QUANTIFYING PREFORM QUALITY THROUGH DEFECT INSPECTION OF IN-FACTORY PHOTOGRAPHS
Topic(s) : Special Sessions
Co-authors :
Umeir KHAN (UNITED KINGDOM), Vincent MAES (UNITED KINGDOM), Robert HUGHES (UNITED KINGDOM), Petar ZIVKOVIC , Turlough MCMAHON , James KRATZ (UNITED KINGDOM)Abstract :
A key challenge facing next-generation aircraft will be meeting increased rate targets at lower cost. Out-of-autoclave processes, namely, dry fibre resin infusion have demonstrated rate-enabling capability through Airbus’ Wing of Tomorrow program [1]. Non-Crimp Fabrics (NCFs) are of prime interest, given their high deposition rates and better mechanical performance over woven formats.
Pre-infusion, NCFs must be shaped from a flat 2D sheet into the required 3D configuration - a process known as preforming. With the drive towards reducing assembly times, integrated aerostructures are becoming more prevalent. This makes part geometry more complex. As a result, during preforming, defects such as in-plane waviness of the ply (Fig. 1) can occur. Understanding the effect of defects for stress analyses is an industrially relevant activity, yet critically relies on a quantitative assessment of preform quality being made.
Various scanning technologies are capable of capturing and quantifying high-resolution data of the preform. However, they can be time-consuming to operate and difficult to integrate, thus negating their potential benefits. Instead, a cost-effective solution is using DSLR cameras. This is motivated by current shopfloor practice, where inspection of defects in photographs is commonplace. However, the data capture is inherently qualitative and, often, waviness measurement remains a manual process. Overcoming these bottlenecks can be achieved with post-processing, i.e. waviness tracing and parameterising, performed in the order of seconds per photograph.
Further complexity arises from the imaging artefacts and high variability between photographs; a consequence of performing data capture on the shopfloor. This makes applying existing image-processing methods [2], to trace waviness, difficult to effectively deploy without manual intervention. To improve this aspect, deep learning is used to segment the individual NCF tows. From this, the outer tow geometry is extracted and defined into amplitude, wavelength, and aspect ratio. The process sequence is applied to all ply photographs captured of a large-scale component in single diaphragm forming [3]; providing numerical data of defects.
The research challenge lies in automating end-to-end characterisation of preform quality. To achieve this goal, a vision-based workflow (Fig. 2) is developed that enables rapid assessment of photographed waviness defects. The workflow is used to analyse a dataset of DSLR pictures to indicate upper bounds of defect severities that might arise during preforming.
Acknowledgements:
This work is supported by Airbus and the Engineering and Physical Sciences Research Council (EPSRC) (EP/W524414/1) and the Programme Grant ‘Certification for Design – Reshaping the Testing Pyramid’ (CerTest, EP/S017038/1). The funding received is gratefully acknowledged.
Pre-infusion, NCFs must be shaped from a flat 2D sheet into the required 3D configuration - a process known as preforming. With the drive towards reducing assembly times, integrated aerostructures are becoming more prevalent. This makes part geometry more complex. As a result, during preforming, defects such as in-plane waviness of the ply (Fig. 1) can occur. Understanding the effect of defects for stress analyses is an industrially relevant activity, yet critically relies on a quantitative assessment of preform quality being made.
Various scanning technologies are capable of capturing and quantifying high-resolution data of the preform. However, they can be time-consuming to operate and difficult to integrate, thus negating their potential benefits. Instead, a cost-effective solution is using DSLR cameras. This is motivated by current shopfloor practice, where inspection of defects in photographs is commonplace. However, the data capture is inherently qualitative and, often, waviness measurement remains a manual process. Overcoming these bottlenecks can be achieved with post-processing, i.e. waviness tracing and parameterising, performed in the order of seconds per photograph.
Further complexity arises from the imaging artefacts and high variability between photographs; a consequence of performing data capture on the shopfloor. This makes applying existing image-processing methods [2], to trace waviness, difficult to effectively deploy without manual intervention. To improve this aspect, deep learning is used to segment the individual NCF tows. From this, the outer tow geometry is extracted and defined into amplitude, wavelength, and aspect ratio. The process sequence is applied to all ply photographs captured of a large-scale component in single diaphragm forming [3]; providing numerical data of defects.
The research challenge lies in automating end-to-end characterisation of preform quality. To achieve this goal, a vision-based workflow (Fig. 2) is developed that enables rapid assessment of photographed waviness defects. The workflow is used to analyse a dataset of DSLR pictures to indicate upper bounds of defect severities that might arise during preforming.
Acknowledgements:
This work is supported by Airbus and the Engineering and Physical Sciences Research Council (EPSRC) (EP/W524414/1) and the Programme Grant ‘Certification for Design – Reshaping the Testing Pyramid’ (CerTest, EP/S017038/1). The funding received is gratefully acknowledged.