Advertisement

Lightweight Design worldwide

, Volume 10, Issue 6, pp 48–53 | Cite as

Application of Evolutionary Algorithms on the Draping Process

  • Florian Brillowski
  • Haoming Zhang
  • Julia Orlik
  • Thomas Gries
Production Draping Process
  • 155 Downloads

The production of fibre reinforced plastics with resin infusion processes requires the usage of textile preforms. The production of the aforementioned is lavish and expensive. Especially the draping steps in the course of the preforming process chain offer significant potential for improvement and are therefore the main focus of two projects at RWTH Aachen University.

Design of the Draping Process

The project Optidrape addresses the geometry specific design of the draping process for fibre reinforced plastics (FRP) components.

In order to be able to produce FRP in large scales, it is necessary to reduce the costs per component by a significant amount. Based upon this, draping is a key process within the preforming process. Draping results influence subsequent production steps as well as the overall quality and strength of the final component. Taking this into account it is problematic that small and medium-sized companies often perform the draping process manually and renounce any simulating software because of high licensing costs. The aftermath is unsteady and deficient draping results as well as high costs for fixing the consequent errors and component scrap.

The algorithm determines an optimal component geometry that guarantees the best possible drapeability.

Determinants of Draping Quality

In this context the pivotal factors for draping quality are material, process and component geometry. The influence of material and process on the draping result has already been examined and discussed extensively in scientific literature. But a systematic analysis of the component geometry is lacking, so that there is still a large room for improvement within this specific area.

Originating from the component geometry, FRP components cause in general higher development expenses than classic construction materials like aluminium and steel. The causes for this circumstance are the complex material properties as well as the state of the art production processes. Up to 33 % of the overall component costs are being caused by the product development while the remaining 67 % are due to the production of the component itself. Additionally the design of lightweight components is often determined by substituting the metallic material with FRP. In the aftermath the great potential of FRP remains unused because of an inappropriate conception regarding the material. There have been first attempts to realize FRP appropriate concepts, but these still neglected the resulting drapeability as well as the overall fit for the preforming process.

In general the evaluation of the component specific drapeability is based upon the experience of the responsible worker or is being determined in arduous and expensive trial and error experiments, Figure 2. This leads subsequently to high scrap rates as well as production costs.
Figure 2

Classic draping process (© RWTH Aachen University)

Aim of the Optidrape Project

Based upon these circumstances it is desirable to be able to illustrate the drapeability of textiles with appropriate simulation software. Thus it is possible to predict draping-effects depending on the technology. Subsequently a reproducible and standardized draping process is feasible. Yet small and medium-sized companies are either lacking experience in Finite Element(FE) simulation or the occurring costs are not economically justifiable for them.

By applying the systematic approach a selection of fitting textiles can be elaborated for a predefined geometry.

Therefore it is the aim of the Optidrape project to improve the draping process in order to be able to pave the way for a material appropriate conception of FRP components. This consequently leads to an increased quality of FRP preforms as well as a reduction of the resulting development costs and time.

Hence a systematic approach in order to select reinforcement textiles will be developed during the course of the project, Figure 3. By applying the approach a selection of fitting textiles can be elaborated for a predefined geometry without being dependent on costly simulation software. It is also possible to determine geometry recommendations for given reinforcement textiles. Initially a data base is being created, in which predefined geometries will be categorised as well as evaluated and classified regarding their drapeability. The evaluation of the drapeability will be dependent on different, commercially available reinforcement textiles. The required material data will be determined using a textile meso-model. The model will be developed at the Fraunhofer-Institut für Techno- und Wirtschaftsmathematik (ITWM) in Kaiserslautern on the basis of textile parameters. These parameters will be measured in various test procedures at the ITA. By using the textile material model, the influence of different load conditions can be quantified. The resulting data is being used as a foundation for determining drape appropriate component geometries. Thereupon an evaluation of the reinforcement textiles in dependency of different, standardised geometries and subsequently the creation of a database (so called drape catalogue) will be carried out at the ITA.
Figure 3

Optidrape approach (© RWTH Aachen University)

Optimisation of Geometry and Drape Result

Yet the selection of the best textile does not warrant an error free draping result. Hence the component geometry is being optimised within the Optidrape approach. For this purpose the Institut für Unternehmenskybernetik (IfU) at the RWTH Aachen University develops an evolutionary algorithm that calculates an optimal component geometry within multiple iteration steps. The necessary data is being determined by an evaluation function through machine learning. Causalities between the input data and the preforming result are being deduced in this context on the basis of preliminary studies as well as the previously created drape catalogue. Due to these causalities, the evolutionary algorithm is consequently able to determine an optimal component geometry that fits the required mechanical properties as well as guarantee the best possible drapeability.

An early implementation of the Optidrape approach enables to reduce the product development time in comparison to conventional approaches. Furthermore the necessity of a costly FE simulation software can be omitted.

Drapefix project aims to reduce the forming forces by 50 %.

Local Binder Application

Another project called Drapefix addresses local binder application for an improved draping of textile preforms. In the course of the production process, the single layers of a preform are being prefixed by using a full surface binder application according to the state of the art. This results in a high contour accuracy as well as good handling properties of the 2-D-preform. However, draping requires high forming forces, because the binder hinders the single layers from sliding off each other. Therefore high investments in plant technologies are necessary in order to apply the corresponding forming forces. Especially when changing geometries often or producing in small amounts, the expensive plant technology can lead to economical inefficiency and might be an exclusion criterion in the course of technology selection. Moreover, the full surface application of binder favors errors like fibre displacements, folds, alleys as well as varying thickness, Figure 5.
Figure 5

Draping by using the full-surface binder application (© RWTH Aachen University)

Aim of the Drapefix Project

Origination from these problems, the Drapefix project aims to reduce the forming forces by 50 % as well as to realise a textile and load appropriate forming of the 2-D-layer packages. Therefore draping errors can be reduced and the usage of smaller presses is being encouraged. Subsequently smaller competitors might be able to diversify further into the FRP segment as well as tap into new markets.

By reducing the amount of binder by 50 % it has been shown in preliminary tests that the shear force could be decreased by more than 30 %, Figure 6.
Figure 6

Decreasing the shear force by reducing the amount of binder (© RWTH Aachen University)

Forming Forces

The reduction of the forming forces is realised by applying the binder locally. Therefore an essential knowledge of the causalities between textile and load appropriate forming, local binder paths as well as the amount of binder is necessary. However, the causalities have not been adequately examined in previous studies and projects, which is why initially these have to be determined in experiments. For this purpose the fracture toughness energy according to DIN EN 6033, the maximum bending stress according to DIN EN ISO 14125 as well as the coefficient of friction according to DIN EN ISO 8295 are being tested at the ITA. In the course of the different testing methods, the used amount of binder, the testing temperature, the binder application system and the orientation of the applied binder will be varied. The considered binder application technologies are the hotmelt and powder application systems. The corresponding characteristic values are being determined with the aid of two measurement points that are placed on the textile. Thereupon the caracteristic values are being transferred into characteristic curves, which describe the forming behaviour depending on the orientation of the binder, Figure 7. Based upon the deduced characteristic curves, an FE simulation model will then be developed in order to determine an optimal component geometry. Therefore the textile and load appropriate forming will be examined iterative and in the course of the simulation.
Figure 7

Experimental setup and determining of the characteristic curves (© RWTH Aachen University)

Optimal Binder Paths

Subsequently the concept of machine learning is applied analogous to the Optidrape approach in order to deduce causalities between input (component geometry, pattern of binder application, pressure) and output (overall quality of the textile preform). These relations are being used for the development of an evolutionary algorithm at the IfU. By using the algorithm it is thereupon possible to determine an optimal local binder path for the predefined component geometry.

By using the algorithm it is possible to determine an optimal local binder path.

Thereafter, a software tool, that automatically outputs an optimal local binder pattern for a predefined component, will be developed, Figure 8. The obtained results will be validated in conclusion by using a demonstration geometry. Therefore the whole process chain (cutting, handling, local binder application, binder activation and forming) is going to be replicated and run through at the ITA preform center. The quality of the preform is going to be monitored with an optical sensor system that is able to register fibre orientation, preform errors as well as contour accuracy. In order to be able to put the obtained results into context, the conventional full surface binder application is used as reference process. This facilitates the immediate identification of technological advantages of the local binder application. Furthermore an economical and technological evaluation of the newly developed approach will be done in the course of the final validation. For this purpose the fibre orientation, the process capability according to six sigma as well as the unit costs and cycle times will be used as evaluation factors.
Figure 8

Improving draping results due to local binder application (© RWTH Aachen University)

Draping — A Key Process of Preforming

Draping describes the fitting of areal reinforcement textiles on curved, three dimensional surfaces. The reinforcement textiles tend to wrinkle especially on surfaces with multiple curvatures, Figure 1. The folds hinder the impregnation of the preform and facilitate draping errors that lead to strength reducing, highly resinous areas within the final component. Therefore the crimp free draping of reinforcement textiles is an important criterion of the construction and design process of FRP-components.
Figure 1

Illustrative wrinkling during the draping process (© RWTH Aachen University)

Evolutionary Algorithms

An evolutionary algorithm is an optimisation procedure that is based upon the natural evolutionary process. Solution candidates are being evaluated, selected, recombined and mutated over several generations in the course of the procedure. The evaluation is done by using a valuation function. Appropriate solution candidates are being chosen on the basis of the valuation function and thereafter the chosen candidates are being mutated or recombined in various iteration steps if desired. (Figure 4) This is carried out in the course of multiple generations of solution candidates until an abort criterion is met. The concept of machine learning is a part of artificial intelligence and describes the iterative learning process of a machine (e.g. computer). At the end of the process the machine is able to identify patterns as well as causalities and generalises them in order to recognise and evaluate circumstances that were not part of the initial data.
Figure 4

Schematic illustration of an evolutionary algorithm (© RWTH Aachen University)

Notes

Thanks

Both projects are being funded via the AiF by the Federal Ministry of Economic Affairs and Energy (BMWi) according to a decision of the German Federal Parliament. We would like to thank all project partners and companies for their support and financial aid. The duration of Optidrape is March 2017 to February 2019, the duration of Drapefix is June 2017 to May 2019.

Copyright information

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2017

Authors and Affiliations

  • Florian Brillowski
    • 1
  • Haoming Zhang
    • 2
  • Julia Orlik
    • 3
  • Thomas Gries
    • 1
  1. 1.Institut für Textiltechnik (ITA)RWTH Aachen UniversityAachenGermany
  2. 2.Institut für Unternehmenskybernetik e.V. (IfU)RWTH Aachen UniversityAachenGermany
  3. 3.Fraunhofer-Institut für Techno- und Wirtschaftsmathematik (ITWM)KaiserslauternGermany

Personalised recommendations