Product condition evaluation using deep learning for a circular economy

Product reuse is one of the core concepts of a circular economy, contributing to lowering demand for (critical) resources and creating more added value for the same efforts of mining, refining and production. To assure the further uptake of such circular economy practices, it is of increasing importance to develop innovative technologies and methods that can increase consumer acceptance and trust in reused goods and, at the same time, increase the economic viability of reuse practices. To date, the condition of products reaching their end-of-first-life is usually evaluated manually, which is time-consuming, costly and highly subjective. Recent developments in deep learning and computer vision have drastically increased the accuracy and speed of object detection and, at the same time, the cost of the required vision systems and related ICT infrastructure continues to decrease. However, to date there is no unified method or framework available that adopts state-of-the-art deep learning for evaluating product conditions. Therefore, the main objective of this research project is to use novel deep learning methods for both (missing) component, wear and defect detection to evaluate the condition of products and the herein contained components. In addition, multimodal fusion techniques will be explored to improve the detection accuracy. Furthermore, the added value of product model identification and the resulting availability of product-specific information from prior observations for component, wear and defect detection will be investigated.