← All insights

Field and maintenance

Industrial maintenance as an operational system

A pump that fails mid-shift is a maintenance job, a gap in the shift report, a misleading sales history and a cost, all at once. Here is one such fault, from the inspection that finds it to the repair that closes it.

Updated 6 min read

Maintenance is often run as a world of its own: a list of jobs, a schedule, a budget. In a physical operation a fault reaches further. A pump out of service changes what a site sells, what its shift report should say, how its sales history reads and what the month's costs look like. The maintenance record has to connect to all of that without becoming a copy of everything else.

This article follows one constructed example through Ancilor Field and the modules around it. The site, times and costs are illustrative.

Pump 2, from finding to repair

On a routine visit, an inspector notices that pump 2's display freezes during a sale and recovers a few seconds later. There is no signal in the back office, so the inspection is completed on a phone offline: the finding, two photographs and a short note. It syncs when the inspector is back in coverage, and the finding is assigned to the site manager.

The site manager raises a maintenance job against pump 2, and a technician takes it. Two days later the technician replaces the display board and closes the job with the part, the supplier and the cost. Because the cost is above the approval limit, a manager has to approve it before the job counts as done. The site manager confirms that the pump is back in service.

One fault, five steps, four people
StepRecordWho acts
FindingInspection with photographs and a noteInspector
RequestMaintenance job linked to pump 2Site manager
RepairWork done, part, supplier and costTechnician
ApprovalDecision on the final costApprover
Return to serviceConfirmation that the pump is selling againSite manager

One pump, one identity

Pump 2 appears in at least four places: the inspection checklist, the maintenance job, the pump controller and the shift report. If each names it slightly differently, its history splits into pieces that nobody can reassemble with confidence. The pump needs one identifier that all four refer to, together with its site, the tank it draws from and the dates on which those relationships held.

Replacement is where identity gets subtle. When a display board is replaced, the pump keeps its history. When the whole dispenser is replaced, the new one should not inherit the old one's faults. That rule needs an owner and a date, not a guess made later from free text.

What the fault changes elsewhere

While pump 2 is out of service, its readings stop moving. The shift report should say that the pump was out of service, with the job attached, rather than show an unchanged reading that looks like a missed entry. The tank's sales history needs the same note: two quiet days on pump 2 are an outage, not a change in demand, and a run-out forecast that cannot tell the difference will mislead the next order.

Connecting maintenance to operations helps the people who plan around it, and it has a cost. A supply planner needs to know that pump 2 is out and when it is expected back; they do not need every supplier price in the maintenance history. Access follows the role and the site: the planner sees the constraint, the approver sees the cost, and pay and personal details stay with the people who manage them.

A hold across the modules

Some faults take equipment out of use: a pump, a tank, a receiving bay. That state deserves a record of its own, a hold, rather than a phrase in a job description. A hold says what is out of use, why, since when, who placed it, who may lift it and when the repair is expected to finish. The expected finish is an estimate and should be shown as one. The hold should be lifted by the person responsible for the repair once the work is confirmed, never because a delivery plan would be easier without it.

In Ancilor, maintenance is not an eighth module. It is a workflow that runs through the modules, and each needs a different part of the same job. Ancilor is designed so that one job is read in each module from its own side:

  • Field holds the finding and the job: who found it, the photographs, the cost and its approval.
  • Supply sees the part or the contractor visit the job is waiting for.
  • Logistics sees a bay or an asset on hold that changes how a delivery can be received.
  • Forecast reads the outage, so that two quiet days on pump 2 are not taken for a fall in demand.
  • People shows who is on duty at the site when the technician arrives.
  • Vision can place a camera event at the equipment beside the job, for a person to review.

The job is designed to keep one owner throughout: other modules read its state and do not change it. The platform page shows how maintenance runs through the modules, and the article on late loads follows a delivery that meets a receiving bay on hold.

Missing readings and doubtful sensors

Before anyone predicts failures, the basics have to hold. A missing reading stays missing, never zero. A reading lower than the one before it should be flagged, because a meter that runs backwards is either a fault or a typing error. A replaced board or a recalibration changes what a trend means, so it belongs on the same timeline as the readings.

Scheduled checks, threshold alarms, condition monitoring and failure prediction solve different problems. A calendar can schedule an inspection. A validated sensor can show that a limit was crossed. Prediction needs a history of failures recorded against the right equipment, and proof that its warnings improve the repair decision. Without clean identities and honest readings, it has nothing sound to learn from.

Standards help with shared terms. ISO 55000 sets out the vocabulary, overview and principles of asset management. ISO 14224 covers the collection and exchange of reliability and maintenance data for equipment in the petroleum, petrochemical and natural gas industries, and is a useful reference by analogy elsewhere. We cite both as reference points, not as a claim of conformity.

Close the job with proof of repair

A completed job and a pump back in service are different facts, and they can need different people. The technician records what was done, which part was used and what it cost. The site manager confirms that pump 2 is selling again, and at the next shift close its readings should agree with the controller. If the display freezes again, the new finding links to the old job instead of starting a fresh story.

Field work has to cope with poor signal. A completion sent twice must not count twice, and a photograph must stay attached to the job it documents. Actions on the job are logged with who took them and when.

Read-only at the pump

Ancilor reads pump totals; it does not send commands to pumps or change the controller's records. AI can summarise the pump's history and earlier findings, and draft the job description. It does not approve work, close jobs or change costs. The technician records the work, the approver approves the cost and the site manager confirms the result. The article on AI-assisted decisions explains where that line sits and why. For contractors and industrial sites, see Site operations, part of Ancilor Field.

Measure the loop before claiming results

A first rollout follows one equipment class through one recurring workflow, from finding to confirmed return to service. Test the cases that break it: a finding attached to the wrong pump, a part that arrives late, a finding reopened after repair, an update made offline, a job cancelled after a part was reserved.

Agree the measures before starting: time from finding to return to service, jobs past their due date, repeat faults on the same pump. Changes in downtime or cost need a baseline, and enough time to separate the system's effect from changes in workload or equipment.

References

Start with one workflow.

Bring one process, the systems it already runs on and the decision your team needs to make.

Discuss your operation