MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202611068871 A) filed by Dr. Sachin Malik; Dr. Prem Shankar Jha; Dr. Sanjey Kumar; and Dr. Surya Kant Pal on June 01, 2026, for Anomaly Detection Model For Real-Time Equipment Failure Prediction Using Multivariate Sensor Data.
Inventors include Dr. Sachin Malik; Dr. Prem Shankar Jha; Dr. Sanjey Kumar; and Dr. Surya Kant Pal.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: Industrial equipment failure prediction from multivariate sensor data is critical for minimizing unplanned downtime and optimizing maintenance schedules. Existing approaches fail to simultaneously capture spatial inter-sensor correlations and temporal degradation patterns with sufficient accuracy for real-time deployment. The present invention discloses a hybrid deep learning anomaly detection model combining one-dimensional convolutional neural network layers for spatial feature extraction across multiple sensor channels with bidirectional long short-term memory network layers for temporal pattern modelling within configurable observation windows. A multi-head attention mechanism provides interpretable sensor and temporal attribution scores identifying failure-relevant channels and time periods. The model outputs a continuous health index score and categorical failure mode classification. Experimental validation on industrial bearing and CNC machining datasets demonstrates detection accuracy of 96.8 percent, false positive rate of 1.7 percent, and average failure prediction lead time of 6.4 hours.
Disclaimer: Curated by HT Syndication.
