Burn-in is the unglamorous part of panel manufacturing. It’s a 48-hour hold on every panel before it ships, in a controlled lab environment, with the panel powered up and exercised under full load. It adds two days to the lead time. It doesn’t produce a thing the customer can see. And it’s where the difference between a 1% DOA rate and a 12% DOA rate actually gets made.
This post is a walk-through of what we do during those 48 hours, why that number is what it is, and what the failure data we collect looks like across thousands of panels.
The bathtub failure curve
Electronic and electromechanical components don’t fail randomly across their service life. The well-known “bathtub curve” — named for its shape when plotted — describes the actual failure pattern:
- Infant mortality (first 24–72 hours): failure rate is high and drops rapidly. Defective parts, latent manufacturing flaws, and marginal solder joints surface here.
- Useful life (years 1–15+ for most BAS components): failure rate is low and roughly constant. Random failures happen at a predictable rate.
- End-of-life wear-out (final years): failure rate climbs again as bearings, capacitors, and other consumables exceed their design life.
The early part of the curve is what burn-in addresses. The components that are going to fail in the first three days of service are going to fail somewhere — either at the factory, where we can replace them before the panel ships, or at the customer’s job site, where the replacement cost is 5–20× higher.
The math on running burn-in is almost purely economic: where do we want the infant-mortality failures to happen?
What we actually test
The burn-in isn’t “leave the panel powered on for 48 hours.” It’s a deliberate stress profile designed to provoke the failure modes we know happen at this stage:
Full-load power cycling
The panel runs at its rated power draw for blocks of 30–60 minutes, followed by full power-down, followed by re-energization. Power-cycling is where heat-stress failures surface. Components that get hot and fail under repeated thermal expansion don’t fail at idle; they fail at load, and they fail at the transitions.
Communication-bus stress
Every protocol the panel speaks gets exercised continuously throughout the burn-in window. For a typical BAS panel, that’s:
- BACnet over MS/TP and BACnet over IP, full message-rate ceiling
- Modbus RTU and Modbus TCP, full register-read cycles
- Any vendor-specific bus the panel uses (Niagara Fox, Distech CanaLink, Honeywell’s serial protocols, etc.)
Bus saturation surfaces marginal terminations and bus-driver chips that pass an ohm-meter check but fail under sustained traffic.
Temperature ramp
Most BAS panels live in mechanical rooms that swing between 5°C and 40°C across seasons. Our burn-in lab cycles the panel between 0°C and 50°C — slightly wider than the field — to give us margin against the worst-case site environment. Components that fail to the cold side (condensation on PCBs, capacitor ESR shifting) and components that fail to the hot side (thermal runaway in voltage regulators, plastic deformation in connectors) both get caught.
Voltage sag and surge
Grid voltage isn’t 120V. It’s 120V plus or minus a noisy 10–15% depending on the building and the time of day. We simulate the upper and lower limits of the panel’s rated voltage envelope across the burn-in window. Power-supply failures, which are the single most common failure mode for BAS controllers, often surface as voltage-margin failures and not as constant-voltage failures.
Why forty-eight hours
The most-asked question we get on the burn-in process is whether 48 hours is overkill, undershoot, or arbitrary. It’s none of the three.
Why not 24?
Twenty-four hours catches most infant-mortality failures, but not all of them. The published failure data on industrial control components shows the failure-rate curve dropping sharply between hour 24 and hour 36, then flattening between hour 36 and hour 72. Stopping at 24 leaves a meaningful tail of failures still to happen — failures that will happen at the customer’s site instead.
Why not 72?
Seventy-two hours catches a few additional failures beyond 48, but the incremental yield gets smaller. Pushing to 72 would add another full day to every panel’s lead time without proportional return. We’ve modeled the trade-off across our actual production data; 48 hours sits at the inflection point where the marginal failure-catch rate stops justifying the additional time.
Why not “until it stops failing”?
Some shops run open-ended burn-ins on a per-panel basis. We don’t, for two reasons. First, predictability of lead time matters to customers: a 48-hour window gives a fixed delivery date. Second, the flat part of the bathtub curve isn’t a meaningful test — running for 30 days isn’t going to catch a useful-life failure that’s coming three years from now, and it dilutes the burn-in resources we have for the infant-mortality phase.
What the failure data looks like
We log every burn-in test, including catches. Across the most recent twelve months of production, the failure-catch breakdown was:
- Power supplies — 41% of all catches. Mostly bad regulators or undersized capacitors. The single most common failure mode by a wide margin.
- Bus driver / communication ICs — 22% of catches. Marginal silicon that passes initial bench tests but degrades under sustained traffic.
- Contactors and relays — 14% of catches. Almost always coil issues — open coils, partial shorts to ground.
- Terminal blocks and connectors — 11% of catches. Stripped threads, off-spec torque tolerance, intermittent contact at temperature extremes.
- Programmable controllers (DDC, PLC) — 7% of catches. Usually firmware-level issues that don’t surface until the controller’s been running its full program for many hours.
- Everything else — 5% of catches. Wire-pull defects on pre-assembled harnesses, occasional enclosure-grounding issues, miscellaneous oddities.
Every catch becomes a record we can search against. If we see a pattern emerging — say, a particular power-supply lot showing higher failure rates than expected — we surface that to the supplier and pull the remaining inventory of that lot. Burn-in isn’t just QC for the panel in front of us; it’s an early-warning system for the supply chain.
What burn-in doesn’t catch
Burn-in is the right answer for infant-mortality failures. It’s not the right answer for everything. The things it doesn’t catch:
- Software bugs in the building-automation logic running on the panel. The panel’s controllers operate during burn-in, but they run a test program — not the customer’s site-specific program. Site-specific program testing happens at the bag-and-tag stage, where the customer’s actual setpoints and sequences get loaded.
- Site-installation issues like incorrect torque on field terminations, missing ground bonds, or improper enclosure grounding. Those are the installing contractor’s responsibility on-site.
- End-of-life wear-out failures projected for 10–20 years out. Those are addressed at the design stage by using long-life-rated components, not at the test stage.
- External system failures — bad sensors, faulty actuators, wrong-sized motors elsewhere in the system. The panel can be perfect and still appear to fail because something attached to it isn’t.
A clean burn-in test is necessary but not sufficient. It’s necessary because the alternative is finding the failures at the customer’s site. It’s not sufficient because some failure modes live outside the panel itself.
What it costs and what it returns
Two days of additional lead time. A dedicated burn-in lab that takes floor space. Power and HVAC overhead to run the test stress profile. Operators to monitor and document the runs. The cost isn’t trivial.
The return is straightforward. Our DOA rate runs below 1%. Field-built panels we’ve benchmarked, against the same component mix, run 8–15%. Every DOA the burn-in catches is a service call the customer never has to make.
When you’re scoping a project and wondering whether the extra lead-time is worth it, the answer almost always comes back to: yes, because the alternative is debugging a dead panel at 4 pm on a Friday with the building owner standing over you. We’d rather find the failure on a workbench in our shop.