[home] [Personal Program] [Help]

---

Session: Session 10: Defect Prediction and Mitigation in AFP
Session starts: Thursday 16 April, 13:10
Presentation starts: 14:30
Room: Main

---

Siddharth Pantoji (TU Delft)
Seán McCarthy (TU Delft)
Martijn van der Voort (TU Delft)
Clifton-John Walle (TU Delft)
Giovanni Zattoni (TU Delft)
Manuel Cruz (TU Delft)
Christos Kassapoglou (TU Delft)
Daniël Peeters (TU Delft)

Abstract:
Gap and overlap defects in AFP composites are formed due to tow position and tow geometry occurring in the manufacturing process. The tow position variation can occur due to path inaccuracy of the robot and tow lateral movement at the compaction roller. The tow geometry variation can occur due to width variation in the supplied tow material and due to compaction of the tow under the roller during layup. Statistical data about these four sources of variations were fit with distributions and then used to simulate sections of an AFP laid tow. These tow sections were assembled in the tow length direction to construct representative synthetic tows which are similar to real tows in their waviness after layup. These synthetic tows were also assembled to create virtual laminae with gap and overlap defects which appeared similar to an experimental lamina. Synthetic tow was simulated using two methods - Monte Carlo (MC) simulations and Markov Chain Monte Carlo (MCMC) simulations. The Monte Carlo simulations were implemented with random samples from the distributions describing the four sources of variations. Tow created using this method had more edge waviness compared to the observations of a real AFP laid tows. This was attributed to the random sampling which lead to drawn variation values which jumped drastically across the distributions rather than following continuous changes like the physical process they describe. This limitation was improved upon in the Markov Chain Monte Carlo simulations which used the Random Walk Metropolis method for sampling. In this case, the drawn variation value is located in the neighborhood of past draw from the distributions and therefore preserve continuity. This leads to synthetic tows whose waviness is more realistic than the synthetic tows created by random sampling. The virtual laminae can be used to identify the optimum programmed shift in adjacent tows to achieve the objective of achieving the right balance of gaps and overlaps. This could help reduce the extensive prototyping that is currently needed in establishing defect allowables. Another use of these simulation methods which are informed by manufacturing variation data is in prioritizing the process improvements required in tightening the AFP layup to the needed levels of quality.