Multi-sensor cyber-physical sorting system (CPSS) based on Industry 4.0 principles: A multi-functional approach
Fotios K. Konstantinidis, Savvas Sifnaios, Georgios Tsimiklis, Spyridon G. Mouroutsos, Angelos Amditis, Αντώνιος Γαστεράτος
Abstract
Products global demand constantly grows, along with drifts of the volume of waste produced, thus having a negative impact on the environment. Besides the advancement of artificial intelligence and cutting-edge sensing and actuation systems, allow for traditional recycling procedures to be digitised and become more efficient. Among these, automated sorting machines allow such efficacy, by using multiple recognition technologies to categorise items, instead of relying on humans to separate them. In this sense, this paper presents a new approach being included in the Cyber-Physical Sorting System (CPSS) capable to be adapted in different industries such as urban wastes, construction & demolition wastes as well as mineral sorting. The materials are laid out over a conveyor belt, where multi-purpose visual sensors (hyper-spectral, industrial and short-wave infrared) acquire images in order to classify and sort them in different bins using robotic arms. The architecture of the proposed CPSS allows the integration of additional sensors, increasing the interoperability with upcoming sensing devices. Furthermore, an intelligent hyper-spectral component adapted to urban wastes was developed that classifies glass, metal, paper, and plastic using multi-spectral data in visual and near-infrared ranges between 470nm - 975nm. Scatter-correction methods and derivatives were applied to extract noise-free features, while an exploratory analysis based on simple and robust Principal component analysis (PCA) was performed to identify the most significant principal components for the training phase. During the evaluation, the neural network performed better when normalisation and Savtizky and Golay (SG) filtering was applied to the dataset against other classifiers.