Living on the Edge: OT and IT for AI and IoT
Industry 4.0 is is often used to describe the data-driven and AI-powered Smart Factory. The critical nature of such industrial applications cannot tolerate the WAN latency of the cloud, so any data must be processed locally at the Edge site where it’s generated from myriad IoT devices. In this scenario, Edge represents the intersection where teams from both IT (Information Technology, responsible for tools necessary for data management and generating insights) and OT (Operational Technology, responsible for control systems and automation around operations and IoT devices) must work together to achieve more efficient operations, improve human-machine interactions, and enhance data analytics.
However, OT and IT both have their own challenges and priorities. Operational technologists want to consolidate and reduce infrastructure sprawl to eliminate IoT silos. They also want management simplicity for planet scale operations, zero-touch on-boarding and easy connectivity to different types of sensors. Above all, systems designed for OT should include built-in infrastructure and data security with multi-tenancy.
Information technologists like the convenience of a SaaS management plane and the flexibility to bring their preferred cloud as well as ML models developed in any domain. Moreover, access to rich data/runtime services is a must-have to execute AI inference at the edge. They want the extensibility of programming frameworks, rich/open APIs, integration with CI/CD pipelines, first-class IDEs, and easy debuggability.
To accelerate the digital transformation journey, organizational structure is evolving with Chief Digitization or Chief Innovation Officers (or other CXOs) overlooking both OT and IT functions so that they can work in lock-step (yin and yang) towards a common goal. CXOs are looking to optimize business processes to gain structural and sustainable competitive advantages while at the same time reducing the costs and lowering the risks.
A quick glance at the hardware portfolios of vendors (also Nutanix partners) like HPE (Edgeline), Advantech, and Supermicro clearly shows hardware convergence is already happening, but software unification as an intelligent edge fabric is severely lacking!
Introducing Nutanix Xi IoT: Intelligent Edge Computing Platform
The Nutanix Xi IoT platform is a 100% software-defined solution that delivers local computing, machine learning, and intelligence for your Edge devices, converging the edge and your choice of cloud into one seamless and delightful application development platform. The Xi IoT Platform eliminates complexity, accelerates deployment for OT, and elevates IT to focus on the business logic. The edge PaaS supports easy-to-use developer APIs, reusable data pipelines, and pluggable machine learning architecture to enable rapid development and global deployment of modern IoT and AI apps.
Let’s take an example from Industry 4.0 real-world Pilot leveraging the Xi IoT Platform.
Industry 4.0 Use Case
The use case is to ensure the quality of packaging for meat products in an automated way. The customer’s previous method was to perform statistical sampling through manual inspection. The implications of error in the manual approach were very high as even a single meat product becoming rotten, owing to bad packaging, would not only pose a health hazard risk for potential consumers, but result in the costly recall of an entire batch.
Image Analytics and Real-time Inference (Quality Check)
The OT team set up an industrial-grade Basler camera, mounted a few feet from the conveyor belt. The camera scans the packaged object and sends high-resolution streaming images over the GigE Vision interface. The camera is connected to the Xi Edge, which is running as VM on existing virtualized hardware at the production site. The native GigE vision connector on the Xi Edge ingests the frames, performs a simple conversion, and places them onto the in-memory data bus.
The IT team leveraged the Data pipeline construct to pick up those frames from the bus and apply a Function, which captures the business logic written in Python. This function also invokes the custom ML model, which was developed by the customer and imported into the Xi Edge Model Library. The industry-standard Tensorflow framework and the run-time environment is built natively into the platform, making the model itself very easy to import. The ML model then uses the Xi “Infer” API to calculate the “quality threshold” based on the package image. An Nvidia T4 Inferencing GPU was leveraged for real-time decisions.
The business logic captures the two possible workflows — Actuation and Re-training.
Robotic Arm Actuation (Quality Control)
Workflow A (as in the image above): If the image is detected with high confidence and the quality threshold is deemed to be low, then the Function outputs to a MQTT topic which is subscribed to by Kepware Kepserverex software. Note that OT team leveraged a Xi Edge generated certificate for secure MQTT communication between KEPServerEX and Xi Edge. The Kepserverex converts the message into a PLC-specific format that actuates a robotic arm to pick up the faulty packaged product from the conveyor belt and put it into a separate bin!
Re-training (Improving the Quality Check mechanism)
Workflow B (as in the image above): If the confidence level itself in the detection of the image is low, then it means that either the package is of a new type or perhaps the model itself is not rich enough to capture the different facets of a previously known package. In either case, the model needs to be re-trained on these new data sets. The global data pipeline construct is used to send the image data from the local Edge to the central (to aggregate over multiple edge sites) GCP Cloud Store instance.
Note that this was the customer’s own public data cloud. In this case, it was GCP, but could have been another public cloud like AWS, or even a private cloud. Google AI Platform was used to re-train the model on this new dataset and the new model was uploaded to the cloud repository. The customer’s CI/CD integration with the Xi IoT Platform made the automatic redeployment of the updated model onto the Edge very simple and seamless.
Assessment (Improving Operations)
While the quality check and control use case was achieved based on the aforementioned steps, it was also deemed very important to investigate the the packaging quality errors so that further operations could be improved. For this improvement, the OT team configured the package data [image and weight] and machine data [from temperature and vibration sensors] to be collected from various sensors via Kepware and sent to Xi Edge as two separate MQTT topics from the same source. The IT team created Xi Edge based Functions in the Xi Edge to correlate the disparate sensor data to one packaged entity and send the aggregated info as JSON to GCP by Data Pipeline for further offline analysis in the cloud.
In a nutshell, you read how Nutanix Xi IoT platform enabled both OT and IT to come together and deliver an end-to-end Industry 4.0 working solution. While we discussed one specific use case in this blog, a similar framework could be leveraged for multiple other use cases or different industrial settings, as deemed fit for your own Industry 4.0 journey. If you are a CXO or representing OT or IT teams, don’t hesitate to contact your local Nutanix Sales team or reach out directly to the Xi IoT team by sending an email to iot [at] nutanix [dot] com. Let’s be on the cutting edge together!
— — — — — — — — — — — — — — — — — — — — — — — — — — — — — —
Fun Stuff: Living on the Edge, literally!