Resin injection into reinforcement with uncertain heterogeneous properties: NDE and control

Home > Research > Hub Core Projects

Resin injection into reinforcement with uncertain heterogeneous properties: NDE and control

Host Institution: The University of Nottingham

Start Date: 1st November, 2019

Duration: 36 months

Lead Investigator: Michael Tretyakov

Co-Investigators: Andreas Endruweit, Andrew Long, Andrew Parsons, Marco Iglesias, Mikhail Matveev


The key requirements of the composite industry are to have repeatable mechanical properties of composite components, to minimise possibility of defects, and to have repeatable production cycles. For the RTM process to be used for high-value components, appearance of dry spots (i.e., defects) should be prevented. Appearance of voids and deviations from the designed filling time are caused by uncertainties in the material and process parameters, which cannot be realistically fully eliminated. To counteract the effect of the deviations, manufacturers use more conservative designs with larger safety factors. These conservative designs increase the total manufacturing and life-cycle costs. Furthermore, defects occurring because of uncertainties can lead to costly rework or higher scrap rate, which again increase the costs. The ability to quantify and control these uncertainties is of great importance for further advances in composites manufacturing. Currently, NDE and rework can take 10% to 30% of the overall manufacturing time in the aerospace industry. Finding new solutions for reducing this time is of the highest significance.

Our vision is to use in-process information from sensors during resin injection for reliable and quick NDE of composite parts and for an active control system to counteract random deviations from the design. Utilisation of in-process data will capture and estimate local deviations from the design for any manufactured part which will create a digital twin for this part. This significant advancement will deliver a major step-change in composites manufacturing by reducing the cost and increasing the robustness of the manufacturing process, thus improving confidence in the parts quality. The project will develop, improve and test innovative Bayesian inversion algorithms to restore permeability of composites components based on data collected from sensors during resin injection into reinforcement. It will also develop and test an active control system based on information from sensors and on physical models together with on-line parameter estimation algorithms to improve resin injection. This control system will minimise occurrence of defects and ensure the process robustness.



  • The research team developed Bayesian inversion algorithms and successfully tested them in virtual and lab experiments to discover and quantify defects in preforms
  • We recently improved the Bayesian inversion algorithms so that they can recover defects of arbitrary, a priori unknown shapes and sizes (e.g. race tracking)
  • We conducted a three month impact exploration (ended in December 2019; £8K of additional impact EPSRC funding was secured for that). Several companies were contacted, and we met with 7 of them. They all expressed an interest in the project and gave us very valuable feedback for further shaping the project towards the industrial needs. With three companies (JLR, Surface Generation and ESI), we are developing further collaboration.
  • With ESI, we applied (in January 2020) for CASE EPSRC studentship for which ESI committed £32K cash and £260K in-kind contribution.

Figure 1: Schematic illustration of Bayesian inversion algorithms.

Figure 2: Illustration of a successful use of a Bayesian inversion algorithm in estimating positions of defects in a tunnel-shaped part.


Guest speaker: Francisco Chinesta, Professor of Computational physics at ENSAM Institute of Technology (Paris), presented his work on data-driven modelling in a recent  seminar “Empowering Hybrid Twins from Physics-Informed Artificial Intelligence”, this can be viewed below.



Copyright © 2020 CIMComp