We're working on the implementation of Nelson Rules in R. I think this is exactly what you're looking for (happy to share, I couldn't find an R implementation anywhere else on the internet):
nelsonr1 <- function(x, m = mean(x), s = sd(x)) {
# Nelson's QC rule 1: detect values outside + or -3 sd
which(abs((x - m) / s) >= 3)
}
nelsonr2 <- function(x, m = mean(x), minrun = 9) {
# Nelson's QC rule 2: detect runs of >= 9 points on the same side of the mean
n <- length(x)
counts <- sign(x - m)
result <- counts
for (runlength in 2:minrun)
result <- result + c(counts[runlength:n], rep(0, runlength - 1))
which(abs(result) >= minrun)
}
nelsonr3 <- function(x, minrun = 6) {
# Nelson's QC rule 3: detect strict increase or decrease in >= 6 points in a row
# Between 6 points you have 5 instances of increasing or decreasing. Therefore minrun - 1.
n <- length(x)
signs <- sign(c(x[-1], x[n]) - x)
counts <- signs
for (rl in 2:(minrun - 1)) {
counts <- counts + c(signs[rl:n], rep(0, rl - 1))
}
which(abs(counts) >= minrun - 1)
}
nelsonr4 <- function(x, m = mean(x), minrun = 14, directing_from_mean = FALSE) {
# Nelson's QC rule 4: 14 points in a row alternating in direction from the mean,
# or 14 points in a row alternating in increase and decrease
n <- length(x)
if (directing_from_mean == TRUE) {
signs <- sign(x - m)
} else {
signs <- sign(c(x[-1],x[n]) - x)
}
counts <- signs
fac <- -1
for (rl in 2:minrun) {
counts <- counts + fac * c(signs[rl:n], rep(0, rl - 1))
fac <- -fac
}
counts <- abs(counts)
which(counts >= minrun)
}
nelsonr5 <- function(x, m = mean(x), s = sd(x), minrun = 3) {
# Nelson's QC rule 5: two out of 3 >2 sd from mean in the same direction
n <- length(x)
pos <- 1 * ((x - m) / s > 2)
neg <- 1 * ((x - m) / s < -2)
poscounts <- pos
negcounts <- neg
for (rl in 2:minrun) {
poscounts <- poscounts + c(pos[rl:n], rep(0, rl - 1))
negcounts <- negcounts + c(neg[rl:n], rep(0, rl - 1))
}
counts <- apply(cbind(poscounts, negcounts), 1, max)
which(counts >= minrun -1)
}
nelsonr6 <- function(x, m = mean(x), s = sd(x), minrun = 5) {
# Nelson's QC rule 6: four out of five > 1 sd from mean in the same direction
n <- length(x)
pos <- 1 * ((x - m) / s > 1)
neg <- 1 * ((x - m) / s < -1)
poscounts <- pos
negcounts <- neg
for (rl in 2:minrun) {
poscounts <- poscounts + c(pos[rl:n], rep(0, rl - 1))
negcounts <- negcounts + c(neg[rl:n], rep(0, rl - 1))
}
counts <- apply(cbind(poscounts, negcounts), 1, max)
which(counts >= minrun - 1)
}
nelsonr7 <- function(x, m = mean(x), s = sd(x), minrun = 15) {
# Nelson's QC rule 7: >= 15 points in a row within 1 sd from the mean
n <- length(x)
within <- 1 * (abs((x - m) / s) < 1)
counts <- within
for (rl in 2:minrun)
counts <- counts + c(within[rl:n], rep(0, rl - 1))
which(counts >= minrun)
}
nelsonr8 <- function(x, m = mean(x), s = sd(x), minrun = 8) {
# Nelson's QC rule 8: >= 8 points in a row all outside the m + -1s range
n <- length(x)
outofrange <- 1 * (abs((x - m) / s) > 1)
counts <- outofrange
for (rl in 2:minrun)
counts <- counts + c(outofrange[rl:n], rep(0, rl - 1))
which(counts >= minrun)
}
For example where the referenced article says "Eight points in a row.." I would like eight to be a parameter.
That's what this does too with the parameter minrun in some functions.
