Decreasing maintenance costs with Deep Learning

Costs for maintenance are a very relevant factor in manufacturing, typically they make up 15% to 40% of total costs. Approximately 30% of maintenance costs occur because of early, maybe unnecessary replacement of machine parts or delayed maintenance work and an increase in unplanned machine outages. The Austrian project COGNITUS aims at decreasing maintenance costs through a combination of datadriven maintenance planning and Deep Learning. One use case within the project focuses on the detection of anomalous pallets in a high-bay warehouse. The use case is conducted together with SPAR Austria, one of the country’s major retailers. Through Exploratory Data Analysis (EDA) the sources for errors in the warehouse were analysed. After finding out, where and when errors occur, cameras with different viewpoints were installed at the crucial sites. The stream of pictures coming from the cameras was progressed with the help of Machine Learning, focusing on the different kinds of errors pallets could potentially arrive with at the warehouse. Through the clustering of pictures and with the help of “fuzzy rules” operators should now be able to choose the necessary quality of the pallets coming into the warehouse. Through a dashboard, the employees can directly observe faulty pallets and the kind of anomalies the pallets are arriving with. This approach should lead to the reduction of maintenance costs in the warehouse. The COGNITUS project will be completed and evaluated in the third quarter of 2022.