The tools in this category are bought by an IT department. That is the part
we are attacking.
We used to say the bottleneck was the network. We checked, and it is
not. More than half of plants already ship telemetry off site, firewall changes run at
tens a week in a normal organisation, and an outage window is not required to read a file.
Claiming to remove a network bottleneck claims to remove something buyers do not experience
as binding.
Two things do bind, and we affect both. The first is the person who knows
what the tags mean. Building the asset model is 40 to 60% of an implementation's effort
by one integrator's published estimate, it is the one step with no tool behind it, and it
ends when they retire. The second is the supplier security assessment. We do not skip
it, and anyone who tells you they do is wrong. We shrink what it has to cover: running
inside the fence with no new conduit leaves an IEC 62443 zone and conduit review with
nothing new to assess.
you python -m auge_plant.ingest ./exports
auge hai-test1.csv: 86 tags, 1637 samples, clock as %Y-%m-%d %H:%M:%S
auge hai-train1.csv: 86 tags, 1643 samples, clock as %Y-%m-%d %H:%M:%S
auge label-test1.csv: 1 tags, 1543 samples, clock as %Y-%m-%d %H:%M:%S
auge connector: hai
auge hai read 84 tags (49%)
auge smelter read 0 tags (0%)
auge substation read 0 tags (0%)
auge 173 tags in, 25 points placed, 91 refused
auge what it will not guess at, which is your first afternoon:
auge hai-test1:P1_PCV02D raw encoding with no sibling or twin to calibrate against
auge hai-test1:P1_PCV02Z raw encoding with no sibling or twin to calibrate against
auge hai-test1:P1_PP04SP no asset: the tag does not say which equipment this belong
auge hai-test1:P1_SOL01D no asset: the tag does not say which equipment this belong
No schema to define, no model to draw, no broker to stand up, no account to
create. No IT department. A folder with exports in it, and the instrument index as a
PDF if one happens to exist. This is a real run on a published laboratory testbed's export,
printed as it came out. We have not yet run at a working plant.
Where does the asset hierarchy come from?
You. The matching is good, but it joins tags to a register the plant is assumed to already have, and most plants do not have one.
The compiler works it out from the plant's own naming, and says which parts it could not and why.
Who works out what the data means?
A person does, by hand. One integrator puts asset modelling at 40 to 60% of the whole implementation effort, and nobody publishes a rate for it.
The compiler does, and it tells you which tags it refused and why.
What happens to the tags nobody can name?
They stay on a list that shrinks as somebody works through it, if they are surfaced at all.
They are published with a reason each, and flagged if they are still moving. A moving one is a fault nobody can be told about.
What do you need before the first reading?
A connection, and therefore a conduit to assess. Three months is the closest comparable product's own published average to go live.
A folder of exports on a machine you own. The security review still happens; it has less to cover.
Where does the answer come from?
A second platform. The pipe delivers data; analysis is another vendor and another contract on top of it.
Here. The model and the agents run on the machine in front of you.
Where does the data go?
To the cloud, by design. That is the value proposition, not a compromise.
Nowhere. The only address the product opens is this machine, and a sealed run proves it rather than promising it.
What does it cost?
Nobody publishes a price. It is licensed per device, so the bill grows every time the plant adds an instrument.
One price per plant, with the anchor on this page. Nothing on the invoice counts your instruments.
And a plant is not one user.
An operator on shift, a planner working a week out and an instrument technician want three
different things from the same model, and may do three different things with it. So the
console asks who you are and then stops showing you the rest. A role is two things and
nothing more: the questions that job asks, and what it is allowed to do. It selects from
the site's policy and can never widen it.
Control room operatorwhat needs a response in the next few hours, and nothing else
Shift supervisorthe short list that survives a handover
Instrument technicianthe readings that cannot be trusted, which is a different list
Maintenance plannerwhat is degrading, corroborated, before a crew is sent
Process engineerwhether the plant still adds up across an area
We are not claiming to be a better pipe. We are claiming that a plant should
be able to find out what we do from a folder of exports, offline, on a laptop, before
anybody opens a port. Everything above is a property of how the thing is built, not a
promise in a slide.