MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111327 A) filed by G Ashwin Prabhu; Mr. R. M. Saravana Kumar; Mr. A. Sam Daniel Fenny; Mr. P. Abbas Kiyasudeen; Mr. Vijaykumar B P; Mr. S. Balaji; Mr. S. Sivakumar; Mr. G. M. Pradeep; and Dr. Dhayanithi J on September 16, 2026, for An Artificial Intelligence Based Predictive Maintenance System For Rotating Mechanical Equipment Using Vibration And Thermal Data Analysis.

Inventors include Mr. R. M. Saravana Kumar; Mr. A. Sam Daniel Fenny; Mr. P. Abbas Kiyasudeen; Mr. Vijaykumar B P; Mr. S. Balaji; Mr. S. Sivakumar; Mr. G. M. Pradeep; and Dr. Dhayanithi J.

The application for the patent was published on September 25, 2026, under issue no. 39/2026.

Abstract: This work proposes an Artificial Intelligence Based Predictive Maintenance System for rotating mechanical equipment, including motors, pumps, fans, gearboxes, and bearings, using combined vibration and thermal data analysis. The system is designed to identify developing faults before functional failure, thereby reducing unplanned downtime, maintenance cost, and secondary equipment damage. Tri-axial vibration signals are acquired at 12.8 kHz, a rate selected to capture bearing and gear-related frequency components up to 5 kHz while satisfying the Nyquist criterion. Thermal data are collected at 1 Hz using an infrared sensor with an accuracy of ±0.5 °C, because temperature changes occur more slowly than vibration variations but provide valuable evidence of friction, lubrication loss, and overload. For each 10-second monitoring window, the system extracts time-domain features such as root-mean-square acceleration, kurtosis, crest factor, and peak-to-peak value, together with frequency-domain features obtained through Fast Fourier Transform analysis. Thermal features include absolute temperature, temperature-rise rate, and deviation from the equipment’s learned operating baseline. A machine-learning model, such as a random forest or lightweight neural network, combines these features to classify normal operation, imbalance, misalignment, bearing damage, and overheating. The proposed prototype targets a fault-classification accuracy above 92%, a false-alarm rate below 5%, and an early-warning period of at least 24–72 hours under gradually developing fault conditions. These targets are justified because industrial users require high detection reliability, limited unnecessary maintenance actions, and sufficient lead time for inspection and spare-part planning. Sensor data are processed in 2-second intervals, enabling near-real-time alerts while maintaining manageable computational demand on an edge controller. The system also generates a health index from 0 to 100 and estimates maintenance priority using fault probability, severity, and temperature trend. By integrating vibration and thermal measurements, the proposed solution is more robust than single-sensor monitoring and offers a scalable and cost-effective approach to condition-based maintenance.

Disclaimer: Curated by HT Syndication.