Services How We Work Sectors Academy About Contact Open the Toolbox →
Tool 2 Guide

FMECA — Failure Mode, Effects & Criticality Analysis explained

A complete guide to the Bluestream FMECA tool. How to rate Severity, Probability, and Detectability; how RPN and the risk matrix compare; how ISO 14224 taxonomy keeps the analysis consistent; and how the FMECA output feeds directly into the RCM decision tree for task selection.

IEC 60812:2018 SAE J1739 MIL-STD-1629A ISO 14224:2016
⚡ TL;DR

FMECA is the structured analysis that identifies every credible failure mode for an asset, describes its local and system effects, and scores it for risk. Bluestream defaults to the RPN method — Risk Priority Number = Severity × Probability × Detectability — with each scale rated 1-10. A risk matrix alternative using Severity × Probability only is also supported for applications where detectability is not meaningful.

The output is a ranked list of failure modes with treatment recommendations. That ranked list is the direct input to the RCM decision tree (Tool 3) — every failure mode in the FMECA becomes one input row to RCM, and the severity category drives the consequence branch of the RCM logic.

What FMECA is

FMECA is a bottom-up analysis that asks three questions for every item in a system: what can fail (failure mode), what happens if it does (effects), and how bad is that combination relative to everything else the system can do (criticality). The output is a table — one row per failure mode — that becomes the foundation of the maintenance programme.

The method is genuinely old. MIL-STD-1629A codified it in the US military in 1980, SAE J1739 formalised it for the automotive industry in 1994, and IEC 60812:2018 is the current international standard. Every reliability-engineering framework — RCM, RBM, RBI, FMEA-MSR, RAMS — uses FMECA (or its parent method FMEA) as its input layer.

What FMECA is not: it is not a risk assessment of the facility, it is not a hazard analysis (that's HAZOP), and it is not a reliability prediction (that's RBD, FTA, or Monte Carlo RAMS work). It is a structured inventory of failure modes with a risk score attached to each, and that narrow focus is what makes it tractable for thousands of failure modes across a production facility.

Inputs to FMECA: the criticality classification from Tool 1 (which assets deserve the analysis at all), the asset's operating context, the functional failure definition, and — ideally — historical failure data from OREDA, vendor bulletins, and the facility's own CMMS. The Bluestream tool populates failure-mode candidates automatically from the FMECA library so assessors don't start from a blank page.

Why it matters — the quantified handoff from criticality to RCM

Tool 1 (Criticality Classification) gave you the dominant class per asset — C1, C2, or C3 — which governs how much effort to spend on each asset. Tool 3 (RCM/RBM) will tell you what maintenance task to apply to each failure mode. FMECA is the bridge: it enumerates the failure modes and rates each one, so the downstream task-selection logic has something concrete to work on.

Concretely, FMECA produces three things that RCM needs:

  1. A list of credible failure modes — one row per failure mode, grouped by item and function.
  2. A consequence rating per failure mode — not just the asset-level dominant class, but the specific severity of each individual failure mode.
  3. A detectability assessment — which determines whether hidden failures are present, and whether the P-F interval concept is even applicable.

Skip FMECA and the RCM tool has nothing to analyse. Do FMECA poorly and RCM generates credible-sounding tasks for the wrong failure modes. The quality of everything downstream — inspection intervals, spare-parts strategy, workforce sizing, turnaround scope — anchors to how seriously the FMECA was done.

FMEA vs FMECA — the "C" is what changes

The terms are often used interchangeably in casual conversation, but the distinction is real and worth keeping straight.

MethodWhat it doesOutput
FMEA
Failure Mode and Effects Analysis
Identifies failure modes, describes their local and system effects. A narrative table: one row per failure mode, describing what happens. No ranking.
FMECA
Failure Mode, Effects, and Criticality Analysis
Everything FMEA does, plus a quantitative or semi-quantitative criticality ranking. A ranked table: failure modes ordered by risk. RPN, risk matrix cell, or criticality number.

The "C" is the ranking. Without it, an FMEA is descriptive — useful for design review, useful for training, but not actionable for prioritising maintenance effort across a large population of failure modes. The criticality ranking is what makes the method scale from "here are the failure modes" to "here are the failure modes you should care about first."

The Bluestream tool produces FMECA output — every failure mode gets an RPN score (or risk-matrix cell if you use the matrix variant), and the results are sortable by risk. If you want just the FMEA narrative without ranking, run the tool and ignore the RPN column; the analysis itself is the same.

System vs Design vs Process FMEA

The literature distinguishes three FMEA variants based on what's being analysed. Knowing which one you're doing matters because the scope of failure modes is different.

VariantScopeTypical users
System FMECA Failure of a system-level function: a pump train, a safety-instrumented function, a compression stage. Reliability engineers, maintenance engineers, RAM analysts.
Design FMECA (DFMECA) Failure of a new design element during development: a component not yet in service. OEM design teams, product development.
Process FMECA (PFMECA) Failure of a manufacturing or operational process step: a machining operation, a calibration procedure, a start-up sequence. Quality engineers, process engineers, automotive (Tier 1 suppliers).

Bluestream's FMECA tool is primarily oriented toward System FMECA — the variant used in operational reliability engineering for asset-intensive industries. You can run Process FMECA with the same tool by selecting the operation as the "item" and the process failure mode as the FM, but the underlying library and benchmarks are tuned for System-level rotating and static equipment failures.

The two scoring methods

FMECA supports two mainstream scoring methods, plus a few less common variants. The Bluestream tool defaults to RPN (S × P × D) but supports the S × P risk matrix as an alternative. Which one to use depends on whether detectability is a meaningful variable for your failure modes.

5.1 RPN method — S × P × D (Bluestream default)

The Risk Priority Number is the product of three 1-10 ratings:

RPN ranges from 1 (trivial, easy to detect) to 1000 (catastrophic, inevitable, undetectable). The multiplicative structure is the point: a failure mode only reaches a high RPN if all three dimensions are bad. A severe failure that is common and undetectable is a different beast from a severe failure that is rare and obvious — the first demands immediate action, the second may be acceptable with a monitoring task.

RPN is the dominant scoring method in automotive (per SAE J1739) and in most asset-reliability applications where failures can be detected before they occur. It is the method the RCM decision tree assumes when asking "can we detect this failure in time to prevent the consequence?"

5.2 Risk matrix method — S × P only

The risk matrix method drops the detectability dimension and scores failure modes on severity × probability only. The result is a 2D cell reference (often 5×5, sometimes 3×3 or 4×4) that maps to a risk band: low / medium / high / intolerable.

The matrix variant is dominant in process industries — IEC 61511 safety-instrumented systems, process-safety event analysis, and most HAZOP/LOPA work. It is also the default in NORSOK Z-008 Annex C Table C.1 (which is itself a consequence-category matrix, not an FMECA matrix, but the underlying logic is identical).

5.3 Choosing between them

Use the RPN method when:

Use the risk matrix when:

Don't mix methods in the same FMECA. If you start with RPN and switch to matrix halfway through the analysis, the rankings are no longer comparable. Pick one per study. If the facility has assets that genuinely need both treatments, run separate FMECAs and document the method choice for each.

FMECA workflow A flow diagram showing the FMECA analysis process. It starts with identifying failure modes for each item, then rating Severity, Probability, and Detectability on 1 to 10 scales. These three ratings are multiplied to produce the Risk Priority Number, which is then classified into a risk band that determines the recommended treatment: accept, monitor, or mitigate. Identify failure modes per item / function S Severity 1-10 P Probability 1-10 D Detectability 1-10 (inverted) RPN S × P × D 1 – 1000 Treatment 1–60 accept 61–119 monitor 120–199 mitigate ≥200 redesign Hand off to RCM (Tool 3) each FM → RCM decision tree ↻ repeat per item assembled table → ranked by RPN → reviewed by team → treatment plan
The FMECA workflow Per item, identify every credible failure mode, then score Severity, Probability, and Detectability on 1-10 scales. Multiply to get RPN. Classify RPN into a treatment band. The ranked table becomes the input to the RCM decision tree for task selection.

The scales

The 1-10 scales are where most FMECAs get loose. The scores are not arbitrary — they should anchor to concrete thresholds the assessor can point to. The Bluestream tool uses the anchoring described below, which follows SAE J1739 with minor adjustments for oil-and-gas context.

6.1 Severity (1-10) — rating effects, not causes

Severity rates the effect of the failure on the system, assuming the failure occurs. It does not rate the failure mode itself — the failure mode is the how; the effect is the what happens next. This distinction is the single biggest source of inconsistency in FMECA scoring.

RatingMeaningAnchor
10CatastrophicLoss of life; major environmental release; destruction of the facility. Barrier defeat under major accident hazard.
9CriticalSerious injury; regulated environmental release; loss of safety function; multi-day facility shutdown.
7-8HighInjury requiring medical treatment; significant production loss; major repair cost; release below reporting threshold.
5-6ModerateProduction reduction; minor repair; requires response within shift.
3-4LowMinor impact; acceptable performance degradation; routine repair.
1-2NegligibleNo operational consequence; customer-unnoticeable; invisible to operations.

Severity anchors should match your Z-008 consequence classes

If the asset is C3 under NORSOK Z-008, at least one failure mode on it should rate Severity ≥ 7 — otherwise the criticality classification and the FMECA disagree, and one of them is wrong. The Bluestream tool surfaces this inconsistency as a warning when you save.

6.2 Probability / Occurrence (1-10)

Probability rates how likely the failure mode is to occur over a defined time period — typically per year of operation for operational equipment, per demand for safety-instrumented functions.

RatingMeaningAnchor (per year)
10Almost certain> 1 failure per year — routine occurrence.
8-9Frequent1 per 1-3 years.
6-7Occasional1 per 3-10 years.
4-5Remote1 per 10-30 years.
2-3Very remote1 per 30-100 years.
1Improbable< 1 per 100 years; no historical precedent.

Anchoring the probability to a per-year rate forces the assessor to look at data rather than gut feel. For oil-and-gas equipment, OREDA data gives λ in failures per 10⁶ operational hours; divide by the annual hours of operation and you have a per-year rate that maps to the scale above.

6.3 Detectability (1-10) — the scale people get wrong

Detectability is inverted relative to the other two scales. Low D means the failure is easy to detect. High D means the failure is hard to detect. This is the single most common FMECA scoring error — assessors treat D like S and P, giving a 10 to something they think is "highly detectable", which then under-scores its actual risk.

RatingMeaningAnchor
10UndetectableNo practical means of detection before the failure manifests. Hidden failures on safety-instrumented functions without proof testing.
8-9Very low detectabilityOnly detected by specialist test or destructive inspection. Proof test interval equals installation life.
6-7Low detectabilityDetected only by detailed inspection at PM intervals. No online monitoring.
4-5Moderate detectabilityRoutine inspection detects it with reasonable reliability. Oil analysis, thickness measurement.
2-3High detectabilityContinuous or very frequent detection. Vibration monitoring, temperature trending, process alarms.
1Certain detectionFailure is immediately obvious on occurrence (loud noise, spray, alarm). Operator cannot miss it.

Detectable is not the same as evident. An evident failure announces itself to the operator when it occurs (noise, alarm, visible leak). A detectable failure can be found before it occurs (vibration trend, temperature rise). A failure can be detectable but not evident — vibration monitoring catches the bearing degradation before the bearing seizes; the seizure is evident, the degradation is detectable. The D scale is about detecting the precursor, not the failure itself. See P-F interval in the glossary.

RPN interpretation & action thresholds

RPN is a number between 1 and 1000, but not every point on that scale carries equal meaning. Bluestream uses the following action bands by default; the operator is free to override per company-specific criteria:

RPN bandClassificationTypical treatment
1 – 60AcceptableNo action required. Run-to-failure is acceptable if operationally tolerable.
61 – 119MonitorPreventive task or condition monitoring. Reassess if score rises after data collection.
120 – 199MitigateMandatory preventive/predictive task. Cannot run to failure. Barrier upgrade likely.
≥ 200RedesignRisk is not acceptable by maintenance alone. Engineering redesign or procedural barrier required.

Don't let RPN drive action alone. A failure mode with Severity 10, Probability 1, Detectability 1 has RPN = 10 — which the band above calls "acceptable". That's dangerous reasoning for a catastrophic failure. Any Severity ≥ 9 failure mode should be reviewed regardless of RPN, because the consequence of being wrong about the probability or detectability is unacceptable. Bluestream surfaces high-severity low-RPN items as a separate "review anyway" list in the output.

Risk matrix — 5×5 with RPN bands overlaid A 5 by 5 risk matrix with consequence severity on the horizontal axis and probability of occurrence on the vertical axis. Cells are colour-coded from green (low risk) in the low-severity low-probability corner to deep red (intolerable) in the high-severity high-probability corner. An annotation explains how the 1-to-10 RPN scale maps into this 5-band view: scales 1-2 roll up to matrix band 1, 3-4 to band 2, and so on. Probability (1–10 → banded) Consequence / Severity (1–10 → banded) 1–2 3–4 5–6 7–8 9–10 1 3–4 5–6 7–8 9–10 Med High Crit Crit Intol Low Med High Crit Crit Low Low Med High Crit Low Low Low Med High Low Low Low Low Med Risk bands Low — accept / monitor Medium — mitigate High — reduce risk Critical — action required Intolerable — redesign RPN ↔ matrix mapping S × P × D gives RPN (1–1000) S × P alone gives matrix cell The Bluestream tool supports both methods; RPN is the default.
The 5×5 risk matrix Consequence bands on the x-axis, probability bands on the y-axis, risk level in each cell. This is the risk-matrix variant of FMECA — detectability is not a dimension here. The cell colour determines treatment priority directly, without computing a numerical score.

Criticality ranking alternatives

RPN and the risk matrix are the two mainstream methods, but they are not the only ones. For legacy or regulatory reasons, you may encounter:

8.1 Criticality matrix

A 2D matrix identical in structure to the risk matrix above but specifically labelled as "criticality" rather than "risk". Common in military and aerospace per MIL-STD-1629A. Functionally equivalent to the Bluestream risk-matrix mode; the terminology differs but the logic is the same.

8.2 Criticality number (MIL-STD-1629A quantitative)

A fully quantitative score computed as Cm = β × α × λp × t, where β is the conditional probability of the failure effect, α is the failure mode ratio, λp is the item failure rate, and t is the operating time. Each failure mode gets a numerical criticality contribution; the sum across all failure modes of an item gives the item's criticality number.

The method is more rigorous than RPN but requires high-quality input data (failure rate λp, mode ratios α) that most operating facilities don't have. It is still used in aerospace and nuclear applications where the data infrastructure supports it. Bluestream does not implement this method in Tool 2, but you can substitute a criticality number by overriding the RPN column if you have the data.

ISO 14224 failure-mode taxonomy

ISO 14224:2016 defines a standard taxonomy for failure-mode coding across the petroleum and natural gas industries. The standard provides:

Why this matters for FMECA: without a standard taxonomy, every assessor invents their own failure-mode vocabulary. "Bearing failure" means different things to different people; "bearing - inner race spalling - fatigue" is unambiguous. ISO 14224 gives you the unambiguous version.

The Bluestream tool's FMECA library uses ISO 14224 failure-mode codes as the default vocabulary. When you select a pump in the asset classification (equipment class PU), the tool presents the ISO 14224 Table B.6 failure modes as the starting list — external leakage, internal leakage, fail to start on demand, vibration, noise, overheating, and so on. You can add custom failure modes, but the ISO-coded ones stay linked to the taxonomy for benchmarking against OREDA and vendor data.

ISO 14224 integration with the CMMS is the real win

Coding failure modes consistently in FMECA is good. Coding them consistently in the CMMS over years of operation is transformative — it turns work-order history into usable reliability data. The FMECA and the CMMS failure-code library should use the same ISO 14224 codes. If they don't, reconciling FMECA predictions against actual failure history becomes an archaeological exercise.

FMECA → RCM: what flows through

The FMECA output feeds the RCM decision tree (Tool 3) one failure mode at a time. Every row in the FMECA table becomes one input row to RCM, carrying the following fields:

FMECA fieldUsed by RCM for
Failure mode descriptionIdentity — the RCM decision applies to this specific FM.
Local effectDetermines whether the failure is evident or hidden.
System effectDetermines the consequence category branch (safety / environment / operational / non-operational).
Severity scoreConfirms the consequence category; triggers the "is it safety-related?" question in the RCM tree.
Probability scoreInforms the age-reliability question (random vs age-related).
Detectability scoreInforms whether condition-based maintenance is feasible (low D = CBM viable; high D = CBM not viable).
Recommended treatment (from RPN band)First-pass task recommendation; RCM decision tree confirms or overrides.

The Bluestream Tool 2 output is structured specifically to carry these fields into Tool 3 without re-entry. When you progress from FMECA to RCM in the platform, the FMECA result is the RCM input — not something you copy, something the tool passes forward automatically.

How the Bluestream tool implements FMECA

Tool 2 on /platform is a failure-mode-by-failure-mode workflow. The sidebar lets you select the asset (carried forward from Tool 1 if you came through the sequence), and the main panel cycles through each failure mode in the ISO 14224 library for the asset's equipment type.

For each failure mode, the tool asks four questions:

  1. Is this failure mode credible for this asset's operating context? If not, skip it — not every Table B failure mode applies to every installation.
  2. What is the Severity? Pick 1-10 using the anchor table; a brief rationale is recorded.
  3. What is the Probability? Pick 1-10; the tool pre-populates a suggestion from OREDA benchmark data where available, which the assessor overrides or accepts.
  4. What is the Detectability? Pick 1-10 using the inverted scale; the tool warns if the score disagrees with the "detection method" free-text field.

Click Score FMECA to consume 1 token and produce the ranked output. The output shows every failure mode with its S/P/D scores, the computed RPN, the treatment band, and any inconsistency warnings (Severity disagrees with criticality class, Detectability disagrees with detection-method text, Probability disagrees with OREDA benchmark). The ranked table exports to the RCM tool for task selection.

Worked examples

Three failure modes showing how the S/P/D scoring lands differently depending on the asset and its context.

Example 1 — Centrifugal pump bearing wear (classic CBM candidate)

RPN 72 · Monitor

An oil-and-gas produced-water injection pump — continuous duty, 500 m³/hr, API 610 OH2. Failure mode: rolling-element bearing wear (ISO 14224 code: WEAR). The pump is in a 2×100% arrangement; the operating context is offshore platform, RED-B redundancy.

Severity (S) = 6 — Pump trips offline; standby pump takes duty. Single-pump operation for days while the failed unit is repaired is acceptable. Moderate impact.
Probability (P) = 4 — OREDA data for pumps of this service gives λ = 14 per 10⁶ hours for bearing failure; ~1 per 8 years per pump. Remote category.
Detectability (D) = 3 — Online vibration monitoring on this pump. Bearing defect frequencies are trended continuously; warning triggers at defined amplitude thresholds, weeks before failure. High detectability.
RPN = 6 × 4 × 3 = 72
Treatment band: Monitor.
Recommended task: Continue vibration-based CBM; no additional action.

This is the canonical CBM case. Moderate severity, low probability, high detectability — the low D makes the RPN tractable. Notice that without the vibration monitoring (D = 6-7 instead), the RPN would rise to ~150-170 and move into the mitigate band. The D score directly pays for the cost of the monitoring system.

Example 2 — PSV fails to lift on demand (hidden failure)

RPN 270 · Redesign/FF

A pressure safety valve on a separator — spring-loaded PSV sized per API 520, protecting against blocked-outlet overpressure. Failure mode: fails to lift at set pressure on demand (ISO 14224 code: FTO). The PSV is the last barrier against vessel overpressure; no upstream protection except the operator-response alarm.

Severity (S) = 9 — Loss of last barrier against overpressure of a hydrocarbon-service vessel. Consequence ranges from major release to vessel rupture depending on the upset. Critical.
Probability (P) = 3 — PSV seat sticking or spring relaxation. OREDA-equivalent rate for failure to open on demand is ~10⁻³ per demand, but the demand rate on the PSV is itself infrequent. Very remote.
Detectability (D) = 10 — This is a hidden failure. The PSV looks fine from outside; there is no routine monitoring that confirms it will lift at set pressure. The only detection is an actual demand event or a removal-and-bench-test. Undetectable in normal operation.
RPN = 9 × 3 × 10 = 270
Treatment band: Redesign.
Recommended task: Failure-finding task (scheduled function test / bench test) at interval determined by acceptable hidden-failure probability. Typical: 3-5 year interval for PSVs in hydrocarbon service.

The hidden-failure problem is the signature FMECA case. S and P are moderate; D pushes the RPN into redesign territory. The only maintenance response is a scheduled failure-finding task — there is no preventive task that addresses the failure itself, because the failure mechanism is sticking of parts that are not accessible during operation. Offline testing is the only option.

Example 3 — Mechanical seal fugitive emission (environment-dominant)

RPN 144 · Mitigate

A mechanical seal on a crude export pump — single-seal arrangement, API 682 Plan 11 flush. Failure mode: external leakage to environment below reportable threshold (ISO 14224 code: ELEX — external leakage). The operator has public methane emission targets under Norway's climate commitments; fugitive hydrocarbon emissions count toward the facility GHG budget.

Severity (S) = 8 — Under the pre-2024 Z-008 framework this rated as medium severity because personnel consequence is low. Under Z-008:2024 with Environment split out as a separate category and Annex E GHG screening, the environmental consequence of sustained fugitive emissions pushes severity higher. High.
Probability (P) = 6 — Mechanical seals on hydrocarbon service have λ ≈ 40 per 10⁶ hours for external leakage; ~1 per 3 years. Occasional.
Detectability (D) = 3 — Optical gas imaging (OGI) surveys at 6-month intervals; acoustic monitoring on the seal vent line. Detectable with standard operator surveillance. High detectability.
RPN = 8 × 6 × 3 = 144
Treatment band: Mitigate.
Recommended task: Continue OGI surveillance; consider upgrade to API 682 Plan 52 dual-seal arrangement on next turnaround to reduce severity.

This example shows how the 2024 Z-008 revision (Environment split out from HSE) actually changes FMECA scoring. A failure mode that was RPN ~108 under the old framework (S = 6 for combined HSE) becomes RPN 144 under the new framework (S = 8 for Environment alone). The higher RPN justifies mitigation that was harder to justify when environmental consequence was lost in the HSE average.

Video walkthrough

A screen-recorded walkthrough of the Bluestream FMECA tool, covering the ISO 14224 failure-mode library, S/P/D scoring with rationale, the RPN vs risk-matrix toggle, and reading the output table.

Common pitfalls

Six FMECA errors that recur across projects. Most of them come from pressure to finish the analysis fast, which reliably produces output that looks like FMECA but fails its purpose.

1. Rating cause instead of effect in severity

Severity is the effect on the system, not the intrinsic "badness" of the failure mode. A shaft fatigue crack rates Severity based on what happens when the shaft breaks (partial trip, full stop, collateral damage to coupling and driver), not on how bad the fatigue itself sounds. Anchor to the effect description every time.

2. Conflating evident with detectable

An evident failure announces itself when it happens; a detectable failure can be found before it happens. The D scale is about advance detection — the P-F interval. A bearing seizure is evident (loud noise, motor trip), but undetectable in the sense that matters for FMECA scoring unless vibration monitoring is in place. Rate D on the availability and reliability of precursor detection, not on whether the failure itself is noisy.

3. Treating hidden failures as D = 5 because "we'll catch them at PM"

Hidden failures are D = 10 unless there is an explicit, scheduled test that exercises the function. Relying on "we'll notice at PM" is not a test — PM inspects the physical condition of the asset, not its function. If the failure is hidden and there is no functional test in the programme, the D score is 10 and the RPN will drive to redesign or failure-finding. Resist the urge to soften that.

4. Single-assessor FMECA

FMECA is a team method. IEC 60812 requires a multi-disciplinary team — typically operations, maintenance, reliability, and a design specialist familiar with the equipment. A one-person FMECA reflects one person's blind spots and will miss failure modes that another discipline considers obvious. If the schedule only allows for a single assessor, schedule a team review afterwards — that's still FMECA. A pure single-person exercise is not.

5. Scoring without failure history

Probability scores derived from intuition drift toward the middle of the scale (3-5). Scores anchored to actual failure rates — from OREDA, vendor bulletins, or CMMS history — are more defensible and spread more widely. Before the team scores Probability, someone should have pulled the relevant data. If the data doesn't exist, that is itself a finding — flag it in the output as "probability based on expert judgement, no benchmark available" so downstream users can discount it appropriately.

6. Letting RPN drive action alone

RPN multiplies three scales, so a zero in any dimension drives the product to zero. A failure mode with Severity 10, Probability 1, Detectability 1 has RPN = 10 — well inside the "accept" band — but its consequence is catastrophic. Every Severity ≥ 9 failure mode should be reviewed regardless of RPN. Bluestream surfaces these as a separate list, but reviewers must actually look at the list. The formula is not a substitute for engineering judgement.

References

Next steps