As Edge AI/ML Hardware and Software Toolsets Mature, 2024 Promises Explosion of Industrial Edge Deployments

One of the major topics at embedded world 2024 concerned deploying Artificial Intelligence (AI)/Machine Learning (ML) at the edge. Industrial computer vendors such as Advantech, ADLINK, Unigen, Eurotech, NexAIoT (NEXCOM), and Cincoze all displayed new or recently released products that will allow customers to run AI/ML inferencing at the edge.

Toolsets for making AI/ML available at the industrial edge have been maturing for the last 2 to 3 years, along a number of axes:

• Hardware accelerators, with vendors like Intel and NVIDIA providing more powerful hardware to power industrial edge computers.
• Model building, libraries, and compression, with vendors such as Amazon Web Services (AWS) IoT Greengrass and Edge Impulse standing out as vendors with growing acceptance for training and building models for industrial computers.
• Edge orchestration and management platforms, with vendors like Dell, Advantech, or Hewlett Packard Enterprise (HPE) building software platforms to manage the training and deployment of AI/ML models and inference on their edge hardware, as well as monitoring containerized deployments.
• Ecosystem development to facilitate the compression of models onto resource-constrained hardware, as well as providing marketplaces with plug-in access to AI/ML applications and model developers.

Edge for Industrial Internet of Things (IIoT) usually refers to platforms and applications running at the server or data center level. This layer lies at the intersection of Operational Technology (OT) and Information Technology (IT) environments, and is where industrial customers increasingly want to run their operations. Vendors such as Litmus Automation natively run at this level of the edge, but industrial vendors such as Emerson, Rockwell, and others have been developing their own edge environments to deploy and manage applications closer to where the data are generated, often with the help of specialist software edge orchestration vendors such as ZEDEDA. The server level is a good point for industrial companies to deploy edge AI/ML because this is the point where IIoT data from across a facility are aggregated after extraction from lower-level controllers.

Increasingly, vendors are also looking to deploy AI/ML one level lower, namely on the machine, production line, or gateway level. This increasingly distributed computing infrastructure allows analysis of streaming data with lower latency responses. It should be no surprise that initial use cases are focused on vision, specifically around object detection, and leading toward defect detection—namely, use cases that create large quantities of data, require low-latency responses, and for which models are comparatively simple to train and customize with existing AI/ML toolsets. While vision use cases are the current sweet spot, time series data analysis at the edge are following rapidly and will be the next development stage for process control, anomaly detection, and maintenance use cases.

embedded world 2024 highlighted the intersection of edge AI/ML toolsets for model and application development on the one hand, with the deployment hardware and software on the other. While deployments of AI/ML models in edge environments are currently limited in the IIoT, the increasingly tight ecosystem will change this. One area where vendors should pay greater attention in 2024 is partnerships with vertical-market specialists, in particular industrial application platforms. These vendors hold a central position in deploying edge AI/ML technologies, with access to large stores of data ingested from the edge, and providing the piping to get data from the field to applications. System Integrators (SIs) will also be paying closer attention, given the current consultative process for building and deploying AI/ML models and applications, creating an opportunity for them to provide an additional professional and managed services layer.