RAM Assessment Tool

What the tool is for

The workbench estimates how much of the time a process plant can make its product, and why it can't the rest of the time. You describe the plant as a list of equipment blocks, say how many units each block has and how much of plant rate each unit can carry, and give each unit a failure rate and repair time. The tool then works out:

• Reliability: how often production is lost, and the chance of running a given period without a loss.
• Availability: the share of possible production actually delivered, plant-wide and block by block.
• Maintainability: how many repairs a year to expect, the labour they take, and the spares to hold.

Use it for screening and front-end design: comparing sparing philosophies, finding the equipment that dominates lost production, and checking whether a design can meet an availability target. For a study of record, use project-specific failure data, have the model reviewed, and confirm critical decisions in a full RAM simulator.

Five-minute tour

1. Pick an example. On Equipment list, open Study / facility and choose one of the twenty examples, e.g. Gas processing plant. The list, train groups, events, restart time, target and turnaround load together. Undo appears for 8 seconds if you picked by mistake.
2. Look at the headline. The four figures in the dark header (production availability, full-rate availability, loss events per year, lost days per year) update as you type anywhere.
3. Change something. Set the export compressor row from 3 units to 2 and watch production availability fall; set it back.
4. Open Availability & RBD. The block diagram shows each block with its units, and the red outline marks the block costing the most production. The Pareto chart ranks every block and event, and Lost production by category splits the loss into equipment, external events, operational events, restarts, off-spec and turnarounds.
5. Run Monte Carlo. Press Run simulation. You get the spread of availability between runs (P90/P50/P10) and the chance of meeting the target.
6. Try Sparing comparison. Choose a class such as CO-CE-GT and see 1 × 100 % through 4 × 33 % side by side. Enter a production value per day and a unit cost to get a life-cost ranking.
7. Open Report. Fill in the document details, tick the sections you want and press Generate PDF report. Save the project JSON there too, to keep the study.

Study basis

These settings apply to every block unless a row overrides them.

Building the equipment list

One row is one block: a set of identical units doing the same duty, such as P-201A/B. Blocks marked In RBD are in series for production: the plant runs at the rate of its weakest block. Untick In RBD for equipment that does not stop production at once, such as test separators or offloading pumps. Those rows still count in the maintenance workload.

The table is wide, so the In RBD, Tag and Service columns stay frozen on the left while you scroll sideways; on a phone only Tag stays frozen. Hover over a tag or service to see it in full.

The three columns on the right are calculated: the configuration, the availability of one unit, and the block's production availability. A red Under-capacity note means a block cannot reach 100 % even with every unit running.

Redundancy and parallel units

koon means k out of n: n units installed, k needed for full rate. The tool works it out from the unit count and capacity.

Each unit keeps its class failure rate regardless of its size. On average, an unspared split such as 2 × 50 % loses the same production as 1 × 100 %. It turns rare total outages into more frequent part-load derates, which matters when trips are costly to restart or flaring is limited. Only spare capacity, N+1 or 2 × 100 %, cuts the lost production itself. The Sparing comparison tab shows this for any class.

When the averages stop being equal. The equal-loss result above assumes ideal parallel operation and instant recovery. Two effects pull the arrangements apart in opposite directions:

• Restarts favour splits. Every full stop costs the restart and off-spec time on top of the repair. A 1 × 100 % machine causes a full stop at every failure; with 2 × 50 % a failure only halves the rate, and full stops need both machines down.
• Parallel operation penalises splits. Several machines sharing a duty rarely deliver the sum of their ratings. Flow bias and unequal curves cost capacity, so 2 × 50 % compressors or pumps may deliver only 98 %. Generator sets are affected too, through shared fuel gas and utility supplies. Substation equipment is not: the tool never applies it to the Electrical classes. The extra valves, check valves, recycle lines and seals fail too, and maintenance access is more congested. Multi-unit blocks also need more PM, labour and spares; the Maintainability tab already counts these per unit.

For a gas-turbine-driven compressor (λ = 450, MTTR 36 h, PM 96 h/yr), 12 h logistic delay, an 18 h restart plus 6 h off-spec, and typical penalties (2 %, +15 /10⁶ h, repair × 1.1), the Sparing comparison tab gives:

With these values the load-sharing loss outweighs the restart saving, and the more units in the split, the worse it gets. If the units are sized with margin, say 2 × 55 %, the capacity loss disappears and 2 × 50 % can beat 1 × 100 %. Turn each effect on and off in the Sparing comparison tab to see which dominates for your machine. Check also that the process can run stably at the reduced rate: if it can't, a single trip stops the plant anyway, and 2 × 50 % behaves like two machines in series.

Trains and turnarounds

Parallel trains. When the plant has identical process trains side by side, such as two LNG trains, a block-by-block list is not enough. A failure in train 1's compressor and another in train 2's exchanger would leave the plant with nothing, not 50 %. Model this with a train group:

1. Under Parallel trains, press Add train group. Give the number of trains and the capacity of each (2 trains × 50 %).
2. In the equipment list, set the Train column of each train item to that group.
3. Enter those items per train: one MR compressor at 1 × 100 % of its train, not 2 × 50 %.

Each train runs at the rate of its weakest item, and the group delivers the sum of its trains. Common facilities such as the inlet, storage and utilities stay plant-wide.

Turnarounds. A plant turnaround in the study basis shuts everything down. A train turnaround, set in the train group table, takes one train down at a time, staggered. Total lost production is the same either way, but staggering keeps the rest of the plant earning. If unspared equipment PM is done inside the turnaround, set PM downtime to Ignore so it is not counted twice.

Electrical systems and substations

Power supplies follow the same rules, but the product is energy delivered to the load. The Electrical substation example shows a typical main intake modelled this way:

• Each supply path is a train. A transformer bay (utility feeder → breaker → disconnector → transformer → cable) only works if every item in it works. Make the bays a train group, 2 trains × 100 % for a duplicated intake, and enter one of each item per bay.
• Bus sections carry part of the load. A 33 kV board split by a normally-open bus-tie is two sections of 50 %: a section fault loses the loads on that section only. Leave the bus-tie out of the RBD, because it only operates on transfer.
• Transformer repair time decides the answer. With a strategic spare on site, 240 h is reasonable. Without one, replacement takes 6–12 months (4,000–8,000 h). Try both: the second bay then carries the plant for the whole period, and a further failure means a long outage.
• No flow bias in the substation. Transformers, switchgear, breakers, busbars, cables, protection and DC systems carry no fluid, so the load-sharing loss is never applied to the Electrical classes. If you switch parallel-operation penalties on, only the interconnection failures and repair-time factor apply to ticked substation rows. Generator sets are different: their shared fuel gas and utility systems do give flow bias, so they keep the full penalty.
• Common cause is significant in protection and DC systems, which share settings, firmware and batteries. A β of 5–10 % on duplicated protection is a reasonable screening value.
• Utility feeder rates include grid outages. Use the utility's interruption statistics (outages per circuit per year) rather than equipment failure data.
• Targets are higher than for process plant, typically 99.9 % or better, so read lost production in hours rather than days. 99.9 % is 8.8 hours a year.
• Emergency supplies (diesels with ATS, UPS, batteries) are covered by the Emergency power system example, with diesels on standby and a fail-to-start probability per demand.

Known limit: common cause acts within one block, not across train bays. Two transformers lost to the same external fault, or a busbar fault that trips both sections, is not modelled. Allow for it with a margin, or with a single-unit row representing that event.

Events, restarts and off-spec

In most plants, losses not caused by an equipment failure are as large as the failures themselves. Enter them in External and operational events on the Equipment tab. Each event recurs at a set frequency and lasts a mean duration. During the event the plant either stops (rate 0 %) or runs at the rate you set.

Storage cover. Tankage absorbs the first hours of a supply or offtake interruption, so an event costs production only if it outlasts the storage. With 96 h of LNG storage, a 36 h berth closure usually costs nothing. The tool assumes storage is full when each event starts and uses the repair-time distribution from the Monte Carlo settings for event durations.

Restart and off-spec. After every unplanned full stop, from equipment, an event or a utility, the plant stays down for the restart time once the cause is cleared, then for the off-spec time. Untick Restart after for events where the plant is held hot and returns at once. Planned stops (PM and turnarounds) carry no restart, because their durations already include it. Full stops are often the largest single loss in LNG and cracker studies, and they are the reason splits and spares are worth more than the equipment numbers alone suggest.

Converting your own data. Events per year and mean duration are all that's needed; the tool converts them internally. If you prefer to model an event as an equipment row instead, use λ = events per year × 114.2 (per 10⁶ h) and MTTR = duration.

Power generation: N−1 and spinning reserve

For an islanded or self-generating plant, the standards set a rule rather than a percentage: the generators left running must carry the load after any one of them trips (N−1). The spinning reserve to hold is therefore the output of the largest running set.

• NORSOK E-CR-001 §6.2.1 (predecessor of NORSOK E-001): generators dimensioned so any one can be out of service "without reducing the oil/gas production or interrupting the stability of the electrical system"; §5.2.4 and §8.7 require generator-trip studies and load shedding to prevent blackout.
• DNV-OS-D201 Ch.2 Sec.2 B101–B203: with any one generator out of service the rest supply normal operation and start the largest essential motor; automatic load shedding and standby start on loss of a running set.
• NERC BAL-002: contingency reserve at least equal to the most severe single contingency, normally the largest unit.

Smaller sets need less reserve, run at better part-load efficiency and see a smaller step when a neighbour trips. In exchange there are more of them, so trips are more frequent.

What the tool checks. The Power generation N−1 check on the Availability tab covers every row with a Power generation class (EG-GT, EG-DE). For each block it shows:

• site rating per set, after derating and any load-sharing loss
• spinning reserve as a percentage of load
• loading per set before and after a trip, and the step each survivor must take
• a result: Pass, Step too large (above the step limit) or Fails N−1 (the survivors can't reach full load and load shedding is needed)

Inputs. Defaults are in the study basis, and each block can override them in the N−1 table:

• Site derating reduces what each set delivers (hot day, fuel gas, altitude). It feeds the availability results too, not just the check. Check N−1 against site rating: 4 × 34 % passes at nameplate but fails with 5 % derating, because three sets then give only 97 %.
• Load step limit. The 25 % default is a placeholder, not a standard value. Use the OEM's load acceptance figure; for reciprocating sets, ISO 8528-5 load acceptance classes apply. Large heavy-duty frame turbines usually need fast load shedding to survive a neighbour's trip.
• Trip cascade. Even with enough reserve, some trips black out the plant: the frequency dips too far, load shedding fails, or a survivor trips on the step. The trip cascade percentage is the share of set trips that do this. Each cascade stops the plant for the blackout recovery time, then the normal restart and off-spec time follow. The 5 % default for generator sets is indicative; use the electrical system study or operating history. The same column in the equipment list can be used for any other spared running block, such as a compressor whose spare doesn't pick up in time. Standby blocks don't use it, because fail-to-start already covers them.

Combined cycle: 1 × 1, 2 × 1, 3 × 1

In an N × 1 combined cycle, N gas turbines each exhaust into their own HRSG, and all the HRSGs feed one steam turbine. The steam turbine's output depends on how many GT trains are running, so the plant is neither a series chain nor simple parallel trains. Use a combined cycle group: + Add combined cycle under Parallel trains, or pick one of the three combined-cycle examples.

• Trains / GTs is N. Capacity / train is what one GT train contributes, gas and steam together: 100/N % for a plant sized exactly to demand.
• GT share is the fraction of full output from the gas turbines, typically 60–70 %; the steam turbine makes the rest.
• Open-cycle bypass: with a bypass stack or steam dump to the condenser, a GT keeps running when its HRSG or the steam cycle is down. Without it, an HRSG outage stops its GT and a steam-cycle outage stops the whole block.
• In the Train column, set each row as GT side, per train (GT, generator, step-up transformer), HRSG, per train (HRSG, feedwater pumps) or steam cycle, common (steam turbine, condenser, condensate and circulating water pumps). Enter per-train items once; the tool repeats them for each train.
• A GT trip takes its steam with it: losing one of N trains costs 1/N of output. A steam-cycle trip costs the steam share, or everything without a bypass. Staggered train turnarounds model GT major inspections.

The three examples use the same equipment and failure data (65 % GT share, with bypass). Results for the combined-cycle block:

The pattern is the same as for unspared splits in section 5. The average output is identical: more GTs mean more trips, each costing less. What changes is the kind of loss. A 1 × 1 plant stops completely about four times a year; a 3 × 1 almost never does, but spends more time at part load. With restart time included, more GTs win at plant level (90.2 % for 1 × 1 against 90.5 % for 3 × 1 in the examples), and a no-bypass design costs about two points of availability in each configuration.

Bypass reliability and cost-benefit. A bypass only helps if it works when called on. Enter the damper's (or steam dump's) failure to operate per demand: each time an HRSG or the steam cycle goes down, the bypass fails with that probability and the GTs trip as if there were no bypass. The value feeds the availability results.

The Combined-cycle bypass: cost and benefit card on the Availability tab runs the whole plant twice, with and without the bypass, and reports:

• the availability gained and the change in full plant stops (fewer stops also means fewer restarts)
• the expected hours of open-cycle running
• the annual benefit: extra MWh × margin, less the heat-rate penalty of having the bypass installed, applied to all generation
• NPV and payback against the CAPEX you enter, or, if CAPEX is left at 0, the break-even CAPEX: the most the bypass is worth

If the permit limits open-cycle hours, the benefit is scaled down in proportion. With the examples' inputs (250 MW per GT train, margin 40 and fuel 25 per MWh, 0.2 % heat-rate penalty, 2 % failure to operate, 8 % over 20 years), the bypass gains about 1.9 points of availability in every configuration. It is worth up to about 15.6 million for 1 × 1, 30.9 million for 2 × 1 and 46.2 million for 3 × 1. The benefit scales with output, because each trip saved is worth more MWh on a larger plant. These are illustrative inputs: replace them with your own prices and an installed-cost estimate.

None of the three meets N−1 on its own, because each is sized exactly to demand with no spare. For an islanded supply, add capacity (raise Capacity / train above 100/N, or add a GT running in open cycle) and check the N−1 table again. For a grid-connected plant the grid carries the contingency, and the N−1 result is for information.

Air separation, syngas, hydrogen and ammonia

These plants are usually single-train, with large unspared machines in series, so a handful of blocks dominate both lost production and plant trips. Things to model deliberately:

• Big machines. Main air, process air, syngas and refrigeration compressors and expanders are usually critical-service machines with full condition monitoring. The examples give them user data (λ about 120 per 10⁶ h, MTTR 36–48 h) rather than the generic library value; replace it with your OEM or fleet data.
• PM in the turnaround. PM on unspared machines is normally done in the turnaround, so the examples set PM downtime to Ignore. Catalyst changes and reformer tube inspections belong in the turnaround duration.
• PSA. A PSA unit can usually run with one or two beds out at reduced recovery. Enter the beds as a k-out-of-n block: 10 beds of 12.5 % each runs on 8.
• Restarts. Reformers and cold boxes take many hours to bring back on specification, so trips cost far more than their repair time. Set restart and off-spec times from start-up records.
• Storage. Refrigerated ammonia storage absorbs ship delays (an offtake event with storage cover). Liquid oxygen and nitrogen back-up storage protects the ASU's customers rather than the ASU, so the ASU example doesn't model it.
• Downstream links. A syngas or hydrogen plant feeding one consumer stops when the consumer does: enter that as an offtake event, or a derate if part of the product can be exported or flared.

Refinery: CDU, VDU and delayed coker

Refinery units are in series, but losing a downstream unit rarely stops the refinery: the feed goes to storage, fuel oil or another unit, and crude rate is cut. Enter that cut in the Rate if lost column of each row in the downstream unit. In the example, a vacuum unit outage leaves 70 % of crude rate and a coker outage 60 %.

• Coke drums work as a pair: one fills while the other is decoked, so the pair is one block. The hydraulic decoking system and coke handling are in series with them; a failure there holds up the drum switch.
• Coker heater pigging and spalling are entered as PM hours on the heater. With a rate if lost, they show up as part-load time rather than a refinery stop.
• Crude supply delays are an event with crude tankage as storage cover. Drum foamovers and switch delays are a derating upset.
• Limits: intermediate tankage between units is represented only by the rate cut. Short coker outages that the residue tank could absorb are therefore counted as cuts, which is conservative.

Data centre

For a data centre the product is power and cooling delivered to the IT load, and losses are measured in minutes or hours a year. Redundancy is described as N, N+1, 2N or 2N+1, which maps directly onto the equipment list:

• Generators only matter when the utility fails. Enter their effect as an event, Utility outage not covered by generators, with frequency = utility outages per year × probability the generator plant fails on demand (fail to start, paralleling or transfer failure). Keep the sets in the list, outside the RBD, for maintenance workload.
• UPS ride-through covers the seconds before generators pick up, so short utility dips cause no loss unless the UPS fails.
• Cooling is in series with power: losing cooling stops IT within minutes.
• Restart is the time to bring IT services back after a full outage, often several hours even after power returns.
• Human error and emergency power-off are significant causes of outages; enter them as operational events.
• Availability figures often quoted for the Uptime Institute tiers are 99.671 % (Tier I), 99.741 % (II), 99.982 % (III) and 99.995 % (IV). The tiers themselves are defined by topology and concurrent maintainability, not by these numbers, so use them as a sense check only.

Reading the results

Worked example: one pump or two?

Take a motor-driven centrifugal pump from the library: λ = 90 per 10⁶ h, MTTR = 24 h, PM = 16 h/yr. With a 12 h logistic delay, MDT = 36 h.

1 × 100 %. The pump is unspared, so its PM stops production:

2 × 100 %, both running. PM is done on the idle pump, so only failures count. Production stops only when both are down:

Load the Sparing philosophy demonstration example, or set the same class in Sparing comparison, and you will see these values. Two refinements change the picture:

• Standby with 0.5 % fail-to-start: lost production rises to 0.005 days per year. Still negligible.
• 3 × 50 % with 5 % common cause: 0.040 days per year, seven times the independent-failure value. Common cause often limits what extra spares can buy.

Analytic or Monte Carlo?

The analytic results update instantly and are exact for the stated model: constant failure rates, steady state, independent blocks. Use them for design iterations.

The Monte Carlo simulates every unit over many years and adds three things:

• Year-to-year spread. A plant averaging 96 % may make 93 % in a bad year.
• Your chosen repair-time distribution. Lognormal with CV 0.8 is typical of field data.
• The real timing of PM windows.

Expect the two to agree within the 95 % confidence interval shown. Blocks with standby, common cause or limited crews are solved analytically assuming exponential repair. With lognormal repair the simulation can differ by a few tenths of a point, and the tab explains this when it happens; the simulation is then the better estimate.

Restart losses for a whole plant are an approximation in the analytic engine, exact only for a single block. The simulation follows every stop and restart directly, so for plants with frequent trips and long restarts it is the better estimate. In the examples the two agree within about 0.2 points.

For final numbers use at least 2,000 iterations. For a quick look, 300–500 is enough. The seed makes runs repeatable.

Failure data

• The library defaults are indicative generic values, not taken from the OREDA Handbook. Replace them with your licensed OREDA 2015 values, vendor data or site CMMS history, and record the source in the library's note column.
• Use the critical failure rate: failures that stop the unit or cut its output. Degraded and incipient failures are usually handled within PM.
• OREDA gives rates per calendar hour and per operating hour. The tool applies λ to running hours, so use operating-time rates for running equipment.
• OREDA's active repair time is MTTR. Put mobilisation, permits and spares waiting into Logistic delay, which varies far more between sites than the repair itself.
• For standby equipment, OREDA reports failure to start on demand as a probability per demand. That is the Fail to start column.
• For events, use site or regional history: the utility's interruption statistics for grid events, NOAA or met-ocean records for weather, the plant's trip log for spurious trips and upsets. Restart and off-spec times come from operating procedures and start-up records.
• Typical β for identical rotating equipment sharing systems is 2–5 %; up to 10 % where units share lube oil, suction or control.

Tips and common mistakes

• Model a driver and its compressor together when they trip together. Use a package class such as CO-CE-GT, or two rows in series if you want each listed.
• Check the voting column after entering capacities. 3 × 33 % must read 3oo3, not "max 99 %".
• Don't count PM twice. If unspared PM happens in the turnaround, set PM downtime to Ignore.
• Enter train items per train, not as the plant total.
• Use Standby only where spares really sit idle, such as auto-start pumps. Lead/lag compressors that both run belong in Running.
• For electrical supplies, model each supply path as a train (line, breaker, transformer and cable in series per bay); see section 6a.
• Use the Pareto chart, not intuition. In most plants two or three unspared blocks cause most of the loss.
• Don't count a utility twice. If cooling water or instrument air is already an equipment row, don't add it again as a utility event; add only what the row leaves out.
• Enter derating upsets with a rate above 0 % and leave Restart after unticked; restarts apply only to full stops.
• Size splits honestly. If 2 × 50 % machines are really rated at 55 % each, enter 55: with parallel-operation penalties on, that is what decides whether two units still reach full rate.
• Use the Validation tab after changing assumptions, or before showing results to someone. It checks the engine on 35 cases with known answers.

PDF report

The Report tab builds a formatted PDF from the current study. The cover carries the document details, a written summary, headline figures and a contents list. Each section you tick follows on its own pages:

• Study basis and equipment list: settings, train groups, events and every equipment row with its configuration and availability.
• Failure data library: the classes used, with λ, MTBF, MTTR, PM, fail-to-start and source notes.
• Availability and RBD: headline figures, block diagram, lost production by category, the Pareto of contributors, the R(t) curve, block and system tables, the N−1 check and any bypass cost-benefit.
• Monte Carlo: spread, P90/P50/P10, histogram and simulated against analytic contributions. If it hasn't been run, it runs first with the current settings.
• Sparing comparison, Maintainability, Validation and Method and assumptions, matching their tabs.

Charts are drawn as vector graphics, so they stay sharp when printed or zoomed. The PDF uses the standard Helvetica font; Greek letters in labels print as words (λ as "lam", β as "beta").

Saving, loading and importing

The study is kept in this browser between visits, but that copy is not a backup. Save it from the Report tab with Save project JSON, then reload it there or with Import on the Equipment tab.

The project JSON holds everything: basis, library, train groups, events and the equipment list. The CSV covers equipment only; enter events in the events table.

To bring in an equipment list from Excel, use Copy CSV template, fill it in, and paste or import it. Columns, in order:

• class can be a library code (PU-CE) or a class name.
• Leave source, λ, MTTR and PM blank to use the library.
• Set source to user and give λ and MTTR to override the library.
• A new train name creates a train group at 2 × 50 %; check it afterwards.
• spare_mode is running or standby, and in_rbd is yes or no.

Limitations

• Storage is modelled only for supply and offtake events, and is assumed full at the start of each event. There is no storage between equipment units, so results are conservative where intermediate tankage absorbs short trips.
• The N−1 check is a steady-state screen. It doesn't simulate frequency or voltage transients; use the electrical system's dynamic study for that and enter the result as the trip cascade probability.
• The open-cycle hours limit scales the bypass benefit in proportion; it doesn't model which outages the operator would choose to ride through in open cycle.
• Trip cascades are modelled as plant-stop events at a rate proportional to unit trips, not tied to the individual trip in the simulation.
• Events are independent of each other and of equipment failures. A hurricane that also damages equipment needs its longer duration entered directly.
• Restart losses for a whole plant are approximate in the analytic results (section 9); the Monte Carlo is exact for them.
• Common cause acts only within a block, not across blocks (e.g. a cooling-water loss tripping several units).
• Repair crews are counted per block, not shared across the plant.
• Failure rates are constant: no infant mortality, wear-out or ageing.
• Production is a single product at a single rate. No product slates or minimum turndown; if the process can't run at the reduced rate a block allows, model that block as stopping instead.
• Results are only as good as the failure data entered.