Why this book matters
Almost everything else on the Reliability Shelf assumes the machine already exists and asks how to keep it running. Practical Reliability Engineering asks the earlier question: how do you engineer reliability into a product or system before it ever reaches your plant - and how do you measure, test and manage it once it does? O'Connor wrote the first edition in 1981; the fifth edition (2012, with Andre Kleyner) is the version that sits behind the ASQ Certified Reliability Engineer exam and most university reliability courses.
For you, the value is the bridge it builds. On one side: statistics - Weibull plots, confidence limits, censored data. On the other: physics of failure - fatigue, wear, corrosion, overstress. Most practitioners live on one side and wave at the other. This book forces you to hold both, which is exactly what separates an engineer who fixes the pump from one who explains why the "identical" pump next door never fails.
Core idea
A part fails when the load applied to it exceeds its strength - and both load and strength are distributions, not numbers. Everything else in the book (derating, safety margins, burn-in, HALT, variation control) is a different way of managing the overlap between those two curves.
The one picture: load-strength interference
If you keep a single mental model from this book, keep this one. Every unit of a component has a slightly different strength (material scatter, manufacturing variation). Every duty cycle applies a slightly different load (pressure spikes, starts, temperature, torque). Failures live where the two distributions overlap.
This one picture quietly explains half the reliability vocabulary you use every day:
- Derating - deliberately running a component below its rating pushes the strength curve far to the right of the load curve. The 15 kW motor on an 11 kW duty is not oversized, it is derated.
- Safety margin - the distance between the means is meaningless on its own; what matters is the distance between the tails. Two designs with the same average margin can have wildly different failure rates if one has more variation.
- Burn-in and screening - weak units are the left-hand tail of the strength distribution. Stress screening removes them before they reach the field.
- Site-to-site mystery failures - the "same pump" that fails every year at one site and runs ten years at another is not the same pump. It sees a different load distribution: water hammer, cavitation, more starts, dirtier product, hotter ambient. Same strength curve, different overlap.
Statistics without self-deception
The statistical half of the book covers the tools you actually need on plant data: the exponential, Weibull and lognormal distributions, how to handle censored data, and how wide your confidence limits really are.
Three warnings from these chapters will save you from most bad analyses:
- Most of your CMMS data is censored. The compressors that have not failed yet are data too - suspensions. Ignore them and every life estimate you produce is pessimistic garbage. Any Weibull fit on plant data must handle suspensions properly (see our Weibull Handbook summary for the field method).
- Small samples mean wide confidence bands. Five failures do not define a wear-out life. The book is blunt about this: quote the interval, not just the point estimate, before you move a shutdown date on the strength of it.
- Models describe, they do not explain. This is the book's most repeated warning. A beautiful Weibull fit with a shape parameter of 3 tells you the failures look like wear-out; it does not tell you what is wearing or why. The distribution is a summary of ignorance until you name the physical mechanism - fatigue, erosion, insulation breakdown, seal face wear.
Field tip
Never present a failure statistic without a mechanism hypothesis next to it. "Beta = 3.2, suspected impeller erosion, confirmed by the last two teardowns" is engineering. "Beta = 3.2" alone is numerology - and O'Connor & Kleyner spend a whole book teaching you the difference.
The bathtub curve, treated honestly
Most training courses draw the bathtub curve as if every machine lives it. The book treats it for what it is: a population-level composite, not the biography of your gearbox.
The classic bathtub is three superimposed processes: decreasing infant mortality (manufacturing defects and installation errors weeding themselves out), a roughly constant random region (external overstress, process upsets, maintenance-induced damage), and an increasing wear-out region (fatigue, wear, corrosion catching up). Add them and you get the bathtub shape - for a large population of identical items.
For complex repairable systems - a crusher, a compressor train, a packaging line - the useful-life region dominates in practice: many failure modes, each with its own clock, superimpose into something close to a constant failure rate for the system as a whole. That is exactly what the aviation data behind Nowlan & Heap's six failure patterns showed, and what Moubray's RCM II turned into maintenance logic: age is a poor predictor for most failure modes, so blanket time-based overhauls mostly miss. O'Connor & Kleyner give you the statistical reasons why; the RCM literature gives you the maintenance consequences.
The Design for Reliability toolbox
The heart of the book is a toolbox for building reliability in, instead of inspecting it in later. Every tool attacks the load-strength picture from a different angle.
| Tool | What it does | Where you feel it on the plant floor |
|---|---|---|
| Design FMECA | Hunt failure modes on the drawing board, when a fix costs a pencil stroke instead of a recall. | The seal flush plan that was corrected before the pump was ever built - the failure you never saw. |
| Derating | Run components well below rated stress to push strength away from load. | The conveyor drive specified with real margin survives the starts and jams the catalogue rating would not. |
| Redundancy | Accept that items fail; make the function survive anyway - where consequences justify the cost. | Duty/standby pumps, dual instrument air compressors. |
| Robust design (Taguchi) | Make performance insensitive to variation instead of fighting variation itself. | A belt-tracking design that tolerates splice and loading variation instead of demanding perfect alignment forever. |
| HALT | Step stress (thermal, vibration, combined) far beyond spec until things break - each break is a weakness found in days instead of years. | The VFD that survives a dusty 45 °C electrical room, because its weak capacitor died in the chamber, not on your wall. |
| ALT | Accelerated life testing: quantified life estimates from elevated-stress testing plus a physics-based acceleration model. | Bearing grease life curves; the basis of many OEM interval claims - ask for the test behind the number. |
| HASS | Production screening with stresses derived from HALT, to catch weak units before shipment. | Fewer dead-on-arrival cards and infant-mortality failures in the first months of an installation. |
| Reliability growth | Track MTBF through test-fix-test cycles (the Duane / Crow-AMSAA idea: improvement follows a measurable trend you can manage against). | Commissioning of a new crusher line: are the fixes actually bending the failure-rate curve, or just moving problems around? |
The last chapters make a point most engineers skip: reliability is a management discipline as much as a technical one. Reliability programs, specifications with teeth, and the cost of unreliability treated as a line item. The argument is simple and quantitative: reliability activities are investments with measurable returns - the money you do not spend on unplanned downtime, secondary damage and warranty claims. If you cannot show that return, you will lose the budget argument to whoever can.
Using it in 2026
The 5th edition is from 2012. The logic has aged well; the toolbox around it has exploded. Here is the honest split.
| Keep as-is | Update with modern practice |
|---|---|
| Load-strength thinking. Still the best single mental model in the discipline. | Measure the load, stop assuming it. Cheap vibration, pressure and current sensors give you the real load spectrum of that pump or conveyor - the distribution O'Connor could only estimate, you can now plot from data. |
| FMECA at design time. Failure modes are still cheapest to kill on paper. | AI-assisted preparation. Pre-draft the FMECA from OEM manuals, standards libraries and CMMS history, then let the team correct instead of type. Days become hours; the engineering judgment stays human. |
| The HALT philosophy. Test to break, not to pass - a weakness found in the chamber is a failure you never chase at 3 a.m. | Digital twins and ML in operation. Derating and margin studies can now run on a model of your actual system, and ML anomaly detection extends the "find weakness early" idea into the operating fleet. |
| Distrust of naked statistics. Models describe, mechanisms explain. | Apply the same distrust to ML. A model that predicts failures without a physical story is the new "beta = 3.2" - useful for triage, dangerous as a conclusion. The 1981 warning transfers to 2026 unchanged. |
Bottom line
This is the reference book of the discipline, not a bedside read. Work through the load-strength and statistics chapters properly once, then keep it within reach as the place where the CRE-level answers live. Combine its rigour with today's data access and you are dangerous - in the good way.
References & further reading
This summary is original explanatory writing. All concepts belong to their authors - go to the sources.
- O'Connor, P.D.T. & Kleyner, A. Practical Reliability Engineering, 5th edition. Wiley, 2012. ISBN 978-0-470-97981-5. Publisher page (Wiley)
- ASQ. Certified Reliability Engineer (CRE) - the certification body of knowledge this book maps to. ASQ CRE page
- Nowlan, F.S. & Heap, H.F. Reliability-Centered Maintenance. United Airlines / U.S. Department of Defense, 1978. Report AD-A066579 - the age-reliability data behind the six failure patterns. Our summary · DTIC record (free)
- NIST/SEMATECH. e-Handbook of Statistical Methods - free companion for the distributions, censored-data and reliability-growth material. Online handbook (free)
- Abernethy, R.B. The New Weibull Handbook - the practitioner's deep dive on life-data analysis. Our summary
Disclaimer. This page is an independent educational summary written entirely in Rob Reliability's own words. It is not affiliated with, sponsored by or endorsed by Patrick D. T. O'Connor's estate, Andre Kleyner, or John Wiley & Sons. No text from the original book is reproduced; all diagrams are our own original illustrations of engineering concepts that are part of the public technical literature. Book titles, trade names and trademarks remain the property of their respective owners and are used solely to identify the work being discussed. If you are a rights holder and have any concern about this page, contact us at hello@robreliability.com and we will address it promptly.
Want this rigour on your fleet?
We bring O'Connor-grade reliability engineering to your fleet: real failure data, real strategies, CMMS-ready.