Taxonomy for Industrial Asset's Q&A

[DhavalRepo18]

Describe the proposed contribution to the taxonomy

We study existing compositional skill and find a big gap for industrial assets. We have a extensive expertise in contributing the taxonomy for industrial assets and Q&A around it.

Here is a simple example:

How to read above tiny taxonomy for compositional skills: Industrial assets have many components and these component fails over a time and that can be detected using various sensor data. We have many nodes to be added progressively.

The attempt is to harmonize the development of taxonomy for Industrial applications specially Industry 4.0.

Input given at the prompt

The fine tune model will be better aligned with end user need for Industrial applications to answer what, where when and how for an asset lifecycle. and reduce a burden of maintaining an in-sample example for building an application.

[lehors]

First, thank you for bringing this up as an issue before submitting any PR on this. This will give us a chance to properly frame the problem and make sure we agree on the best way forward.

Then, it ought to be said that this the kind of input we are hoping to get: something that addresses a specific gap and that contributors can address drawing from their own domain of expertise. So, thank you, your proposal is definitely interesting!

With that being said, while you talk about skills it seems likely that this would include or combine some knowledge as well. We're not quite set up for handling knowledge yet so you'd have to clearly separate the two and only submit skills if possible. Otherwise you may need to wait to be able to add the necessary knowledge first, before being able to submit the related skills.

The proposed tree structure looks fine. You'll have to build it up incrementally as you add skills though, one PR at a time, rather than the whole structure as a single PR. This is so that our backend process can evaluate the impact of the proposed skills, one by one, on a PR basis.

I would suggest to refer to this issue in each related PR so that reviewers can get the broader picture.

@obuzek may have more to add.

[DhavalRepo18]

@lehors Thank you for your detailed feedback. Actually, here is the detailed path we may feel more reasonable:

The compositional skill can be contributed back to base model, but knowledge may be internal property of respective product. So knowledge will come in step 2 but that knowledge will not reach to base model as it may have proprietary data. The following diagram will cover. And yes, in some cases we will be able to get the knoweldge that may be public.

The point you made (internal test), we were also wondering what is it?

@obuzek Waiting for your feedback.

[lehors]

The backend process I was referring to is a system that trains the base model with the proposed added skills and tests whether it actually leads to an improvement. You can use the lab CLI to do some testing locally on your machine but this is just an approximation because running the real tests requires much more powerful hardware.

It's great if you can propose skills without knowledge but would they be of any use for someone who doesn't have access to the knowledge?

[juliadenham]

Revisiting this now that it's been 5 months!