Präventive Überlastungsvermeidung von Kugelmühlen mittels Machine Learning & KI
Wir haben ein maschinelles Lernsystem entwickelt, das Überlastungen von Kugelmühlen 30–40 Minuten vor ihrem Auftreten vorhersagt. Dies verschafft den Bedienern die nötige Zeit und Übersicht, um die Mühlenlast und Betriebsparameter anzupassen. Zuvor hatte der Kunde mit 9,1 Tagen jährlicher Ausfallzeit und übermäßigem Energieverbrauch zu kämpfen. Das neue prädiktive Modell stabilisierte die Produktionszyklen erheblich, reduzierte die Ausfallzeiten um 30 % und senkte den Energieverbrauch von 530 kWh auf 390–420 kWh.
01 —The Impact
The results, up front.
30% reduction in downtime Energy usage decreased from 530 kWh to 390–420 kWh Annual impact of $0.85M–$1.16M Improved mill stability and operator confidence More predictable and efficient production cycles
02 — The Challenge
Where they started.
The mill operation lacked reliable predictive visibility. Operators had to react based on lagging indicators, leading to sudden overloads, emergency shutdowns, and increased wear. Energy consumption remained very high due to instability in mill loading. We started by analyzing sensor data, energy profiles, operating modes, and historical overload events. Our team engineered features across vibration, torque, feed rate, ball-loading, and acoustic signatures to build a robust predictive dataset. A supervised ML model was trained to identify precursor patterns signaling overload conditions. Once deployed, the model continuously monitored live streams and triggered operator alerts 30–40 minutes in advance through an integrated dashboard. We also optimized operating regimes by identifying ball-loading configurations that reduce overload probability. The project combined data science, operations optimization, and real-time monitoring to deliver sustained performance improvements.
04 — Approach & Methodology
How we got there.
The ML system surfaced hidden variable correlations previously impossible to detect manually. Overloads strongly correlated with specific amplitude behaviors in vibration signals and certain ball-loading ranges. Predictive insight enabled proactive load adjustments and optimized mill operating modes.
With the overload risk dramatically reduced, operators transitioned from firefighting to proactive cycle management — stabilizing output and reducing both energy intensity and mechanical stress.
05 —In Practice
Project samples.
Keep exploring Renascence
Your turn
Ready to write your own success story?
Book a discovery call and see what behavioral CX can do for your business.