Recurring plant inspection rarely fails because nobody knows how to do it — it fails because somebody has to do it. A platform that drives the same route every night delivers something a person cannot: an unbroken record.
Every maintenance plan sets intervals. In practice those intervals compete with faults, shutdowns and holidays — and lose. A round scheduled weekly then happens monthly, and what occurred between two inspections is anyone’s guess.
On top of that comes the shortage of skilled staff. In the energy sector the worldwide shortfall of qualified personnel is estimated at several tens of thousands of positions. Anyone who has such people is reluctant to spend them on a check that ends without a finding 99 per cent of the time — rather than on the repair that follows from it.
And finally there are areas nobody likes to enter: narrow ducts, hot plant sections, areas with a gas hazard, night rounds across a large site. Every one of those inspections is a risk assessment in its own right.
An inspection round is a sequence of defined measuring points. Every one of them can be captured by sensors — the skill lies in capturing them reproducibly, from the same position and the same angle.
Overheating bearings, pumps running hot, faulty terminals in switchgear, insulation faults on cables. Temperature trends show up long before a component fails.
Methane, hydrogen sulphide, carbon monoxide and combustible gases along the route. Escaping gas is picked up where it originates — not only once it reaches the fixed sensors.
Reading analogue pressure gauges and meters, checking levels, documenting valve positions, spotting corrosion and traces of leakage. In darkness too, with lighting that switches on as needed.
Laser scanning of the plant as a by-product of every round. The result is an as-built model that is immediately available for modifications, shutdown planning and contractor briefings.
The decisive point is reproducibility: because every measuring point is captured from the same position, two recordings weeks apart are directly comparable. That is precisely what a round on foot never quite achieves.
The real gain is not the hour saved. It is that irregular individual observations turn into a consistent, high-frequency data set. A single temperature reading says little; the same measuring point over six months shows a drift that announces a failure.
That data set also survives a change of staff. Experience — which pump always runs a little warmer in summer, which terminal has been conspicuous twice already — otherwise exists only in the head of the person who has walked the round for twelve years. And leaves with them.
For the business case that means: counting only personnel hours falls short — and without a baseline measurement beforehand there is no solid basis for the calculation anyway. The benefit comes from unplanned downtime avoided.
We have already worked on the task underlying this page under competition conditions. At the Robotics Challenge staged by the International Atomic Energy Agency together with CSIRO Data61 in Australia, the task for unmanned ground vehicles was: navigate a storage area autonomously, identify and count the containers held there, read their markings, measure radiation, and enter position and reading into a database. The course was a test setup with mock containers — not a real storage facility.
This is plant inspection under sharpened conditions. Every measuring point has to be assigned to an object beyond doubt, every result documented traceably — in an environment where nobody stays longer than necessary. That chain of approach, recognise, measure, assign, document is exactly what an inspection round in an industrial plant consists of. The difference lies in the reading, not in the procedure.