Empowering Industrial IoT with Edge Intelligence: a comparative study

The Industrial Internet of Things (IIoT) is revolutionizing the landscape of industry 4.0 by interconnecting devices, operators, and processes to drive digital transformation.

This transformation, however, encounters its set of challenges when solely reliant on cloud computing due to limitations in bandwidth consumption, data sovereignty, and privacy concerns.

The evolution of Edge Intelligence (EI) emerges as a solution, blending the capabilities of edge computing with artificial intelligence to process IIoT data closer to the source of data generation.

This approach not only addresses the limitations posed by cloud-centric deployments but also opens up additional opportunities for industrial applications, particularly in safety-critical tasks based on computation- and bandwidth-intensive operations such as video analytics.

Industrial use case: safety-critical task monitoring

The pivotal use case under consideration involves a mechanical factory workshop equipped with temperature sensors, video cameras, and IoT wearable devices for workers.

This setup aims to monitor environmental conditions and ensure worker safety by identifying potential hazards such as fire outbreaks and sending out an alert system to workers in proximity of potential hazards through their personal wearable devices.

In the following sections, we explore and compare two deployment strategies: one based on cloud computing and the other on EI, both utilizing a RaspberryPi4 as the IoT device and Amazon Web Services (AWS) for their computing needs.

AWS, in particular, offers a broad deployment ecosystem, including services such as computing power, storage, networking, and security.

Cloud-based deployment

Proposed Cloud-based deployment

This architecture leverages the cloud for data processing and storage. It involves collecting temperature data through sensors connected to a Raspberry Pi4, which then transmits this data to the cloud for analysis.

Upon detecting an anomaly, such as a temperature spike indicating a fire, video streams from the workshop are processed in the cloud using machine learning models to identify the fire source and alert nearby workers through their wearable devices.

Edge-based deployment

Proposed EI-based deployment

Conversely, the edge-based deployment brings intelligence closer to the data source. It utilizes AWS IoT Greengrass to extend cloud services to the edge, enabling local processing of temperature data and video analytics.

This approach ensures prompt detection and response to safety hazards by processing data on-site without relying on cloud connectivity, thereby enhancing response times and operational efficiency.

Comparative results analysis

A comparative analysis of both deployments highlights the advantages and trade-offs of each deployment:

Responsiveness

The time needed to analyze a frame of video is fastest in the Cloud-based system, at only 0.01 seconds per frame. The RaspberryPi4, used in both systems, is slower at 0.28 seconds per frame. However, incorporating a Google Coral Board into the Edge system significantly speeds up processing to less than 0.07 seconds per frame, thanks to its specialized Edge TPU.

Additionally, uploading video to the cloud takes 4 seconds, but alerting wearables happens quickly in both setups, in under a second.

Bandwidth usage

The Cloud-based deployment requires sending video to a remote server, consuming significant bandwidth, especially with high-resolution footage. This can be a major issue when many cameras are in use, potentially slowing down the network. Compressing video can reduce bandwidth use but might lower analysis accuracy by up to 35%. In contrast, the Edge-based system uses minimal bandwidth, just a few KBs, as most processing is done locally.

Energy Footprint

The RaspberryPi4 in the Edge setup uses about 3.1 watts for reading temperatures, up to 5.5 watts when fully operational. This is considerably less energy than what's consumed by cloud servers, which not only use much more power but also require extra energy for cooling, networking, and supporting systems, even with efficient power management.

Conclusion

EI promotes moving computational and analytical capabilities to the data's source rather than relying on centralized cloud computing or remote servers.

As shown in the comparative results analysis for the use case shown in this article, such paradigm shift in IIoT deployments offers a strategic advantage in terms of responsiveness, reliability, privacy, and operational efficiency.

By processing data at the edge, industrial applications can achieve faster decision-making and enhanced safety protocols, pivotal for the advancement of Industry 4.0, paving the way for a future where industrial operations are safer, more efficient, and resilient.

If you’re interested in the field of EI for IIoT, particularly around industrial safety and operational efficiency, we recommend the paper "Edge Intelligence for Industrial IoT: Opportunities and Limitations", by Claudio Savaglio, Pasquale Mazzei, and Giancarlo Fortino.

Published in the "Procedia Computer Science,", DOI https://doi.org/10.1016/j.procs.2024.01.039 This paper offers a thoroughly detailed version of the comparative analysis of cloud-based versus edge-based deployment strategies explored in this article, underlining the transformative potential of EI in advancing Industry 4.0.

Read now in ScienceDirect