Reliability — MTTF, MTBF, exponential failure model: Visual walkthrough

This is the whole story of the exponential failure model told in pictures. We start from nothing — no formulas assumed — and build, one picture at a time, until the famous curve $R (t) = e^{- λ t}$ falls out on its own. Then we squeeze the average lifetime (MTTF) out of that same curve.

If you have never seen an integral, a derivative, or the letter $e$ before — good. We earn every one of them here.

Step 1 — What we are even measuring: the survival fraction

WHAT. Imagine we build 1000 identical components and switch them all on at once. As time passes, they die one by one. At any moment we count how many are still alive and divide by the 1000 we started with. That fraction is our reliability $R (t)$ .

WHY. Probability is slippery for a single part — it either works or it doesn't. But a fraction of a big crowd is a real, drawable, measurable number. This is the honest picture behind the word "probability of survival."

PICTURE. In the figure, blue dots are alive, red dots are dead. Early on almost all are blue; later, more turn red. The height of the blue curve at time $t$ is $R (t)$ .

Figure — Reliability — MTTF, MTBF, exponential failure model

Step 2 — The one assumption: a constant death rate $λ$

WHAT. We now make one modelling choice. In each tiny slice of time $d t$ , the same fraction of whoever is currently alive will die. Call that fraction-per-unit-time $λ$ (Greek letter "lambda"). So in a slice $d t$ , the number of deaths is $λ \times (those alive) \times d t$ .

WHY this assumption and not another. Real parts have three life stages (see Bathtub Curve): early "infant" failures, a long flat middle, and old-age wear-out. During the flat middle the death rate genuinely is roughly constant — a working chip is "as good as new" no matter how long it has already run. That memoryless middle is exactly where a single number $λ$ works, and it is where spacecraft spend their mission.

WHY a rate, not just a count. A raw death count depends on how big the crowd is. A rate — deaths per survivor per hour — is the crowd-independent fingerprint of the part itself.

PICTURE. Same-size bites are taken out of the shrinking blue bar each second. Because the bar shrinks, each equal-fraction bite removes fewer actual dots than the last — the seed of the slowing decay.

Step 3 — Turning the picture into a rule of change

WHAT. Let's write "how fast the survival fraction drops." The change in $R$ over a tiny time $d t$ is written $\frac{d R}{d t}$ — read it as "the slope of the blue curve." From Step 2, the drop equals $λ$ times whoever is currently alive, which is $R$ itself:

$how fast R falls \frac{d R}{d t} = - death rate λ who’s left to die R$

WHY the derivative tool. We don't just want how many have died — we want the instantaneous slope of the curve, because the whole behaviour is about how the drop keeps changing. "The slope right now" is precisely what the symbol $\frac{d R}{d t}$ means: rise over run as the run shrinks to zero.

WHY the minus sign. $R$ goes down, so its slope is negative. The minus sign carries that fact; with it, $λ$ stays a clean positive number.

PICTURE. At each point on the blue curve a small orange tangent arrow shows the slope. Where the curve is high, the arrow is steep (lots alive → lots dying). Where the curve is low, the arrow is shallow (few alive → few dying). The slope is proportional to the height. That single sentence is the equation.

Step 4 — Solving it: separate, then add up the slices

WHAT. We rearrange so all the $R$ -stuff sits on one side and all the $t$ -stuff on the other. This is called separating variables:

$tiny % drop in R \frac{d R}{R} = - λ tiny time step d t$

Then we add up every tiny slice from the start ( $t = 0$ , $R = 1$ ) to now ( $t$ , $R (t)$ ). "Add up infinitely many tiny slices" is exactly what the integral sign $\int$ means:

$\int_{1}^{R (t)} \frac{d R}{R} = - λ \int_{0}^{t} d t$

WHY separate first. We can only integrate one variable at a time. Putting each variable on its own side lets us total up each side independently — like sorting laundry before folding.

WHY the logarithm appears. The left side asks: "what function, when I take its slope, gives $\frac{1}{R}$ ?" The answer is the natural logarithm $ln$ . So $ln$ isn't pulled from a hat — it is forced on us as the antidote to dividing by $R$ . The right side is easy: adding up a constant $λ$ over time gives $λ t$ .

$left side total ln R (t) - ln 1 = right side total - λ t$

Since $ln 1 = 0$ (raising anything to the zero power gives 1), this is just $ln R (t) = - λ t$ .

PICTURE. The shaded area under the flat $- λ$ line from $0$ to $t$ is a rectangle of height $λ$ and width $t$ — its area is $λ t$ . The integral is literally that grey area.

Step 5 — Undo the log: meet $e^{- λ t}$

WHAT. We have $ln R (t) = - λ t$ . To free $R$ we apply the exact opposite of $ln$ : raising the number $e$ to a power. Because $e^{l n x} = x$ (they cancel — that's what "opposite" means):

$R (t) = e^{- λ t}$

WHY the number $e$ . $e \approx 2.718$ is the one base whose own growth rate equals its own value — the unique fixed point of "slope proportional to height." That is precisely the property Step 3 demanded, so $e$ is not a convenience; it is the only base that fits.

PICTURE. Here is the payoff curve. It starts at $R (0) = 1$ , dives steeply while many are alive, then flattens as the crowd thins — the exact behaviour we sketched in Step 3, now exact.

Step 6 — Edge and degenerate cases (never leave the reader stranded)

WHAT & WHY. A model is only trustworthy if it behaves sanely at its extremes. We test all four corners.

Input	Formula gives	Meaning — does it make sense?
$t = 0$	$R = e^{0} = 1$	Everyone alive at launch. ✅
$t \to \infty$	$R = e^{- \infty} = 0$	Given forever, everything eventually dies. ✅
$λ = 0$	$R = e^{0} = 1$ for all $t$	A part that cannot fail lives forever. ✅
$t = 1/ λ$	$R = e^{- 1} \approx 0.37$	At the "average life," only 37% still alive — not half!

PICTURE. The curve is shown with all four checkpoints marked. Stare at the green marker at $t = 1/ λ$ : the height there is $0.37$ , not $0.5$ . This is the single most common misread of the whole topic.

Step 7 — Squeeze out the average lifetime (MTTF)

WHAT. The Mean Time To Failure is the average of all the individual death-times. Averaging a continuous spread means: weight each possible death-time $t$ by how likely a death is at that instant, then add up. That weight is $f (t)$ , the failure density, which is just the downward slope of $R$ :

$f (t) = - \frac{d R}{d t} = λ e^{- λ t}$

$MTTF = \int_{0}^{\infty} a death time t its likelihood λ e^{- λ t} d t = \frac{1}{λ}$

WHY this equals the area under $R (t)$ . There is a beautiful shortcut: for anything that starts at 1 and only falls, the average lifetime equals the total area under the survival curve. So MTTF is simply the area under $e^{- λ t}$ , which is $1/ λ$ .

PICTURE. The shaded area under the survival curve is a "filled-in" exponential; its area equals the width of the equal-area rectangle of height 1 — that width is exactly $1/ λ$ .

The one-picture summary

The whole journey on one canvas: crowd of parts → constant death fraction $λ$ → slope = $- λ R$ → separate & integrate → $ln$ appears → undo with $e$ → $R (t) = e^{- λ t}$ → area underneath = MTTF $= 1/ λ$ .

Recall Feynman retelling — say it back in plain words

Line up a thousand identical gadgets and turn them on. Every second, the same slice of whoever's still running quits — that slice-per-second is $λ$ . Because the crowd shrinks, the actual number quitting shrinks too, so the survival curve nosedives at first and then eases off. Writing "the drop equals $λ$ times who's left" and adding up all the tiny drops forces a logarithm to show up; undoing the log with the number $e$ hands you $R (t) = e^{- λ t}$ . Test the corners: full at launch, zero after forever, and — the sneaky one — only $37%$ left at the "average" lifetime, not half. Finally, the average lifetime itself is just the area sitting under that survival curve, which comes out to a clean $1/ λ$ . That's the entire model, no memorising required.

Recall Quick self-check

Why does a constant failure rate still give a slowing death count? ::: Because the constant is a fraction of survivors; as the survivor pool shrinks, the same fraction removes fewer and fewer actual parts. What forces the logarithm into the derivation? ::: Integrating $\frac{d R}{R}$ — the log is the function whose slope is $1/ R$ , so it's the only antidote. What is $R (t)$ at $t = 1/ λ$ ? ::: $e^{- 1} \approx 0.37$ , so 63% have already failed by the mean lifetime. What is MTTF geometrically? ::: The total area under the survival curve $R (t) = e^{- λ t}$ , equal to $1/ λ$ .

Prerequisites & neighbours: Probability and Statistics Fundamentals · Poisson Process · Weibull Distribution · Bathtub Curve · Fault Tree Analysis · Availability vs. Reliability · Mission Design Constraints · Safety-Critical Systems