MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111023 A) filed by Pragati Engineering College on September 16, 2026, for Attention-Based Anomaly Detection In Industrial Iot Sensor Streams.
Inventors include Dr. Prasanth Varasala; Mr. V. V. N. Sarath; Mr. Kotikalapudi Siva Sankar; and Mrs. P. Ramya Krishna.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: The Industrial Internet of Things (IIoT) supplies unrelenting information of sensor data from machines, production lines, electrical system, environmental monitoring and industrial control processes. Such sensor streams are useful for inferring information about the condition of the industrial asset, but can also include abnormal patterns that can be indicative of equipment degradation, sensor failures, communications failures, cyberattack, unexpected operating conditions, or process disturbances. Correct early identification of such anomalies is vital in minimizing equipment downtime, enhancing the safety conditions of the industrial environment, ensuring production quality, and also to facilitating predictive maintenance. Typical methods for anomaly detection rely on fixed thresholds, statistical models or popular machine learning methods, and these methods have a tendency to fail to detect complex and time-varying abnormal patterns in multivariate industrial sensor streams. This work introduces an Attention-based Anomaly Detection Framework for Industrial IoT Sensor Streams, which integrates the temporal sequence modelling and attention mechanism to detect the important sensor observations and time intervals of abnormal system behaviours. The proposed framework preprocesses the continuously gathered sensor data which includes data cleaning, data normalization, missing value filling and time window generation. A deep learning sequence model is then learned to capture temporal relationships between other sensor variables, and an attention mechanism is applied to give more attention to the observations that play a significant role in anomaly detection. The model calculates an anomaly score for each time window, and an adaptive decision mechanism categorizes the window as normal or anomalous. The framework can be used to identify both completely abnormal events and changes over time from normal operating behavior. Several metrics can be used for experimental evaluation, including accuracy, precision, recall, F1-score, ROC-AUC, false alarm rate and anomaly detection delay. The proposed attention-based approach aims to achieve (1) better detection performances and (2) provide interpretability of the anomaly decision by highlighting which sensor features and temporal regions contributed to the decision. The framework thus offers a practical, AI-based solution for intelligent monitoring of an IOT environment, which can be applied to early warning, predictive maintenance, equipment health monitoring and industrial safety.
Disclaimer: Curated by HT Syndication.
