Hello!
Looking at your post carefully, I see some things that may prove helpful going forward...
• You've had an unusually large number of failures for a single owner
• All the failures were front lights. What about the taillights?
• The lights have been from two different makers
• All have been used with the same Shimano dynohub, a DN72
• You're a fast rider
• All lights had overvoltage protection -- assuming it has not failed, that is supposed to protect the lights for fast riders.
• Lights have always been on even if turned off (not sure I understand this...perhaps the switch isn't working properly?)
A couple questions/thoughts...
• How do the lights fail...do they simply fail to light? Or, do they not turn off and then eventually no longer work?
• How are the lights wired? Do they use a dual lead directly from the dynohub, or does one path use the bike frame as a return to the dynohub source?
• Is there a short in the wiring, most likely at the dynohub plug (I've seen some cases where a single strand came free of the plug and shorted out against something metal).
• How are the headlights mounted? On their factory mounts, or on something custom?
• If you're not running a slave-powered taillight (one attached to and supplied by the headlight), what about the unused leads/lugs on the headlights? Could something be shorting across them, say when turning the handlebars?
First thing I'd do is mount the bike in a stand so the front wheel can spin freely and then attach a voltmeter to see how much power is being output by the hub. If all is well there, I'd backtrace from the dynohub, looking for faults. A cheap voltmeter would be a good investment compared to another headlight that could fail in turn.
I do remember when B&M first introduced the IQ Cyo, there were a number of failures traceable to the switch assembly. Depending on age and model, might yours be from an older, faulty batch?
Sorry to not offer anything definitive, but hopefully something in the above will prove helpful in tracing this frustrating and expensive problem.
Best,
Dan.