Optimizing Industrial Systems: AI and Kalman Filters in Pump Fault Detection

Our exhibit shows how combining AI techniques and Kalman filters makes pump fault detection more accurate. Our demonstration uses sensor data from ITT Gould Pumps. Visitors can observe a simulated pump displaying real-time vibration and flux signals, with live data visualizations of time-waveform and FFT graphs. To emphasize how raw sensor data becomes noisy during faults, a pump will switch between normal operation and common failure modes, like cavitation, misalignment, soft foot, or dry running. A split-screen contrasts raw signals against data preprocessed with a Kalman filter, highlighting its ability to stabilize measurements by filtering noise. Visitors will see a deep learning (DL) model predict the pump’s condition, using both data types. A side-by-side comparison shows the model’s predictions using raw versus Kalman-filtered data, demonstrating how preprocessing with Kalman filters improves failure detection accuracy. This hands-on demo makes it easy to see the power of combining classical state estimation with modern AI for tackling sensor noise in industrial systems. Through clear visualizations and direct comparisons, visitors learn how preprocessing sensor data enhances AI-driven predictive maintenance. A side-by-side comparison highlights how using Kalman-filtered data boosts prediction accuracy.