Machine Learning (ML) for efficient optimisation of mould-filling processes in composite manufacturing

Machine Learning (ML) for efficient optimisation of mould-filling processes in composite manufacturing

    

Motivation

Continuous-fibre reinforced plastics (CoFRP) are increasingly being used in automotive and aerospace engineering to reduce the weight of load-bearing structures. Among others, the component properties are strongly depend on defect-free manufacturing and optimized process parameters. During component design, manufacturing simulations (e.g. draping or mold filling based on meshed computational methods) are used to ensure and optimize manufacturability.

These simulations show a high degree of reliability, but require long computation times, especially during iterative optimisation. Recently, Machine-Learning (ML) techniques are applied to reduce the computational load during optimisation. These models learn the underlying process dynamics from process samples and can then give recommendations for optimal process parameters.

Thesis subject and goals

The aim of this project is to train an ML-algorithm to estimate optimal injection port positions (process parameters) for variable part geometries in RTM-processing. To this end, an existing ML-framework from will be adapted to interact with an RTM-simulation environment. During ML-training, different geometries of variable complexity will be fed to the simulation environment. Thereby, the ML-algorithm learns, which component design requires which optimal injection port positions. After the training, the algorithm will be able to estimate optimal port positions for a new component without the need for computation intensive optimization.

The thesis is based on existing scripts, which shall be advanced and extended and will be conducted in cooperation with  AMT of TU Delft. The thesis’ structure is as follows:

  • Literature research on the state of the art in ML-assisted engineering.

  • Acquaintance with existing scripts on RTM-Process simulation and the ML-framework.

  • Enhancement of scripts, parameter studies and benchmark with classical optimization

  • Thesis composition and documentation of results

 

Discipline:   Mechanical or Industrial Engineering or comparable

 

Type of Thesis:   Simulation, Optimization, Numerics

 

Requirements:

  • Interest in Engineering Simulation and numerics

  • Strong analytical skills and independent working style

  • Programming experience is advantageous (ideally Python)

 

Language:  English

 

Begin:  direct start is possible

 

Contact: 
Dr. Bariș Çağlar                             Dipl.-Ing. Clemens Zimmerling

TU Delft | AMT                                KIT| FAST-LBT

+31 6 27 19 29 80                          +49 721 / 608 45409

B.Caglar∂tudelft.nl                        clemens zimmerlingDwt9∂kit edu