (e.g., an ATL machine, an AFP machine, etc.) via a network 120 to facilitate the layup of one or more laminates 140. For example, the manufacturing control system 110 can provide NC programs to the fiber placement machine 130 to direct the operation of the fiber placement machine 130. Each NC program 115 includes instructions to direct the fiber placement machine 130. These instructions can indicate, for example, where the head 134 pays out a strip of tow, the feed / layup speed, the temperature of the heater of the fiber placement machine 130, where to pay out or cut off the tow in a given strip, etc. A tow is a single continuous piece of prepreg tape and a strip is a group of continuous pieces of prepreg tape that are fed by one pass of the head 134 of the fiber placement machine 130.

[0018] The fiber placement machine 130 may lay up the laminate 140 according to instructions in the received NC program, and may operate sensors 132 (e.g., laser sensors, actuator sensors, cameras, IR sensors, etc.) to take measurements of the laminate 140 during and / or after layup. The measurements may indicate where the tows / tapes are placed in the laminate 140. These measurements may also indicate how the tows are physically positioned relative to one another in the layup, whether the tows are separated by gaps or include overlapping areas ("laps"), the location of twisted or unbonded tows, or other information. These measurements are transferred to the manufacturing control system 110 and stored as measurements 119 for evaluation by the neural network 117. For a large composite part, such as an aircraft wing, millions of measurements (e.g., 8 million measurements) may be taken of a single laminate and used as input to evaluate the NC program.

[0019] The manufacturing control system includes a controller 112, an interface (I / F) 114, a memory 116, and a display 118. The controller 112 manages the operation of the manufacturing control system 110 based on instructions stored in the memory 116. For example, the controller 112 can modify, train, or manipulate a neural network 117 stored in the memory 116 to evaluate the NC program stored in the memory 116. The controller 112 can also manipulate the I / F 114 to provide the NC program to the fiber placement machine 130. The controller 112 can be implemented, for example, as a custom circuit, a hardware processor executing programmed instructions, or a combination thereof. The I / F 114 can include an Ethernet interface, a Universal Serial Bus (USB) interface, a Serial Attached SCSI (SAS) interface, a wireless interface compatible with the IEEE 802.11 wireless protocol, or the like.

[0020] The neural network 117 includes a combination of layers, each containing nodes, and connection weights that link the nodes in the different layers. The neural network 117 may include, for example, one to three hidden layers, and each layer may include a set of hidden layers. The neural network 117 may be implemented as a shallow network having up to 20 neurons in each layer. For example, the neural network 117 may include a single layer network having 10 neurons. Although only one neural network 117 is shown, the memory 116 may include a variety of neural networks, each of which may be trained on a different model of fiber placement machine 130 or a different type of composite part (e.g., an aircraft wing or fuselage).

[0021] Laminate 140 includes a series of uncured tows of prepreg tape laid up in strips. As used herein, the combination of tows physically arranged for a single layer is referred to as a "ply sequence" and the combination of tow lengths in a single layer is referred to as a "ply shape." A ply sequence may be, for example, one ply shape, Or it may include multiple ply shapes.

[0022] The operation of manufacturing system 100 will now be described in detail with reference to Figures 2A-2B. In this embodiment, it is assumed that neural network 117 is currently untrained and that the designer of laminate 140 wishes to train neural network 117 in order to evaluate an NC program for laying up laminate 140.

[0023] 2A is a flow chart illustrating a method 200 for predicting out-of-tolerance manufacturing errors for NC programs used in a manufacturing environment in an example embodiment. The steps of 200 are described with reference to the manufacturing system 100 shown in Figure 1, but one of ordinary skill in the art will appreciate that method 200 may be performed in other systems. Not all steps are included in the flowcharts described herein, and other steps not shown may be included. The steps described herein may also be performed in a different order.

[0024] The method 200 is divided into a training and evaluation phase. During the training phase, the neural network 117 is trained based on measurements 119 of one or more layups (and / or resulting composite parts) already manufactured according to a known NC program to enable the neural network's 117 predictive capabilities. The training uses the known NC program as input and the measurements to validate its output. During the evaluation phase, the trained version of the neural network 117 is used to evaluate one or more NC programs. Thus, in many embodiments, the training phase is performed only once or periodically (e.g., at known time intervals or each time a new layup is measured), and the evaluation phase is performed multiple times after training is completed.

[0025] Steps 201-206 represent the learning phase. At 201, an engineer sets up a predecessor NC program. At step 202, the fiber placement machine 130 is operated according to one or more predecessor NC programs to lay up a laminate. Each of the predecessor NC programs may include an NC program for manufacturing a laminate. However, in many embodiments, it may be preferable for the predecessor NC program to be used to direct the same robot or manufacture a laminate of the same design as a new NC program that is subsequently evaluated by the neural network 117.

[0026] In step 204, tow information and parameters are extracted from the advanced NC program. In step 205, the controller 112 obtains measurements of out-of-tolerance manufacturing errors of the laminate laid up according to the advanced NC program. In one embodiment, the measurements are pre-processed to indicate the location and type of out-of-tolerance manufacturing errors detected in the laminate (e.g., based on size or degree), and can also report the time spent laying up the laminate, the conformance of the laminate to design parameters, etc.

[0027] It is understood that compliance with design parameters will vary depending on the NC program being used. For example, a design may indicate that a region should be filled with tows of prepreg tape having a given fiber orientation. However, the width of the region may not be evenly divisible by the width of the tows, so the NC program designer may have to choose between slightly overfilling or underfilling the region with tows. This can result in gaps within the region (e.g., at the region's boundaries), tows extending beyond the region's boundaries, or both.

[0028] The measurements may also include results of inspection procedures performed on the resulting composite part, such as non-destructive imaging (NDI) techniques performed by laser, camera, infrared, or ultrasonic inspection. These results may be reported on a per-site basis to indicate where manufacturing errors were found. The controller 112 may further associate the received measurements with instructions in the NC program used to design the layup. For example, the controller 112 may determine that an instruction in the NC program is for layup at a site of the layup and associate the measurements at that site with the NC program instructions. The measurements may also indicate the presence or absence of out-of-tolerance manufacturing errors (e.g., by applying a binary flag indicating whether an out-of-tolerance condition occurs at each of a number of sites). Parameters in the NC program that may be analyzed include tool path centerlines, individual start and stop positions, feed rates, heater settings, etc.

[0029] In step 206, the controller 112 trains the neural network 117 based on the extracted tow information and measurements (e.g., based on training data including measurements and prior NC programs). Training the neural network includes adjusting weights between nodes in the neural network based on measurements. For example, inputs to the neural network 117 may include, for each section of the layup, the shape or curvature of the layup, the distance from the current position to the end of the tow, the number of plies at that position, the fiber angle in the tow, the temperature of a heater on the fiber placement machine, the feed rate or lay-up rate, the relative placement of the tows in the strip, the placement of the strips in the plies, and other inputs used to predict the likelihood of gaps, overlaps, folds, kinks, and / or wrinkles. That is, the neural network 117 may be operated to generate results for a first section using a first set of inputs associated with the first section, and then additionally to generate results for a second section using a second set of inputs. In an environment where a large number of measurements are taken per stack (e.g., several thousand or more), the neural network 117 may be trained using measurements for a small number of stacks (e.g., 1-4 stacks) having a large number of measurement sites. In further embodiments, the neural network may be trained using data from a large number of stacks (e.g., using a subset of the measurements from each stack).

[0030] During training of the neural network 117, the output of the neural network (e.g., in the form of an estimated percentage likelihood that the manufacturing error is out of tolerance at each of a number of locations) may be compared to known locations of manufacturing errors contained in the training data. If the predicted locations of manufacturing errors in the output are inaccurate, the connection weights between nodes in the multiple layers of the neural network 117 are adjusted. For example, inaccurate predictions by the neural network may be associated with a high cost as specified in a cost function, which causes the connection weights to be adjusted during training. The output generated by the neural network may include, for each of a number of locations (e.g., a number of equally spaced points along a grid across the surface of the laminate), the likelihood of manufacturing errors, such as, for example, inter-tow gaps exceeding a predetermined dimension and being out of tolerance, inter-strip gaps exceeding a predetermined dimension, inter-tow overlaps exceeding a predetermined dimension (e.g., overlaps of any dimension), inter-strip overlaps exceeding a predetermined dimension (e.g., overlaps of any dimension), slow production / lay-up rates, tow wrinkles, tow twists, tow separations, fill percentages less than a predetermined amount, etc. These various parameters may also be aggregated into a single likelihood value if desired.

[0031] After training is complete, the neural network 117 may be validated to ensure that its output accurately indicates the likelihood of manufacturing errors. This may be accomplished through cross-validation techniques. For example, the neural network 117 may be operated to make predictions for NC programs that were not included in the training data. The predictions made by the neural network 117 may be compared to the actual results of layup operations performed according to the NC programs. The neural network 117 may be validated by statistically determining the accuracy of the predictions for each part. If the neural network 117 fails validation, steps 204-206 are performed again on additional training data to retrain the neural network.

[0032] Once training of the neural network 117 is complete, the neural network 117 can be used to evaluate new NC programs without the need to lay up new layups in accordance with the new NC programs. Thus, the method 200 transitions from the training phase to an evaluation phase, in which the one or more new NC programs are evaluated using the neural network 117 to predict locations of out-of-tolerance manufacturing errors that may occur when the composite part is laid up using the one or more new NC programs. The out-of-tolerance conditions are evaluated. The parameters used to define each of the out-of-tolerance conditions can be varied. For example, selecting a large parameter value for an out-of-tolerance condition may affect the desired properties of the final composite part, while selecting a small parameter value for an out-of-tolerance condition may increase production time, etc. The designer then generates at least one new NC program for evaluation. The new NC program may be, for example, an NC program for laying up a laminate according to the same design as the previous NC program and intended to be used by the same fiber placement machine as was used to lay up the laminate according to the previous NC program.

[0033] In step 207, a new NC program is designed to direct the layup of tows by the fiber placement machine to produce a layup for curing into a composite part. In step 208, the controller 112 loads the new NC program. The new NC program directs the layup of tows by the fiber placement machine to produce a layup for curing into a composite part. The new NC program is intended to create, for example, a similar composite part or the same composite part as used in the previous NC program. In step 210, the controller 112 extracts tow information (e.g., head 134 operating parameters related to or facilitating the feeding of tows) and parameters from the loaded new NC program. This may include identifying sets of instructions for each strip in the new NC program and determining inputs to the neural network 117 based on these sets of instructions. For example, the instructions may be used to calculate the inputs to the neural network 117 described above with respect to step 206.

[0034] In step 212, the controller 112 applies input from the new NC program (e.g., based on the tow information and parameters) to the neural network 117. As described above, the neural network 117 has already been trained based on measurements for a previous NC program that describes the locations of out-of-tolerance manufacturing errors in the tow placement and / or one or more other laminates that have already been laid up. Thus, the neural network 117 determines the likelihood of out-of-tolerance manufacturing errors at the locations indicated by the new NC program based on the instructions and other information in the new NC program used as input. For example, the neural network can generate location-specific results for the new NC program being evaluated. The locations analyzed may be uniformly distributed across a grid, may be centered around a predefined region of interest, or may be spaced according to a desired resolution level.

[0035] The output from the neural network 117 determined for each part can provide a likelihood of a manufacturing error. Thus, step 214 includes reporting one or more likelihoods of a manufacturing error based on the output of the neural network 117. For example, the likelihoods can be shown in a text format (e.g., a list of the likelihoods and parts), in a graphic format (e.g., a heat map of the stack), etc. This data is considered in an iterative loop in which a new NC program is produced and evaluated. If the likelihood is below a threshold range (e.g., 5 percent) (step 216), the controller 112 reports that the new NC program is ready for production (step 218). Alternatively, if the likelihood exceeds the threshold, a new NC program is redesigned and the process proceeds to step 208.

[0036] Method 200 is a technical advantage over conventional systems and techniques because it allows the output from different NC programs (for laying up laminates of the same design) to be compared from a manufacturability perspective, allowing designers to quickly evaluate NC programs for a layup design without having to consider the need to manufacture a new laminate for each NC program.

[0037] 2B is a flow chart illustrating a method 250 for predicting out-of-tolerance manufacturing errors in an exemplary embodiment. Method 250 differs from method 200 in that an operator adjusts tolerance conditions and determines the likelihood of out-of-tolerance manufacturing errors based on those tolerance conditions. In this manner, a designer can see the effect of tightening or relaxing tolerances in the design process. This method provides designers with more flexibility and can be implemented as part of an iterative design process.

[0038] According to method 250, in step 252, the controller 112 obtains definitions characterizing out-of-tolerance conditions of the laminate. For example, the definitions may indicate the shape and size of each category of out-of-tolerance condition (e.g., twists, folds, overlaps, gaps) or may indicate measurements indicative of the out-of-tolerance condition.

[0039] In step 254, the controller 112 trains the neural network 117 based on the definitions. For example, the neural network 117 may be trained based on definitions applied to measurements of previously laid-up stacks. In step 256, the controller 112 operates the neural network 117 to identify a first set of predicted locations for the new stack that are out of tolerance as defined by the numerical control (NC) program. A report indicating the first set of predicted locations is then provided to an engineer for review.

[0040] An engineer may be interested in how the number, type, and location of predicted out-of-tolerance conditions change with a revised tolerance definition. Thus, step 258 includes the controller 112 revising the definition, and step 260 includes retraining the neural network based on the revised definition. In step 262, the controller 112 operates the retrained neural network to identify a second set of predicted locations of out-of-tolerance conditions for the new laminate, and in step 264, the controller 112 generates a report comparing the first set of predicted locations and the second set of predicted locations. The report may be provided to the engineer to provide insight and feedback on the design. The report may also be provided as a heat map.

[0041] FIG. 2C illustrates a method 280 for training a neural network in an exemplary embodiment. FIG. 2C includes designing a neural network architecture (number of layers, number of nodes in each layer, and connections between nodes in different layers) in step 282, initializing weights for connections between nodes in step 284, calculating predicted values ​​based on the input in step 286, and comparing the predicted values ​​with measurements in step 288. The method further includes determining whether the correspondence between the predicted values ​​and the measurements is acceptable in step 290. If an acceptable correspondence exists, the learning may be terminated. Alternatively, if the level of correspondence is not acceptable, in step 292, the weights are updated to strengthen the correspondence between the predicted values ​​and the measurements.

[0042] Together with the explanation relating to the method for evaluating the NC program described above, Figures 3-6 show the production of a laminate using the NC program, and Figures 7-8 show various additional components of a system for evaluating the NC program.

[0043] 3 is a diagram illustrating an AFP machine 300 mounted on a support 370 in an exemplary embodiment. The AFP machine 300 includes any system or apparatus for laying up tows 352 comprised of material for curing into a composite part. 300 includes a head 380 that provides a tow 352 of curable build material (e.g., CFRP) during (e.g., simultaneously with) layup. The tow 352 is laid up to form a layup 350, which includes one or more layers of material that are cured to form a single, integral composite part. In this embodiment, the layup 350 includes an aircraft fuselage section and is held in place by a rotating holder 360. A sensor 390 measures the layup as it is being laid up and may utilize any of the sensing techniques described above. The sensor 390 may obtain measurements (for training data as described above).

[0044] As the AFP machine 300 operates to lay up tows 352 in the layup 350, the AFP machine 300 moves straight toward / away from the layup 350 along an X-axis 366, moves vertically up and down along a Y-axis 364, and moves laterally along a Z-axis 362. The layup 350 can also be rotated about its central axis and moved axially relative to the AFP machine 300. As used herein, when the AFP machine 300 lays up multiple tows 352 simultaneously during a single "pass" of the head 380, the tows 352 are collectively referred to as a single "course." A set of non-overlapping strips that are laid down sequentially is referred to as a ply. As plies are added to the layup 350, the strength of the resulting composite part is effectively increased.

[0045] Laying up material for a large laminate 350, such as a fuselage section, is a time-consuming and complex process. The operation of the AFP machine 300 is controlled by an NC program so that the tows 352 are laid up quickly and efficiently. In one embodiment, the NC program provides instructions for aligning / repositioning the AFP machine 300, moving the head 380, and laying up the tows 352 for the laminate 350, strip by strip. In this manner, by executing instructions from the NC program, the AFP machine 300 produces the laminate for curing into a composite part.

[0046] FIG. 4 is a top view of a plurality of strips 450 laid up by a fiber placement machine in an exemplary embodiment, corresponding to the arrow 4 shown in FIG. 3. As shown in FIG. 4, the head 380 advances in a direction 420 while laying up the strips 450 on the stack 400. Each strip 450 includes a plurality of tows 352. There are gaps 440 between the tows 352 within a single strip, and there are gaps 430 between the strips. If the gaps 440 or 430 exceed a predetermined limit, an out-of-tolerance manufacturing error may have occurred in the stack. Additionally, a manufacturing error may also occur if the strips 450 do not fully occupy (or over-occupy) the desired area as indicated in the tow 352 placement design. If there is a gap between the edge of the area and the edge of the adjacent tow 352, an out-of-tolerance condition may exist.

[0047] 5-6 show two different laminates laid up according to two different NC programs in an exemplary embodiment. As shown in FIG. 5, laminate 500 has gaps 530 between strips 550 and gaps 540 between tows 352 within the strips. Meanwhile, laminate 600 has a different combination of gaps 630 and gaps 640 for strips 650 laid up in a different arrangement.

[0048] 7 illustrates a heat map 700 showing the likelihood of manufacturing errors at different locations in an exemplary embodiment. The shape of the heat map 700 corresponds to the shape of the laminate that it illustrates. The heat map 700 is generated by applying inputs to the neural network 117 for each of the various locations in the laminate that will be cured into an aircraft wing, and color coding each location to indicate the likelihood of manufacturing errors. In this embodiment, the heat map 700 shows areas 710 that are prone to out-of-tolerance manufacturing errors and areas 720 that are prone to manufacturing errors. and an area 720 where errors are less likely to occur.

[0049] FIG. 8 is a diagram 800 illustrating the conversion of a strip 810 into a binary image 820 in an exemplary embodiment. A binary image 820 may be generated for each strip, with each row (or column) representing an individual lane of the strip, and the data in these rows (or columns) indicating the start and end positions of the tows in the respective lanes. As shown in FIG. 8, the individual tows 812 extend at different distances along the X direction. In each row 822 of the binary image 820, a "1" indicates the presence of a tow and a "0" indicates the absence of a tow in a particular length unit (e.g., 1 inch, 1 centimeter, etc.). Thus, in one embodiment, each strip is represented by a binary image having a predetermined width equal to the maximum number of tows in the strip. In the binary image, one pixel row corresponds to each tow. The binary image can be recast into a vector and used as input for a neural network in the form of a pattern recognition network. Also, additional input parameters can be added to this vector. In this manner, the controller 112 can operate a pattern recognition network based on the binary image 820 to predict the likelihood of out-of-tolerance gaps between strips, and between tows within a strip.

[0050] 9 is a flow chart illustrating a method 900 for making predictions for an NC program used in a manufacturing environment in an example embodiment. In step 902, the method 900 includes laying up a layup using a fiber placement machine. Step 904 includes obtaining measurements of the layup, the measurements indicating locations that are out of tolerance. Step 906 includes training a neural network based on the measurements, and step 908 includes analyzing the NC program for laying up a new layup via the neural network. Step 910 includes generating a heat map based on output from the neural network that indicates the likelihood of locations on the layup that exhibit an out of tolerance condition. <Example>

[0051] In the following examples, additional processes, systems, and methods are described in the context of a manufacturing environment for predictively analyzing NC programs to identify the likelihood of manufacturing errors.

[0052] Referring more specifically to the drawings, embodiments of the present disclosure may be described in relation to an aircraft manufacturing and service method 1000, as shown in FIG. 10, and an aircraft 1002, as shown in FIG. 11. Prior to production, the method 1000 may include specification and design 1004 of the aircraft 1002 and material procurement 1006. During production, the aircraft 1002 undergoes component and subassembly manufacturing 1008 and system integration 1010. The aircraft 1002 then undergoes certification and delivery 1012 and enters into service 1014. During its customer service life, the aircraft 1002 undergoes routine maintenance and service 1016, which may include modifications, reconfigurations, refurbishments, and the like. Apparatus and methods embodied herein may be employed at any one or more suitable stages of manufacturing and service described in method 1000 (e.g., specification and design 1004, material procurement 1006, component and subassembly manufacturing 1008, system integration 1010, certification and delivery 1012, in-service 1014, and maintenance and service 1016) and / or any suitable components of the aircraft 1002 (e.g., airframe 1018, systems 1020, interior 1022, propulsion system 1024, electrical system 1026, hydraulic system 1028, environmental system 1030).

[0053] Each step of the method 1000 described above may be performed or implemented by a system integrator, a third party, and / or an operator (e.g., a customer). Note that a system integrator may include, but is not limited to, any of the aircraft manufacturers and major system subcontractors. A third party may include, but is not limited to, any number of vendors, subcontractors, and suppliers. An operator may be an airline, a leasing company, a military entity, a service organization, etc.

[0054] 11 , an aircraft 1002 produced by method 1000 may include an airframe 1018 with a number of systems 1020 and an interior 1022. Examples of systems 1020 include one or more of a propulsion system 1024, an electrical system 1026, a hydraulic system 1028, and an environmental system 1030, as well as any number of other systems. Additionally, while illustrated in the aerospace industry, the principles of the invention may be applied to other industries, such as the automotive industry.

[0055] As previously discussed, the apparatus and methods embodied herein may be employed during any one or more stages of manufacturing and service described in method 1000. For example, the parts or subassemblies corresponding to part and subassembly manufacturing 1008 may be similarly fabricated or manufactured as parts or subassemblies manufactured during the in-service life of the aircraft 1002. Also, one or more apparatus embodiments, method embodiments, or a combination thereof may be used, for example, during subassembly manufacturing 1008 and system integration 1010 to substantially speed up or reduce the cost of assembly of the aircraft 1002. Similarly, one or more apparatus embodiments, method embodiments, or a combination thereof may be used during the in-service life of the aircraft 1002, for example, but not limited to, during maintenance and service 1016. For example, the techniques and systems described herein may be used in material procurement 1006, component and subassembly manufacturing 1008, system integration 1010, in-service 1014, and / or maintenance and service 1016, or may be used in the airframe 1018 and / or interiors 1022. These techniques and systems may also be utilized in multiple systems 1020, including, for example, propulsion system 1024, electrical system 1026, hydraulic system 1028, and / or environmental system 1030.

[0056] In one embodiment, the parts are part of the airframe 1018 and are manufactured during part and subassembly manufacturing 1008. The parts are then installed into the aircraft during system integration 1010 and used during in-service 1014 until they are worn out and are no longer usable. The parts are then scrapped and replaced with newly manufactured parts during maintenance and service 1016. Components and methods of the present invention may be utilized during part and subassembly manufacturing 1008 to evaluate new NC programs that can be used to manufacture the parts.

[0057] Any of the various control elements (e.g., electrical or electronic components) shown in the figures or described herein may be implemented as hardware, a processor executing software, a processor executing firmware, or a combination thereof. For example, an element may be implemented as dedicated hardware. A dedicated hardware element may be referred to as a "processor," "controller," or other similar terminology. When provided in the form of a processor, the functionality may be provided by a single dedicated processor, a single shared processor, or multiple individual processors, including those that may be shared. Furthermore, the explicit use of the terms "processor" or "controller" should not be construed as referring solely to hardware capable of executing software. These terms may implicitly include, but are not limited to, digital signal processor (DSP) hardware, network processors, application specific integrated circuits (ASICs) or other circuitry, field programmable gate arrays (FPGAs), read only memory (ROM) for storing software, random access memory (RAM), non-volatile storage, logic circuits, or other physical hardware components or modules.

[0058] Additionally, a control element may be implemented as instructions executable by a processor or computer to perform the function of the element. Some examples of instructions include software, program code, and firmware. The instructions run when executed by a processor and direct the processor to perform the function of the element. The instructions may be stored in a processor-readable storage device. Examples of storage devices include digital or solid-state memory, magnetic storage media such as magnetic disks or magnetic tapes, hard drives, or optically readable digital data storage media.

[0059] Furthermore, the present disclosure also includes examples according to the following appendices.

[0060] Appendix 1. A method for facilitating manufacturing of a composite part, the method comprising: loading a numerically controlled (NC) program that directs the layup of tows by a fiber placement machine to produce a layup for curing into a composite part; extracting tow information written into the NC program; applying inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other laid-up laminates; and reporting a likelihood of out-of-tolerance manufacturing errors based on an output of the neural network for the NC program.

[0061] Clause 2. The method of clause 1, further comprising subdividing the laminate into a plurality of sections, wherein for each of the plurality of sections in the laminate, a likelihood of an out-of-tolerance manufacturing error is reported.

[0062] Appendix 3. The method of Appendix 2, wherein reporting includes providing a heat map visually illustrating the laminate, wherein the likelihood of out-of-tolerance manufacturing errors in each of the plurality of portions is reported as a color in the heat map.

[0063] Addendum 4. The method of any one of Addendums 1-3, wherein the input is selected from the group consisting of the curvature of the laminate at a given location, the length of a tow, the distance from a current position to an end of a tow, the number of layers at a given location, the fiber angle within a tow, the temperature of a heater on the fiber placement machine at a given location, the relative placement of tows within a strip, and the placement of strips within a layer.

[0064] Addendum 5. The method of any one of Addendums 1-4, further comprising generating a binary image for each strip, the binary image having a row for each lane in the fiber placement machine.

[0065] Appendix 6. The method of any one of appendices 1 to 5, wherein the manufacturing error is selected from the group consisting of a gap between tows exceeding a predetermined dimension, a gap between strips exceeding a predetermined dimension, an overlap between tows exceeding a predetermined dimension, an overlap between strips exceeding a predetermined dimension, a slow production rate, tow wrinkles, tow twists, tow separation, and a fill rate less than a predetermined amount.

[0066] Addendum 7. The method of any one of Addendums 1 to 6, further comprising modifying the NC program and repeating the steps of extracting, applying, and reporting.

[0067] Appendix 8. An aircraft part assembled according to the method of any one of appendices 1 to 7.

[0068] Appendix 9. Embodying programmed instructions that, when executed by a processor, cause the processor to perform the method of any one of appendices 1-7 to facilitate manufacturing of a composite part. Non-transitory computer-readable medium.

[0069] Appendix 10. An aircraft part assembled according to the method defined by the instructions stored on a computer readable medium according to Appendix 9.

[0070] Clause 11. An apparatus for facilitating the manufacture of a composite part, the apparatus comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by a fiber placement machine to produce a layup for curing into a composite part; and a controller that performs the method of any one of clauses 1-7.

[0071] Clause 12. An apparatus for facilitating manufacturing of a composite part, comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by a fiber placement machine to produce a layup for curing into a composite part; and a controller that identifies tow information written into said NC program, applies inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other layups that have been laid up, and reports a likelihood of out-of-tolerance manufacturing errors based on an output of the neural network for said NC program.

[0072] Addendum 13. The apparatus of Addendum 12, wherein the controller subdivides the laminate into a plurality of sections and reports the likelihood of an out-of-tolerance manufacturing error for each of the plurality of sections.

[0073] Addendum 14. The apparatus of Addendum 12 or 13, wherein the input is selected from the group consisting of the curvature of the laminate at a given location, the length of a tow, the distance from a current position to an end of a tow, the number of layers at a given location, the fiber angle within a tow, the temperature of a heater on the fiber placement machine at a given location, the relative placement of tows within a strip, and the placement of strips within a layer.

[0074] Addendum 15. The apparatus of any one of Addendums 11 to 14, further comprising a sensor for acquiring the measurements.

[0075] Appendix 16. Manufacturing of aircraft parts using an apparatus according to any one of appendices 11 to 15.

[0076] Appendix 17. A method for facilitating manufacturing of a composite part, comprising: laying up a layup using a fiber placement machine; obtaining measurements of the layup indicative of locations that are out of tolerance; training a neural network based on the measurements; analyzing a numerical control (NC) program for laying up a new layup via the neural network; and generating a heat map indicative of the likelihood of locations on the layup exhibiting an out of tolerance condition based on output from the neural network.

[0077] Addendum 18. The method of Addendum 17, further comprising training the neural network, wherein training the neural network comprises adjusting weights between nodes in the neural network based on the measurements.

[0078] Addendum 19. The method of any one of Addendums 17 to 18, wherein the layup of the laminate further comprises laying down tows of fiber reinforcing material.

[0079] Addendum 20. The method according to any one of Addendums 17 to 19, wherein the shape of the heat map corresponds to the shape of the laminate.

[0080] Appendix 21. An aircraft part assembled according to the method of any one of appendices 17 to 20.

[0081] Clause 22. An apparatus for facilitating the manufacture of a composite part, the apparatus comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by a fiber placement machine to produce a layup for curing into a composite part; and a controller that performs the method of any one of clauses 17-20.

[0082] Appendix 23. A method for facilitating manufacturing of a composite part, the method comprising: obtaining a definition characterizing an out-of-tolerance condition for a layup; training a neural network based on the definition; operating the neural network to identify a first set of predicted locations of out-of-tolerance conditions for a new layup as defined by a numerical control (NC) program; modifying the definition; re-training the neural network based on the modified definition; operating the re-trained neural network to identify a second set of predicted locations of out-of-tolerance conditions for the new layup; and generating a report comparing the first set of predicted locations and the second set of predicted locations.

[0083] Clause 24. The method of clause 23, wherein the definitions refer to dimensions and shapes that are out of tolerance.

[0084] Addendum 25. The method of Addendum 23 or 24, wherein the neural network is trained based on the definition applied to measurements of an already laid-up laminate.

[0085] Addendum 26. The method of any one of Addendums 23 to 25, wherein the report is provided as a heat map.

[0086] Appendix 27. An aircraft part assembled according to the method of any one of appendices 23 to 26.

[0087] CLAIM 28. A non-transitory computer readable medium embodying instructions programmed when executed by a processor to perform a method for facilitating manufacturing of a composite part, the method including: loading a numerically controlled (NC) program that directs the layup of tows by a fiber placement machine to produce a layup for curing into a composite part; extracting tow information written into the NC program; applying inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other laid-up laminates; and reporting a likelihood of out-of-tolerance manufacturing errors based on an output of the neural network for the NC program.

[0088] Addendum 29. The medium of Addendum 28, further comprising subdividing the laminate into a plurality of sections, wherein for each of the plurality of sections in the laminate, a likelihood of an out-of-tolerance manufacturing error is reported.

[0089] Addendum 30. The medium of Addendum 29, wherein the reporting includes providing a heat map visually illustrating the laminate, wherein the likelihood of out-of-tolerance manufacturing errors in each of the plurality of portions is reported as a color in the heat map.

[0090] Addendum 31. The input includes the curvature at a given portion of the laminate, the length of the tow, the distance from the current position to the end of the tow, the number of layers at a given portion, the fiber angle in the tow, 31. The medium of any one of claims 28 to 30, wherein the temperature of a heater provided on the fiber placement machine at the position, the relative arrangement of tows within a strip, and the arrangement of strips within a layer are selected from the group consisting of.

[0091] Addendum 32. The medium of any one of Addendums 28-31, further comprising generating a binary image for each strip, the binary image having a row for each lane in the fiber placement machine.

[0092] Addendum 33. The medium of any one of Addendums 28-32, wherein the manufacturing error is selected from the group consisting of a gap between tows exceeding a predetermined dimension, a gap between strips exceeding a predetermined dimension, an overlap between tows exceeding a predetermined dimension, an overlap between strips exceeding a predetermined dimension, a slow production rate, tow wrinkles, tow twists, tow separation, and a fill rate less than a predetermined amount.

[0093] Addendum 34. The medium of any one of Addendums 28-33, wherein the method further includes modifying the NC program and repeating the steps of extracting, applying, and reporting.

[0094] Appendix 35. An aircraft part assembled according to the method defined by the instructions stored on a computer readable medium according to any one of Appendixes 28 to 34.

[0095] Although specific embodiments have been described herein, the scope of the disclosure is not limited to these specific embodiments, and is defined by the following claims and their equivalents.

Claims

1. 1. A method for facilitating manufacturing of a composite part, comprising: laying up a laminate using a fiber placement machine; obtaining measurements of the laminate indicative of areas that are out of tolerance; training a neural network based on the measurements; and analyzing a numerical control (NC) program for laying up a new laminate through said neural network; and generating a heat map based on output from the neural network that indicates the likelihood of locations on the laminate exhibiting an out-of-tolerance condition.

2. The method of claim 1, further comprising training the neural network, wherein training the neural network comprises adjusting weights between nodes in the neural network based on the measurements.

3. The method of claim 1 or 2, wherein laying up the laminate further comprises laying tows of fiber reinforcing material.

4. A method according to any one of claims 1 to 3, wherein the shape of the heat map corresponds to the shape of the laminate.

5. A method according to any one of claims 1 to 4, wherein the measurement value is obtained by a sensor.

6. An apparatus for facilitating the manufacture of a composite part, comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by the fiber placement machine to produce a layup for curing into a composite part; A controller for carrying out the method according to any one of claims 1 to 5.

7. An apparatus for facilitating the manufacture of a composite part, comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by the fiber placement machine to produce a layup for curing into a composite part; a controller that identifies tow information written into the NC program, applies inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other laminates in a layup, and reports a likelihood of out-of-tolerance manufacturing errors based on the output of the neural network for the NC program.

8. The apparatus of claim 7, wherein the controller subdivides the laminate into a plurality of portions and reports the likelihood of out-of-tolerance manufacturing errors for each of the plurality of portions.

9. The apparatus of claim 7 or 8, wherein the input is selected from the group consisting of curvature at a given location of the laminate, length of a tow, distance from a current position to an end of a tow, number of layers at a given location, fiber angle within a tow, temperature of a heater provided on the fiber placement machine at a given location, relative positioning of tows within a strip, and positioning of strips within a layer.

10. An apparatus described in any one of claims 7 to 9, further comprising a sensor for acquiring the measurement value.

11. Manufacturing of aircraft parts using an apparatus described in any one of claims 7 to 10.

12. A system for facilitating manufacturing of a composite part, comprising: a memory storing a numerically controlled (NC) program that directs the layup of tows by the fiber placement machine to produce a layup for curing into a composite part; a controller that identifies tow information written into the NC program, applies inputs based on the tow information to a neural network trained with measurements indicative of out-of-tolerance errors for other laminates in a layup, and reports a likelihood of out-of-tolerance manufacturing errors based on the output of the neural network for the NC program.

13. The system described in claim 12, wherein the controller subdivides the laminate into a plurality of portions and reports the likelihood of out-of-tolerance manufacturing errors for each of the plurality of portions.

14. The system of claim 12 or 13, wherein the input is selected from the group consisting of curvature at a given location of the laminate, length of a tow, distance from a current position to an end of a tow, number of layers at a given location, fiber angle within a tow, temperature of a heater on the fiber placement machine at a given location, relative positioning of tows within a strip, and positioning of strips within a layer.

15. A system described in any one of claims 12 to 14, further comprising a sensor for acquiring the measurement value.

Images