Automated fibre layup techniques are often applied for the production of complex structural components. In order to ensure a sufficient component quality, a subsequent visual inspection is necessary, especially in the aerospace industry. The use of automated optical inspection systems can reduce the inspection effort by up to 50 %. Laser line scan sensors, which capture the topology of the surface, are particularly advantageous for this purpose. These sensors project a laser beam at an angle onto the surface and detect its position via a camera. The optical properties of the observed surface potentially have a great influence on the quality of the recorded data. This is especially relevant for dark or highly scattering materials such as Carbon Fiber Reinforced Plastics (CFRP). For this reason, in this study we investigate the optical reflection and transmission properties of the commonly used Hexel HexPly 8552 IM7 prepreg CFRP in detail. Therefore, we utilise a Gonioreflectometer to investigate such optical characteristics of the material with respect to different fibre orientations, illumination directions and detection angles. In this way, specific scattering information of the material in the hemispherical space are recorded. The major novelty of this research are the findings about the scattering behaviour of the fibre composite material which can be used as a more precise input for the methods of image data quality assessment from our previous research and thus is particularly valuable for developers and users of camera based inspection systems for CFRP components.

Automated Fibre Placement is a common manufacturing technique for composite parts in the aero-space industry. Therefore, a visual part inspection is required which often covers up to 50% of the actual production time. Moreover, the inspection quality of this manual step fluctuates significantly. A camera-based automated inline inspection is capable of increasing the inspection efficiency and accuracy. However, the interpretability of the acquired data strongly depends on the sensor configuration and the inspected material. Thus, this paper introduces methods for modelling and assessing an imaging sensor on the example of a composite material reflecting a spot laser to a camera sensor. In this context, the reflection properties of the material are incorporated into a simulation and validated in comparison to real camera images from the experimental setup. The EMVA 1288 sensor model in combination with the Cramér–Rao lower bound indicates a feasible estimability of the beam propagation, but shows limitations in the predictability of the number of incident photons. The laser spot analysis indicated that the laser spot can deviate from an exact oval shape but its peak value is suitable for robust spot identification in an image. The outlined methodology is also adaptable to other imaging sensors, illumination sources and materials. Thus, the findings can be useful for other fields and manufacturing processes.

Automated fibre layup techniques are commonly used composite manufacturing processes in the aviation sector and require a manual visual inspection. Neural Network classification of defects has the potential to automate this visual inspection, however, the machine decision-making processes are hard to verify. Thus, we present an approach for visualising Convolutional Neural Network (CNN) based classifications of manufacturing defects and quantifying its robustness. Our investigations have shown that especially Smoothed Integrated Gradients and DeepSHAP are particularly well suited for the visualisation of CNN classifications. The Smoothed Integrated Gradients technique also reveals advantages in robustness when evaluating degraded input images.

In the aerospace industry, the Automated Fiber Placement process is an established method for producing composite parts. Nowadays the required visual inspection, subsequent to this process, typically takes up to 50% of the total manufacturing time and the inspection quality strongly depends on the inspector. A Deep Learning based classification of manufacturing defects is a possibility to improve the process efficiency and accuracy. However, these techniques require several hundreds or thousands of training data samples. Acquiring this huge amount of data is difficult and time consuming in a real world manufacturing process. Thus, an approach for augmenting a smaller number of defect images for the training of a neural network classifier is presented. Five traditional methods and eight deep learning approaches are theoretically assessed according to the literature. The selected conditional Deep Convolutional Generative Adversarial Network and Geometrical Transformation techniques are investigated in detail, with regard to the diversity and realism of the synthetic images. Between 22 and 166 laser line scan sensor images per defect class from six common fiber placement inspection cases are utilised for tests. The GAN-Train GAN-Test method was applied for the validation. The studies demonstrated that a conditional Deep Convolutional Generative Adversarial Network combined with a previous Geometrical Transformation is well suited to generate a large realistic data set from less than 50 actual input images. The presented network architecture and the associated training weights can serve as a basis for applying the demonstrated approach to other fibre layup inspection images.

The aerospace industry has established the Automated Fiber Placement process as a common technique for manufacturing fibre reinforced components. In this process multiple composite tows are placed simultaneously onto a tool. Currently in such processes manual testing requires often up to 50% of the manufacturing duration. Moreover, the accuracy of quality assurance varies significantly with the inspector in charge. Thus, inspection automation provides an effective way to increase efficiency. However, to achieve a proper inspection performance, the segmentation of layup defects need to be examined. In order to improve such defect detection systems, this paper performs a comprehensive ranking of segmentation techniques. Thus, 29 statistical, spectral and structural algorithms from related work were evaluated based on nine substantial criteria as assessed from literature and process requirements. For reasons of determinism and easy technology transferability without the need of much training data, the development of new Machine Learning algorithms is not part of this paper. Afterwards, seven of the most auspicious algorithms were studied experimentally. Therefore, laser line scan sensor depth maps from fibre placement defects were utilised. Furthermore noisy images were generated and applied for testing algorithm robustness. The test data contained five defect categories with 50 samples per class. It was concluded that Adaptive Thresholding and Cell Wise Standard Deviation Thresholding work best yielding detection accuracies mostly > 97 %. Noteworthy is that influenced input data can affect the detection results. Feasible algorithms with sensible parameter settings were able to perform reliable defect segmentation for layed material.

The Automated Fiber Placement process is established in the aerospace industry for the production of composite components. This technique places several narrow material strips in parallel. Within current industrial Automated Fiber Placement processes the visual inspection takes typically up to 50% of overall production time. Furthermore, inspection quality highly depends on the inspector. Therefore, automation of visual inspection offers a great improvement potential. To ensure reliable defect detection the segmentation of individual defects must be investigated. For this reason, this paper focusses on an assessment of defect segmentation algorithms. Therefore, 29 structural, statistical and spectral algorithms from related work were assessed, theoretically, using the 12 most relevant criteria as assessed from literature and process requirements. Then, seven most auspicious algorithms were analysed in detail. For reasons of determinism, Neural Network approaches are not part of this paper. Manually labelled prepreg defect images from a laser line scan sensor were used for tests. The test samples contain five defect types with 50 samples of each. Additionally, layups without defects were analysed. It was concluded that Adaptive Thresholding works best for global defect segmentation. The Cell Wise Standard Deviation Thresholding performs also quite well, but is very sensitive to grid size. Feasible algorithms perform reliable defect segmentation for layed up material.