Industrial AutomationField engagement

Adaptive control on a machine that cannot be taken out of service, where the only test environment is the running equipment.

Control systems lead.
live
PLC, in service
Modbus TCP
over an industrial VPN
no
model in the system
setpoint hold as the operating point moves
adaptive law, holds across the rangefixed gain, drifts as load and wear move the plant
What was at stake

A hydraulic top-drive system in active service. The equipment earns while it runs, so it cannot be taken offline to accommodate engineering work, and there is no test rig equivalent to it. Whatever gets written has to be correct against the real machine the first time it touches it. That is a different discipline from most software work: no staging environment, no safe rollback window, and a feedback loop for a mistake that runs through several tonnes of moving hydraulic machinery.

The constraint

A fixed-gain controller tuned at one operating point drifts out of specification as load, temperature, and wear move the plant away from it, and a live machine never sits still. The controller does not fail loudly when this happens; it just holds the setpoint less well, in conditions nobody is watching for. Two more constraints bind below the control law: the loop deadline is part of the specification and the protocol round-trip to the PLC happens inside it, and the controller behaves as it implements Modbus rather than as the standard describes, discoverable only against this controller.

ObserveDetect driftTuneDeploythe live system, over time
A standing loop around the live system. Watch, catch drift while it is small, tune, and deploy, before a customer ever sees a problem.
The fork

The reflex, and the fix.

Road not taken

A fixed-gain PID

Pull

Faster to deploy, well understood, and fine in a demonstration.

Why not

It cannot hold across the operating range of live equipment, and its degradation is silent, which means nobody finds out until the machine is out of specification.

Road taken

An adaptive control law

Accepted

More design and tuning work, carried out against a running machine with conservative bounds and a fallback to the known-good controller.

Bought

Stability held across the range of conditions the machine actually meets rather than only at the point it was commissioned at.

Decision

Adapt to the operating regime, because the plant will not hold still and a silent drift out of specification is the worst available failure mode.

How it was built
01Read over Modbus
02Estimate regime
03Adapt gains
04Write setpoint
05Verify
01

Protocol and transport as ground truth

Read and write the controller over Modbus TCP through an industrial VPN gateway, treating the controller's observed behavior as ground truth rather than the standard's description of it. The round-trip is measured and budgeted as part of the loop, not assumed.

02

A gain-scheduled law rather than fixed gains

The controller carries a tuning per operating regime and moves between them, so stability holds across the range of load, temperature, and wear the machine actually meets rather than only at the commissioning point. The scheme runs inside a safety-instrumented basic process control system, which bounds what the adaptation is permitted to do.

03

The deadline as part of the specification

The loop closes inside its deadline with the protocol round-trip accounted for and headroom reserved, because a control loop that meets its deadline on average and misses it occasionally is a control loop that does the wrong thing occasionally.

04

Correct against the running machine, first time

The only available test environment is the equipment in service, which raises the bar on everything upstream: staged rollout, conservative adaptation bounds, and the ability to fall back to the known-good controller at any point.

How it was measured

A single headline number hides where a system fails. This work was scored on the dimensions that actually decide whether it holds in production, measured on real, held-out cases rather than the demo path.

Setpoint hold across the operating rangeLoop timing including protocol round-tripBehavior at the adaptation boundsFallback reliability
figures

What it produces
Without this discipline

A controller tuned beautifully at commissioning that quietly stops holding specification a few months later, in operating conditions nobody instrumented, with no alarm and no obvious cause.

This system

Stable control held across the machine's real operating range, on equipment that stayed in service throughout, over an industrial protocol path with the round-trip budgeted inside the loop deadline.

adaptive control lawModbus TCP over an industrial VPNdeadline includes the round-tripno model in it
The operating envelope

What it owns, and what it hands to a person.

Handled with confidence
The operating range the law was designed and bounded for
Live operation with no service interruption
Flagged for review
Operation approaching the adaptation bounds
Out of scope by design
Mechanical faults
Unbounded adaptation
The honest limit

Adaptive control widens the range over which a controller holds. It does not remove the need for the plant to stay inside the envelope the law was designed for, and a mechanical fault is still a mechanical fault. The adaptation is bounded deliberately, because an unbounded adaptive law on live hydraulic equipment is a worse failure mode than a slightly detuned fixed one.

What it generalizes to

There is no machine learning in this engagement. The work is a control law, a timing budget, an industrial protocol, and the discipline of being correct against hardware you cannot stop. It sits directly alongside the real-time closed-loop work in robotics and the fail-closed architecture in defense, and the same discipline appears in all three: the deadline is part of the specification, the hardware is ground truth rather than the document, and what the system does at its boundaries is designed rather than discovered.

How we engage

You have a system like this one.
Tell us where it stands.

Whether it is failing, not yet built, or about to meet a scale it has never seen, we can tell you what we see.

Start a conversation
mostafa@opulion.dev · Response within 24 hours · By inquiry