A drop of more than 15 % in working-plane irradiance at unchanged electrical power points to the lamp or the optics, not the supply. Track mW/cm2 per 100 W of input: a xenon arc typically runs 500–1,000 h, and end of life is usually defined at 70–80 % of initial output, not at failure.
Diagnosis needs a baseline, and brightness by eye is not one. Record three numbers on day one, at the working plane rather than at the lamp port: irradiance in mW/cm2 at a marked distance, lamp voltage and current at the operating setpoint, and cumulative lamp hours. Then compute the ratio of optical to electrical input — for example 180 mW/cm2 at 300 W, or 0.6 mW/cm2 per watt. That single ratio separates supply faults from optical ones, because a failing power supply moves the denominator and a dirty window does not.
Two cautions. The meter drifts too — an uncorrected photodiode loses 3–5 % a year, so keep one reference meter and cross-check annually. And 10 mm of axial movement in a diverging beam shifts the reading by 5–10 %, the same size as the fault you are hunting.
Xenon short-arc lamps erode at the electrodes, so the arc gap widens and the arc wanders: output falls slowly at constant current, the hotspot drifts, and the UV end weakens faster than the visible as the quartz envelope solarises. A lamp at 600 h may hold 85 % of its visible output but only 70 % of its 365 nm output — enough to halve an apparent quantum yield in a UV-driven experiment while the bench still looks bright.
Ignition behaviour is the other tell: a healthy lamp strikes within a second or two, while one that needs repeated attempts or only strikes warm is near the end of its electrode life. Replacement is covered in replacing a xenon lamp bulb without breaking alignment, and the warm-up discipline that protects a new lamp in how long a xenon source should warm up.
Supply faults change the numbers on the front panel, which is what makes them the easiest to rule out. Watch current and voltage under load. If the current sags below setpoint, the supply has hit its voltage ceiling because the arc gap grew — a bulb symptom. If current is on setpoint but output fell in a step rather than a slope, suspect the igniter, a capacitor bank or a cooked connector.
Ripple is the fault most often missed. Point a photodiode at the beam and scope the trace: a healthy supply holds ripple below 3 %, a degraded one shows a 100 Hz component at 5–15 %. That is fast enough to average away on a slow power meter and slow enough to bias a photoelectrochemical measurement by the same 10 %.
Optics fail in a pattern rather than uniformly. A dust film on an ellipsoidal reflector costs 5–10 % of throughput, outgassed silicone grease on a window or fibre end costs more in the UV than the visible, and a light guide with a browned bundle loses its outer fibres first — a bright core and a dim edge on a uniformity scan.
The distinguishing test is spectral: measure at 365 nm and 550 nm with the same meter. A ratio that has shifted toward the visible means UV absorption — contamination, solarised quartz, or a filter left in the beam. An unchanged ratio at uniformly lower output means geometric loss: misalignment, defocus, or a displaced reflector after maintenance.
| Observation | Most likely cause | Test in ten minutes | Decision threshold |
|---|---|---|---|
| Slow fall over weeks, power unchanged | Bulb aging | Irradiance vs lamp-hour log | Below 70–80 % of initial: replace |
| UV falls faster than visible | Envelope solarisation or UV absorption | 365 nm / 550 nm ratio against baseline | Ratio down > 10 %: replace lamp or optics |
| Bright core, dim edge | Damaged fibre bundle or defocus | Five-point scan across the spot | Non-uniformity > 25 %: realign or replace |
| Step change overnight | Supply, igniter or connector | Panel current and voltage under load | Current off setpoint > 3 %: service supply |
| Flicker or unstable reading | Ripple, arcing, loose terminal | Photodiode on a scope | Ripple > 5 %: service supply |
| Hard ignition, then normal | Electrode erosion | Count strike attempts over five starts | More than two attempts: replace lamp |
| Uniform loss after maintenance | Misalignment or displaced reflector | Compare spot photo with commissioning image | Centroid moved > 2 mm: realign |
The log makes diagnosis cheap: 2 % per 100 h is normal ageing, 12 % in a week is a fault, and without it the two look alike. Meter practice is covered in optical power meters in photocatalysis, and the preventive schedule in installing and maintaining photochemical light sources.
Compare two meters at the same plane, or a stable LED reference. If both agree and both read low, the source has decayed; if they disagree by more than 10 %, one calibration is suspect.
Yes, but output below 70–80 % of initial invalidates quantum-yield comparisons, and an old lamp risks envelope failure under pressure. Replace on measured output, not the calendar.
For contamination, yes — 5–15 % is typical when a visible film is removed from a window or reflector. For solarised quartz or an eroded arc, no amount of cleaning helps, and the 365/550 nm ratio will tell you which case you are in.
Because the loss is after the lamp: a window, a filter left in the path, a fibre connector not seated, or a reactor wall of borosilicate that cuts everything below 320 nm. Measure at the inner window face to close that gap.
