Automated fibre layup techniques are widely used in the aviation sector for the efficient production of composite components. However, the required manual inspection can take up to 50 % of the manufacturing time. The automated classification of fibre layup defects with Neural Networks potentially increases the inspection efficiency. However, the machine decision-making processes of such classifiers are difficult to verify. Hence, we present an approach for analysing the classification procedure of fibre layup defects. Therefore, we comprehensively evaluate 20 Explainable Artificial Intelligence methods from the literature. Accordingly, the techniques Smoothed Integrated Gradients, Guided Gradient Class Activation Mapping and DeepSHAP are applied to a Convolutional Neural Network classifier. These methods analyse the neural activations and robustness of a classifier for an unknown and manipulated input data. Our investigations show that especially Smoothed Integrated Gradients and DeepSHAP are well suited for the visualisation of such classifications. Additionally, maximum-sensitivity and infidelity calculations confirm this behaviour. In future, customers and developers could apply the presented methods for the certification of their inspection systems.

In the aerospace industry, automated fibre laying processes are often applied for economical composite part fabrication. Unfortunately, the current mandatory visual quality assurance process takes up to 50% of the entire manufacturing time. An automised classification of manufacturing deviations using Neural Networks potentially improves the inspection's effectiveness. Unfortunately, the automated decision-making procedures of machine learning approaches are challenging to trace. Therefore, we introduce an approach for evaluating the classifiers response for this use case. For this purpose, we present a parallel classification approach of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) with suitable intermediate checking stages between both classification processes. The particular novelty of this study is this intermediate comparison to trace the behaviour of the two classifiers along their image processing chains and to project the results back to the input image. With respect to the SVM, we analyse their extracted input features via t-Distributed Stochastic Neighbor Embedding calculations and parallel coordinates plots. Moreover, the classification score of the SVM as well as the feature vector distances within the SVM are investigated. For the CNN, the outputs of its first joined convolutional layer are correlated with the raw input images of different classes using Structural Similarity Index Measure metrics. Additionally, also the CNN's classification rates are analysed. Accordingly, a suitable uncertainty confidence interval for the CNN is determined on the bases of its neural activations. Finally, the relevance of individual pixels for the CNN decision is determined through Smooth Integrated Gradients and linked to the manually extracted image features for the SVM Classifier. The results of this paper are particularly valuable for developers and users of visual inspection systems in safety-critical domains.