The prime killer of tungsten filament lamps is not hours of operation, it is the repeated application of instantaneous inrush current from the on/off cycle. Each inrush current surge (from a conventional wall toggle switch, for example) creates a hot spot somewhere in the filament structure. This is because at a very subtle level, the filament’s cross-section is not perfectly uniform. These non-uniformities of cross-section mean that the resistance of the tungsten varies with each dimensional variance at that point in the filament. Since in effect the sections of variable individual resistance are connected in series, this means some areas carry more current than others. The effect should be obvious. Some spots on the filament will run hotter than others. These hot spots develop crevasses - which further reduce the effective cross section of the material at that point. and, with each inrush event, the thinning and crevasse development gets worse and worse.
After a fixed number of cycles, the inrush current of the total filament load is sufficient to turn the thin spot into the equivalent of a fusible link. When this happens, there’s is a bright flash and the bulb is kaput.
This phenomenon was well documented by vacuum tube manufacturers more than seventy years ago. It is called the Miller Larson effect. You can read about it rather extensively in the RCA Vacuum Tube Designer’s Handbook.
The solution is simple.
1.) do not select switches that allow instantaneous current inrush. I recommend slide dimmers with the off/on switch at the beginning of the dimmer travel.
If this solution doesn’t work for you - as in theatrical fixtures, then I suggest you keep a pre-warm current on the filament so as to reduce the strain during the move to full brightness. Vacuum tube people do this all the time…no reason conventional tungsten filament lamps can’t enjoy the same advantage.
I have used both strategies. I have typical Edison base tungsten halogen bulbs in more than twenty locations that are so protected. These lamps are currently sixteen years old, without a single filament failure, and are known to have over ten thousand hours’ burn time.