Iron Lake Reliability

Weibull is pronounced WYE-bull (hard I).

Iron Lake Reliability tutorial

Iron Lake Reliability closes the reliability loop for a system: 1 Reliability Prediction (MIL-HDBK-217F parts count / part stress, NSWC-11) → 2 Reliability Block Diagram → 3 Bayesian Analysis → 4 Monte Carlo Simulation against the requirement → 5 Close the Loop (run the loop, design what-ifs) → back to the prediction. The Weibull tools follow: life data on the Weibull Workbook, the Weibull Calculator, Fleet Prediction and RDT. This page is the public tutorial. The in-app Tutorial tab adds live math callouts, Rank models, and the same walkthrough next to the workbook chrome.

Open the Iron Lake Reliability workbook Open Weibull Calculator Open the interactive tutorial

Product walkthrough

A guided tour of Iron Lake Reliability on this site. You follow one example project — example data, for learning, not a measurement of a product — from a parts prediction to a system decision: the module cards, the shared File menu, Reliability Prediction, the block diagram, a short stop in the Weibull workbook, Bayesian analysis, Monte Carlo, Close the Loop, fleet prediction and reliability demonstration, then a longer look at the workbook generator for failure modes, effects, and criticality. Same self-hosted file as the in-app Tutorial tab.

Start the closed loop

Hi, and welcome to Iron Lake Reliability. We are going to follow an example project all the way from a parts prediction to an actual system decision. Quick heads-up: every result and every bit of life data here is example data, for learning. It is not a measurement of your product.

The idea is a closed loop. Prediction feeds the block diagram, life data and Bayesian updating sharpen it, Monte Carlo checks it, and then you improve the design. On the intro page, the module cards let you pick the analyses you actually need. Choose yours, and click Jump into analysis. The reliability block diagram stays on; it is the core.

File → Load example

The File menu is shared across every tab. Before anything else, save your current work, because loading an example replaces the whole project, not just the sheet you are looking at. Then go File → Load example and pick the built-in example. That brings in the prediction, the block diagram, the workbook datasets, the Bayesian analyses, and the fleet and test inputs, all at once.

Reliability Prediction

The example is broken into subunits: battery modules, the inverter, the motor, control electronics, and a few supporting circuits. The handbook basis is MIL-HDBK-217F, plus NSWC-11. Do take a second to read the environment and assumptions before you compare numbers.

The summary reports FIT, failures per billion hours, with parts count and part stress side by side. One thing that trips people up: prediction is evaluated at 100 percent duty. The real duty cycle gets applied later, in the block diagram’s mission profile.

For your own project, Actions opens the Import Wizard. Look over the mapping and the proposed changes before you apply anything. Amend existing Prediction lets you touch just the subunits you choose and leave the rest alone. Check the source notes on custom rates, especially for mechanical parts or battery cells, since those rates usually come from a different basis than the electronics handbook.

Expand BOM gives the parts table some breathing room. Under File, the parts library does a reviewed round trip. Update library from Prediction shows you what is new, what changed, and what did not, before it applies anything. What you really want out of this stage is a traceable rate for each subunit, with quantities, conditions, and assumptions you would be comfortable defending.

Reliability block diagram

The example mixes series elements with redundant groups. In the Mission panel, check life in years, operational duty, and dormancy. Together, those set the real mission exposure. The main result is probability of success, shown next to the requirement and whether you meet it. A probability without a stated mission is basically a horoscope, so keep the mission and requirement in view whenever you talk about the number.

Probability of success by year shows how the system holds up over its life. Compare it against the dashed requirement line, on a fixed 0 to 100 percent axis. The importance table helps you find the weakest link. Rank one is your leading contributor in that view. A bearing or a battery wearing out behaves very differently from a constant failure rate.

The legend only shows the symbols you are actually using. You can lock selected values and press Rebalance allocations, then see whether the rolled-up allocation closes on the requirement. In Block data sources, check whether each block is using prediction, manual values, test data, or field data.

Weibull workbook

The Weibull workbook, under Other Reliability Analysis Tools, is the bridge from prediction to observed life. The example datasets cover a battery module, a motor bearing, an inverter, and a motor. Keep different failure mechanisms separate. Mixing them into one population is how you get a beautiful fit to nothing in particular. Then carry that evidence into Bayesian analysis, or import the fit straight into a block. Chapters 7–13 below are the written reference for the fit math; this tour does not linger there.

Bayesian analysis

The prior is your baseline belief, usually from the subunit prediction. The data section adds test or field evidence, and you get the posterior. The example shows why that matters: observed behavior can make one subunit look worse than predicted while another one actually improves. Keep the prior, the data, and the posterior clearly separate when you explain it. An update changes the evidence behind a block. It does not quietly relax the requirement so the design passes.

Monte Carlo simulation

Monte Carlo simulation runs the live block diagram, a million times by default. Compare the simulated probability of success, and its interval, with the analytic result and the requirement. If you get a disagreement warning, treat it as a nudge to go back and review the diagram and its assumptions.

Close the Loop

Close the Loop is where it comes together. Run a cycle across prediction, the block diagram, Bayesian analysis, and Monte Carlo. Check the requirement status, look through the ranked design what-ifs, and compare against your saved cycle history. The real question is which change actually fixes the weak part of the system, not which number looks prettiest. A proposed rate reduction is just a scenario until there is an engineering change behind it. Wishful thinking, sadly, has no failure rate.

Fleet prediction and reliability demonstration

Fleet prediction turns the selected life model into expected failures over a warranty period, for a whole population. Check the life model source, fleet size, warranty time, and units before you read the forecast. An expected count is a planning number. It is not a promise that every fleet returns exactly that many.

Reliability demonstration sizes a test for a reliability claim. Look at the target, confidence, allowed failures, and life source together. When the Weibull life-stress fit supplies an acceleration factor, the field shows where it came from. You can override it, or press Reset to computed. A bigger acceleration factor only means something if the test and the field are exercising the same mechanism. Otherwise you have just found a faster way to be wrong.

Workbook generator

This part covers failure modes, effects, and criticality, using MIL-STD-1629A. On that tab, under Criticality definitions, open the workbook generator. Pick your subunits, the content, and the format, and it writes one workbook for that selection.

The example workbooks have a summary report up front, a sheet for each subunit, and an interface worksheet. The summary is where a reviewer starts: failure modes by severity, criticality totals, and the top items to look at.

The interface worksheet considers, for every external connector pin, open, short to ground, short to power, and short to the adjacent pin. Connectors, as anyone who has done field work will tell you, are where good designs go to wiggle loose, so review these rows carefully.

On a subunit sheet, the circuit and part failure-mode rows are prefilled from the bill of materials, with reference designator, subunit, and circuit function. Review those generated rows against the real design. The failure-mode distributions are editable. Then fill in the effects at the subunit, unit, and system levels, so it is clear what actually happens at each level when something lets go.

Task 102 criticality updates along with the prediction, and the criticality matrix lets you compare severity against criticality. There is one Export menu, with PowerPoint, Excel, PDF, and CSV. Excel gets the full tables for each subunit, and the summary report also exports to Excel or PDF for the review package.

From here, go back around the reliability loop as your evidence improves, and make the next design decision.

Product in one paragraph

Iron Lake Reliability is a client-side reliability workbook for hardware. It fits two-parameter Weibull (β, η) or lognormal (μ, σ) to right-censored times, projects Arrhenius or inverse-power accelerated life to use conditions, sizes an RDT, models systems with reliability block diagrams, predicts failure rates from a parts list, and exports each tab to PowerPoint, Excel, PDF, or CSV. It is not a certified assessment and it is not an Omni-style PDF/CDF curve tool.

Reliability Analysis Loop overview

The top row follows the loop: stages 1, 2 and 3 (Reliability Prediction, RBD and Bayesian Analysis), stage 4 (Monte Carlo Simulation) and, last, the Close the Loop tab itself, which runs the loop and shows all four stages and their status. The Other Reliability Analysis Tools drop-down holds the Weibull Workbook, the Weibull Calculator, Fleet Prediction and RDT. Each stage tab links to the loop with Close the Loop →.

  1. 1 · Reliability Prediction. Reliability Prediction estimates the failure rate λ from the parts list with MIL-HDBK-217F Notice 2 parts count or part stress, and NSWC-11 for mechanical parts.
  2. 2 · Reliability Block Diagram. Assign from prediction links blocks on the Reliability Block Diagram to prediction Subunits; model the system with series, parallel, k-out-of-n, and cold-standby groups.
  3. 3 · Bayesian Analysis. The prediction becomes the prior; test and field data update it to a posterior that the linked blocks follow, with a credible band.
  4. 4 · Monte Carlo Simulation. Simulate the system with parameter uncertainty and compare Ps and its confidence interval with the requirement.
  5. Design what-ifs. The Close the Loop tab ranks the top contributors and what-ifs (redundancy, a better quality level). Apply one, update the loop, and the next cycle starts again from the prediction.

Getting started. The intro page first asks you to Customize your reliability analysis closed loop: each analysis module is a card. Click anywhere on a card to switch it on or off; selected cards are highlighted with a check in the corner. The Reliability Block Diagram (RBD) is the core and always on. Then click Jump into analysis. Change the modules any time with ⚙ Customize analysis (or File → Analysis setup (modules)…).

Fastest way in: click Load the example loop at the top of the in-app Tutorial. Cycle 1 lands at about 0.973, below R ≥ 0.99; a redundant Controller lifts cycle 2 to about 0.998, which meets it. Everything works signed out.

The example project on first visit. On a first visit (nothing saved in this browser) every tab opens on one preloaded example project. The results and the life data are example data, for learning, not a measurement of a product. A quiet note under the tool tabs reads “Example project: Electric motorcycle (preloaded) — New project”: that label is the built-in parts list, not a product story. New project asks first, then returns every tab to blank defaults (empty prediction, one-block RBD at τ = 1,000 h with the default requirement and FIT (FPBH) units, empty workbook, no Bayesian analyses, default Monte Carlo, Fleet and RDT inputs, no loop history). Load example brings the example back. Saved or autosaved work and files you open are never replaced. Each tab’s File → This tab clears only that tab.

Reliability Prediction

The Reliability Prediction tab estimates an equipment failure rate from its parts list. Electronic parts use MIL-HDBK-217F Notice 2 (1995): the Appendix A parts count tables or the full part stress models (Sections 5–23) with junction temperatures; bearings, gears, motors, springs, and couplings use NSWC-11 (2011). 217F is not maintained: results are estimates for comparison, not field predictions. It works signed out and autosaves in your browser. Load example loads example data (the built-in parts list, labeled an electric motorcycle: nine subunits at Ground, Mobile; preloaded on a first visit), with R ≥ 0.95 over a 5-year life (about 149 ride-hours a year, τ ≈ 1,606 effective hours). Each subunit’s 217 / NSWC-11 result is the baseline prior of a Bayesian analysis whose test or field data come live from the Weibull Workbook (battery fade, motor bearings and pilot-fleet datasets); the posteriors drive the RBD (3-of-4 battery modules, redundant 12 V DC-DC, a 2-of-3 throttle-sensor vote, redundant brake-light circuits), Monte Carlo with a constant-rate vs wear-out hazard chart, Close the Loop, Fleet Prediction (2,000 battery modules over the 5-year warranty) and the Reliability Demonstration (RDT). Each part row has a note for the source of its rate (vendor datasheet, vendor L10 or cycle-life data, field data, test); a pasted or dropped image (for example a datasheet screenshot) can be attached and is saved with the project; custom rates without a source are flagged and the notes go into the reports.

Environment and handbook basis

  1. Enter a Prediction name and Analyst. Keep Handbook revision at MIL-HDBK-217F Notice 2.
  2. Pick the Environment (default), one of the 14 handbook environments (GB … CL); a new prediction starts on SF (Space Flight), saved predictions keep theirs, and lines can override it.
  3. The Mission time τ is the RBD mission time (read-only, Set in RBD →; the mission reliability and report follow it live). Enter Prior operating hours T0; the prediction is at 100% duty (duty is set on the RBD). Leave Industry Accepted Modifiers at None (native 217) for pure handbook values.

Method, temperatures, and part stress

  1. Pick the Method: Parts Count (MIL-HDBK-217F N2 Appendix A) uses λg·πQ and ignores temperature; Part Stress (MIL-HDBK-217F N2 Sections 5–23) evaluates each part with its section model λp = λb·πT·πQ·πE·(section factors). The method is saved with the project and printed in every report.
  2. Enter the Interface / baseplate temperature (°C) from your thermal analysis.
  3. Give each BOM line a ΔT (°C): the rise from the baseplate to the temperature the handbook uses for that part (junction Tj for microcircuits and semiconductors; case for resistors; ambient for capacitors and relays; hot spot for inductors; insert for connectors). The Tj / T column shows T = Tbaseplate + ΔT. ΔT can be imported from an Excel / CSV column or set with Global change; it belongs to the prediction row and is never copied from the Parts library (neither are reference designators or stresses).
  4. πT comes from each section’s own equation and activation energy at T, e.g. πT = exp(−Ea/k · (1/(T+273) − 1/298)) for transistors (§6.3), resistors (§9.1) and capacitors (§10.1); microcircuits use λp = (C1·πT + C2·πE)·πQ·πL (§5.1, πT from §5.8). Relays, motors and synchros carry temperature inside their own λb equations. Click a part to see every factor with its section and page and to edit the stress inputs (power, voltage stress, contact form, complexity…); each default is shown with its source.
  5. Assumptions: a blank ΔT or baseplate uses the Appendix A default temperature for the environment. Part types without an implemented part stress model (SMT boards, §16.2) fall back to parts count and are flagged. A ⚠ marks temperatures above a section’s model range or a typical maximum rating. The summary shows both totals side by side.

Add parts by hand or from the library

  1. + Add line: reference designators, part number, description, quantity, subunit.
  2. Set the part type (217F family and subtype, quality level πQ, πL years for microcircuits, or an NSWC-11 mechanical type). Each line shows λg, πQ, πL, unit and line FIT (FPBH), share, and the handbook citation.
  3. Part numbers in your library fill in automatically (matched by part number, then category and keywords). + Custom part takes your own λ. Unmatched lines are flagged; select them and use Bulk assign.

Import an Excel / CSV BOM

  1. Open Actions › Import Wizard and Import File Management. Download Excel BOM template gives you the recognized columns: reference designators, part number, description, manufacturer, quantity, category / part type, quality level, subunit, environment override, years in production, notes.
  2. Choose .xlsx, .xls, or CSV (or paste). Pick the sheet; check the detected header row.
  3. Map columns to fields (common names are guessed). Remember this mapping reuses it for files with the same columns.
  4. Choose Replace current lines or Append, then import. Fix unmatched lines with Bulk assign.

Parts library

  1. File › Parts library › Edit parts library… lists your standard parts, assignments, and keywords. Add or edit entries, or use Save to library on a BOM line. Update library from Prediction… previews New / Changed / Unchanged entries before applying (Undo reverts); Re-match BOM… or an import then uses them.
  2. Share the library through a shared library file (File › Parts library › Library location…).

Industry Accepted Modifiers

  1. Actions → Industry Accepted Modifiers… holds named modifier sets. Choose one for the prediction next to the environment; None (native 217) keeps native handbook values and applies no ground rules. Lines can override the set.
  2. Ground rules switch on by default when a set is selected and can be toggled one by one. Rule 1, commercial passives → R-level quality: Lower / commercial resistors and capacitors use the 217F ER R πQ (0.1); inductors, crystals and filters have no R level and keep theirs. Rule 2, microcircuits → minimum B-level quality: parts worse than Class B use the Class B πQ (1.0). Rule 3, commercial semiconductor FETs → 0.1× failure rate: commercial Si FETs (MOSFET or JFET) and GaAs FETs have final λp multiplied by 0.1, and a part-stress MOSFET uses that same 0.1 once (not a second reduction). Commercial discrete semiconductors (diodes, transistors, optoelectronics) use JANTX πQ 1. Commercial electromechanical relays use High quality, πQ 1.5; commercial solid-state and time-delay relays use πQ 1. Ferrite beads (fixed coils / chokes) use πQ 1. Commercial switches use πQ 1. Commercial general connectors use πQ 1, are modeled as Rectangular / General, use mating cycles 0 to 0.15 per 1,000 h (the range above 0.05 to 0.5, πK 1.5), and on a modular circuit card assembly count as a single connector (half a mated pair). SMT plastic BGAs use lead configuration πLC 100 and ceramic BGAs use πLC 50. Commercial crystals and oscillators use High quality, πQ 1; oscillators are the quartz crystal in Miscellaneous.
  3. Order: quality rules set πQ first (one quality rule per part). Connector type, mating cycles and SMT lead configuration adjust the model inputs, then the set’s rows and line overrides, then the single-connector factor and rule 3 on the final λp. A commercial FET can take both the discrete-semiconductor πQ rule and the FET multiplier. Affected lines are tagged, and the report and exports list the active rules.
  4. Optional rows replace or multiply πQ, πE / environment λg, πL, a λ multiplier, or an NSWC-11 C-factor for one family, all 217 parts, all NSWC-11 parts, or all parts; the most specific row wins. No row values ship with the app.

Global change

  1. Global change…: choose a scope (all parts, selected rows, current filter, categories, part families, or a subunit).
  2. Choose the field (quality level, environment, πL years, NSWC-11 temperatures, duty, adjustment set, λ multiplier, or a C-factor) and set or multiply a value.
  3. Preview the lines changed and λ before → after, then Apply. Undo reverses it.

Highlight adjusted parts

  1. Tick Highlight adjusted parts to badge every line whose factors differ from the native handbook default; the badge names each factor and its source.
  2. Filter with Adjusted only. The report lists them in an Adjusted parts table.

NSWC-11 mechanical parts

  1. Set a line to Bearing (ball / roller), Gear, Electric motor, Spring (helical compression), or Mechanical coupling.
  2. Fill in the model inputs. A bearing takes load rating and applied load (or rated L10 life), speed, reliability basis, lubricant viscosity ratio, water content, temperature, service condition, and contamination. Each cites its NSWC-11 equation.
  3. The line shows the base rate and each C-factor; the λ adds to the total.

Summary, pie, and Pareto

  1. The summary gives λ per 10⁶ h and FIT (FPBH), MTBF in hours and years, mission reliability R(t) = exp(−λ·t·duty), and P(failure).
  2. The pie shows failure rate by category, filtered to all, electronic, or mechanical parts, with small categories folded into Other.
  3. The Pareto ranks BOM lines by FIT (FPBH) (quantity included), largest first; the top contributor is red, and hovering a bar shows its share and cumulative share. The downloaded PNG and the exports add a cumulative-share line and a dashed 80% marker. Both charts have Download PNG.

Report and exports

  1. The tab’s Export drop-down builds PowerPoint, Excel, PDF (print), and BOM with λ CSV with only the prediction: summary and assumptions, pie and Pareto images, category and subunit summaries, adjusted parts, flagged lines, and the full BOM.
  2. Linked RBD results are left out unless you tick Include linked RBD results (off by default). Export → Report options… sets the title, project, and analyst.
  3. File › Export BOM writes BOM CSV or BOM XLSX. Report options… has Subunit checkboxes for PowerPoint / Excel, and Excel carries the schematic pages; the prediction is saved with the project (File → Save, .ILR).

Link to the RBD

  1. On the RBD, select a block and use Assign from prediction to link it to one prediction Subunit.
  2. Blocks stay linked and update when the BOM or library changes; unlink to freeze a value.

Subunits and the FIT summary

  1. The parts table opens grouped by Subunit with every Subunit collapsed; open one with its + or use Expand all, and rename one with its ✎ (or double-click the name): all its lines move, and linked RBD blocks follow. Group by switches to category, unit or no grouping.
  2. The summary shows the failure rate and the Parts Count and Part Stress totals in FIT (FPBH), with 2–3 significant digits.
  3. Columns ▾ picks and reorders columns and closes on any click outside it or on Esc.
  4. Drag a BOM column header sideways to move the column, as in Excel: a blue bar shows where it will land. The order is saved with the project, and Reset column order restores the default.
  5. Each chart title names what it plots: the Unit (project) name for the whole system, or the Subunit / Subassembly you focused on. The name updates when the focus changes and is also in Download PNG and the exports.

Worked example: parts count check

Parts count check (λ = Σ N·λg·πQ)
Lineλg (GF)πQλ per 10⁶ h
40 × film chip resistor (RM)0.016106.4
MTBF at 50 per 10⁶ h10⁶ / 5020,000 h

Worked example: part stress at Tj

The example’s 32-bit MOS microcontroller, Space Flight (SF, the default environment), commercial quality, 40 °C baseplate with the example’s nominal microcircuit ΔT of +30 °C, so Tj = 70 °C (§5.1 p. 5-3, §5.8 p. 5-13, §5.9 p. 5-14, §5.10 p. 5-15).

λp = (C1·πT + C2·πE)·πQ·πL
FactorValueSource
C1 (32-bit MOS microprocessor)0.56§5.1
πT at 70 °C, Ea = 0.35 eV0.1·exp(−0.35/8.617×10⁻⁵ · (1/343 − 1/298)) = 0.60§5.8
C2 (hermetic package)0.053§5.9
πE (SF) · πQ (commercial) · πL0.5 · 10 · 1§5.10
λp(0.56·0.60 + 0.053·0.5)·10 ≈ 3.61 per 10⁶ h

Reliability Block Diagram (RBD)

The Reliability Block Diagram tab models a system from its parts. It opens with a sample pump skid (two pumps in parallel → controller → 2-out-of-3 pressure sensors) wired left to right from Start to End; a second sample, Standby power, has a one-shot start battery and a cold-standby genset pair. It works signed out and autosaves in your browser.

Build the diagram: blocks and structures

  1. Load a sample or click New. Type a Diagram name; it names every export.
  2. From Building blocks, drag a chip onto the diagram or click it to add next to the selection: Component (one block with its own model), Series (every branch must work), Parallel (active redundancy: any one branch), k-out-of-n (at least k of n; set k in the group properties), or Cold standby (spares dormant until switchover; set the Switch reliability per switchover).
  3. Drop on a block to place before or after it; drop on a group frame to put it inside. Groups nest. Undo reverses the last edit.
  4. A long chain wraps onto the next row to fit the screen: each row reads left to right, and a return line leads into the next row. A group wider than the screen is drawn smaller on its own row. Exports use the same layout.
  5. Shortcut: select a block and use Make redundant → Parallel pair, 2-out-of-3, or Cold standby pair.
  6. A parallel pair (or 2-of-3, or standby pair) is not made a Unit automatically: it stays where you put it, at the level of its members. Set its Level to Unit (or use Group into Unit) to make it a Unit.

Block life models

  1. Click a block to open Block properties: name, icon (library, uploaded image resized in the browser, or a colored badge), and model.
  2. Choose Weibull (β, η, optional γ), a constant failure rate λ or MTBF (exponential), or a fixed probability p for one-shot items. Constant-rate blocks default to failure rate λ and the diagram labels them in FIT (FPBH) (λ 20,000 FIT (FPBH)); switch Enter as to MTBF to label them MTBF 50,000 h. Saved projects keep their choice.
  3. Import β / η from workbook fit copies the current workbook fit; Copy model to siblings applies the model to the rest of the group.

Mission time τ and prior age T0

  1. Pick the Time unit (hours, cycles, miles, km) and enter Mission time τ.
  2. Enter Requirement R ≥ next to mission time (the system reliability needed at τ; new diagrams start at 0.95). This is the only place the requirement is set: it is saved with the diagram, the results show Meets or Below requirement, and every other tab reads it and τ live and read-only with Set in RBD → (Bayesian Analysis: each analysis's required R(τ) or p, P(meet) and verdict; Monte Carlo, including the component-list mission time; Close the Loop and the loop status; the Reliability Prediction tab's check of the prediction at τ); no tab keeps its own copy, and it updates as soon as it changes.
  3. Failure rate units: FIT (FPBH) (default) or MTBF (hours). One display setting for the whole project (RBD blocks and results, Close the Loop, Bayesian summaries, history and exports); values and math do not change. Set only on the RBD Mission panel, saved with the diagram; older projects open in FIT (FPBH).
  4. The Mission profile box (on by default for hours diagrams) takes the life requirement (years), operational duty cycle and a dormancy factor (default 3%) that applies only to the off time: τ = teff = years × 8760 × duty + years × 8760 × (1 − duty) × dormancy (on hours at the full rate plus off hours scaled). Example: 10 years, 30% duty, 10% dormancy → 26,280 h + 61,320 h × 0.1 = 32,412 h. The box shows the formula and the on / off breakdown; at 100% duty τ = years × 8760 h.
  5. Optional Prior age T0: each Weibull block is known good at T0 and uses R(T0 + τ) / R(T0). Exponential blocks are memoryless; fixed p ignores time; cold-standby spares start new.

Results, block FIT (FPBH), and importance

  1. The Ps R(τ) card gives Rsys(τ) next to the Requirement card (✓ Meets / ✗ Below), with P(failure) = 1 − Ps, system MTTF (mean residual life when T0 > 0) and B10 life. Probability of success by year shows years 1–10 as bars or lines; a legend under the diagram explains its symbols.
  2. Reliability allocation splits the requirement over Units and Subunits; the Closure line shows whether the rolled-up allocation closes on the requirement, with blocks below allocation in red.
  3. The Per-block reliability and importance at τ table lists each block by rank with R(τ), FIT (FPBH), and Criticality. FIT (FPBH) is failures per 10⁹ h: λ × 10⁹ = 10⁹ / MTBF for exponential blocks, and the average over the mission, −ln R(τ) / τ × 10⁹ (marked “avg over τ”), for Weibull and fixed-p blocks. Criticality is the block’s share of the system mission unreliability; the bar shows it.
  4. Rank 1 is the weakest link, the block the system is most sensitive to, outlined and tagged WEAKEST on the diagram. A redundant block can have a high FIT (FPBH) and still low criticality.
  5. The Importance ranking chart and the R(τ) / F(τ) and hazard charts each have Download PNG.
  1. Use Assign from prediction on a selected block and pick one prediction Subunit.
  2. Linking renames the block to the Subunit name, adding “(2)” when another block has that name. Relinking renames it again; you can rename it afterwards, and unlinking keeps the name. A block that still carries the automatic name follows a renamed subunit or prediction.
  3. Linked blocks (🔗) are exponential with λ from the prediction (100% duty; the RBD applies the block’s duty cycle). With Auto-update when the prediction changes on they follow BOM and library edits; otherwise out-of-date blocks wait for Refresh / Refresh all.
  4. Open in prediction jumps to the lines; Unlink (freeze value) keeps the current λ.

Save and export

  1. The diagram is saved with the project (File → Save / Open project…, .ILR). File → Export has Diagram (.svg), Diagram (.png), and Results (.csv) (summary, blocks with FIT (FPBH), curve).
  2. The tab’s Export drop-down builds PowerPoint, Excel, PDF (print), and Importance CSV with only the RBD: system result, diagram, importance chart, reliability curve, and the block, structure, and importance tables (the curve data goes to Excel only).

Block data sources

  1. At the bottom of the RBD tab (and on the Bayesian Analysis tab) the Block data sources table lists every block once (identical blocks combined ×N): source (Prediction, Test data, Field data or Manual), linked dataset, and prior / posterior FIT (FPBH). An unlinked block reads “Manual · unlinked”.
  2. Link to… links a block to a prediction scope or to test / field data; Unlink keeps the current value as a manual entry. Both tabs share one state, saved in the .ILR.
  3. A summary strip gives the count and share of blocks and the share of system FIT from each source. Without the Bayesian module, the test / field data and posterior parts are hidden.

Worked example: pump skid

Pump skid at τ = 1,000 h (from new)
PartModelR(1,000 h)FIT (FPBH)
Pumps, 1-of-2Weibull β = 1.8, η = 8,000 h each0.9994523,683 avg over τ each
Controllerλ = 20,000 FIT (FPBH) (MTBF 50,000 h)0.9802020,000
Sensors, 2-of-3Weibull β = 1.2, η = 30,000 h each0.9991716,883 avg over τ each
System Psseries of the three≈ 0.9788

The pumps have the highest FIT (FPBH), but they are redundant. The single-string controller carries criticality ≈ 0.93, so it ranks 1 as the weakest link.

Bayesian Analysis

Validation pending: results in this section have not yet been independently validated. Verify before relying on them.

The Bayesian Analysis tab combines a prior (engineering judgment, a reliability prediction, or earlier data) with new test or field data to give a posterior: parameter estimates with credible intervals, R(t) bands, BX life, and the probability of meeting a requirement. It opens with three EXAMPLE analyses and works signed out. On a wide screen the read-only block diagram sits on the left: blocks that follow a Bayesian posterior carry a B badge and a purple outline, and clicking one opens its analysis and shows its posterior summary.

Pick a model and prior

  1. Weibull (β, η): priors as 95% ranges (lognormal) or bounds (uniform); posterior by MCMC (4 chains, R̂ and effective sample size reported) or exact grid quadrature, in a background worker.
  2. WeiBayes (β known): exact gamma update of θ = η^−β; with a vague prior and zero failures it reproduces the classical 63.2% WeiBayes bound.
  3. Exponential / Poisson: gamma prior on λ; Gamma(a, b) with r failures in time T gives Gamma(a + r, b + T). The prior source defaults to the Reliability Prediction λ (217F / NSWC-11) whenever a prediction exists, the sample prediction included; with none it starts at an MTBF estimate with strength. Saved projects keep their saved prior source.
  4. Binomial: beta prior on the success probability; Beta(a, b) with s successes in n trials gives Beta(a + s, b + n − s).

Read the posterior

Cards show R at the mission time (median and credible interval), P(meet requirement), and BX life; the table and charts compare prior and posterior. Worked check: the EXAMPLE controller’s prior is the prediction (59,400 FIT (FPBH), MTBF 16,800 h) with strength 2, Gamma(2, 33,700 h); two failures in 120,000 h give Gamma(4, 153,700 h), posterior mean MTBF 38,400 h. With an MTBF estimate of 50,000 h as the prior instead: Gamma(2, 100,000 h) → Gamma(4, 220,000 h), MTBF 55,000 h.

Assign the posterior to RBD blocks

Assign to blocks… links the posterior to new or existing blocks (B badge) that follows the posterior. From any RBD block, Update with test/field data (Bayesian) opens the tab with that block’s model as the prior. With Bayes-linked blocks the RBD shows Ps with a 90% credible band from posterior draws.

Step by step: a prior from the prediction, updated with field data

  1. Click Load the example loop at the top of the in-app Tutorial: the Controller prior is the current Reliability Prediction (the EXAMPLE electric motorcycle on first launch: 59,400 FIT (FPBH), MTBF 16,800 h; with no prediction it loads the EXAMPLE controller board), and it sets up all of this in one click.
  2. On Bayesian Analysis, pick “EXAMPLE Controller — field data”, or New analysis… with Model = Exponential / Poisson.
  3. Prior on λ: Prior source = Reliability Prediction λ, Prediction scope = Whole assembly, Strength = 2 equivalent failures (a weak prior, so the data dominate).
  4. Test / field data: 2 failures in 120,000 total operating hours. Requirement: read-only from the RBD Mission panel (set τ = 1,000 h and Requirement R ≥ 0.97 there; every Bayesian analysis follows it live), credible level 0.9.
  5. Read the posterior: R(1,000 h) ≈ 0.976 (90% credible 0.951–0.991); P(meet requirement) (the Controller’s share of R ≥ 0.97, shown as R ≥ 0.971) rises from 26% (prior) to about 66%; posterior median MTBF ≈ 41,900 h. The density chart shows the posterior against the dashed prior; the R(t) chart shows the 90% credible band.
  6. Under “Assign to RBD blocks (Reliability Analysis Loop)”, click Assign to blocks… and tick a block or add a new one. The block gets the B badge and the RBD shows Ps with its credible band.

Monte Carlo Simulation

Validation pending: results in this section have not yet been independently validated. Verify before relying on them.

The Monte Carlo Simulation tab simulates 10,000 to 1,000,000 missions of the live RBD (series, parallel, k-out-of-n, cold standby with switch reliability, prediction- and Bayes-linked blocks) or of a component list (Weibull, exponential, lognormal, normal, fixed p). It reports simulated Ps with a Clopper–Pearson interval next to the analytic value, MTTF, BX, a time-to-failure histogram, empirical R(t) against the analytic curve, and a failure-cause Pareto; Show on RBD highlights the top contributors. Parameter uncertainty samples the Bayesian posteriors each run; an optional spares mode estimates fleet spares. A fixed seed makes results reproducible. Sampling error: SE = √(P(1 − P)/N), about 0.00014 for P = 0.98 and N = 1,000,000. On a wide screen the read-only block diagram sits on the left with the top failure contributors of the last run outlined; click a block for its share of the simulated mission failures.

Step by step: simulate the EXAMPLE pump skid

  1. Load the EXAMPLE loop (Load the example loop at the top of the in-app Tutorial) or build a diagram on the RBD tab. On Monte Carlo Simulation, System defaults to the live RBD.
  2. Set 100,000 runs (the default is 1,000,000) and seed 12345, tick Parameter uncertainty so Bayes-linked blocks draw from their posteriors, and click Run simulation.
  3. Compare simulated Ps ≈ 0.973 (95% CI 0.972–0.974) with the analytic point-model value ≈ 0.973 and the RBD posterior band (≈ 0.973, 90% credible 0.949–0.991).
  4. Check the time-to-failure histogram (dashed line = mission time) and the empirical R(t) against the analytic curve.
  5. The failure-cause Pareto shows the single Controller causing about 95% of mission failures. Show on RBD highlights the top contributors on the diagram.

Close the Loop tab (closed loop)

The Close the Loop tab (last in the top row, after Monte Carlo Simulation) runs one cycle across the Prediction, RBD, Bayesian, and Monte Carlo tabs: the prediction λ is the prior → test and field data give the posterior → linked RBD blocks take it (the Posterior FIT (FPBH) card names the blocks it covers, e.g. Covers RBD blocks: Pump A, Pump B; the FIT (FPBH) is per block, one unit each, not a combined figure) → Monte Carlo gives system reliability with uncertainty → compared with the requirement, which is set only on the RBD Mission panel and shown read-only next to Run full loop (Set in RBD →). Each stage recomputes when its inputs change, and the loop shows Stale until you press Update loop (or turn on Live). Every cycle is stored (timestamp, what changed, prior → posterior, RBD and Monte Carlo results) with a trend chart, saved in the project and in exports. After a run, the Close the Loop tab lists the top contributors and what-ifs (add redundancy, improve a block) with confirm and Undo, and points prediction-linked blocks to Global Change.

Step by step: close the loop on the EXAMPLE pump skid

  1. Click Load the example loop at the top of the in-app Tutorial and confirm. It loads the EXAMPLE pump skid, controller and pump analyses, sets the RBD requirement to R ≥ 0.99 at 1,000 hours, and runs cycle 1.
  2. Cycle 1 (prediction prior 59,400 FIT (FPBH), the current prediction): RBD Ps 0.973 (90% credible 0.949–0.991, P(R ≥ 0.99) = 0.06) and Monte Carlo 0.973 (95% CI 0.9728–0.9734): below the requirement (the whole interval is under 0.99).
  3. Feedback lists the Controller as the top contributor; the top what-if, Add a redundant Controller, lifts Ps from 0.973 to 0.998. Click Apply to RBD… and confirm (Undo is offered).
  4. The loop shows Stale. Press Update loop (or turn on Live): cycle 2 lands at about 0.998 (95% CI 0.9978–0.9980, entirely above 0.99; RBD P(R ≥ 0.99) = 1.00) and meets the requirement.
  5. The history table and the trend chart show both cycles against the requirement line. Undo next to “EXAMPLE loop loaded” restores your previous work.

Fleet mode: math and worked example

Fleet mode answers: if the life model is true, what fraction of units survive the mission or warranty life Tw, and how many failures should N identical units expect? R(Tw) = exp[−(Tw/η)β], F = 1 − R, expected failures N·F, P(≥ 1) = 1 − e−NF (Poisson) or 1 − (1 − F)N (binomial), and B-lives tx = η[−ln(1 − x/100)]1/β. These are point predictions, not confidence statements. The charts add a 95% Fisher-matrix band for the workbook fit only: y = β ln(t/η), Var(y) = ℓ²Vβ + (β/η)²Vη − 2ℓ(β/η)Cβη with ℓ = ln(t/η), Rlo, hi = exp[−ey ± z√Var(y)]. The life model source can be the workbook fit (with the ALT use case, ηuse = ηtest · AF), a Bayesian posterior point model (posterior median β, η or posterior mean λ; no band) or a MIL-HDBK-217 prediction (λ = λpmh/10⁶ × duty, β = 1, η = MTBF).

Worked example (bearing sample, Tw = 400 h, N = 250). 12 failures and 7 suspensions give the MLE β = 2.168257, η = 716.212 h. Then (400/716.212)2.1683 = 0.282795, R(400 h) = 0.753675, F = 0.246325, N·F = 61.58 expected failures, P(≥ 1) ≈ 100%, and B1 / B10 / B50 = 85.8 / 253.7 / 604.8 h. The Fisher covariance (Vβ = 0.2774, Vη = 9297.1, Cβη = −7.535) gives Var(y) = 0.1528 at 400 h, so the 95% band is F 12.32%–45.58% and N·F 30.8–113.9. Switching the source: the EXAMPLE 217 controller board (λ = 2.4154 × 10⁻⁵ /h) gives R(2000 h) = 0.952840 and N·F = 11.79; the EXAMPLE Bayesian exponential controller (prior = the 217 prediction, strength 2; posterior Gamma(3, 142,802), λ̄ = 2.1008 × 10⁻⁵ /h) gives 0.958854 and 10.29; the EXAMPLE Bayesian Weibull bearing (β = 1.9498, η = 740.33 h) gives R(400 h) = 0.740013 and N·F = 65.0. Note: the default first-load workbook is the Arrhenius ALT demo (Ea = 1.30 eV, AF ≈ 3.5 × 10⁸), so F at use at Tw = 2000 h is about 2.5 × 10⁻¹⁹ and shows as 0.000%. That is correct, but it makes a poor teaching example, which is why the worked example uses the bearing sample.

Limitations: units are independent and identical; the Fisher band is first order and drawn only for the workbook fit; Tw has no calendar or duty conversion; Bayesian results are flagged validation pending.

Fleet results flow back into Close the Loop: with Tw and N filled in, the Close the Loop tab shows a 5 · Fleet card and a Fleet results report (N, Tw, life model source, R(Tw), expected failures N·F, P(≥ 1 failure), B-life, and the fleet life model’s R(τ) against the RBD requirement), with feedback such as the η needed to meet the requirement and the spares to plan. It updates live, is stored with each loop cycle, and appears in the exports.

Weibull tools walk-through

The Weibull tools analyze life data and feed the same diagram.

  1. Workbook grid. Time, F/S, quantity, optional STRESS. Open Weibull Workbook and try Load bearing sample (single stress) or the first-load ALT demo (500 K / 523 K / 573 K).
  2. Fit. Default Distribution model is Arrhenius-Weibull, Fit method MLE. Read β and η (or μ, σ). Check the probability plot before you believe a B-life.
  3. Calculator. Weibull Calculator evaluates R(t), F(t), BX% (B10), mean life (MTTF), remaining life, h(t), and AF from that LifeFit. See the B10, MTTF, and hazard guides for the identities.
  4. Fleet. Warranty life and fleet size → expected failures. For the life requirement (years), duty cycle and dormancy use the RBD’s mission profile (below). B1 / B10 / B50 at use stay under Test vs use on the workbook after ALT.
  5. RDT. Reliability Demonstration sizes a success-run (or related) demonstration at a stated life, with optional AF from the life curve. Its Life model source can also take the shape from a Bayesian posterior or the 217 prediction (β = 1).
  6. Export. Every tab has its own Export drop-down at the bottom: PowerPoint, Excel, PDF (print), and CSV with only that tab’s charts and tables. Sign in is optional on the website if you want saved analyses.

Export any tab

Each tab’s Export drop-down (under “Export: <tab>”, for example “Export: Weibull Workbook”) builds PowerPoint, Excel, PDF (print), and CSV files with only that tab’s results: charts as high-resolution images with legends, plus its tables. Plot data goes to Excel and CSV only. Files are named after the tab, the analysis, and the date. Every chart also has Download PNG. Exports work signed out.

What’s new in this version

Weibull identities side by side

Once a two-parameter fit exists, B10, MTTF, and hazard are not three extra opinions. They are three questions asked of the same (β, η). Reliability is R(t) = exp[−(t/η)β]. The time at which a fraction x/100 has failed is tx = η [−ln(1 − x/100)]1/β, so B10 uses x = 10. The mean is MTTF = η Γ(1 + 1/β), which equals η only when β = 1. The hazard among survivors is h(t) = (β/η) (t/η)β−1. Characteristic life η is the time where F = 1 − e−1 ≈ 0.632, for every shape. None of these is a substitute for the others.

A round wear-out model, β = 2 and η = 1 000 h, is short enough to recompute by hand. −ln(0.90) = 0.105360516 and its square root is 0.324593, so B10 = 324.6 h, which rounds to 325 h. Γ(1.5) = √π / 2 = 0.886227, so MTTF = 886.2 h. At 500 h the cumulative hazard is 0.25, R(500) = exp(−0.25) = 0.7788, and h(500) = 2 × 500 / 1 0002 = 0.001 / h. The mean is below η, B10 is far below the mean, and the hazard at 500 h equals the exponential hazard 1/η only because 500 h is η/2. At 1 000 h the same hazard has doubled to 0.002 / h. Full arithmetic, including the bearing-sample fit, is on the Weibull analysis, B10, MTTF, and hazard rate pages.

The bearing sample in the workbook (12 failures, 7 suspensions) fits at β = 2.168257 and η = 716.212 h. The same identities then give B10 = 253.7 h (254 h to the nearest hour), MTTF = 634.3 h, R(500 h) = 0.6321, and h(500 h) = 0.001989 / h. Load that sample and open Weibull Calculator if you want the workbook to recompute them. The point of the four guides is the algebra; the point of the calculator is to evaluate your own LifeFit without retyping the gamma function.

Worked example (bearing sample)

Load bearing sample, Recalculate. You should see about β = 2.17, η = 716 h (wear-out). Then:

Same fit the calculator will report
AskWhereResult
B10Calculator → BX% = 10≈ 254 h
MTTFCalculator → Mean life≈ 634 h
R(500 h)Calculator → R(t)≈ 0.632
h(500 h)Calculator → Failure rate≈ 0.00199 / h

Step-by-step Weibull identities for this dataset are on Weibull analysis. Rank / recommend distribution (device type, physics of failure, close-call Hybrid weights) lives on the in-app Tutorial so it can read the live grid.

FAQ

Is this the in-app tutorial?

This URL is a crawlable guide. The interactive tutorial — math callouts, Rank models, and the walkthrough video — still lives in the app at the Tutorial tab (#tutorial). Use this page to learn the flow; use the tab when you want the live workbook beside the text.

What should I enter first?

Times greater than zero, F or S, and quantity on the Weibull Workbook grid. Start with Load bearing sample for a single-stress endurance set, or the ALT demo if you need temperature cells. Then Recalculate and read β and η.

Where are B10, MTTF, and hazard?

After a fit, open Weibull Calculator. BX% life, mean life, and failure rate h(t) all read the current LifeFit. Fleet expected failures and RDT sample size are separate tabs — they are not hidden on the probability plot.

Does Iron Lake Reliability only plot PDF and CDF curves?

No. The product is a reliability workbook: censored life data, 2P Weibull / lognormal / exponential, Arrhenius and inverse-power ALT, RDT, fleet risk, and export. F(t) is computed from a fit, not sketched as a distribution overlay.

Open the Iron Lake Reliability workbook Open Weibull Calculator