SME Review - Workshop outline

[DawnmarieDesJardins]

Please place all feedback to the workshop outline in this issue by continuing the conversation below.

[KalyanBondugula]

Kalyan Bondugula: [Dawnmarie – Here is my high level feedback.

• Overall, the connected factory will resonate more in the “factory floor” setting – say in discrete manufacturing. Not sure if Oil field is a great example of that. You should check few connected factory case studies – Jabil/Sandvik etc. • The use case introduction does not talk about any data collection/devices related point. Focused only on the Machine Learning aspect. Looks more like intro for ML workshop than IoT. • We need to talk about “Edge” processing and “OPC UA” framework related points which are key for Manufacturing/Oil & Gas customers. • SaaS is one solution option, customers may also be interested in “Build” with “PaaS”. Should be considered as an option for solution]

[mpeder]

@DawnmarieDesJardins here is some high level feedback from me as well:

• agree with @KalyanBondugula that this intro sounds more like an ML workshop and very little IoT.
• we need to mention the capabilities and selling point of using SaaS IoT Central as much as ML and Databricks.
• Would exercise 8 also include how to actually notify someone? If not then I think it should :-) In other words let's me clear on what " Integrating the predictive maintenance model into the solution" actually means
• I think a key point to make something like this real is how they will host the models later, manage and deploy them? The lab you point to in the bottom uses AMLS and Azure DevOps and it us unclear whether this will also be part of this workshop?

[olivierbloch]

I like the outline and only have a few suggestions:

• as discussed in review meeting on 5/30 I agree that we should have pre-trained models and available notebooks to reduce the complexity of the data science part and focus more on the IoT part (collection of data, continuous export, maybe add an alert on some telemetry threshold to trigger some action, control of the pumps
• IoT Central now supports exporting and importing app templates (preview announced today), which allows to create and make available a rich set of dashboards that the student can simply import as a template. This will give the IoT Central app the look and feel of a final product.

[krishna0815]

Few suggestions from my side.

1. Remote field services esp in Oil and Gas / Offshore exploration typically merits edge analytics. Can we morph a part of this into edge story

2. ASA offers built-in ML based functions for Anomaly Detection - both in cloud and edge based on now open sourced ML.net offering. Meaning, for simple A/D scenarios customers do not need to build and train their models. We should try and include this as a part of the crawl-walk-run approach within the ML spectrum.

3. Geospatial capabilities - ASA has built in Geospatial functions, again on cloud and edge, to continuously detect distance between assets, generate alerts if an asset enters or leaves a geofence etc. For oil fields etc, based on real customer scenarios safety is critical. So customers like Petrofac (now in prod) want to generate alerts if a worker with inadequate training enters a high risk site. Essentially, they have a matrix of workers (single coordinates) and allowable sites (geofences) that ASA deploys in the cloud and edge.

4. Talking of general analytics, i recommend introducing audience to simple concepts like time windows, native output in parquet for downstream batch processing and ML training with Spark, late arrival policies etc. These are very commonly used capabilities in many IoT scenarios.

Happy to help guide on the implementation of these features.

[krishna0815]

Please accept these views as applicable to the 'Securing IoT end to end ' workshop as well.

[krishna0815]

In effect.. -

• ML and Data filtering on the Edge: compute happens here in real-time with low latency
• ASA runs in cloud for dashboarding and cross-device analytics
• ML model trained in the cloud with Databricks