Community Research and Development Information Service - CORDIS


FibreMap Report Summary

Project ID: 608768
Funded under: FP7-NMP
Country: Austria

Final Report Summary - FIBREMAP (Automatic Mapping of Fibre Orientation for Draping of Carbon Fibre Parts)

Executive Summary:
The FibreMap project aims at the development of an automatic quality control and feedback mechanism to improve draping of carbon fibre material on complex parts. There is a strong need in the automotive industry for automatic systems that perform quality control and improve draping processes in order to allow high production volumes. The technology that has been developed in the project includes a new sensor system (the “FScan” sensor) for robust detection of fibre orientation combined with a robotic system to scan complex parts. The sensor is based on a new technology that uses reflection models of carbon fibre to solve the problems encountered with earlier vision-based approaches.
The data coming from the inspection system will be fed into draping simulation to improve the accuracy of the simulation tools. Draping is the process of converting 2D woven carbon material to complex 3D parts (preforms) with the goal of having the fibres oriented along specific directions as designed through finite element calculations. This is done to maximize the strength-to-weight ratio of the part.
There is a strong trend in the automotive industry towards lightweight parts to increase fuel efficiency, also considering the needs of electrical vehicles. Setting up the draping process for a complex part takes up to 50 preforms for trial-and-error improvements. Current production processes thus require simulation and quality control tools to reach the expected volumes of several 100.000 parts per year.
The two main goals that were pursued in the FibreMap project are:
(A) A fibre angle measurement system to acquire a dense mapping of fibre directions on carbon fibre parts of complex shape.
(B) Computational methods that use data generated by the fibre angle measurement system to improve draping simulation.
At the end of the project the integrated demonstration could successfully show dense fibre measurement on a 3D part, using the camera-based “FScan” sensor for orientation measurement and the “Workcell Simulator (WCS)” software to automatically plan an inspection path for the robotic system. Scanning of 1 m² with a resolution of 50um per pixel took less than 3 minutes. The resulting data were fed into the PAM-FORM simulation software to compare fibre orientation measurements to the draping simulation results. A series of process optimization cycles run with the help of the simulation tools could demonstrate the effectiveness of the simulation tools in predicting draping results.

Project Context and Objectives:
Carbon fibre parts are more and more frequently used in the automotive industry and in the aerospace industry. During the production process flat carbon material is placed in a mold in several layers. Because of the complex shape of the part, the flat material will distort to match the 3D shape. However, the positioning of the fibres on the part is of high relevance for structural parts, where the mechanical strength of the parts depends on fibre orientation, and for decorative parts, where visible distortions of the fibres have to be avoided. The measurement of fibre orientation is often done manually with the help of various technical devices to establish a common coordinate system. This is difficult for parts of complex shape and it is not possible to measure a dense mapping of fibre orientations in this way.
To enable the setup of the manufacturing process, draping simulation is used to avoid experimenting with different process parameters. Draping simulation predicts how the material will distort when being placed in the mold. However, it is difficult to compare the results of draping simulation that provides a map of shearing angles between warp and weft fibres, and actual measurements on the part. As before, measurements can only be done point-wise and thus do not allow an adequate comparison over the whole part.
Finally, the increasing volume of carbon fibre parts, mainly in the automotive industry, requires adequate means for quality control. Manual inspection and point-wise measurements cannot be easily scaled to volumes of several 100.000 parts per year. Also for this purpose, automatic inspection and measurement tools are required.

The FibreMap project addresses the above mentioned challenges, by developing two technologies:
(A) A fibre angle measurement system to acquire a dense mapping of fibre directions on carbon fibre parts of complex shape.
(B) Computational methods that use data generated by the fibre angle measurement system to improve draping simulation.

The system for fibre orientation measurement uses a robot that is continuously moving a sensor across the surface of the part to scan the whole part. The sensor is based on machine vision using photometric stereo, which proved to be suitable for handling different kinds of carbon and glass fibre material, despite their difficult optical properties (black, highly reflective). The field of view of the sensor is about 50x50mm², which requires that it is moved over the part to fully scan the surface of the part. This can be easily done by a robot. The challenge, however, is to determine a path for the robot that fully covers the part, does not lead to collisions and is reachable for the robot. Based on a simulation of the whole robotic workcell and using a model of the image acquisition process, such a path can be obtained. During the scanning process, the robot motion and the image acquisition by the sensor need to be synchronized, so that fibres can be accurately mapped to the 3D shape of the part.

For draping simulation there exits two approaches. In the simple case only the geometry of the part is considered and the flat carbon material is virtually placed on the shape. This kind of simulation can be done very quickly and leads to reasonable results for simple geometries. In the more complex case material properties need to be considered and the shearing of warp and weft during the draping draping process becomes relevant. This e.g. requires that certain friction coefficients are known. Currently, such coefficients are found by experiments, such as the picture frame test, for each type of material that is used. In order to avoid these experiments and to obtain information about the overall accuracy of the draping simulation, a comparison has to be made between the fibre orientation as simulated and as measured on the part. Therefore the measurement data, that contain a very dense mapping of fibre orientations, need to be converted into a grid, that is more suitable for comparison with the simulation, and they also need to be reduced to represent the shearing angle only, to correspond to the main output of the simulation. This comparison then helps to on the on hand optimize the draping process and on the other hand to determine the material parameters. The latter can be achieved through inverse methods that minimize the different between simulation results and experiments.

Project Results:
The main results of the project are tightly linked to the main objectives of the FibreMap project:

(A) A fibre angle measurement system to acquire a dense mapping of fibre directions on carbon fibre parts of complex shape.
(B) Computational methods that use data generated by the fibre angle measurement system to improve draping simulation.

The fibre angle measurement system (A) is based on a robotic system that moves a sensor across the whole surface to fully scan the part and to measure fibre orientations on a dense 3D grid. To create such a system, several developments were needed.
The fibre orientation measurement is based on the “FScan” sensor. The sensor is using a camera and specific illumination with a photometric stereo approach to determine fibre orientation in 3D for each pixel of the camera. This creates a dense map, where fibre orientation is measured on a grid of 50x50µm². The technology proved to be useful for static measurements, where the sensor is not moving during image acquisition, and it could be shown to work well, independent of the carbon material that is being used. For the use on a robotic system, the sensor had to be extended to allow continuous scanning. This required improvements of the illumination and of the image acquisition process. Two lasers were integrated into the illumination system to acquire local 3D information and thus enable a more accurate calculation of fibre orientation. Also the scanning process had to be made quicker. By using particular feature of the camera, a frame rate of 1000 images per second could be achieved, which allows the sensor to scan at speeds of up to 1m/s. A key issued that had to be solved was the fact that photometric stereo requires the combination of several images taken under different illumination. In that static case, these images can be combined on a pixel-by-pixel level, however, if the sensor is moving, each image will be taken from a slightly different position, which requires an accurate compensation of the motion between the single images. The final sensor could be shown to achieve an accuracy of less than 0.5° in a static setting on flat material and of less than 1°, when using it in continuous scanning mode with a robot on 3D parts.

In order to fully scan a 3D part, the robot has to move the sensor along a path that has to fulfill several constraints. The path has to be chosen in such a way that all relevant areas on the part are covered, i.e. no “holes” should remain on the parts, so that fibre orientation is measured in all locations where this is physically possible. This is achieved by including an accurate model of the image acquisition process. This model includes properties such as the working distance (+/- tolerances) and the angle relative to the surface normal (+/- tolerance) and the field of view that can be covered from a particular position. It should be noted that due to the complex 3D shape of the part, the actual field of view does not necessarily have to be of rectangular shape. Actually it can have an arbitrary shape, depending on the local curvature of the surface. Fully coverage, however, is not the only criterion. Additionally, any kind of collision between the robot, the sensor, the part or any other object with the reach of the robot has to be avoided. This is achieved by creating a simulation model of the whole robotic work cell. Those positions for which collisions are detected, will be re-planned, to find a different way of covering the missing surface area. Finally, the planned path has to be compatible with the robot’s kinematic structure. All points along the path need to be reachable and singularities are to be avoided. Once these boundary conditions are fulfilled, the path needs to be optimized to minimize the time needed for the scanning process. This requires an approximate solution of the travelling salesman problem and is achieved by using two heuristics. The first uses a penalty term that favors positions that are close to the previous ones, so that the robot remains within area of the part before proceeding to other areas. The second heuristic penalizes sharp turns and thus leads to a smoother path that can be processed more quickly. With these two parameters, the length of the path and the time needed for scanning could be controlled and optimized. All of these methods were integrated in the “Workcell simulator” software and during the demonstration it could be shown that an area of about 1m² could be scanned in less than 3 minutes.

Another challenge that had to be resolved, was the accurate mapping of the fibre orientations on the 3D part. During the scanning process thousands of images are acquired at high speed. Using a time-stamping method, these images were synchronized to the robot motion, so that for each image the exact joint angles of the robot were known. Through calibration of the whole system, including the sensor, the robot, the part and the whole workcell, a projection of the fibre orientations on the 3D part was made possible with high accuracy. Because the sensor is using a matrix camera, the single image patches had to be aligned to each other, to avoid discontinuities at the overlap between the single images. This was needed to compensate for any remaining inaccuracies in the whole transformation from the image pixels to the 3D position on the part. The whole processing chain could be implemented, so that is processing the data in real-time. This enables to user to obtain the data almost immediately after the scanning process is finished. Currently, the processing is limiting the maximum speed, if post-processing of the data is considered, then the scanning process itself can be made substantially faster. During test on a 3D part of complex shape, an accuracy of less than 1° could be established on most areas of the part.

For the simulation of the draping process, existing software for draping simulation based on “PAM-FORM” was used and extended to allow the integration of the measurement results. Draping simulation uses a finite element approach that analyses how the flat fabric will distort as it placed in a 3D mold. Various material parameters, such as the friction coefficient between warp and weft, are needed for an accurate simulation. The output of the simulation model is the shearing angle between warp and weft for each finite element. To allow the integration of the fibre orientation measurement coming from the scanning system, specific software had to be developed. Whereas the scanning system provides the fibre direction of either warp or weft, depending on what is visible on the particular location on the surface, the simulation model provides the change of the angle between warp and weft (the shearing angle). Additionally, the grid provided by the simulation software is much coarser than the 50x50µm² grid of the fibre orientation measurement system. By using adaptive averaging within the each finite element the measurement data could be converted in to a single measurement of the shearing angle per finite element. This made it possible to directly compare simulation results to experimental measurement. Within a series of production test, the high accuracy of the simulation models could be confirmed through this comparison.
A second aspect that was covered in the project is the need to determine material parameters of the carbon fabric to allow accurate simulation. These material parameters need to be found through mechanical experiments that are to be done for each single type of material. To avoid such experiments, methods were developed to directly calculate material parameters by comparing the simulation results with the experimental data. Specific inverse methods were implemented to perform numerical optimization. It could be shown, that the ability to determine material parameters from fibre orientation measurements depends on the shape of the parts. Essentially, if a quick draping simulation, that only considers the geometry of the parts, also leads to good simulation results, then fibre orientation does not contain enough information to determine material parameters. Within the experimental work of the project is was found, that – aside from simple cases - such as a spherical shape, the complexity of the parts is sufficient to determine material parameters. Through tests with simulation results using different material parameters, it could be shown that the optimization methods converge to the correct set of parameters. For real-world measurement data, various pre-processing steps will be needed to remove noisy areas from the data.

Throughout the project all test were done on three parts of different complexity. The first part was mainly flat, with a hemispherical “bump” in its center, the second part was an arrangement of 3 elevated rectangular areas, and the third part was a complex 3D shape resembling a real-world part as used in the automotive industry. The developments of the FibreMap project were tested on all of these parts. In the final year of the project, three experimental cycles were conducted, where process parameters were determined through simulation, the parts were manufactured using these parameters, the resulting pre-forms were scanned and the measured fibre orientation were compared to the simulation results.
Potential Impact:
Based on the exploitation strategy already identified prior to the start of the project the main outputs were broken down into single exploitable results. The results were defined so that independent exploitation of each result is possible. The following five results were identified:
(1) Sensor for fibre orientation measurement (“FScan” sensor)
(2) Robotic scanning system
(3) Draping simulation software (PAM-FORM software)
(4) Improved draping design process
(5) Software for workcell simulation and path planning (WCS software)

An exploitation strategy was developed along two different paths. For results (3) and (4) the developments of the FibreMap project will be integrated into existing software for draping simulation (“PAM-FORM”). This software will then have the ability of importing fibre measurement data and comparing these data to simulation results for the same part. PAM-FORM is currently being sold as a product by project partner ESI and consequently, the results will be offered as an additional feature to the customers of ESI. Exploitable result (4) will be implemented through the use of the simulation software. This will save long-lasting experimentation with different process parameters to find an optimal draping process. This exploitable result will be the main benefit for the end-users, who are manufacturing carbon fibre parts using a draping process.

The exploitation strategy for results (1), (2) and (5) has to follow a substantially different path. Result (2), the inspection robot, will obviously include results (1) and (5), but all of the results can be offered on the market independently. To facilitate the implementation of the robotic scanning system in industrial environments, a system integrator is needed, who has got some background in robotics, machine vision and inspection systems. Consequently, the search for cooperation partners in the field of system integration was already started during the project and potential partners have been identified. Based on a detailed market survey and feedback obtained during presentation of the results at relevant fairs, the consortium expects sales of about 10-15 systems per year, which will result in 15-25 additional jobs at the partners’ organizations.

The main presentation of the FibreMap project took place in March 2016 at the JEC’16 fair in Paris. The JEC fair is the worldwide key fair in composite parts manufacturing. About 50 contacts could be established to potential customers and cooperation partners. In the months following the fair, four of these contacts resulted in concrete discussions about the implementation of the FibreMap technology. Several other fairs have been attended with different dissemination material (poster, video, flyers) being shown at these exhibitions, e.g. HMI, Robobusiness, EMO2015, MECSPE.

In terms of scientific publications the consortium took the opportunity to present the results of FibreMap on several occasions. In the course of the project the conferences ECCM, MetroAerospace, IC3, IAS-13, EASFORM and CAE were used as a platform to disseminate information about FibreMap. In addition to this journal publications for Composite Structure, Composite Science and Technology and T-ASE were in preparation.

In terms of addressing the wider public, press releases were prepared by Profactor and distributed to relevant media. A major demonstration took place in April 2015 during the “Lange Nacht der Forschung”. This Austrian-wide event is basically an “open-door-day” of industrial and academic research organizations and departments. During this day about 300 participants visited Profactor, where the FibreMap project results were demonstrated. A similar activity also took place during the European Robotics Week in November 2015.

The FibreMap project also took a leading role in the organization of a cluster of projects formed during the FoF Impact Workshop in Brussels. In March 2015 a workshop was organized, involving several of the projects that were submitted to the same topic as FibreMap. The workshop was mainly used for a detailed exchange of information between the projects. Furthermore, Profactor also prepared the cluster presentation for the FoF impact workshop in Brussels in April 2016.

Environmental impact
During the final year of the project a series of 3 cycles of experiments was done to assess the accuracy with which draping results for different process parameters can be predicted. By a direct comparison between shearing angles calculated with the simulation tools and experimental data measured with the robotic scanning system, it could be shown that there is a very good match. Consequently, these simulation tools have the potential of eliminating the need for experimental production of several parts during the setup of the production process. Sometimes up to 50 (scrap) parts need to be produced, before satisfactory process parameters are found. By using draping simulation, this can be reduced by at least 90% to 4-5 parts.

A detailed assessment of the market potential of the robotic scanning system indicates a potential of 10-15 systems per year and a potential of 15 to 25 new jobs in systems’ engineering and software development.

Spin-off companies
The idea of setting up a spin-off company for selling the robotic scanning system on the market was investigated in detail by the project partners. However, for various reasons it was decided to exploit the results jointly within the existing organizations of the project partners.

List of Websites:

Profactor GmbH
Dr. Christian Eitzinger
Im Stadtgut A2, 4407 Steyr-Gleink, Austria
Tel: +43 7252 885250

Reported by

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top