For two decades, I've specified, installed, and cursed at distance and proximity sensors in environments that would make a datasheet weep. Forget the pristine lab conditions and marketing fluff. Real-world performance is a negotiation between physics and harsh reality. Let's cut to the chase.
Experience Correction: Where Theory Meets the Grime
The published repeatability and accuracy figures? They assume a perfect, static target. In practice, thermal drift in laser triangulation sensors is routinely 2-3x worse than spec when mounted on machinery cycling from 10°C to 45°C. I've logged data showing a 0.1mm-accuracy unit drifting 0.25mm over an 8-hour shift in a non-climate-controlled factory. The fix isn't in the brochure: you must implement automated re-referencing against a physical datum at defined intervals, not just at startup.
For ultrasonic sensors, the stated beam angle is a half-power point. The actual acoustic energy extends much wider, causing phantom echoes from machine frames or conveyors the marketing images conveniently omit. I've seen this cause false triggers on "empty" production lines. The solution is empirical mapping during commissioning: chart the actual dead zones and echo profiles in situ, not on the test bench.

Boundary Conditions: When to Walk Away
These technologies have hard failure modes. Knowing them prevents catastrophic design flaws.
1. Laser Triangulation on Clear or Molten Targets: It will fail. The laser passes through clear glass or plastic, or plunges into a molten polymer surface, returning a signal from a background object or the container bottom. I've witnessed this ruin a high-speed bottling line. For these applications, confocal chromatic or specific polarized laser sensors are the only viable choice, despite their cost.
2. Ultrasonics in Vacuum or High-Velocity Airflow: No medium, no measurement. This is obvious. Less obvious is performance degradation in turbulent, high-velocity air streams (e.g., near exhaust vents), which refract and scatter sound waves, causing signal loss or massive noise. If airflow exceeds 5 m/s across the transducer face, reliability plummets.
3. Inductive/Capacitive Proximity in Slurry or Flakes: Inductive sensors detecting metal through a non-metallic medium work until that medium becomes conductive. Wet wood pulp, carbon-rich slurry, or metallic dust accumulation creates a parasitic capacitance or eddy current path that either masks the target or causes permanent false detection. I specify a minimum 5:1 ratio of required sensing distance to contaminant layer thickness as a rule of thumb. If you can't guarantee that, use a mechanically isolated, wiper-sealed mechanical limit switch instead.
Counter-Intuitive Conclusion: Sometimes Less Resolution is More
The industry's relentless drive for higher resolution and faster response can be a trap. In vibration-heavy environments (mills, presses, stamping), a 10µm-resolution laser sensor operating at 5kHz will output a torrent of meaningless noise, overwhelming controllers and causing system instability. The intuitive response is to add dampening or shielding, often with limited success.
The data-driven, counter-intuitive fix is to deliberately degrade performance. By configuring a sensor to 50µm resolution and applying a heavier internal digital filter (increasing response time from 0.2ms to 8ms), you filter the mechanical noise at the source. The system sees a stable, slightly lagged signal, enabling robust control. I've applied this to steel coil diameter measurement on a vibrating mandrel, improving overall line stability by 40%. The "inferior" sensor specification created a superior result. The lesson: match the sensor's capability to the signal-to-noise reality of the application, not the other way around. The highest spec is often the least stable.