Class 13 · Sensors and Actuators

A sensor answers a question you have to ask precisely. An actuator does something to the world, and the world includes people.

Asking the right question#

The hardest part of sensing is not the electronics. It is turning a production requirement into a measurable quantity.

"Is the actor in position" is not measurable. What is measurable: is there something within 30 cm of this point, has this pressure mat been stood on, has this beam been broken, has this magnet come within 5 mm of this reed switch.

Each of those answers a slightly different question and each fails differently. The pressure mat also triggers for a dropped prop. The beam also triggers for a cable swinging through it. The magnet only works if the actor is wearing the right shoe.

Asking the right question#

Choosing a sensor is choosing which failures you can live with. State them, out loud, before you buy anything.

The sensor families#

Switches and contacts#

The simplest and the most reliable: a mechanical contact that opens or closes.

• Limit switch. Mechanical, robust, definite. Used at the ends of travel of anything that moves.
• Reed switch. A magnet closes a sealed contact. No wear, works through non-magnetic material, ideal for a door or a lid.
• Pressure mat. Large area, cheap, and triggers for anything heavy.
• Micro switch. Small, precise operating point, used where repeatability matters.

Switches and contacts#

The wiring convention that matters: wire safety-critical switches so that the safe condition is a closed circuit, and the unsafe condition is open. This is normally-closed logic, and its virtue is that a cut cable, a corroded contact or a pulled connector produces the unsafe reading, so the system stops.

Wired the other way, a broken cable reads as "safe" forever, and the guard that is supposed to stop the machine has silently stopped existing. This is the single most important sentence in the class, and Class 15 builds on it.

Analogue sensors and dividers#

Analogue sensors and dividers#

Many sensors are variable resistances: a thermistor, a light-dependent resistor, a potentiometer, a force-sensing resistor. On its own a resistance is not readable by an ADC, so you put it in a divider with a fixed resistor and read the voltage at the junction.

V_out = V_in × R2 ÷ (R1 + R2)

Two design decisions:

Which arm the sensor goes in determines whether the reading rises or falls with the quantity. Both work; pick one and label it, because a sensor whose reading falls as light increases will confuse everybody including you in three months.

The fixed resistor should be about the geometric mean of the sensor's range. If an LDR runs 1 kΩ bright to 100 kΩ dark, use about 10 kΩ, which puts the useful part of the curve in the middle of the ADC's range instead of squashed at one end.

And from Class 11: keep the source impedance the ADC sees under about 10 kΩ, or the reading is low and you will chase it for an hour.

Position and motion#

• Potentiometer. Absolute position, cheap, wears out, fine for a knob and poor for anything that moves constantly.
• Rotary encoder. Two out-of-phase signals; counting the edges gives direction and distance. Incremental, so it does not know where it is at power-up and must be homed.
• Hall effect sensor. Detects a magnetic field. No contact, no wear, works through a panel.
• Time-of-flight / ultrasonic. Distance without contact. Ultrasonic is cheap and fooled by soft surfaces and by angles. Time-of-flight laser is more precise and fooled by black and by glass.

Position and motion#

Incremental against absolute is the distinction that catches people. An encoder that has counted 4,000 pulses knows it has moved 4,000 pulses, not where it is. Lose power and the count is gone. Every incremental system needs a homing routine and a limit switch, and a machine that has not been homed since power-up must not be allowed to move at speed.

Current and presence#

• Current transformer. Clips around a live conductor and reports the current without any electrical connection. This is how you know whether a lamp is actually drawing power, which is a far better test of "is the practical on" than trusting that you sent the cue.
• Optical beam break. Emitter and receiver; something crossing between them breaks the beam. Fast, reliable, and the standard way to detect a person entering a hazard zone.

Wire it before you wire it#

Raw counts first, then the map, then the averaging. Notice that the LDR's useful range is a fraction of the ADC's range, which is true of nearly every real sensor and is the thing that makes a naive map() feel unresponsive at both ends.

Actuators, and the responsibility that comes with them#

Actuators, and the responsibility that comes with them#

The three that need warnings#

Solenoids overheat. A solenoid is a coil designed for a short pull, and many are rated for a duty cycle rather than continuous use — twenty-five per cent, say. Hold one on continuously and it will get hot enough to burn, and then it will fail. If a solenoid must hold, either pick one rated for continuous duty, or drive it hard to pull in and then drop to a lower PWM duty to hold. That technique is standard and it is worth knowing by name: peak and hold.

Steppers lose steps silently. A stepper commanded to move 200 steps believes it moved 200 steps. If the load jammed at step 40, it still believes it. Nothing in the stepper reports the error. Every stepper system in a show needs a home switch, and something that verifies position at least once per cycle, or it will drift over a run and be discovered on the night it matters.

Anything that moves can trap someone. This is where the class stops being about electronics. A motor with enough torque to move scenery has enough torque to break a finger and it does not know the difference. That is Class 15, and it is why nothing in this class is allowed to be built with mains-powered motion.

The three that need warnings#

The chain, end to end#

The chain, end to end#

Sensor to decision to actuator. Every link adds delay, and the total is what the audience experiences as responsiveness.

The chain, end to end#

Driving a motor in both directions#

A MOSFET switches a motor on and off in one direction. Reversing it needs four switches arranged in an H, which is why the part is called an H-bridge and why you buy it as a module rather than building it.

Close the top-left and bottom-right and current flows one way; close the other diagonal and it flows the other. That is the whole idea, and everything else is about the two ways it goes wrong.

Shoot-through is closing both switches on the same side at once, which is a short circuit across the supply through two transistors. A real driver chip includes a dead time — a few hundred nanoseconds where both are off during the changeover — and that is a large part of what you are paying for. It is also why you do not build one from four MOSFETs on breadboard.

Braking against coasting. Open all four switches and the motor coasts, still turning, driven by its own momentum. Close both bottom switches and the motor is shorted to itself: its own generated voltage drives a current that opposes the rotation, and it stops sharply. A scenic piece coasting to a stop and one braking to a stop are visibly different, and which you want is a design decision, not a default.

Driving a motor in both directions#

And the motor is a generator. Decelerating scenery pushes current back into the driver. On a small load the driver absorbs it; on a large one it raises the supply rail until something fails. That is why serious motion control has a braking resistor, and it is the reason a scenic automation supplier's schematic has a component on it you did not expect.

Measuring force and position properly#

Measuring force and position properly#

Two sensors worth knowing because they answer questions a switch cannot.

A load cell is a metal element with strain gauges bonded to it, wired as a bridge. Deforming the metal changes the resistances by a fraction of a per cent, and the bridge turns that into a differential voltage of a few millivolts. It needs a dedicated amplifier — an HX711 or similar — because a few millivolts on top of a couple of volts of common mode is not something an ordinary ADC can read.

What it buys you in a theatre: knowing what a flown piece actually weighs, whether a counterweight set is balanced, and whether the thing that should be resting on the deck is resting on the deck. That last one is a genuinely good interlock, because it measures the condition itself rather than a proxy for it.

An inductive proximity sensor detects metal without contact and without a magnet, at a few millimetres, and is unbothered by dust, paint and stage haze. Where a reed switch can be defeated with a magnet and an optical sensor can be defeated with haze, a proximity sensor mostly cannot, which is why they appear on machinery rather than on props.

Both give you the same thing: a measurement of the physical condition rather than a proxy for it, which is the difference between "the cue was sent" and "the thing actually moved".

At the bench#

Two hours.

Block A — Build three dividers (30 min)#

An LDR, a thermistor and a potentiometer, each in a divider, each read by the ADC and printed.

For each: calculate the expected voltage at both extremes before building, choose the fixed resistor by the geometric mean rule, and compare your calculation to the measurement. Then swap the sensor to the other arm and confirm the reading inverts.

Block B — Normally closed, proved (25 min)#

Wire a limit switch normally open and a second normally closed, both read by the microcontroller, both printed.

Then cut each cable in turn, with the system running, and record what each reads. Write one sentence about which arrangement you would use on a guard and why.

This is a two minute experiment with a permanent result, and it is the single most examinable idea in the class.

Block C — The encoder that does not know where it is (25 min)#

Read a rotary encoder and count. Turn it fifty clicks. Note the count.

Now power-cycle the board without moving the encoder. The count is zero and the encoder is still in the same place, and the system now believes something false.

Write a homing routine using a limit switch. Prove it works from three different starting positions.

Block D — Peak and hold (20 min)#

Drive a solenoid at full power continuously. Measure the current, and measure its temperature every thirty seconds with the IR thermometer for four minutes. Plot it.

Now drive it at full power for 100 ms and then at forty per cent PWM to hold. Measure the current and repeat the temperature log. Confirm it still holds.

Two curves on one axis: that is your evidence, and it is the argument you will make to a designer who wants a latch held closed for a whole act.

Block E — Measure your own chain (20 min)#

Sensor to actuator, with the scope on both ends. Trigger on the sensor, measure to the actuator's first movement. Record the total.

Then insert delay(250) somewhere plausible in the loop and measure again. Then remove it and reduce the debounce from 50 ms to 5 ms and measure again, and count the false triggers over twenty presses.

That last measurement is the trade-off in this whole class, in numbers you produced.

Numbers from this class#