Enclosures, Thermal, and Mechanical Design
The enclosure is part of the circuit. It changes the antenna, the temperature, and what the user can break.
🎯 By the end you'll understand
Why a box is an electrical component: how plastic near a PCB antenna shifts its resonance, how a steel mounting surface eats your link budget, and how a LiPo pouch under the antenna is the worst placement decision available. What an IP rating actually certifies, why IP54 is a specific pair of tests rather than a vibe, and why a ventilation slot and an IP rating are mutually exclusive unless you buy a membrane. How to do the thermal arithmetic that decides whether a sealed box turns your humidity sensor into a self-heating thermometer — for Beacon the answer is "not during normal operation, and catastrophically during an over-the-air update." How to mount a board, leave room for a battery to swell, and design a button boss that does not hold the button down. And what Beacon's printed shell cost, in rupiah and in decibels.
KEY INSIGHTThe box is a component you did not put on the BOM
Everything the enclosure does to your product, it does invisibly. It detunes the antenna without a warning light, it delays the humidity reading without an error flag, and it holds the button down without telling anyone. There is no DRC for mechanical design.
One of the twenty-five pilot units went flat on day four. Its battery was a 1000 mAh cell that should have lasted, by the canon's own projection, something on the order of three years. On the bench it was fine: charge it, and it ran. The cause turned out to be 0.3 millimetres of plastic. The enclosure's button boss — the little pillar that transmits a press from the outside surface down to the tactile switch on the PCB — was long enough to hold the switch closed when the two shells were clipped together. Beacon's firmware treats a held button at wake as a request to enter the over-the-air update path, which brings up Wi-Fi. So the unit woke once a minute, brought up Wi-Fi, scanned for ten seconds, found nothing, and slept — averaging about seventeen milliamps instead of forty microamps, and emptying the cell in under three days.
No schematic review catches that. No PCB design rule check catches it. It is a mechanical interference of a third of a millimetre that turns into a firmware state that turns into a dead battery, and it is entirely typical of what enclosures do. The lesson is not "be careful with bosses." It is that the enclosure participates in the electrical, thermal, and RF behaviour of the product, and it deserves the same treatment the schematic got: predictions written down, tolerances stated, and a review with someone who will argue with you.
Four couplings turn a box into a component. RF: dielectric near the antenna shifts its resonance and metal near it destroys the pattern. Sensing: a sealed volume means your sensor measures the box's air, not the room's, with a time constant set by how the air exchanges. Thermal: whatever the electronics dissipate has to leave through the enclosure, and the temperature rise is that power times a thermal resistance you did not calculate. Mechanical: tolerances stack, plastic creeps, batteries swell, and every one of those becomes an electrical event. Design the box against those four, and the aesthetics will still be fine.
COMPARISONOptions, and the cost curve that decides between them
Four routes are realistic for a small hardware project, and the choice is almost entirely a function of volume. The table below uses indicative Indonesian prices for a Beacon-sized shell — roughly 45 × 35 × 14 mm, two parts, four heat-set inserts.
| Route | Tooling / setup | Per unit at 25 | Per unit at 2,500 | What you actually get |
|---|---|---|---|---|
| FDM 3D print (PLA or ABS) | None — a CAD file | 32,000 | 32,000 | Any geometry, tomorrow. Layer lines, anisotropic strength, no real sealing surface without post-processing, and the cost never falls because printing does not scale. |
| SLA / resin print | None | 55,000 | 55,000 | Crisp features and a genuine sealing surface. Most resins are brittle and degrade in UV; not a product finish, but excellent for a pilot that needs a gasket groove. |
| Off-the-shelf ABS box, machined | A drilling fixture | 55,000 | 48,000 | Proper injection-moulded plastic and an existing gasket design, at the price of a box that is the wrong size and has a logo on it. Machining every unit is labour that does not go away. |
| Injection-moulded ABS | ≈ IDR 87 M for a simple two-cavity tool | 3,484,500 | 39,300 | The real answer at volume: a few thousand rupiah per part, snap fits, textured surfaces, moulded-in bosses. The tool is a five-to-eight-week lead time and a design freeze. |
| Machined aluminium | Fixturing | 210,000 | 165,000 | Beautiful, rugged, thermally excellent — and an RF disaster for a device whose entire purpose is to transmit. |
The injection-moulding row is the one worth staring at. At 25 units the amortised tool costs three and a half million rupiah per unit, which is more than twenty times the whole electronic bill of materials. At 2,500 units it is thirty-nine thousand, competitive with printing. At 25,000 it is under eight thousand and nothing else is close. That crossover is not a detail; it is the shape of hardware economics, and Lesson 6.8 generalises it to every non-recurring cost in the project. The practical consequence for a pilot is simple and slightly unromantic: print it. A printed shell is the correct engineering decision at 25 units even though it is the wrong one at 2,500, and pretending otherwise costs a tool that will be obsolete by the time Rev B lands.
The trap in printing is that FDM lets you draw geometry a mould cannot make: undercuts, uniform-thickness-violating ribs, zero draft, snap features that only work because the layer lines happen to flex the right way. If you expect to mould eventually, keep the discipline early — draft angles of a degree or two on vertical walls, wall thickness kept roughly uniform (thick sections sink and warp when moulded), no undercuts that would need a side action, and ribs kept to about 60 % of the wall thickness they attach to. None of that costs anything while printing, and all of it saves a redesign when the tool is quoted.
DEFINITIONIP ratings: two digits, two specific tests
The IP code, defined in IEC 60529, is two digits and each one is a distinct test performed on a real sample. It is a measurement, not a design intent, and "IP54 enclosure" on a datasheet means somebody put a sample in a dust chamber and under a spray rig and it passed.
| Digit | Meaning for Beacon's IP54 | How it is tested, roughly |
|---|---|---|
| First digit: 5 | Dust-protected. Not dust-tight. Ingress of dust is not prevented entirely, but not enough enters to interfere with operation or safety | The sample sits in a talc-dust chamber, usually with the interior held at reduced pressure to draw dust in, then is opened and inspected |
| Second digit: 4 | Splashing water. Water splashed against the enclosure from any direction has no harmful effect | An oscillating spray tube or a hand-held spray nozzle wets the sample from all angles for a specified time |
Two things follow that people get wrong constantly. First, the digits are not cumulative guarantees of everything below them — IPX4 says nothing about immersion, and a device rated IP54 is not "nearly waterproof," it is splash-resistant and will die in a bucket. Second, and more relevant to Beacon: a ventilation slot and an IP rating do not coexist. The moment you cut a slot so the humidity sensor can breathe, you have a path for both talc and spray, and the enclosure is unrated. Requirement R9 lists IP54 as optional, which was a sensible hedge at the time and turns out to be the requirement doing real work, because the sensing requirement and the ingress requirement point in opposite directions.
An expanded-PTFE membrane vent is a small disc of microporous fluoropolymer, adhesive-backed or press-fitted into a hole. Its pores pass water vapour and air but not liquid water, because liquid water's surface tension cannot force it through pores of that size below a specified water-entry pressure. Manufacturers publish two numbers that matter: the airflow (which sets how fast your enclosure's humidity equilibrates) and the water-entry pressure (which sets what ingress rating the vent will support). A well-chosen vent lets a sealed product keep an IP rating while still measuring the outside air — at a cost of ten to fifteen thousand rupiah per unit at pilot quantities, and a hole you must not cover with the mounting adhesive. It is the only honest way to have both, and it is why Beacon's second enclosure revision has one.
HOW IT WORKSThe enclosure is part of the antenna
The ESP32-C3-MINI-1's antenna is a printed structure on the module's own PCB, tuned by the manufacturer with a specific environment assumed: the module at the edge of a host board, the antenna region overhanging the host's ground plane, and a keep-out on every layer. Lesson 5.7 placed Beacon's module to satisfy that, and the module's regulatory grant depends on it. What Stage 5 could not address is what happens after the board goes in a box.
Three mechanisms, in descending order of how much they will hurt you.
Metal near the antenna: not a degradation, an off switch
A conductive surface a few millimetres from a 2.4 GHz antenna reflects and shorts the near field. An aluminium enclosure does not reduce your range; it removes your product's function, and no amount of tuning recovers it. The same applies to a steel mounting surface, a magnet mount on a machine tool, or a metal bracket that happens to run past the antenna end. If the product must be metal, the antenna has to be outside it — an external antenna, a plastic window, or a separate radome. There is no third option, and discovering this after tooling is a very expensive week.
Plastic near the antenna: dielectric loading, a shift not a block
Common enclosure plastics have a relative permittivity of roughly 2.5 to 3.2 — ABS, PLA, and polycarbonate all sit in that band. Material in the antenna's near field raises the effective permittivity the antenna sees, which lowers its resonant frequency and worsens the match at the frequency you actually use. The result is a loss of a decibel or two, not a catastrophe, and it is proportional to how close the plastic is and how thick it is. Keeping a couple of millimetres of air between the antenna and the wall costs nothing in a design and buys most of the difference back.
The battery under the antenna: the mistake that looks tidy
A LiPo pouch cell is a stack of foil electrodes in an aluminised laminate pouch. It is, electrically, a sheet of metal. Tucking it under the module because that is where the space is puts a conductor directly in the antenna's near field, and it costs several decibels — enough to turn a comfortable link into a marginal one. The cell goes at the far end of the enclosure, always, and if the layout makes that awkward, the layout changes. On Beacon this constraint is why the JST-PH connector is at the opposite edge from the module.
Putting numbers on it helps. Bluetooth Low Energy on the 1 Msym/s PHY has a receiver sensitivity in the region of −95 dBm for a decent implementation, and Beacon transmits at 0 dBm, giving a nominal 95 dB of budget. Free-space loss at 2.44 GHz over one metre is about 40 dB, and indoors the path loss grows faster than free space — a path-loss exponent near 3 is a reasonable planning figure for a cluttered building. Subtract about 4 dB for one plaster wall and hold back 10 dB for fading, and roughly 81 dB of usable budget remains, which the log-distance model turns into something over twenty metres. That is why requirement R2's "10 m through one plaster wall" looked so comfortable on paper. Now subtract 6 dB for a battery under the antenna and 8 dB for a steel mounting surface, and the same model gives about seven metres. The requirement fails, and nothing about the electronics changed.
SEE ITTrade the enclosure off — and watch four requirements move together
Every control below is a mechanical decision. Every readout is an electrical or sensing consequence. Try to satisfy requirement R2 (ten metres through a plaster wall), requirement R3 (humidity to ±3 %), requirement R9's optional IP54, and a sane cost, all at once — and notice that you cannot, without spending money on exactly one specific part.
Four experiments. Choose aluminium, because it is beautiful and rugged, and watch the estimated range collapse to a couple of metres — the product no longer does its job at any price. Move the battery under the module to save space and lose six decibels for nothing. Seal the box properly for IP54 with no vent, and watch the humidity time constant go to ninety minutes, which means the tag reports the room's humidity from an hour and a half ago. Then fit the membrane vent: the lag falls to about twelve minutes, the ingress claim survives, and the cost rises by twelve thousand rupiah — the only line on this page that buys two requirements at once.
CORE CONCEPTHeat, and the sensor that has to live in it
Thermal design for a device like Beacon is one equation and one honest surprise. The equation is that a steady dissipation P raises the internal temperature above ambient by P × Rth, where the thermal resistance from the inside of the enclosure to the surrounding air is, for a small sealed plastic box under natural convection, somewhere in the region of forty to sixty degrees per watt. That number is an estimate you should measure rather than trust, but its order of magnitude is what makes the arithmetic interesting.
The surprise: normal operation is thermally invisible
Beacon's duty cycle is brutal in the right direction. The active window is about 45 ms at roughly 40 mA average, once a minute, and deep sleep draws about 8 µA by the canon's budget. At 3.3 V, the active window delivers 0.045 s × 0.040 A × 3.3 V ≈ 5.9 mJ per cycle. Spread over 60 seconds that is about 0.10 mW. Sleep adds roughly 8 µA × 3.7 V ≈ 0.03 mW. Total: around 0.13 mW. Multiply by 55 °C/W and the self-heating is seven thousandths of a degree — three orders of magnitude below the SHT40's own accuracy. Every intuition about "sealed boxes cook their sensors" is wrong here, and it is worth knowing why: because the duty cycle is 0.075 %, and duty cycle beats everything.
The catastrophe: the one mode that is not low-duty
There is exactly one state in which Beacon is not a low-duty device, and it is the over-the-air update path from Lesson 4.9. Wi-Fi association, download, and verification run the radio and the core continuously, drawing on the order of 100 mA at 3.3 V — about 330 mW, more than two thousand times the average. Run that for six minutes in a sealed printed shell with a thermal time constant near nine minutes, and the internal air rises by P × Rth × (1 − e−t/τ) ≈ 0.33 × 55 × 0.49 ≈ 9 °C.
Nine degrees does not sound alarming until you remember what relative humidity is. RH is the ratio of the water vapour actually present to the amount the air could hold at that temperature, and the saturation vapour pressure climbs steeply with temperature. Heating a fixed amount of water vapour by 9 °C from 28 °C multiplies the saturation pressure by about 1.66, so the measured relative humidity falls to about 60 % of its true value. A workshop at 61 %RH is reported by the tag as 37 %RH, and it stays wrong for tens of minutes while the box cools. Nothing is broken. The sensor is accurate, the firmware is correct, and the reading is nonsense, because the sensor is measuring the air in the box and the box just ran a heater.
You cannot cool the box, and you should not stop doing over-the-air updates. What you can do is stop reporting numbers you know are wrong: suppress or flag humidity and temperature reports for a defined recovery window after any Wi-Fi session, and set a bit in the advertising payload's flags byte so the gateway knows why the data is missing. That converts an invisible data-quality problem into an explicit, diagnosable one — which is exactly the argument Lesson 6.7 makes about what belongs in eight bits of telemetry. It also means that when someone later asks "why did unit 09 report 37 % on a humid morning?", the answer is in the record rather than in a hypothesis.
Sensor placement, and the other heat source
Where the sensor sits inside the box matters more than the total power. Three placement rules earn their keep. Put the SHT40 as far from the module and the regulator as the board allows, and on Beacon it is at the opposite corner. Put the vent directly over it, so the air the vent exchanges is the air the sensor reads rather than air that has to circulate through the whole box. And if the board layout permits, cut a slot in the PCB partly around the sensor — a routed slot interrupts the copper and the FR-4 that would otherwise conduct heat from the module straight into the sensor's pads, and it is the standard technique for exactly this problem.
The last heat source is the one you do not control: whatever the tag is stuck to. A Beacon on a lathe's motor housing is measuring the motor, not the room, and its humidity reading will be correspondingly wrong. That is not a design defect — it is a deployment question, and the honest answer is a mounting instruction and, ideally, an entry in the fleet dashboard that flags units whose temperature diverges persistently from their neighbours'.
IN PRACTICEMounting, swelling, buttons, and light
Holding the board
Four standoffs and four M2 screws into heat-set brass inserts is the boring, correct answer for a printed shell. Screwing directly into printed plastic works twice and then strips; heat-set inserts, pressed in with a soldering iron at around 200 °C, give a metal thread that survives service. Do not clamp the board at more points than you can keep coplanar — a board held at four corners on a slightly warped shell is a board under permanent bending stress, and bending stress is what cracks the multilayer ceramic capacitors that Lesson 6.2 warned about, weeks later, in the field.
The battery pocket
A LiPo pouch cell is not a rigid component and must never be treated as one. Three rules, all of them safety-relevant rather than merely tidy. Leave clearance for swelling — cells gain thickness with age and with high-temperature storage, and a cell compressed by a rigid pocket is a cell being pressed against its own separator; a millimetre of clearance on the thickness is a sensible allowance for a cell of this size. Never let a screw boss, a rib, or a sharp edge bear on the pouch, because a puncture in a charged LiPo is a fire and not a metaphorical one. And support the cell so it cannot move, using foam or a moulded pocket, because the wire leads and the JST-PH connector are the weakest mechanical link in the product and a cell sliding around a box works them until they fail. That last one is not hypothetical: one pilot unit went silent on day eleven because its battery connector had worked loose on a bench that a CNC machine vibrates all day.
Strain relief
Every wire that enters or leaves needs its stress taken by the enclosure rather than by a solder joint. On Beacon that is one wire pair, from the cell to the JST-PH header, and the answer is a moulded channel that grips the twisted pair plus a dab of flexible adhesive at the connector. The general rule: the last mechanical constraint before a joint should be closer to the joint than any load path, so that pulling the wire pulls on the plastic.
Buttons that feel like buttons and do not stay pressed
A tactile switch has a specified actuation force and travel; a typical small SMD tactile switch actuates at a couple of newtons with a few tenths of a millimetre of travel. The enclosure's job is to transmit a press without pre-loading the switch. That means a boss whose length leaves a defined gap at rest — 0.2 to 0.4 mm is a workable target — with a rib or shoulder that limits over-travel so a hard press bottoms out on plastic rather than on the switch's dome. Then check the stack: the boss length, the shell's wall thickness, the standoff height, the PCB thickness, and the switch's own height all have tolerances, and printed parts have generous ones. If the worst-case stack closes the gap, the button is held down, and on Beacon a held button at power-up is two separate failures at once — the boot-mode trap on GPIO9 and the over-the-air path. Unit 14 found both.
Light
The WS2812B-2020 is a surface-mount part looking upward at the inside of a shell. Getting its light outside takes a light pipe: a rod of clear material from just above the LED to the outer surface, with the outer end frosted so the eye sees a glowing spot rather than a point source. A printed shell can use a short length of clear acrylic rod or an insert printed in clear PETG. Two details are easy to miss. Keep the pipe's entry close to the LED — a millimetre of air gap loses most of the light to the inside of the box, which then glows in a way that looks like a fault. And do not run the pipe past the antenna keep-out, because a fat rod of dielectric is exactly the loading the previous section warned about.
BEACONBeacon's printed shell: what it cost, and what it changed
The pilot enclosure is an FDM-printed two-part PLA shell, 45 × 35 × 14 mm, four M2 heat-set inserts, a silicone gasket in a moulded groove, a light pipe of clear PETG, and a battery pocket at the opposite end from the module with a millimetre of swell clearance. Twenty-five sets were printed over two nights. The first revision had no vent. Cost per unit, with gasket and inserts, was about IDR 41,000 — which is worth noting alongside the electronics, because requirement R5's IDR 150,000 target explicitly excludes the enclosure, and a reader who forgets that will under-estimate the product by twenty-seven per cent.
| Property | Bare board | Sealed shell, v1 | Shell + membrane vent, v2 |
|---|---|---|---|
| Received signal at the gateway, 3 m line of sight | −60 dBm | −62 dBm | −62 dBm |
| Humidity time constant, moving between two rooms | < 60 s | ≈ 90 min | ≈ 12 min |
| Internal rise after a 6-minute over-the-air session | — | ≈ 9.0 °C | ≈ 6.2 °C |
| Humidity reported during that rise (true value 61 %RH) | — | 37 %RH | 43 %RH |
| Ingress rating claimable | None | IP54 achievable, untested | IP54 achievable, untested |
| Cost per unit at 25 | — | IDR 41,000 | IDR 53,000 |
The RF result was the pleasant surprise: two decibels, no more. That is the reward for a decision made two lessons ago in layout — the module at the board edge with its keep-out respected, the battery at the far end, and two millimetres of air between the antenna and a two-millimetre PLA wall. Had the cell gone under the module, the same measurement would have been six to eight decibels worse and the pilot's coverage problems would have been substantially self-inflicted.
The humidity result was the unpleasant one, and it is the reason this lesson exists. A sealed box with a ninety-minute time constant does not produce wrong humidity readings in a steady room; it produces readings that are ninety minutes stale, which is invisible until something changes and then looks like a broken sensor. The membrane vent brought that to about twelve minutes, at twelve thousand rupiah a unit, and the honest framing that went into the log is worth repeating: requirement R3 specifies ±3 %RH for the sensor, and nothing in the requirements ever specified a response time for the product. That omission was not caught by the review in Lesson 5.9 because it is not a circuit property. It goes on Rev B's list as a requirements change, not a hardware one.
Unit 14's shell came off the printer with a button boss at the top of its tolerance band, and the switch's actuator at the bottom of its. The stack closed. Clipped together, the shell held the button pressed; at every wake the firmware saw a held button and entered the over-the-air path, brought Wi-Fi up for a ten-second scan, found no update server, and slept. Averaged over a minute that is roughly 17 mA, against a design average of 40 µA — four hundred times the budget — and a 1000 mAh cell lasted under three days. Two fixes went onto Rev B with this evidence attached: shorten the boss by 0.4 mm and add a shoulder that limits travel, which is the mechanical fix; and require the button to be released and pressed again within a window after wake before the over-the-air path is entered, which is the firmware fix that makes a stuck button harmless rather than fatal. The second one is the better change, because it removes a class of failure rather than one instance, and it costs nothing to manufacture.
WEAKNESSWhere this bites
This is the default sequence and it is backwards for any product with a radio, a sensor, or a battery. The enclosure decides where the antenna can be, where the cell can be, where air can enter, and where a user's thumb will land — and every one of those is a PCB constraint. Beacon got away with it because the module was already at a board edge for RF reasons and the battery connector was already at the far end, so the mechanical constraints happened to be satisfied. That was luck dressed as process. The cheap fix is to draw the enclosure's outline, the mounting face, the vent, and the button in the mechanical layer of the PCB project before routing, so that the two documents cannot silently disagree.
"IP54 enclosure" on a spec sheet, when what you mean is "there is a gasket," is a claim about a laboratory result you do not have. It matters because the claim propagates: a customer mounts the tag in a wash-down area, water gets in past a print's layer lines, and the conversation that follows is about your documentation rather than about physics. The honest formulation is what Beacon's log uses — IP54 achievable, untested — which tells a reader exactly what has been done and exactly what has not. If the rating matters commercially, buy the test; it is not expensive next to the cost of being wrong.
Printed parts are dimensionally generous, injection-moulded parts shrink by a per-cent or so and shrink differently in different directions, PCB thickness has a tolerance, and switch heights have a tolerance. Adding the nominal numbers and getting a positive gap tells you nothing about whether the worst case has a positive gap. The discipline is to write the stack down with each contributor's tolerance, add them in the worst direction, and make the design work there. Unit 14 is the entire argument: every part was in specification, the assembly was not, and the failure presented as a dead battery four days into a pilot.
KEY INSIGHTHow this connects
- RF Layout: Antennas, Keep-Outs, and the Board Edge (5.7) — the keep-out that lesson defended in copper is defended here in plastic and air. Both are needed; either one alone is worthless.
- EMC, Noise, and Certification Basics (6.4) — the enclosure is part of the unit under test, and the module's regulatory grant is conditional on the antenna environment that this lesson's box defines.
- Reliability and Field Failure Modes (6.6) — next: connectors that work loose, cells that swell, and the honest limits of what a 25-unit pilot can tell you about any of it.
- Firmware in the Field: Logging, Telemetry, and Remote Diagnosis (6.7) — the post-update humidity error becomes a flag bit there, which is how an invisible data-quality problem becomes a diagnosable one.
- Cost, BOM, and Scaling From 25 to 25,000 (6.8) — the injection-mould crossover in this lesson's cost table is the same curve every non-recurring cost follows.
Mechanical design is where a product stops being a board and becomes a thing someone holds. It is also the discipline with the fewest automated checks, which is exactly why it deserves a written prediction and a review.
GLOSSARYKey terms introduced
- IP code (IEC 60529)
- Two digits describing tested protection against solids and liquids. IP54 means dust-protected (ingress not fully prevented but harmless) and protected against splashing water from any direction. It is a test result, not a design intention.
- ePTFE membrane vent
- A microporous fluoropolymer disc that passes air and water vapour but blocks liquid water below a specified water-entry pressure. It is the only way to equalise a sealed enclosure's humidity without losing its ingress rating.
- Dielectric loading
- The shift in an antenna's resonant frequency and match caused by material with a permittivity above 1 in its near field. Common enclosure plastics sit near εr 2.5–3.2 and typically cost a decibel or two; metal does not load the antenna, it disables it.
- Thermal resistance (Rth)
- Degrees of temperature rise per watt dissipated, from a source to ambient. For a small sealed plastic enclosure under natural convection it is on the order of tens of °C/W — an estimate to measure, not to trust.
- Duty cycle
- The fraction of time a load is active. Beacon's 45 ms every 60 s is 0.075 %, which is why its self-heating is thousandths of a degree in normal operation and nine degrees during a continuous Wi-Fi session.
- Saturation vapour pressure
- The maximum water-vapour pressure air can hold at a given temperature. It rises steeply with temperature, which is why heating sealed air by 9 °C drops the measured relative humidity to about 60 % of its true value with no water going anywhere.
- Tolerance stack
- The accumulated dimensional variation of every part in a chain. A design is correct only if it works at the worst-case stack, not at the nominal one; unit 14's held button was a stack that closed.
- Heat-set insert
- A knurled brass sleeve pressed into printed or moulded plastic with a heated tool, giving a metal thread that survives repeated assembly where a screw driven into plastic would strip.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1Beacon's self-heating in normal operation is about 0.007 °C, yet the same box produces a 24-point humidity error. Reconcile those two statements.›
They describe different operating modes. In normal operation the device is awake for 45 ms once a minute at roughly 40 mA, an average of about 0.13 mW; against a thermal resistance near 55 °C/W that is seven thousandths of a degree, which is nothing. But the over-the-air update path runs Wi-Fi continuously at around 100 mA — about 330 mW, more than two thousand times the average — and six minutes of that in a box with a nine-minute time constant raises the internal air about 9 °C. Relative humidity is a ratio against the saturation vapour pressure, which rises steeply with temperature, so a 9 °C rise from 28 °C multiplies the denominator by about 1.66 and the reading falls to roughly 60 % of the true value: 61 %RH is reported as 37 %. The general lesson is that thermal problems are about the worst-case mode's power and its duration relative to the thermal time constant, never about the average.
2A colleague proposes an aluminium enclosure for the pilot: rugged, cheap to machine locally, and it would improve the thermal resistance by a factor of three. Argue against it in one paragraph.›
The thermal improvement solves a problem Beacon does not have — self-heating in normal operation is seven thousandths of a degree — while destroying the one function the product exists for. A conductive surface millimetres from a 2.4 GHz printed antenna shorts its near field; the loss is not a few decibels of degradation but tens of decibels, and the estimated range falls from over twenty metres to a couple. Requirement R2 becomes unachievable at any transmit power the module supports. There is also a regulatory consequence: the ESP32-C3-MINI-1's modular certification is conditional on the antenna environment described in Espressif's integration guidance, and burying the antenna in metal is outside it. If a metal housing were genuinely required — for impact or for a customer's aesthetic — the antenna would have to leave the box: an external antenna on a connector, a plastic window over the antenna region, or a separate radome. Each of those is a different product with different costs, and none of them is a machining decision made late.
3Why can a sealed enclosure with a ventilation slot not be rated IP54, and what does the alternative cost?›
Because both IP digits are tests of ingress through every opening, and a slot is an opening. The first digit, 5, requires that dust entering the enclosure be insufficient to interfere with operation, and the test draws talc dust in; a slot admits it directly onto the board. The second digit, 4, requires that water splashed from any direction have no harmful effect, and a slot passes spray. The gasket around the seam is irrelevant if there is a hole in the lid. The alternative is a microporous ePTFE membrane vent, which passes air and water vapour while blocking liquid below a specified water-entry pressure, so the enclosure equalises humidity and still keeps a rating. It costs roughly twelve thousand rupiah per unit at pilot volumes, plus a hole and the discipline not to cover it with mounting adhesive. On Beacon it also cut the humidity time constant from about ninety minutes to about twelve, which is the second requirement it buys.
4Unit 14's shell held its button pressed. Two fixes were proposed: shorten the boss, and change the firmware so the over-the-air path needs a fresh press. Which is the better change, and why?›
Both should be done, but the firmware change is the better one because it addresses the class rather than the instance. Shortening the boss by 0.4 mm fixes the specific tolerance stack that closed on unit 14; it does not help if a user leans the tag against something, if the shell warps in a hot van, if a later print run comes off a different machine, or if a customer sticks tape over the button. The firmware change — requiring the button to be observed released and then pressed again inside a short window before entering the update path — makes a continuously-held button indistinguishable from no button at all, so every one of those scenarios becomes harmless. It also costs nothing to manufacture and can be deployed to units already in the field, whereas the mechanical fix requires new shells. The general principle: when a mechanical variation causes a firmware state, prefer the firmware change that makes the state unreachable by accident, and treat the mechanical fix as defence in depth rather than as the answer.