Detecting deviations in cooking cycles

Hi ​@TIM,

Fingerprint is the right tool, but the bit that actually makes "automatically compare every future cycle" work is the trigger, and it's the part most people miss. Full mechanic:

1. Find and rank your good cycles, then build the fingerprint from them as overlaid layers. Start with a value-based search that returns every cooking cycle (your CycleActive search). Then attach a quality calculation so each cycle gets a score you can sort on final product temperature, energy per batch, peak-temp-in-spec, whatever "good" means for you — and refine down to the best dozen or so in Event Analytics. (This is exactly how the product-loss use case narrowed to ~15 batches: a VBS for the runs, then calculations like max concentration and average temperature, sorted in a histogram / parallel-coordinates plot.) Load that refined set into the focus chart as layers (each aligned at its start), add the tags that define healthy behaviour temperature, steam pressure, valve position — and create the fingerprint. The hull is simply the min/max of all those layers at each timestamp from the start: the band your good cycles stayed inside as the cycle progressed (fingerprints).

One catch that matters here: a fingerprint monitor only evaluates the visual tags. Hidden tags get checked by the on-demand Diagnose → Fingerprint deviations tool but are ignored by the monitor so any tag you want to alert on must be a visible tag in the fingerprint.

2. Make a cycle-start search this becomes the trigger. A deviation monitor needs a trigger that defines where each comparison begins; the trigger and the fingerprint are then aligned from that point. Reuse the cycle-detection search from your cycle-identification work a VBS on your CycleActive tag, or a digital-step search on the step tag marking "cooking begins." That search's start is where the fingerprint gets laid down for each new cycle (setting up a monitor).

3. Set up the deviation monitor. Promote the fingerprint, choose Detect deviations, and point it at that trigger search. Every 2 minutes it evaluates live data the deviation score is a similarity measure built from the sum of residuals to the hull, and the live threshold is computed assuming the rest of the cycle stays perfectly inside the band. That's what gives the early warning: a cycle drifting out mid-way trips before it finishes, not after. On a trip it emails the team with the parameter that left the band.

For the full build end-to-end, the Craft and Adopt Your Golden Batch (Profile) VILT recording walks through fingerprint creation, deviation detection, and the monitor setup.

Kind regards
Frederik