Neuromorphic Edge Analytics for Industrial IoT

Abstract

An ever-growing quantity of on-the-move data from all facets of operations and a growing number of Industrial Internet of Things (IIoT) systems are making that demand for smart, energy-efficient processing – closer to the edge, with low latency – more and more pressing. This information needs to be sent to the cloud leading to costly communications, latency and privacy problems of traditional cloud based analytics. To this end, the authors propose a novel data analysis framework, called Neuromorphic Edge Analytics (NEA), which can process data in real-time, adaptively and can further decrease the resource usage in industrial settings. To realize the event-driven and efficient neural computing in human brain, the proposed framework combines the concepts of the neuromorphic computing and Edge Intelligence based on the SNNs.

The NEA design approach consists of four stages: Data Acquisition, Neuromorphic Processing in which concept the model is based on SNNs and used for temporal pattern recognition and anomaly detection, Edge intelligence in which it becomes a solution for local decision making and predictive analytics and Cloud integration in which model updates, data storage, system optimization etc. take place. It's not just the architecture, but an ideal one where the three following criteria could be added: low consumption, inferential speed, want to have distributed industrial systems.

In the field of predictive maintenance and fault detection, in real application, the experimental results have demonstrated high level of performance of the NEA framework in terms of latency, and bandwidth consumption with the same accuracy performance compared to the conventional deep learning methods. Moreover, it is an event-driven architecture – favorable for change in the industrial environment.

To achieve intelligent, autonomous and sustainable industrial operations, the study highlights the need for next generation of IIoT systems to adopt the neuromorphic edge analytics.