PlantOps BI

XmR Chart

A process behavior chart answers the question a trend line can't: is this week different, or is it noise? Limits computed from the 53 weeks starting in 2024 flag 13 weeks of 2025 as a real change. They also show the line was never steady to begin with.

BOI-03 weekly efficiency (units produced / target). Shaded: the 2024 baseline. Dashed: natural process limits. Filled points are outside them.
40%45%50%55%60%202420252026Upper 57.2%Center 53.0%Lower 48.9%0Range limit 5.1%Moving range
40%50%60%202420252026Upper 57.2%Center 53.0%Lower 48.9%0Range limit 5.1%Moving range

How the Limits Work

The center is the mean of the baseline weeks: 53.0%. The average moving range, the mean absolute change from one week to the next, is 1.6%. The natural process limits are the center plus and minus 2.66 average moving ranges, here 48.9% to 57.2%. The 2.66 is 3 / 1.128, the constant that turns an average range of two points into three standard deviations.

Using the moving range, not the overall standard deviation, is the point. A shift in the middle of the data inflates the standard deviation and widens the limits until the shift hides inside them. Week-to-week ranges barely notice a sustained shift.

Reading This One

Start with the baseline, because the chart does. 19 of the 53 weeks starting in 2024 fall outside limits computed from those same weeks. A predictable process rarely does that. The demo data builds in two causes that move a line without anyone touching it: seasonal output (a push in the autumn and a summer dip) and a slow month-to-month drift. 6 of the 9 high weeks of 2024 fall from September on. Seasonality is an assignable cause too, and the chart is right to flag it.

Against that, the 2025 trouble still stands out. From Feb 3, 2025 to Apr 28, 2025, 13 consecutive weeks sit below 48.9%. They average 43.5%, and the worst, 40.5%, is lower than any week of 2024 (the lowest was 44.9%). That is the equipment problem showing up as a signal, not an opinion.

After the rebuild, 53 of the 70 later weeks run above the 2024 center and 17 sit above the upper limit, but 12 of those 17 fall from September on, the same season that lifted 2024. Before calling it a step up, compare like with like: the same weeks a year apart. If the gain holds outside the busy season, start a new baseline from the new weeks and write the date of the change on the chart.

Fix the baseline and leave it until the process changes on purpose. Limits that recompute over all the data drift toward the problem they should flag.

The DAX

The measures go on a line chart with dim_date[week_start] on the axis and the line picked by a slicer. The baseline measures clear the visual's date filter and apply their own, so every week sees the same limits. The line filter stays.

Efficiency % =
DIVIDE (
    SUM ( fact_production_hourly[units_produced] ),
    SUM ( fact_production_hourly[units_target] )
)

-- Baseline: the weeks starting in 2024 (the last one ends 5 Jan 2025).
-- Limits stay fixed while new weeks arrive.
XmR Center =
IF (
    NOT ISBLANK ( [Efficiency %] ),
    CALCULATE (
        AVERAGEX ( VALUES ( dim_date[week_start] ), [Efficiency %] ),
        REMOVEFILTERS ( dim_date ),
        dim_date[date] >= DATE ( 2024, 1, 1 ),
        dim_date[date] <= DATE ( 2025, 1, 5 )
    )
)

XmR Avg Moving Range =
CALCULATE (
    VAR Weeks =
        FILTER (
            ADDCOLUMNS ( VALUES ( dim_date[week_start] ), "@v", [Efficiency %] ),
            NOT ISBLANK ( [@v] )
        )
    VAR Ranges =
        ADDCOLUMNS (
            Weeks,
            "@mr",
                VAR ThisWeek = dim_date[week_start]
                VAR PrevWeek =
                    MAXX ( FILTER ( Weeks, dim_date[week_start] < ThisWeek ), dim_date[week_start] )
                VAR PrevValue =
                    MAXX ( FILTER ( Weeks, dim_date[week_start] = PrevWeek ), [@v] )
                RETURN
                    IF ( NOT ISBLANK ( PrevWeek ), ABS ( [@v] - PrevValue ) )
        )
    RETURN
        AVERAGEX ( Ranges, [@mr] ),
    REMOVEFILTERS ( dim_date ),
    dim_date[date] >= DATE ( 2024, 1, 1 ),
    dim_date[date] <= DATE ( 2025, 1, 5 )
)

-- Center and limits are blank on weeks with no data, so the lines stop where the data does
XmR Upper Limit =
IF (
    NOT ISBLANK ( [Efficiency %] ),
    [XmR Center] + 2.66 * [XmR Avg Moving Range]
)

XmR Lower Limit =
IF (
    NOT ISBLANK ( [Efficiency %] ),
    [XmR Center] - 2.66 * [XmR Avg Moving Range]
)

-- 1 on weeks outside the limits, for a marker or conditional color
XmR Signal =
VAR v = [Efficiency %]
RETURN
    IF ( NOT ISBLANK ( v ) && ( v > [XmR Upper Limit] || v < [XmR Lower Limit] ), 1 )

Inside XmR Avg Moving Range, AVERAGEX skips the first week's blank range, so the divisor is one less than the number of weeks, as the formula needs. The figures on this page come from the same math in the site's code, and its tests check the limits against a hand-worked example.

Downtime Pareto · OEE across lines · About the demo dataset