Open Access

Timo Schroth, Vera HummelORCiD, Konrad von LeipzigORCiD, Jan SchuhmacherORCiD

• The industrial sector is evolving towards increased customization, diminishing batch sizes, and shorter product lifecycles, affecting intralogistics, which faces challenges in managing an expanding variety of parts and variants. This diversification leads to a decline in efficiency owing to the complexity in pick and stow operations, as traditional systems, digital solutions, and optimization methods mainly rely on historical data without incorporating near-real-time process information. Conventional approaches separate pick and stow operations in both process and workforce, culminating in extended process durations. Instead, data-driven AI-based methods offer a solution by clustering and combining pick and stow operations into optimized bundles, considering travel distance and time. The research employs AI algorithms to streamline picking and stowing, aiming to enhance logistics performance by reducing travel distance and time. Due to the absence of real data, a simulation-based procedure to generate synthetic test and training data is adopted. The real-world logistics system of the learning factory Werk150 is modeled in AnyLogic simulation software to carry out picking and stowing in a 3D warehouse layout. This database is leveraged to train an unsupervised machine learning model using the data analytics software TensorFlow by applying algorithms focused on clustering and combination. A comparative study of these algorithms is conducted to pinpoint optimal strategies for improving logistics performance. Future research will target this methodology, which will be enriched by experimental tests in Werk150 involving near-real-time data, practical investigations, and the use of real data to conclude with an analysis to validate the optimization strategies' effectiveness.