The deviation model — finding the odd cylinder

Sixteen exhaust temperatures overlaid on one axis is sixteen lines in a ribbon: you can see the ribbon move, but not which line is wrong. The Deviation model view answers the other question — which member of this family is not behaving like the rest of it.

What a family is

A family is a set of channels whose names differ only by a number — “Cyl 1 Exhaust Temp” through “Cyl 16 Exhaust Temp”, “TC-1” through “TC-8”, “Rod Load 1” through “Rod Load 4”. Overtrace finds them from the channel names in the log you loaded; nothing has to be configured, and a family needs at least three members before it is offered.

1. Open the View menu in the top bar and choose “Deviation model”.
2. Every family in every loaded log is laid out one under the next — scroll to the one you want; there is nothing to select first.
3. Each family's header names it, counts its members, and says whether anything is outside the band. Use the caret to roll a family up.
4. Read the number in each tile's top-right: that member's average distance from the family mean over the visible time range.
5. Tiles outside the band are outlined and coloured — red for running hot, blue for running cold.
6. Click any tile to plot that channel on the Trend view and go straight there.

The two y-axis modes

• Shared range — every tile is drawn on one pooled range clamped to the 1st–99th percentile of the whole family, with the family mean behind each member as a grey line. Amplitudes are directly comparable between tiles.
• Δ from mean — each tile shows the member minus the family mean around a zero line. Better for spotting a member that drifts apart only under load.

How the band is decided

Each member's average deviation is compared against two standard deviations of the spread across the family, with a floor at 3% of the pooled range. The floor matters: a genuinely even bank has a spread near zero, and without it every member would be flagged over rounding noise. A healthy bank flags nothing.

The baseline — the second, more sensitive verdict

Comparing a cylinder against its peers has one blind spot, and it is the important one: the peers include the cylinder that is going bad, so the mean walks along with the fault. If every member of the bank creeps up ten degrees over a month, a peer comparison sees nothing at all. So each member is also compared against ITSELF.

• Overtrace finds the earliest steady running window in the log — every member sitting still, no drift across the window, and the unit actually turning — and stores each member's average there as its baseline. The window is shown in the family header.
• That baseline is compared against the most recent steady running window in the range you are looking at. Two steady windows, so days of downtime in between cannot skew the number — and on a unit whose fingerprint holds to half a degree, the drift band is a fraction of the peer band.
• “Δ from baseline” draws each member against its own baseline with the drift band dotted in. Members that have drifted are marked amber; a member that is peer-clean but drifting carries a small corner mark in the other modes too.
• “Re-baseline to view” takes the baseline from the range on screen, for when you know a particular stretch was normal — after a valve job, for instance.

What it is measuring, and what it is not

• Only the visible time range is modelled. Zoom into a load step and the numbers recompute for that window.
• The comparison is against the other members in the same family on the same module — not against a factory spec, and not across modules.
• Averages come from bucketed samples, so a 300,000-row log paints as fast as a 3,000-row one. The peak Δ line under each tile is the largest excursion found in the range, with the time it happened.
• “Copy summary” puts the whole table — every member, its average Δ, its peak Δ, and which members are outside the band — on the clipboard for a report or an email.

FAQ