An instrument enclosure and a distribution cabinet look alike and behave nothing alike. A distribution cabinet is judged on current capacity and temperature rise, where 65 K is acceptable. An instrument enclosure holds a grating compartment, a chromatography detector or an XRF sensor, and those parts want ±0.5 K, micrometre stability and 60 dB of shielding — three orders of magnitude apart, in one steel box.
They also fight each other: more airflow means more noise and vibration, a tighter EMC seam means worse heat rejection, thicker panels mean weight and lead time. So the order is fixed — settle the detector's allowable swing and the site noise limit, choose the cooling path, then use structure and isolation for what is left.
1. Three constraints, one envelope
Most reworks come from starting the drawing before the inputs were fixed. Get four numbers first.
- Heat load by compartment — not one total; 350 W in the supply bay and 400 W spread evenly are different problems.
- Allowable swing at the detector — ±0.5 K for spectroscopy heads, ±0.1–0.2 K for scientific CCDs, ±3 K for a plain electronics bay.
- Site ambient and noise ceiling — a lab runs 22–26 °C and often caps 55 dB(A) at 1 m; a shop-floor cabinet may see 40 °C and dust.
- EMC standard — laboratory, medical and industrial limits are not interchangeable, and the target dB follows from it.
Put all four in the RFQ. Fan size, gasket type, panel gauge and isolator are consequences, not choices.
2. Form factor first, because it caps the heat budget
| Form factor and envelope | Construction and use |
|---|---|
| Benchtop case, 450 × 400 × 180 mm | 1.2–1.5 mm steel, folded; single-unit analysers |
| 19-inch subrack, 482.6 mm panel, 3U–9U | 1.2 mm galvanised steel; modular instruments |
| Floor-standing, 600 × 800 × 1800 mm | 1.5–2.0 mm folded frame; optics plus electronics |
| Wall-mount / in-line, 600 × 500 × 250 mm | 1.5 mm, IP54/IP65 sealed; online monitoring |
Run the heat flux density before committing. A 120 W load in a benchtop case has under 0.6 m² of wetted surface, about 200 W/m², and natural convection will not hold it; the same load in a floor cabinet is comfortable. Above 100 W/m², reserve space and cut-outs for forced convection at concept stage.
3. Thermal: heat flux density before fan curves
For a sealed painted steel enclosure in still air, use 5–6 W/(m²·K): ΔT = Q / (5.5 × A). A 400 W load on 4 m² gives 18 K, so 25 °C ambient becomes 43 °C inside. If the detector must stay under 30 °C, more louvre area will not fix it.
For ventilation, size on V = 3.1 × Q / ΔT (m³/h): 400 W at a 10 K rise needs 124 m³/h, so specify 160 m³/h. That ΔT is air-on to air-off, not the average internal rise, which runs 30–50% higher — the margin is not optional.
| Cooling path and capacity | Control / noise | Use when |
|---|---|---|
| Natural convection, ≤150 W | ±5 K, silent | Low flux indoors; louvres break IP |
| Forced convection, 150–1000 W | ±2–3 K, 45–65 dB(A) | Default for electronics bays |
| Air-to-air exchanger, 300–800 W | ±3 K, 50–60 dB(A) | Sealed cabinet, keeps IP54 |
| Cabinet A/C, 500–3000 W | ±1–2 K, 55–70 dB(A) | Hot shop floor; needs drain |
| Peltier or liquid loop, 100–500 W local | ±0.1–0.5 K, 40–55 dB(A) | Detector or optics cavity only |
One rule that catches projects late: the noise ceiling can invalidate the cooling choice. At 55 dB(A) at 1 m a standard axial fan above 2000 rpm is out of spec, so you need a larger fan at lower speed or an external plenum with acoustic foam — decided before the prototype, not after.
4. Pulling sensitive parts off the heat path
Holding ±0.5 K is rarely about capacity; it is about thermal short circuits. Classic case: detector bracket and power modules share one 2 mm base plate, 80 W conducts straight across, and the detector tracks load instead of setpoint however the airflow is tuned.
- Break the bridge — at least 5 mm of POM, phenolic or FR-4 under a sensitive bracket; stainless bracket, not aluminium, which conducts about five times better.
- Separate by height, not width — power high or in a rear bay, optics low and forward; hot air does not migrate down.
- Series airflow, not parallel — cold air enters the sensitive bay first, then the power bay; and never blow it across a grating, because impingement modulates refractive index and shows up as baseline noise. Baffle the air along the wall.
- Move the roughing pump out — on a mass spec it is the largest heat and vibration source at once; put it outside or in an acoustic bay on a bellows.
Field target: keep the internal gradient under 2 K and long-scan drift disappears. Above 3 K a precision spectrometer climbs slowly over eight hours at the customer site, never on the bench.
5. EMI: the seam decides, not the gauge
Shielding effectiveness is set by the longest seam, and only weakly by thickness. Keep the longest aperture dimension under one-twentieth of a wavelength for roughly 20 dB — 15 mm at 1 GHz. A 1.5 mm steel panel is worth over 100 dB alone; bolted at 100 mm pitch it measures 20–30 dB.
- Seams — screw pitch from 100 mm to 50 mm buys 10–15 dB; above 60 dB, gasket the seam or weld it continuously and dress flush.
- Gaskets — conductive fabric at 30–50% compression, beryllium copper fingers at 20–30%; under-compression leaves resistance high, over-compression takes a set.
- Powder coat masking — 60–80 μm of powder is an insulator, so tape the door rebate, gasket land, ground stud and filter face before coating. If the drawing does not show it, every face gets coated.
- Apertures and windows — honeycomb waveguide at three times the cell size in depth gives 40–60 dB at 60–80% open area, two offset perforated layers 20–30 dB; 100–200 mesh screen in laminated glass gives 40–50 dB at 60–75% transmission and ITO glass 80% at double the cost.
- Cable entry — clamp the shield 360° to the enclosure; a pigtail braid is inductive at high frequency and does nothing.
Bonding and filtering. Gaskets, filter housings and shields all land on the enclosure, which reaches earth through one M6 stainless stud on bare metal — a screw biting through powder is not a bond. An IEC inlet filter shunts common-mode noise through its own housing, so its mounting face must be bare and clamped on all four screws; on a coated panel it filters nothing. Keep power and signal 100 mm apart, crossing at 90°. Agree the test basis early: IEC 61326-1 for laboratory equipment, CISPR 11 Class A or B for industrial and medical; enclosure-level figures mean IEEE 299 in a chamber.
6. Vibration: find the source, size the isolator
Nine of ten cabinet vibration problems start inside the cabinet — fans, roughing pumps, compressors, coolant pumps. Reduce at source, isolate second, stiffen third. Isolator choice turns on one ratio: natural frequency below one-third of the excitation frequency for about 80% isolation. Ignoring damping, T = 1 / |1 − (f/fn)²|; a 1450 rpm pump excites at 24 Hz, and fn at 8 Hz gives T ≈ 0.13. Rubber pads sit at 8–15 Hz, springs at 2–5 Hz.
- Frame — torsion under 0.1°/m; any bench carrying optics should have its first mode above 80 Hz.
- Panels — 1.5 mm folded with stiffener pitch of 300 mm or less; a panel that drums in the airflow radiates the noise you just paid to remove.
- Feet — four-point levelling with damping pads, centre of gravity inside the support polygon, ±15 mm adjustment for floors that are never level.
- Draw the boundary — for an interferometer or high-resolution grating the cabinet is not the vibration solution; mount optics on granite with air isolation and let the enclosure handle guarding, shielding and environment.
7. Build, test and sign off
Instrument work is low volume and high mix: 1.2–2.0 mm cold-rolled steel for coated parts, galvanised where an internal part must conduct, frames folded from 1.5–2.0 mm C-sections, or 40 × 40 extrusion while the design is still moving. Extrusion modifies fast but is less stiff than a folded frame. Prefer folding and clinch fasteners at these volumes; welded assemblies need flattening or panel flatness misses 0.5 mm/m. Specify polyester powder, not epoxy, where panels are wiped with IPA.
Put eight items on the drawing and most of the back-and-forth disappears: heat load by compartment and internal maximum; ambient ceiling and IP rating; noise limit with measuring position; EMC standard, target dB and who tests; faces that must conduct and be masked; aperture schedule with waveguide vents and shielded windows; ground stud location and finish; isolation method, transport lock and lifting points.
Then test four things. Temperature: four hours at full load, thermocouples at detector and power bay, pass at ±0.5 K and under 2 K across the cabinet. Noise: A-weighted at 1 m, background 10 dB below the reading. Shielding: one enclosure-level test before volume build, near-field probes on critical seams. Vibration: 1–1000 Hz spectrum at the optic mounting face with pumps running. Electrical safety to IEC 61010-1 is mandatory.
8. FAQ
Can you hold ±0.5 K across the whole cabinet?
Not economically, and rarely needed. That figure normally applies to the detector or cavity: build a small sub-compartment with a Peltier element or dedicated air loop and run the rest at ±3 K.
Can conductive bonding be added after coating?
Yes, by locally stripping the coating and adding gasket or conductive adhesive, but contact resistance must then be measured against 10 mΩ. Masking before coating costs nothing.
What panel thickness gives 60 dB?
Thickness is not the lever. 1.2 mm steel is already far beyond 60 dB; the limit is seams, apertures and cable entries. Gasket the seams, 50 mm screw pitch, waveguide vents, 360° cable clamps.
The fan is too loud but a slower one will not cool enough.
Re-measure the load first — it is often 20–30% below the design figure. If it is real, use a larger fan at lower speed (6–10 dB(A) at equal airflow) or move the fan into an external plenum and keep only ducting inside.
Alignment shifts after shipping. Is the frame too flexible?
Usually an unlocked optical stage, not the frame. Optics and isolation tables need transport locks, and the drawing should show lifting points and lock locations; frame problems show up as doors that will not close or panels out of flat.
Do you take 10–30 piece orders?
Yes, that is the normal batch. Laser cutting, CNC bending and clinch fastening need no tooling. Prototype in 7–15 days, then 7–15 days after first-article approval, to drawing or sample.
About Xingyuan Electric
Xingyuan Electric (Suzhou) Co., Ltd. builds custom sheet metal enclosures with a 30+-person team, a 3,000+ m² production base and 20+ laser cutting, CNC bending and welding machines. Lead time is 7–15 days and a 10–30 piece batch is normal, because the fold-and-clinch route needs no tooling. Budget a premium over a distribution cabinet of the same size — shielding hardware alone runs 8–20% of cost. We work from your drawings or samples, with structural review, cooling and shielding advice and first-article approval. Contact sale@xingyuanep.com or WhatsApp +8615221234007.