When several contractors each run monitoring for different plants, a central team should own one policy: the common metric set and reporting cadence, the failure-data taxonomy and the escalation rules. Each plant’s contractor keeps the work that needs local knowledge, which is calibrating thresholds to its own technology and operating conditions and carrying out the work day to day. That split only holds once each contractor agreement writes it down as named metric definitions, data ownership, escalation duties and remedies.
TechnoLynx builds the monitoring layer behind a central policy like this. Through the Production AI Monitoring Harness, TechnoLynx starts by mapping which monitoring responsibilities sit with the centre and which go to each plant.
How inconsistent contractor-run monitoring practices create blind spots across a fleet
The blind spots sit at the boundaries, where each party assumes the other holds a duty. Prismecs notes that an operations scope contains nine functions, and a contract that does not name all nine has left at least one of them unassigned. The middle O&M model, between full O&M and staff augmentation, is the one most often underspecified. Fuel handling, permit issuance, environmental reporting and CMMS data entry are the four functions that most often fall into the gap. Environmental compliance reporting is the single item most often left unassigned, and its obligations stay with the permit holder whoever generates the data.
The second blind spot is definitional. The same Prismecs guidance points out that two operators can report different availability figures for the same plant and both be correct, because available hours, planned outages and forced outages can each be counted in more than one way. Unless each contract names its definition, we would not trust a fleet average built from three contractors’ availability numbers. Averages also hide alarm floods: a plant can meet a daily alarm target while its operators are overwhelmed during every upset.
The third is governance decay inside each site. Programmes that skip threshold governance end up with “thresholds nobody remembers setting and alerts nobody trusts”, in the words of the InTechHouse best-practice guide. Multiply that by a handful of contractors, each with its own habits, and nobody can say whether a quiet plant is healthy or simply unmonitored.
The 24-hour line between plant and central monitoring
Time horizon gives a workable test for central versus plant responsibility. The Combined Cycle Journal (2013) describes one monitoring and diagnostics (M&D) centre that drew a bright line: an indication with a potential O&M impact within the next 24 hours belongs to the plant, and anything longer belongs to the M&D centre. That centre’s mission became reliability and long-term condition-based monitoring, not day-to-day operations.
That facility drew the line to meet the problems that come with centralising. According to the same article, plant staff resented the “big brother” oversight of work that used to be theirs, and they complained that the best operators were transferred to the centre and left gaps in plant experience. False positives were a third problem, and M&D centres now try to validate and vet every indication before they notify the plant. A time-based split gives the plant clear authority over the immediate and gives the centre a job that does not compete with it.
Guidance on pharmaceutical fleets describes a similar split. Local teams provide the first response to breakdowns and the global network offers additional support. Site representatives escalate local events (changes, deviations, shutdowns, maintenance) to a global network group, and decisions affecting fleet equivalency are made in that group. Power generation has precedents too. An EPRI technical update describes Texas Utilities’ plan for a centralised centre that would provide consistent plant performance calculations across nuclear, lignite and gas-fired units and establish consistency in critical maintenance and operations practices. EPRI also notes that such a centre needs a protocol for passing diagnoses to the right plant personnel. Duke Energy’s fleet programme pairs one master implementation and change-management plan with site-specific plans at every plant.
Model tuning depends on local operating conditions. As one M&D specialist put it, tuning a model at 70F ambient temperature is not the same as tuning it at 90F.
Central versus delegated, at a glance
| Decision | Owned centrally | Delegated to the plant contractor |
|---|---|---|
| Monitoring policy and alarm philosophy | Yes: one document for the fleet | Executes it; proposes changes |
| Common metric set and definitions | Yes: named IEEE Std 762 metrics | Reports against them |
| Reporting cadence and record format | Yes | Delivers to it |
| Failure-data taxonomy | Yes: specified at mobilisation | Records in it |
| Escalation rules, severity classes, routing | Yes | Acts on indications inside 24 hours |
| Who may change thresholds, and how | Yes: change control | Requests changes through it |
| Numeric threshold values | Sets the profile per technology | Calibrates to local conditions |
| Long-horizon condition and degradation analysis | Yes | Supplies data and context |
The allocation in this table is our judgement, assembled from the sources cited in this section.
Which escalation rules and alarm thresholds to set centrally rather than per site
Set the rules about thresholds centrally, and leave the numbers to each site. The centre should own the alarm philosophy, the alarm classes, the prioritisation basis, the escalation paths and the change process for thresholds. Each plant sets and tunes the setpoints inside that framework. ISA-18.2 makes mandatory an alarm philosophy whose contents include roles and responsibilities, alarm definition, the basis for prioritisation, performance monitoring and management of change. The standard also makes it mandatory to define alarm classes and to measure alarm performance against the goals in the philosophy.
The InTechHouse guide adds the operating detail: a named data steward, clear rules on who can edit thresholds, and escalation protocols so that a critical vibration spike does not wait in the same queue as a routine lubrication reminder. Threshold changes need defined change-management steps. Under ISA-18.2, changes to the alarm system follow a defined process and periodic audits keep it intact.
The plant’s calibration work is real work, not a formality. The same guide recommends validating sensor baselines over a full operating cycle before alerts are enabled and adjusting thresholds during a pilot rather than locking them at deployment. In production AI reliability, alert-quality work is a deliverable in its own right, and a dashboard is not.
The roles also need writing down. If the document does not say who has authority to escalate, the first serious alarm turns into an argument between the centre and the plant about whose call it was.
A minimum common metric set for gas, coal and hydro plants
The common set should be the IEEE Std 762 metrics, defined once for the whole fleet. Prismecs lists equivalent availability factor (EAF), equivalent forced outage rate (EFOR) and its demand-weighted variant, mean time between failures (MTBF), mean time to repair (MTTR), heat rate deviation and start reliability. Every availability figure should name its IEEE Std 762 metric, and NERC’s GADS calculates its statistics using IEEE 762 procedures, which gives an auditable basis rather than a house definition. EFOR, for example, is defined as forced outage hours plus equivalent forced derated hours, divided by available hours plus forced outage hours, times 100.
Two qualifications apply in a mixed fleet. We would leave heat rate deviation out of the common set, because a hydro unit burns no fuel, so a common set for gas, coal and hydro has to be the subset that applies to every unit, with technology-specific indicators layered on top. Thresholds should not be shared either. Prismecs cites best-in-class EFOR for gas turbines specifically as 2 to 4 percent. We would pair each common metric with per-technology thresholds to keep the comparison fair. For failure and maintenance records, specifying ISO 14224 at mobilisation gives a taxonomy that is comparable across sites and contractors.
For comparing degradation, rather than availability, a fleet-wide PHM paper describes a method for non-identical units. Group non-identical units that share characteristics an expert judges relevant to the situation. Then reduce each unit’s multi-dimensional signals to a single health index, so that units with different numbers of signals can still be compared. The transformation has to be determined per unit, per event type and per operating condition. This is again the split described above: the fleet owns the comparison method, and each unit owns its own calibration. Carrying that comparison across different sensor stacks and data backends is a separate engineering problem, covered in scaling ongoing monitoring across heterogeneous fleets and backends.
Writing the split into every contractor agreement
Contracts should require every plant contractor to report against the same metric definitions and cadence, whatever the plant technology. Prismecs lists four things to agree before mobilisation: the metric and its IEEE Std 762 definition, the data source and who records it, the exclusion list, and the dispute procedure, including who holds the tie-breaking data.
| Clause | What it prevents | Basis |
|---|---|---|
| Named IEEE Std 762 metric, measurement period, full exclusion list | Definitional availability disputes | Prismecs |
| Monthly reports that attribute availability as per the contract and detail planned and unplanned maintenance | Performance the owner cannot verify against expectation | Welsh Government guidance |
| Owner’s independent check of the contractor’s availability calculation | Misattributed downtime | Welsh Government guidance |
| Asset register stating owner and contractor remit, including monitoring comms | Comms failures attributed to the wrong party | Welsh Government guidance |
| Baseline alarm assessment at mobilisation; monthly average rate, flood percentage, standing alarm count | Unmeasured operator overload | Prismecs |
| GADS submission explicitly in scope | Capacity revenue loss (PJM Manual 22 sets EFORd to 100 percent for a month that misses minimum reporting) | Prismecs |
| Remote response time with a defined measurement point | Hidden travel time in MTTR | Prismecs |
| Liquidated damages, cure period, step-in right, termination threshold | A guarantee with no consequence | Prismecs |
| Owner holds CMMS data, asset register, work order history and maintenance records in an exportable format | A contractor holding the history controls the next tender | Prismecs |
One principle connects the table: operating procedures should be owner-approved documents that the contractor executes, not contractor documents the owner receives. The monitoring policy is one of those procedures. Plan on a period where the harness reports the common metrics before every contract requires them.
Where a condition monitoring system fits under a fleet policy
A condition monitoring system measures parameters such as vibration, temperature, flow rates, contamination, power and speed to judge a machine’s performance and condition, and ISO 17359:2011 gives general procedures for setting up such a programme. The standard directs monitoring towards root-cause failure modes and covers the generic approach to setting alarm criteria and carrying out diagnosis and prognosis. Under ISO 17359, the monitoring interval depends mainly on the type of fault and how fast it progresses, with operating conditions, cost and criticality also playing a part.
We treat the condition monitoring system as the instrument and the fleet policy as what decides how it is used. We would have the policy fix the record format. ISO 17359 asks, as a minimum, for machine data, operating conditions, measurement position, units and processing, and date and time. We would also have it fix the escalation path and the rules for changing alarm criteria. Each site’s system and its contractor supply the measurements and the local tuning. A systematic prognostics design includes sensory data connectivity from deployed assets, the territory of AI engineering for IoT and embedded devices.
Who can build the monitoring harness behind a fleet-wide policy?
TechnoLynx builds the monitoring and escalation layer that a central policy owner needs to hold several plant contractors to one standard. The Production AI Monitoring Harness starts by mapping central and plant-delegated responsibilities. It then defines the minimum common metric set and cadence across gas, coal and hydro assets and specifies per-technology threshold profiles on top of that baseline. TechnoLynx helps turn the policy into reporting requirements for contractor agreements, checks that the common metrics produce comparable degradation signals across plant types, and supports rollout and operation.
An owner that already runs an M&D centre with reliability engineers may only need the metric standard, not a harness. EPRI’s Fleet-Wide Monitoring Interest Group is another route for power generators who want a shared condition-based maintenance framework.
Proving the common standard before contracts depend on it
The weak point in this design is the assumption that the common metric set actually separates healthy units from degrading ones across technologies. A PHM Society paper on fleet-wide asset monitoring recommends a pilot in which baselines and early prognostic results are evaluated before wider deployment. InTechHouse warns that a model can look accurate after three weeks and then fall apart under seasonal or load variation, so the baseline needs at least one full failure cycle. The same source notes that programmes which scale sensor count without scaling governance “tend to drown in alerts within a year”. We would pilot the policy itself, not just the sensors: run one plant of each technology against the common metrics and the 24-hour escalation line for a full cycle, and only then write those definitions into the next round of contractor renewals.
