TLDR; At Signify we are investigating automated visual quality control of our products to improve our quality and to lower the amount of rework. One approach is to use the images of manufactured products to automate visual quality inspection based on innovative digital technologies. The aim for Signify is to obtain suggestions (including a proof of concept) for industrially implementable solutions.
Project Description: The goal of this assignment is to explore methods for automated visual quality control based on images of products after they have been manufactured. The proposed approach is to train a model on reference images of good products to identify anomalies (e.g. surface defects) and decide whether a new product should be accepted, rejected, or inspected by an operator. The assignment focuses on the development of algorithms that can be used for an automated visual quality control system that does not require human intervention and that can outperform humans in accuracy and efficiency.
This assignment will build on earlier work (see also ) done on this topic at Signify. As such, some preliminary data for this task is already available. More data will be provided by the company. The previously investigated approach uses deep generative models (e.g. VAE, GAN) to assess the likelihood of an image displaying a correct product, and potentially localize defects. A suggested follow-up approach is to investigate density estimating models (see e.g. NADE , MADE ) to target this problem in a more principled manner.
Company Description: Signify is the world leader in connected LED lighting products, systems and services. Through our innovations, we unlock the extraordinary potential of light for brighter lives and a better world. We serve professional and consumer markets, transforming urban spaces, communities, work places, stadiums, buildings, shopping centres and homes. Our products, systems and services help our customers to maximise energy use, drive efficiencies and deliver new experiences and services.
 Li, J.: Visual Inspection of 3D Printed Products. Master’s thesis, Eindhoven University of Technology (2018)
 Santos Buitrago, N. R., Tonnaer, L., Menkovski, V., Mavroeidis, D.: Anomaly Detection for imbalanced datasets with Deep Generative Models. arXiv:1811.00986 (2018), https://arxiv.org/abs/1811.00986
 Larochelle, H., Murray, I.: The neural autoregressive distribution estimator. In: Gordon, G., Dunson, D., Dudk, M. (eds.) Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics. Proceedings of Machine Learning Research, vol. 15, pp. 29-37. PMLR, Fort Lauderdale, FL, USA (11-13 Apr 2011), http://proceedings.mlr.press/v15/larochelle11a.html
 Papamakarios, G., Pavlakou, T., Murray, I.: Masked Autoregressive Flow for Density Estimation. ArXiv e-prints (May 2017)