Predictive Maintenance
Challenge Polteknik Sp. z o.o. provides metal processing machinery, including essential sheet metal bending machines used across various sectors such as automotive, appliances, and metalworking. Unexpected downtime incurs substantial costs—both direct (loss of production, service) and indirect (delays, penalties, and lower quality). The project's goal was to establish a system for detecting early signs of component wear or machine failure risks, transitioning from reactive to predictive maintenance, with the requirement that the solution be scalable and deployable globally. Solution The initiative was conducted in partnership with Polteknik Sp. z o.o., StatSoft Polska, and STIGO Sp. z o.o. Polteknik managed data collection from the machines, pinpointed vital measurement points, and installed sensors to track operating parameters. The collected data were integrated into an analytical framework to correlate deviations with the actual condition of the machines. StatSoft Polska oversaw the analytics process and developed predictive models using PCA, KNN, MCD, and autoencoder algorithms. An anomaly score was created to indicate potential failure risks, and supervised models were assessed to predict future risk trends. The system architecture was designed to be adaptable, facilitating the easy integration of new machines, sensors, and data sources. STIGO created user-friendly software that allowed for automated data gathering, continuous monitoring, and real-time alerts. Results Maintenance could be scheduled based on actual data. Components were replaced preemptively before they failed. The likelihood of unplanned downtime diminished, and service teams gained enhanced insights into machine conditions. The machine manufacturer improved its competitive edge. Summary This project illustrates how advanced data analytics and machine learning facilitate the transition from reactive to predictive maintenance, enhancing machine reliability and providing long-term benefits for both manufacturers and their clientele.