Sensor fusion combines readings from multiple sensing channels into a single assessment, so that patterns spanning several channels become detectable.
Some hazards exist only in the relationship between channels. A machine failing shows heat, vibration and sound changes together, often with no single channel breaching anything. Independent detectors report nothing wrong, correctly and uselessly.
Channels are sampled on a shared timebase at a shared location, normalised so they are comparable, then combined - by rule, by weighting, or by comparing the whole vector against known patterns.
Rising temperature plus a changed vibration signature plus a changed acoustic profile, on one machine, in one minute.
All channels sit on one board sampling together, which is what makes correlation between them meaningful rather than approximate.
