Which control chart should I use?
SenSight offers five chart types and the choice is not cosmetic — the wrong chart either misses a real problem or cries wolf.
I-MR — you measure one piece at a time (a batch temperature, a daily reading). Sigma is estimated from the moving range between consecutive points. This is the default and it is right whenever subgrouping is impossible.
X̄-R — you measure several pieces at each check (5 parts every hour, 3 samples per shift). The chart tracks the subgroup means, and a second chart tracks the range inside each subgroup. This is the most used chart in manufacturing, because averaging cancels measurement noise and makes real shifts visible much earlier than I-MR.
X̄-S — same idea, but for larger subgroups (n > 8). With many measurements the standard deviation is a better spread estimate than the range.
EWMA and CUSUM exist for one reason: a Shewhart chart (I-MR, X̄-R) is deliberately blind to small shifts. A drift of half a sigma can run for months without a single point crossing 3 sigma. EWMA weights recent points and slowly bends; CUSUM accumulates the deviation until it crosses a decision limit. Use them when the process is stable but you suspect slow drift — tool wear, sensor ageing, a slowly clogging filter.
A warning about subgroups: a subgroup must be a set of measurements taken under the same conditions, close in time. If you group five parts made on five different machines, the range inside the subgroup contains the machine-to-machine difference — the limits widen, and the chart stops detecting anything. Group by what is common, then compare groups.
I-MR — you measure one piece at a time (a batch temperature, a daily reading). Sigma is estimated from the moving range between consecutive points. This is the default and it is right whenever subgrouping is impossible.
X̄-R — you measure several pieces at each check (5 parts every hour, 3 samples per shift). The chart tracks the subgroup means, and a second chart tracks the range inside each subgroup. This is the most used chart in manufacturing, because averaging cancels measurement noise and makes real shifts visible much earlier than I-MR.
X̄-S — same idea, but for larger subgroups (n > 8). With many measurements the standard deviation is a better spread estimate than the range.
EWMA and CUSUM exist for one reason: a Shewhart chart (I-MR, X̄-R) is deliberately blind to small shifts. A drift of half a sigma can run for months without a single point crossing 3 sigma. EWMA weights recent points and slowly bends; CUSUM accumulates the deviation until it crosses a decision limit. Use them when the process is stable but you suspect slow drift — tool wear, sensor ageing, a slowly clogging filter.
A warning about subgroups: a subgroup must be a set of measurements taken under the same conditions, close in time. If you group five parts made on five different machines, the range inside the subgroup contains the machine-to-machine difference — the limits widen, and the chart stops detecting anything. Group by what is common, then compare groups.
Try it in the app
Try: load "Steel — Extended", pick X̄-R, set subgroup size 5, and compare with I-MR on the same column. Then switch to EWMA and watch what appears that the Shewhart chart could not see.