Several thousand megawatts of North American wind capacity run on turbines whose manufacturer no longer exists. Clipper Windpower built roughly 400 Liberty 2.5 MW machines before it stopped. DeWind, Nordic Windpower and Fuhrländer are gone. Suzlon’s older S88 fleet and Mitsubishi’s MWT-1000s rely on thin service organizations far from the factory. Towers, blades and gearboxes on these machines have mature third-party supply chains. The electronics do not.
A converter or control cabinet in a 2008 turbine holds a dozen or more circuit cards. There are DSP boards, gate drivers, I/O and communications cards, and power supplies. A contract engineering firm designed them for the OEM. A contract manufacturer built them. Nobody ever sold them on the open market. When one fails, the operator has three choices. Pull a card from a turbine that is already down. Buy a used one of unknown history, if one exists. Or find someone who can repair or rebuild it. This article covers the third option, because operators know it least.
Repair first
Most failed cards are repairable. The typical wind-converter failure is a power-supply section, a driver stage, a bank of dried capacitors, or corrosion at a connector. It is rarely the processor. A bench that has seen the card before can usually diagnose and repair it in two to three weeks. It then tests the card and returns it with a multi-year warranty. The cost is a fraction of a replacement. On a fleet with a modest failure rate, this is enough. A rolling stock of repaired spares covers the fleet indefinitely.
We currently repair eleven card types for one Clipper Liberty fleet operator. They include DSP slave and master controllers, firewire and communications cards, gate-driver boards and power supplies. We price each by the complexity of the card. Every repair carries a three-year warranty. We charge nothing for any card we cannot repair. That last term matters with orphaned hardware. The operator can send us doubtful cards without risk.
When repair is not enough
A repair program runs out of road in two situations. The first is attrition. Cards fail in ways that destroy them, or the population shrinks over fifteen years until there are fewer working cards than turbines. The second is a component that no longer exists. A 2005 design might carry a DSP that went out of production a decade ago. It might carry a one-time-programmable FPGA whose maker has moved on. When the last of those parts leaves the broker market, no one can repair the card at any price.
The answer is to build new cards. That is a very different job from repairing them. You need the original fabrication data for the printed circuit board. You need a bill of materials with manufacturer part numbers. You need a schematic. You need the firmware and programming files for every programmable device. And you need a test procedure that proves a new card behaves exactly like the original. In our experience with orphaned fleets, the fabrication data usually survives. The schematic sometimes survives. The bill of materials, firmware files and test procedures almost never do. They lived on a server at a company that no longer exists.
What reverse engineering actually involves
Where the documentation is missing, someone has to recreate it from working hardware. An engineer measures a known-good card and identifies its components. She cross-references each one to a current or aftermarket part number. She captures the schematic net by net. She reads out the programmable devices where the technology allows it.
That last step carries the real risk. Some device families, notably antifuse FPGAs, are designed so that no one can read their contents back. If the original programming file cannot be found in an archive, the engineer has to understand the logic from the outside and re-implement it in a modern part. This is possible. But it is engineering rather than copying, and it should be priced and scheduled as such.
Done properly, the output is a build package. It holds fabrication data, a bill of materials, a schematic, programming files and a test procedure. A contract manufacturer can turn that package into new cards in lots of twenty-five or fifty. The new cards are tested, coated and warrantied like the originals. The one-time engineering cost is real, typically tens of thousands of dollars per card type. Against a turbine that cannot return to service, it is a small number. Operators can spread it across the fleet. Where several operators run the same machine, they can share it.
A word on obsolete components
The last step in sustaining an orphaned fleet is buying the components that are running out before they are gone. A build package identifies the discontinued devices on each card. The operator can then make one consolidated purchase from authorized aftermarket sources, sized to the fleet’s expected life. That beats paying broker prices card by card as failures occur. Operators skip this step most often. It is the step that most often ends a fleet.
Where to start
Do you operate an orphaned or thinly supported fleet? Start with an inventory. List the card types in the cabinet, the working spares you hold, and the documentation you have for each. Then get the repairable cards repaired and warrantied, so your spares are proven rather than assumed. Then ask, card by card, whether the documentation exists to build more. Where it does not, decide now whether to recreate it, while working reference cards still exist. The cost of reverse engineering rises sharply once the last good card is gone.
Powergenics repairs and remanufactures control and converter electronics for power plants and wind fleets. That includes Clipper Liberty (Magnetek-built) converter cards. We coordinate reverse engineering and production builds for card types no OEM supports. Three-year warranty on every repair. No fix, no fee. Richmond, Virginia. 800-861-9773.