I-Corps: Development of decentralized anomaly detection for industrial facilities

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of technology that will aid digital controllers in detecting and identifying anomalies in industrial systems. Every industrial sector relies on industrial control systems to function. There is significant commercial interest in technologies which reduce the risk of unplanned failure and improve worker knowledge and operational effectiveness through data-driven insights. Existing solutions share common issues around cybersecurity, cost of installation, workforce knowledge on how to use the solution, and the ability to explain unusual observations. By separating from the cloud and focusing on the autonomous and explainable capabilities of the technology, technicians using this technology may be able to perform their jobs rapidly and more effectively without needing be experts in artificial intelligence / machine learning (AI/ML) themselves. Adopting a “plug-and-play” approach will reduce the barrier to entry for deploying analytics on smaller systems, thereby enabling better awareness of facility states and avoiding facility degradation before it becomes urgent. The technology will also benefit underrepresented entities in the current industrial solutions landscape as the solution may reduce the cost of implementation and the infrastructure redesigns. This technology may reduce industrial downtime and improve overall facility and process reliability, thereby reducing costs of US-based manufacturing. This I-Corps project is based on the development of an autonomous, edge-based analytics technology for decentralized and scalable fault detection. The core technology is an efficient software which can detect faults in an industrial control system autonomously using on a very small computational footprint. Such a solution will make the deployment of analytics possible even for small-scale systems which are normally overlooked in analytics initiatives. The proposed technology autonomously collects, analyzes and builds analytical AI models based on data from digital industrial assets in the field and then uses these models to continuously monitor for signs of anomalous behavior corresponding to equipment degradation or malfunction, all with no connection to the cloud. The software is derived from kernel-based regression methods, such that all predictions and decisions can be directly introspected and explained. These capabilities are paired with a simple and easy-to-understand user interface which can be interactively used to identify the root signals driving a given alarm system. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.