A health score exists to answer one question fast, across a fleet of hundreds of assets: which ones need attention first? Rather than reading through failure logs asset by asset, a maintenance planner scans a sorted list and starts at the bottom.
The inputs that make a health score useful are the same ones a good engineer would weigh mentally — asset age relative to expected life, recent failure frequency, count of open faults, PM compliance, and (where available) live condition data like vibration or temperature trend. A score that ignores live telemetry is really just a static criticality rating with a number attached.