Health score(MHS)

A number from 0 to 100 that sums up how well a sensor or a machine is behaving. 100 is perfect. It falls when readings drift away from what the model expects, when the forecast is uncertain, or when the readings get close to the safe limit.

State(HEALTHY, WATCH, DEGRADED, CRITICAL)

Healthy (score 85 or more), Needs watching (65 to 85), Degrading (40 to 65) and Critical (below 40). No data means there are not enough recent readings to analyse the signal.

Expected range(P10–P90, control limits)

Where the model expects the reading to be: 80 out of 100 readings should fall inside it. A reading that stays outside it for long enough raises an alarm.

Safe range(physical limits)

The limits within which the equipment can work without risk. They are set for each sensor in Configuration (or come from the plant data) and are not learned by the model.

Rule on readings

A condition set for a sensor in Configuration, such as staying above a value for a while or rising quickly. It is checked every minute on the readings that arrive, not on the forecast, and raises an alarm as soon as it is met.

Peak risk(PoV)

The highest chance, at any moment of the forecast period, that the reading goes outside its safe range.

Context window(context size)

How many hours of past readings the model looks at before it forecasts. A longer window sees slower trends but takes longer to notice a change in how the machine works.

Forecast length(prediction length, horizon)

How many hours ahead the model forecasts. It is the longest advance warning a forecast can give. It cannot be longer than the context window.

Forecast refresh(cycle, stride)

How often, in minutes, the forecast is calculated again. Every machine follows the refresh set for the whole tool unless it has one of its own. It cannot be longer than the forecast length, or part of the future would have no forecast.

Forecast step(stride, grid)

The time each forecast point covers. It is worked out from the forecast length so that a forecast has about 60 points; you do not set it.

Remaining life(RUL)

The estimated time until the accumulated wear reaches the failure level, with a likely range. It needs the wear model of the sensor to be calibrated, and it is best used to compare sensors and to follow a trend rather than as an exact date.

Wear accumulated(damage, FHT)

How far the sensor has drifted from its healthy behaviour, as a share of the failure level. 100 % is the level the model treats as failure.

Forecast differs from reality(RDI)

How far the readings have been from what the model predicted. A signal that behaves as forecast scores low here; a sustained difference is the early sign of degradation.

Forecast is uncertain(BWGR)

The model is less sure about the next hours (a wide expected range). It says more about the forecast than about the machine.

Little margin to the limit(Cpk)

The expected readings are close to the safe limit, even if they have not crossed it.

Chance of leaving the safe range(PoV)

The forecast gives a real probability that the reading will cross the safe limit.

Critical and secondary sensors(criticality)

A critical sensor defines the health, remaining life and critical sensor of its machine. A secondary one is monitored and reported on its own but does not downgrade the machine by itself. If a machine has no critical sensor, every sensor counts.

Critical sensor (of a machine)(driver)

The signal that most limits the machine: the one that sets its remaining life or, while the life is not estimated, the one with the lowest health.

Open alarm(alert)

Something that needs a person. An alarm stays open until a user acknowledges it, even if the condition that raised it has already gone.