Using ChatGPT for AVEVA Intouch, System Plaftorm and OMI

Hi Rainer,

As you can imagine we've also experimented with ChatGPT quite a bit  .

I am definitely interested in getting some more formal solutions, but I've had great success with scripting in particular. I just taught ChatGPT as best I could on how to write QuickScript.NET, and the results were pretty solid if not perfect. I learned early on to tell it to only support .NET 1.1 to avoid it trying to use things like generics and lists etc. I also splurged on the paid-for version of ChatGPT, which allowed me to feed it the Scripting.PDF document. One of the early successes I had was when I wanted to integrate weather information into objects in System Platform. ChatGPT did not only write the script for me, but helped me get access to the correct (free ) website that would supply the necessary information for me. At the time it took me less than 10 minutes to implement and I was really impressed!

One of the things that we want to improve in an upcoming version is to allow more text-based exports of our configuration. This would include an XML-based export/import of Industrial Graphics (you can already do this today using the GR Access toolkit, but I want this to be a menu option). One of the things that this would allow us to do is to teach ChatGPT what the inside of that XML looks like and of course how to create one on its own... I believe that would lay the groundwork for a GPT that would allow us to create industrial graphics!

We are also working on exporting of other configurations in text-based formats including Templates and Attributes which should further allow us to teach ChatGPT how to create a model and probably more importantly the templates used to create that model.

Up until that point we would still be using ChatGPT in its native format which as you know can sometimes result in hours of experimentation to attempt to find the ultimate prompt . But once all of these things are in place, I think it would just be a short step to create a custom GPT that we can teach up front to do a lot of these things. One of the challenges may be to find examples to feed it but I am sure that if we could overcome this, we could build a special "System Platform GPT" that can be a very quick leg-up when creating a new solution.

Of course, I am looking at this through my "System Platform & OMI" eyes and other products may see other opportunities to leverage it.

Hi Rainer,

ChatGPT does an impressive job leveraging public information to provide answers to text based questions. But much of what you are describing above needs a better understanding of the private data used in AVEVA applications (specific models and data in). That is exactly where AVEVA Industrial Assistant is heading. Initially it is targeted as a support tool for end customers to get more value from their production data. But we have also begun exploring how this capability could transform the configuration and management of systems. Expect to hear much more about this soon as the team is gearing up to put this in production later this year!

AVEVA previews industrial AI assistant in collaboration with Microsoft at Hannover Messe

We are also looking into how AI would apply to Enterprise SCADA. Things like, automatically summarise the previous shift for a handover report, automatic screen comparisons as part of Parallel Operation Verification step that most of our pipeline users must do as part of their regulated change management process. 

We are also exploring the potential for AI to automate routine tasks performed today by the operator following multiple steps. The Enterprise SCADA Automation Bundle for example could receive an Action Sequence that has been proposed by the AI, for the operator to approve and launch. Keeping the Human in the Loop will be essential here, even if we were able to do things like a What if analysis which would perform the action sequence against a simulator to show how the asset will likely respond to the sequence of commands.

Other avenues to explore would be HMI screen translations to streamline the adoption of our next-gen HMI, and code completion as part of our SDKs. However, AI hallucinations remain part of this landscape today, so human inspection and comprehensive testing would be required before putting anything into production.

Asset Inferencing is another promising avenue of investigation: a way to keep up to date the various mappings between different data models as part of the Unified Namespace / Common Data Model solution that would facilitate Common Configuration.

Lastly, I have been suggesting that customers think ahead about what kind of data to collect given that we dont yet know what data the AI would need to be trained on. So, having as much time-corrolated, clean data as possible in a central historian that is normalised as much as possible to a master asset model will be invaluable in the future, once a few years of such data is available to train on. This should include data from outside the control system as well, such as maintenance and commercial data. 

A general question, does anyone have any privacy or security concerns about submitting their data, manuals, display, configuration, etc. to be added to ChatGPT's model? Or is this something that we should be comfortable with?

Hi Ernst, this is what we are looking for, adding any data to make our developer lives easy. Your example for weather data is exactly what we will be using ChatGPT for.  Bang on on the export and import as an easy option, thats exactly what we need. 

Honestly we would like to use ChatGPT in its native format rather than a product, maybe you can have an add on within ChatGPT if you need to incentivise on it , riding the wave of what you suggested, i would like some more things:

1) Creation of tags as a JSON file that i can import within AVEVA , if chatgpt can create this JSON from excel input, that would be amazing

2) Simple xml to change some graphics

3) Scripting ( but you already touched on it )

4) Understanding logs, say if put the logs from OCMC can it tell me what happened

I think even if we have some videos on how to use ChatGPT in its native format with Intouch/ AVEVA Portfolio that would be amazing, product intergation can come much later

Hi John, for some of us , creation of tags, finding scripts or some ways to implement like weather data example, we don't have problems with data privacy. A product like Industrial assistant would be amazing but im not sure how much of that is reachable to a common system integrator/ end user who doesn't have a huge budget or doesnt have Flex. We would want something thats more reachable even if it's ChatGPT in its native format or some plug in charges

Hi Jake, thanks for your reply, I'm working for an AVEVA Distributor and i have seen Enterprise SCADA but I'm embarrased to say I don't understand Enterprise SCADA, initially thought it was System Platform Enterprise but when i went through it said about Oil and Gas Pipeline SCADA and much more. When I see this, I'm thinking Enterprise SCADA is beyond the reach for most customers but I would definitely like to understand more. We have big customers but none use this, would like to know what it does and where it fits in

Hi Rainer - yes, it is not surprising that Enterprise SCADA is not as widely known as System Platform. Formally known as OASyS from the Telvent company, it specialises in Oil & Gas, but also very large regional water districts, and has been used in the Transportation, Electrical and Weather industried to a lessor extant in the past.

Its strengths are suited for very large systems (geographically dispersed) with high availability requirements (99.999% uptime), with multiples layers or arbitration and redundancy, and a fully distributed operating model with a flexible deployment architecture.

Out of the box it supports all the current North American standards for Midstream transportation, as well as downstream distribution, particularly around the Control Room experience and hydrocarbon measurement.

While waiting for others to respond - we are well aware of the concerns with private data being sent to public systems like OpenAI. This would be a great question to ask Jim Chappel at either AWC in Paris, or the PIpeline Summit in Kanaskis this year.