Deep Learning aplicado a inspeção visual da presença de um componente de conjunto de eixo

identificação ou falta retenção montado eixo veicular partir de imagens. rede neural convolucional foi utilizada para aprender as características das imagens e realizar a classificação. O sistema foi avaliado utilizando uma base de imagens coletada ambiente real de uma Apesar desbalanceamento conjunto de dados, o método produziu resultados máximos sensibilidade, especificidade e F1-score. disso, arquitetura rede Abstract The evolution of industrial processes based on the concepts of smart factory in Industry 4.0 and the need to perform decision-making tasks less human-dependent should increasingly demand the industrial application of machine learning. In this sense, this work proposes the use deep learning to identify the presence or lack of a retaining ring at a vehicle axis end from images. A convolutional neural network was used to learn features from images e to perform classification. The system was evaluated using a dataset of images collected in a real industrial environment. Despite the dataset imbalance, the method yielded maximum results in sensitivity, specificity and F1-score. Thereafter, the neural network architecture was optimized (90% reduction of the number of parameters) to increase computational efficiency.