Quick estimates for how things are built
What it means
Build it stronger than the maximum expected load by a multiplier.
Typical factors
Buildings: ×2–3. Bridges: ×4. Aircraft: ×1.5 (weight-critical). Pressure vessels: ×4. Elevators: ×10.
Why not just ×1?
Materials vary, loads are uncertain, manufacturing isn't perfect, and failures are catastrophic.
An elevator cable rated for 1,000 kg is actually strong enough to hold 10,000 kg. This is why buildings don't collapse from one extra person, why bridges don't fail when a lorry hits a pothole, and why planes don't break apart in turbulence. The safety factor quietly saves millions of lives daily.
In practice
80% of bugs come from 20% of the code. 80% of heat loss from 20% of the building envelope. 80% of failures from 20% of components.
Action
Find the 20% first. Fixing it gives 4× more return than spreading effort equally.
Also true:
The last 20% of performance improvement costs 80% of the effort. "Good enough" is often the right engineering answer.
Vilfredo Pareto noticed in 1896 that 80% of land in Italy was owned by 20% of the population — and the ratio kept appearing in unrelated systems. Engineers use it to decide what to fix first, what to optimise, and when to stop. A bridge optimised for the last 5% of strength costs twice as much for 5% gain.
What it is
A quick estimate using round numbers that gets you within 10× of the real answer, done in your head in minutes.
Classic example
"How many piano tuners in Chicago?" Fermi: population ÷ pianos per family × tuning frequency ÷ hours per tuner = ~125. Actual: ~150.
Engineering use
Before spending a week calculating: "Is this even in the right ballpark?" If your estimate says it would need 10× more power than available, don't detail-design it.
Enrico Fermi famously estimated the yield of the first nuclear bomb test by dropping scraps of paper and watching how far they moved in the blast wave — getting within a factor of 2 of the measured result. Engineers use this constantly: the first answer should come in minutes, not days. Only if it looks promising do you do the full calculation.
What it means
Nothing is made perfectly. A "10 mm" rod is actually 9.95–10.05 mm. The ±0.05 mm is the tolerance.
Clearance vs interference fit
Clearance: shaft is smaller than hole (moves freely — bearings). Interference: shaft is larger (press-fit — never comes apart). Transition: could be either.
Why it matters
A jet engine has thousands of parts with tolerances of ±0.005 mm. Stack tolerances wrong and parts don't fit, or jam, or vibrate to destruction.
Tight tolerances cost money — exponentially. A part machined to ±1 mm costs €5. The same part to ±0.1 mm costs €50. To ±0.01 mm costs €500. The engineering discipline is: use the loosest tolerance that still works. Over-specifying tolerances is one of the most common and expensive rookie engineering mistakes.
The rule
A design error caught at the drawing stage costs €1 to fix. At prototype: €10. In production: €100. After delivery: €1,000+.
Why
As a project progresses, more decisions have been built on top of the error. Fixing the foundation means rebuilding the house.
Applies to software too
A bug found in code review: 1 hour to fix. Found in testing: 1 day. Found in production: 1 week + reputation cost.
This is why engineering reviews, design audits, prototypes, and testing exist. The money spent catching errors early is always cheaper than correcting them later. The Boeing 737 MAX disasters were partly attributed to software changes late in the design process that weren't subjected to the same scrutiny as the original design.
What it means
Have one more than you need. If you need 1 pump, install 2. If you need 2 engines, use 4.
Examples
Commercial aircraft: 2+ engines, 3+ hydraulic systems, 4+ flight computers. Nuclear plants: multiple coolant loops. Data centres: dual power, UPS, generators.
The maths
If each component fails with probability p, two independent components both fail with probability p². If p=0.01, two units: 0.0001 (100× safer).
A single-engine aircraft failing means a forced landing. A twin-engine commercial plane losing one engine continues to destination safely. This is why critical infrastructure is never designed to exactly minimum requirements — always with capacity to survive failures. RAID storage, dual power supplies, backup generators, emergency brakes — all redundancy.
Feedback loop
Output affects input
A thermostat senses temperature (output) and adjusts the heater (input). Without feedback, systems overshoot or drift. Negative feedback = stability (thermostat). Positive feedback = runaway (microphone squeal, population explosion, compound interest).
Precision vs accuracy
Repeatable ≠ correct
Accurate = close to truth. Precise = consistent, even if consistently wrong. A scale that always reads 2 kg too heavy is precise but not accurate. A scale that gives random readings near the true value is accurate but not precise. You need both.
Failure mode analysis
"What can go wrong?"
FMEA (Failure Mode and Effects Analysis): list every possible failure, how likely it is, how bad the consequence is, and how detectable it is. Product of these three = risk priority number. Fix the high scores first.
The weakest link
System = its weakest part
Strengthening the strongest part of a system achieves nothing. A chain with one weak link fails at that link regardless of how strong the other links are. Always find and address the bottleneck first.
Diminishing returns
Each improvement costs more
Getting from 0% to 90% efficiency is often easier than 90% to 99%. The last 1% of fuel efficiency in a car might cost more to achieve than the first 30%. Decide in advance what "good enough" is.
First principles thinking
Build from fundamentals
Instead of "we've always done it this way," ask: what are the physical constraints? What does physics actually allow? Elon Musk famously used this to challenge rocket costs: what are the raw materials? Why does the assembled rocket cost 100× more? This mindset drives engineering breakthroughs.
Tacoma Narrows Bridge (1940)
Resonance ignored
The bridge oscillated at its natural frequency in a 64 km/h wind and collapsed after 4 hours. Lesson: all structures have resonant frequencies. Match external forces to them and they amplify to destruction. Changed bridge design permanently — aerodynamic testing became mandatory.
Challenger Space Shuttle (1986)
Known failure mode, ignored
O-ring seals known to fail at low temperatures. Launch temperature: -2°C. Engineers objected; management overrode. 73 seconds after launch: disaster. Lesson: safety culture matters as much as engineering. Known risks must be escalated, not suppressed.
Millennium Bridge, London (2000)
Positive feedback oscillation
Opened June 2000; wobbled so severely it closed within 2 days. Pedestrians involuntarily walked in sync with the bridge's natural lateral frequency, amplifying the oscillation. Lesson: human interaction with structures creates unexpected feedback loops. Cost £5M to fix.
Mars Climate Orbiter (1999)
Unit conversion error
One team used imperial units (pound-force seconds), another metric (newton-seconds) for thruster data. The $328M spacecraft entered the atmosphere and burned up. Lesson: unit consistency is safety-critical. Always check units explicitly at system interfaces.
1. A floor is designed to hold 500 kg/m² with a safety factor of 3. What is the maximum safe load, and why isn't the safety factor just 1?
2. Estimate (back-of-envelope): how much steel is in a typical 10-storey office building?
3. Why does a plane with 4 engines not need to be 4× as powerful as a plane with 2 engines to carry the same load?
4. A part is specified as "25.00 ± 0.05 mm." What does this mean, and what happens if you tighten the tolerance to ±0.005 mm?
5. A system has three components in series, each with 99% reliability (1% failure rate). What is the system reliability?