Technical notes / Healthy baseline
A healthy-machine baseline is only half of condition monitoring
What baselines and σ-based thresholds can show in condition monitoring, and what they cannot.
A baseline is a record of how a machine behaves when it is known to be healthy. New measurements are compared with it. In my thesis prototype, baselines were recorded with no load at discrete spindle speeds, because vibration depends strongly on speed.
From baseline to threshold
A common way to turn a baseline into a rule is to use its mean μ and standard deviation σ. Thresholds at μ + 2σ and μ + 3σ, as defined in the thesis, flag values that are unusually high compared with the healthy record. Z-score normalisation expresses a value as the number of standard deviations from the mean.
If healthy values were roughly Gaussian, a one-sided μ + 3σ threshold would be exceeded by about 0.13% of healthy samples, and a μ + 2σ threshold by about 2.3%. Choosing the threshold is a trade-off between false alarms and missed events.
What a healthy baseline cannot show
The prototype was tested on a healthy machine only. A baseline shows when behaviour departs from healthy. It does not show what a particular fault looks like, and real-fault detection was not demonstrated. To go further, you would need recordings from real fault conditions.
The prototype applies rule-based diagnostic logic to the features and trends them in CSV files. It is a prototype for a university lab machine, not a production monitoring system.