Burn-in is a long process: 8 hours at 40 °C with a load and a monitor. The bottleneck is rarely the chamber. It is the loading and unloading, which is manual, repetitive and prone to connector damage.
The starting point
The customer had four burn-in chambers and 12 racks per chamber. Each rack held 24 power supplies. Operators loaded the units into the racks by hand, pushed the racks into the chamber and connected a multi-pin connector. The process worked, but loading took 25 minutes per rack and the connectors were damaged about 1% of the time.
The target was to double throughput without adding a chamber. The chamber capacity was sufficient; the handling was not.
The retrofit
We added an infeed conveyor with a singulator, a loading robot and a rack buffer. The robot picked the unit from the conveyor, oriented it and inserted it into the rack. The rack was held in a fixture that located it precisely, so the robot did not need to search.
The existing racks were retained, but we modified the connector entry with a chamfered guide to reduce damage. The robot used a force sensor to stop if the connector did not seat, rather than pushing harder.
| Metric | Before | After |
|---|---|---|
| Loading time per rack | 25 min | 9 min |
| Connector damage | 1.0% | 0.15% |
| Units per shift | 1,200 | 2,400 |
| Operators on loading | 3 | 1 |
Test data integration
Each unit was scanned before loading, and the chamber controller recorded the test result against the serial number. The data was forwarded to the customer's MES. The operator could see a failed unit on a monitor and remove it at unload.
The integration was the hardest part, because the chamber controller used a proprietary protocol. We added a gateway that read the controller's log and converted it to the MES format. The gateway also buffered data if the network was down.
What went wrong first
The first robot gripper was a parallel jaw design that gripped the housing. It worked for one model but marked the plastic on another. We changed to a vacuum gripper with a soft pad, which handled both models without marks.
The second issue was the rack buffer. We sized it for 30 minutes, but the chamber loading cycle was 45 minutes. The buffer ran out and the robot stopped. We added a second buffer lane, which gave 60 minutes.
Lessons for similar retrofits
Keep the existing racks if they are structurally sound; the cost of new racks is often higher than the cost of modifying the connector entry. Invest in the gripper and the force sensing, because that is where the damage occurs. Size the buffer to the longest chamber cycle, not the average.
Do not automate the unloading first. Loading is usually the bottleneck because it requires orientation and connection. Unloading is simpler and can remain manual until the loading is stable.
Frequently asked questions
Can the robot handle different power supply sizes?
Yes, with a gripper that can change or a vacuum pad that accommodates the range. The rack fixture also needs to be adjustable.
How was the 8-hour burn-in time handled?
The burn-in time did not change. The retrofit increased the loading rate, so more racks could be processed in the same chamber time.
What was the ROI?
About 11 months, based on labour savings and scrap reduction. The customer also gained capacity without a building expansion.
If you are planning a burn-in or test-line retrofit, send the rack drawing and the chamber cycle to our engineers through the contact page.
