ROBOTECA FREE SAMPLE
Dashboard

A free lesson from Engineering Foundations (Maths, Physics & Chemistry): the whole module, nothing cut short.

LESSON · Algebra & Functions

Exponential growth, decay and the time constant

Turn 1 45 min LESSON

ALearning Material

The functions of the earlier lessons in this topic put the unknown in the base: , , a polynomial. An exponential puts the variable in the ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍exponent instead:

Moving the variable into the exponent changes the behaviour completely. In , moving from to doubles the input and quadruples the output. In , moving from to doubles the input and multiplies the output by 1024. A polynomial grows by a fixed factor when you scale the input; an exponential grows by a fixed factor when you add to the input. This distinction is why an exponential eventually outgrows every polynomial, whatever its degree.

The physical statement behind the notation is that an exponential is what results when the rate of change is proportional to the amount present. Money earning interest grows in proportion to how much is there. A hot object loses heat in proportion to how far above ambient it is. A capacitor discharges through a resistor at a current proportional to the voltage still on it, which is proportional to the charge still on it. Radioactive decay, population growth, the settling of every first-order system in engineering: all follow the same rule, and the exponential is its solution.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍The form used in engineering is written around a characteristic time instead of as :

Here is the value at , is the time constant in the same units as , and is the natural base. The reason for instead of 2 or 10 belongs to calculus and is given there. In brief, is the base that makes the rate of change equal to the amount present with no extra constant, so every formula in this lesson stays simple.

The time constant is the most useful number in the subject, and it can be measured with a stopwatch: is the time taken for the gap to shrink to ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍, about 36.8%, of what it was. Substituting gives , and this holds from any starting point, not only from . An exponential has no memory of when the clock started. Wait another and 36.8% of that remains, which is 13.5% of the original, and so on.

The following table is used constantly and is better memorised than derived each time:

Elapsed Fraction remaining Fraction complete
36.8% 63.2%
13.5% 86.5%
5.0% 95.0%
1.8% 98.2%
0.7% 99.3%

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍The table gives a practical rule: a first-order system is settled to within 1% after about five time constants, and nothing useful happens after that. When a data sheet quotes a settling time, or a control engineer waits before taking a reading, or a filter's response is described as slow, five is the number being used.

Most real systems take a third form, because most quantities do not decay to zero but approach a final value. A cooling object approaches ambient; a charging capacitor approaches the supply; a heated tank approaches its setpoint. The rule is that the gap decays exponentially, not the quantity itself:

Read it as: the final value, plus the initial gap shrinking away. This one expression covers charging and discharging, heating and cooling, and using it correctly is mostly a matter of identifying and .

A thermocouple with a time constant of 4.0 s sits at 20 degrees C and is plunged into a bath at 80 degrees C. After 6.0 s it reads:

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍After 6 s, more than a full time constant, the sensor still reads 13 degrees low. This is the sensor-lag problem that makes a fast control loop chase a stale measurement, and it is why the response time of a thermowell is an important specification. Reading within 1 degree of the bath requires , which needs , and 1/60 is between (0.0183) and (0.0067), so about s. The five-time-constant rule gives 20 s without that arithmetic.

Growth works the same way with the sign flipped, and the natural companion number is the doubling time. If , doubling needs , so . The reverse question, the half-life of a decay, is the same arithmetic: ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍. So half-life and time constant differ by a factor of about 0.7, and confusing them gives a 44% error that looks plausible.

A shortcut for growth is accurate enough for mental arithmetic. For a growth rate of percent per period, the doubling time is approximately periods. At 7% per year, doubling takes about 10 years; at 2%, about 35. It comes from , which is why appears in finance and in reliability work alike.

Exponential growth is the most over-applied model in engineering. Nothing grows exponentially for long. The model holds while the resource being consumed is effectively unlimited, and every real growth eventually meets a constraint: the population runs out of food, the market saturates, the amplifier hits its rails, the buffer fills. An exponential fitted to early data and extrapolated gives confident predictions with large errors. An exponential growth model should be used over a stated interval, with the constraint that will end the growth named.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Two terms appear in every data sheet. A decade is a factor of 10 and an octave is a factor of 2, and both are ways of counting multiplications rather than additions. Saying a signal falls by 20 dB per decade, or that a resistor range spans three decades, is counting on an exponential scale. That scale is the subject of a Turn 2 lesson in this course, and the logarithm that makes it computable is the next lesson here.

Formulas & method.

  • The defining property. An exponential is the solution to "the rate of change is proportional to the amount present". Adding to the input multiplies the output.
  • Decay to zero. . Growth. .
  • Approach to a final value (the one most systems take): . The gap decays, not the quantity.
  • The time constant. is the time for the gap to fall to of whatever it was, ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍from any starting point.
  • The settling table. 63.2% / 86.5% / 95.0% / 98.2% / 99.3% complete at 1 to 5 . Five time constants is settled, for engineering purposes.
  • Half-life and doubling time. . They are not the same as , and confusing them is a 44% error.
  • Rule of 70. At percent per period, doubling takes about periods.
  • An RC circuit. , in seconds when is in ohms and in farads.
  • The limit of the model. Nothing grows exponentially for long. State the interval and name the constraint that will end it.

Where this shows up.

  • Robotics. A motor's current rises with the time constant and its speed settles with a mechanical time constant, and the ratio of the two decides whether a current loop can be tuned independently of a speed loop. Battery discharge, sensor filtering and the decay of a complementary filter's memory are all first-order.
  • ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Automation. Every thermowell, every level transmitter with damping, and every analogue filter in an instrument is first order, and the sensor lag above is why a loop cannot be tuned faster than its measurement. Process gains and dead times are quoted alongside for this reason, and a step test to find is the standard field identification.
  • Electronics. is used constantly: debounce timing, filter corner frequencies, power-on reset delays and coupling capacitors. A capacitor is 99% charged after five time constants, which is how a supply sequencing delay is chosen.
  • Digital design. A net's rise time is set by its RC, and the delay through a stage is quoted at the 50% crossing, which is , the same number as the half-life. Leakage, retention and the discharge of a dynamic node are all exponential decays.

Why it exists. A very large class of physical systems obeys one sentence: the rate at which something changes is proportional to how much of it there is, or to how far it is from where it is heading. Exponentials are the functions that satisfy that sentence, so the exponential is the necessary shape of first-order behaviour. With the notation and the time constant, a large family of systems (thermal, electrical, mechanical, financial) can be described, measured and compared with a single number.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Mental model. An exponential approach is a gap that always loses the same fraction of itself in the same time. It never quite arrives, because a fraction of something is never nothing, but it approaches so quickly that after five time constants the remainder is below anything that can usually be measured. Growth is the same picture run the other way: the same fraction added each period, so the increments get larger while the proportion stays fixed.

Common misunderstandings.

  • "The time constant is how long it takes to finish." It is how long it takes to get 63.2% of the way, and the process is not done until about five of them. Quoting as a settling time understates it by a factor of five, which in a commissioning schedule or a filter specification is the difference between a working system and a chronically early reading.
  • "Half-life and time constant are the same thing." , so they differ by about 30%, and using one where the other belongs is a 44% error in the opposite direction. Neither is wrong, since they answer different questions, and the data sheet states which one is quoted.
  • ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍"This is growing exponentially, so it will keep doing so." Exponential growth is always a description of a period during which the limiting resource was not yet binding. Every real exponential turns over. Extrapolating one past the end of its data produces a confident and very large error.

Connections. The base and exponent notation is from the powers lesson in the numbers topic, and is that notation with the variable moved into the exponent. Getting out of the exponent requires the logarithm, which is the next lesson and is the exact inverse of this one. Rearranging to choose a capacitor was a worked example in the rearranging lesson. The claim that the rate of change is proportional to the amount is a differential equation, which the calculus and differential-equations topics make precise, and the same reappears as the pole of a first-order system in the control lessons.

B‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Immediate Active Recall

QUERY

State the physical sentence that an exponential is the answer to.

REVEAL
ANSWER

The rate of change is proportional to the amount present (or to the gap remaining). Interest on a balance, heat loss proportional to the temperature difference, a capacitor discharging at a current proportional to its remaining voltage: one sentence, one family of solutions.

Did you recall it?
QUERY

Give the exact meaning of the time constant, including the part that is often omitted.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍REVEAL
ANSWER

is the time for the gap to fall to of what it was, measured from any starting point, not only from . An exponential has no memory of when the clock started, which is why the same describes every stretch of the curve.

Did you recall it?
QUERY

Give the settling table and the engineering rule that comes out of it.

REVEAL
ANSWER

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍63.2% / 86.5% / 95.0% / 98.2% / 99.3% complete at 1, 2, 3, 4 and 5 time constants. The rule: five time constants is settled to within 1%, and nothing useful happens after that.

Did you recall it?
QUERY

Write the form for a quantity approaching a final value, and say what actually decays.

REVEAL
ANSWER

. The gap decays exponentially, not the quantity. Read it as "the final value, plus the initial gap shrinking away", and it covers charging and discharging, heating and cooling, in one line.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Did you recall it?
QUERY

How do half-life and time constant differ, and how large is the error of confusing them?

REVEAL
ANSWER

, so the half-life is about 30% shorter. Using one for the other is roughly a 44% error in the other direction, and it looks entirely plausible because both are times of the right order.

Did you recall it?
‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍QUERY

A thermocouple with s goes from 20 to an 80 degree bath. What does it read at 6.0 s, and what does that mean for a control loop?

REVEAL
ANSWER

degC, still 13 degrees cold after one and a half time constants. A loop tuned faster than the sensor is chasing a stale measurement, which is why a thermowell's response time is an important specification.

Did you recall it?

CConceptual Questions

Answer in your own words first, then compare with the model answer.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍PROMPT

A manager sees three months of 8% monthly growth in a metric and projects it two years forward. Explain what is wrong, without claiming the growth is not real.

REVEAL MODEL ANSWER
MODEL ANSWER

The growth may be real and the arithmetic of the projection flawless, and the projection may still be worthless. At 8% per month, the rule of 70 gives a doubling roughly every nine months, so two years is a bit under three doublings and the metric would grow about eightfold. Nothing in the data refutes that. What is wrong is the assumption hidden in the extrapolation: that the mechanism producing the growth will continue to operate unchanged. Exponential growth happens when a system is consuming a resource that is, for now, effectively unlimited, and every real system eventually meets the boundary of that resource. A market has a finite number of customers; a process has a maximum throughput; a server has a bandwidth limit; a population has food. The exponential is not the long-run shape, it is the early part of an S-shaped curve whose upper half has not happened yet, and three months of data is almost always entirely within that early part. The projection is therefore a statement of what would happen if no constraint existed, not an estimate of the future, and its value depends completely on how near the constraint actually is. The constructive response is to make the extrapolation testable: name the constraint that will end the growth, estimate where it sits, and say when the curve is expected to bend. If the limit is a million customers and the current figure is ten thousand, the exponential has plenty of room and two years may be defensible. If the limit is a hundred thousand, the growth ends inside the projection window and the eightfold figure is wrong. This turns a projection that cannot be refuted into one that can be checked.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Compared to the model answer - did you get it?
PROMPT

Two first-order stages are put in series: a sensor with a 2 s time constant feeding a filter with a 2 s time constant. A colleague says the combination has a 4 s time constant. Assess that.

REVEAL MODEL ANSWER
MODEL ANSWER

The claim is wrong in a specific way. Time constants do not add, because the combination is no longer a first-order system at all. One first-order stage responds to a step with the familiar ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍, which starts moving immediately at its steepest and then eases off. Two in series produce a response that starts flat, because at the instant of the step the first stage's output has not yet changed, so the second stage has nothing to respond to. The curve then accelerates, reaches an inflection, and only afterwards settles like an exponential. That S-shape is the signature of second-order behaviour, and no single time constant describes it. What can be said quantitatively is more useful than "4 s". The 10% to 90% rise time of one stage with is about s; for two identical stages in series it is roughly s, shorter than the 8.8 s that naive addition would suggest, but much longer than the 4.4 s of a single stage. The settling time behaves similarly: the tail is still governed by the slower pole, so five time constants of the dominant stage remains a reasonable rule, but the start of the response is delayed by an amount that behaves very like dead time. And that delay is the reason the colleague's error matters. A control loop tolerates a large time constant far better than it tolerates dead time, because a lag it can see is a lag it can compensate, while an interval in which nothing happens gives the controller no information and directly eats phase margin. Cascading two filters to "get more smoothing" therefore gains smoothing at the cost of something a loop handles much worse, and the right question is how much apparent dead time the second stage has added, not what the combined time constant is.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Compared to the model answer - did you get it?
PROMPT

Why is the base used for engineering exponentials when powers of 2 and 10 are easier to think about, and does the choice change any physical prediction?

REVEAL MODEL ANSWER
MODEL ANSWER

The choice changes no physical prediction. Any exponential can be written in any base: ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍. The curve is the same curve, and a measurement cannot tell you which base someone wrote it in. What changes is only which constants are clean and which are cluttered. The reason wins in engineering is that it is the base for which the rate of change of the function equals the function itself, with no leftover multiplier. Write a decay in base 2 and its rate of change carries a factor of ; write it in base 10 and the factor is . Since the physical law that generates every exponential in this lesson is a statement about the rate of change being proportional to the amount, the base that makes that proportionality constant exactly one is the base in which the law and its solution have the same shape. Everything downstream inherits the cleanliness: the time constant is simply the reciprocal of the rate constant, the pole of a first-order system is , and no stray 0.693 propagates through a derivation. Where another base is more convenient, it is used. Half-life in base 2 is more natural for anything measured by halvings, which is why nuclear and pharmacological work quotes it. Decades in base 10 are more natural for frequency response and for anything plotted on a log axis, which is why filter roll-off is quoted in dB per decade. The practical skill is converting between them, and the conversion is always a factor of or sitting where the units change. Choosing a base selects the arithmetic, not the physics.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Compared to the model answer - did you get it?

DPractice Problems

P1 (easy). A capacitor charges through a resistor with ms from 0 V towards 5.00 V. Give the voltage at (a) 50 ms, (b) 100 ms, (c) 150 ms; (d) state the time at which it is within 1% of 5.00 V, using the rule rather than a calculation.

P2 (medium). A tank of water at 85 degrees C cools towards a room at 21 degrees C. After 12 minutes it reads 62 degrees C. (a) Find the time constant. (b) Predict the temperature after 40 minutes. (c) Find how long until it reaches 30 degrees C. (d) State one assumption the model makes that a real tank breaks, and which direction the error goes.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍P3 (harder). An RC low-pass filter is placed in front of a 4-20 mA input to reject noise. The process has a step change that must be detected within 2.0 s. (a) If the filter's is 1.5 s, what fraction of the step has appeared at 2.0 s, and is that enough to trip a threshold set at 80% of the step? (b) Find the largest for which 80% of the step appears within 2.0 s. (c) The noise is at 50 Hz. Explain, using the time constant alone, roughly how much a filter of that attenuates it, and state the trade you have just made explicit.

Solutionsclick to reveal

P1. Use , the approach form with and ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍. (a) At one , 63.2% complete: . (b) At two , 86.5%: . (c) At three , 95.0%: . (d) Within 1% is five time constants, so . Parts (a) to (c) use the table instead of the exponential because the table is the tool used in practice, and it is exact to the precision quoted.

P1Compared to this solution - did you get it right?

P2. (a) The gap form: . At , , so . Taking the natural log of both sides, which is the next lesson's tool used here for the first time, , so . (b) . (c) , so , giving and ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍. Sanity check against the table: 30 degrees is of the gap remaining, which is just under the 13.5% at two time constants, so a little under min. That agrees, and doing the check costs nothing. (d) The model assumes the cooling rate is proportional to the temperature difference and nothing else, which is Newton's law of cooling. A real tank breaks it in at least two ways: radiation grows as the fourth power of absolute temperature, so it contributes more at 85 degrees than the linear model allows, and evaporation removes heat at a rate that falls off sharply as the surface cools. Both mechanisms are stronger early, so the real tank cools faster than the model at the start and then converges to it. The practical consequence is that a fitted from an early measurement is too short and under-predicts the later temperatures; fitting from two late points gives a better estimate of the long tail.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍P2Compared to this solution - did you get it right?

P3. (a) The fraction of a step that has appeared after time is . At and : , so 73.6%. That is not enough to trip a threshold at 80%, so the detection would be late. How late is worth computing: 80% needs s, so the trip comes 0.41 s after the deadline. (b) Set , so and ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍, giving . Any filter slower than that misses the deadline. (c) A first-order filter passes signals whose period is long compared with and attenuates those whose period is short. Its corner is at Hz. Fifty hertz is times above the corner, and a first-order filter falls at a factor of 10 in amplitude per factor of 10 in frequency, so 390 times above the corner is an attenuation of about 390 times, roughly 52 dB, leaving under 0.3% of the interference. The trade now stated explicitly: this filter is enormously more than is needed for 50 Hz, and the binding constraint is the 2.0 s detection deadline, not the noise. A of even 0.1 s would still attenuate 50 Hz by a factor of about 31 and would detect the step in 0.16 s. So the right design is to make as small as the noise rejection requires and keep the detection margin, rather than choosing a slow filter and discovering the deadline problem in commissioning. A first-order filter's speed and its rejection are set by one number, so one cannot be improved without giving up the other.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍P3Compared to this solution - did you get it right?

EFeynman Exercise

Explain to somebody who understands powers but has never met a time constant: what makes exponential behaviour different from everything else in this topic, what the time constant physically means, and why a system is called settled after five of them. Write your explanation, then compare.

REVEAL MODEL ANSWER
MODEL ANSWER

The earlier lessons put the unknown at the bottom: squared, cubed, raised to a fixed power. An exponential puts it in the exponent, and that change produces a different kind of behaviour. With a square, multiplying the input multiplies the output: doubling the input quadruples the output. With an exponential, ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍adding to the input multiplies the output. Every extra second, or every extra year, multiplies by the same factor again, which is why an exponential eventually overtakes any power, and why compounding seems slow at first and then very fast. Exponentials appear throughout engineering because they solve one common physical rule: the rate at which something changes is proportional to how much of it there is. Money earns interest in proportion to the balance. A hot cup loses heat in proportion to how much hotter it is than the room. A charged capacitor pushes out current in proportion to the voltage still on it. All of these give the same shape, because the rule requires it. The time constant sets the scale. It is the time for the remaining gap, the distance between the current value and the final value, to shrink to about 37% of what it was, measured from any starting moment. Starting a stopwatch at any point and waiting one time constant leaves 37% of the gap that existed at the start. The curve has no memory of when the measurement began, so one number describes every stretch of it and characterises the whole response. The five-time-constant rule follows from this. After one time constant, 37% of the gap remains. After two, 37% of that, about 13%. After three, 5%. After four, under 2%. After five, under 1%. Five has no special property: it is where the remainder falls below what can usually be measured, so it is used as the working definition of finished. The time constant is therefore not the time the process takes. Quoted as a settling time, it understates the answer by a factor of five, an error that appears in filter specifications and commissioning schedules as a reading taken before the measured quantity has arrived.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Compared to the model answer - did you get it?

FError Analysis Framework

The mistake: treating the time constant as the settling time, and applying a decay formula to the quantity when it is the gap that decays.

Why it happens: is the number that appears in the formula and on the data sheet, so it is the number that reaches memory, and "the time constant of this loop is 2 seconds" sounds like a complete statement about how long it takes. The second error comes from the simple form being taught first, so it is the one that is reached for. It is correct only when the final value is zero, which is true of a discharging capacitor and false of a charging one, false of a cooling object approaching room temperature, and false of almost every process quantity.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍How to recognize it: a predicted final value that is zero when the physics says the system approaches something else, which is the immediate signature of the wrong form; a reading taken at the moment a timer expires that is consistently about a third short of where it should be, which is one time constant used as five; a settling time quoted in a specification that the measured system misses by a factor of about five; a fitted that changes depending on which part of the curve was used, which means the process is not first order; a temperature or level that approaches something other than zero being modelled with a formula that has no in it.

How to correct it: write the approach form every time, even when happens to be zero, so the general case is the habit and the special case is the exception. Identify and explicitly before substituting anything. Use the settling table rather than the exponential for anything at a whole number of time constants, since it is faster and removes an arithmetic step. Quote settling times as in any specification, and when reading somebody else's specification check which of the two they meant. And where a fitted drifts between early and late data, stop and ask whether the system is really first order before trusting either value.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍GMini Challenge

A 24 V supply charges a 470 uF capacitor through a 10 kohm resistor into a circuit that must not be enabled until the rail is above 21.6 V, which is 90% of final. (a) Find the delay. (b) The designer wants to halve the delay and proposes halving the capacitor. Evaluate that, and give one alternative that does not change the capacitor. (c) Discharge is through a separate 1 Mohm bleeder when the supply is removed. Find the time for the rail to fall below 5 V, the safe-touch threshold, and say whether that is acceptable for a panel that is opened for maintenance.

REVEAL MODEL ANSWER
MODEL ANSWER

(a) The charging rail is with s. Reaching 90% needs ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍, so . This lies between two and three time constants, consistent with the table's 86.5% and 95%, which is a free sanity check. (b) Halving the capacitor to 235 uF halves to 2.35 s and therefore halves the delay to 5.4 s, so the proposal works arithmetically. Whether it is acceptable depends on what the capacitor is for, which is the substance of this part. If it is there as a timing element only, halving it is fine. If it is a bulk decoupling or hold-up capacitor, which a 470 uF part on a 24 V rail very likely is, then halving it halves the stored energy and halves the time the rail survives a supply dropout, which is a functional change disguised as a timing tweak. Two alternatives leave it alone. Halve the resistor to 4.7 kohm, which halves identically and costs only a doubling of the steady-state current through it, from 2.4 mA to 4.8 mA, which on a 24 V rail is trivial. Or, better, stop using an RC as a threshold detector at all and use a supply-supervisor part with a specified threshold and delay, since the RC approach makes the delay depend on the resistor tolerance, the capacitor tolerance, which is often plus or minus 20% on an electrolytic, and the comparator's threshold, so the real delay might be anywhere from about 7 to 15 s. (c) Discharge through the bleeder is a decay to zero, so with s, which is 7.8 minutes. Falling below 5 V needs ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍, so . That is not acceptable for a panel opened for maintenance without further measures. The common standards for enclosed equipment expect stored energy to fall to a safe level within a stated time after isolation, typically 1 s for an accessible point and 5 minutes for one behind a tool-secured cover, and 12 minutes exceeds both. The fixes are straightforward: reduce the bleeder to about 100 kohm, giving s and a fall below 5 V in 74 s, at a cost of 0.24 mA of continuous standby current; or add an active discharge triggered by loss of the supply; or, at minimum, fit a warning label stating the discharge time and keep the cover tool-secured. The same appears in both halves of this problem with very different values, because the resistors differ by a factor of a hundred, and the safety requirement constrains the slow time constant as firmly as the functional requirement constrains the fast one.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Compared to the model answer - did you get it?

Quiz Check

Answer without looking back.

QUIZAuto-graded check · feeds your mastery score
  1. An exponential is the solution to which physical statement?

  2. After one time constant, the fraction of the change that has been completed is:

  3. A quantity approaching a final value is modelled by:

  4. ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍The half-life of a decay relates to its time constant by:

  5. A metric has grown 8% a month for three months. Extrapolating exponentially for two years is unsound because:

This is a free sample

Progress and the spaced-repetition reviews are part of the course. The full track continues from here.