edcc (version 1.0-0)

ecoCusum: Economic design for the CUSUM control chart

Description

Calculate the optimum parameters of n(sample size), h(sampling interval), k(reference value) and H(decision interval) for Economic Design of the CUSUM control chart. For more information about the reference value see 'Details'.

Usage

ecoCusum(h, H, n, delta = 2, lambda = 0.01, P0 = NULL, P1 = NULL, C0 = NULL, C1 = NULL, Cr = 20, Cf = 10, T0 = 0, Tc = 0.1, Tf = 0.1, Tr = 0.2, a = 0.5, b = 0.1, d1 = 1, d2 = 1, nlevels = 30, sided = "one", par = NULL, contour.plot = FALSE, call.print = TRUE, ...)
  echCusum(h, H, n, delta = 2, lambda = 0.01, P0 = NULL, P1 = NULL, C0 = NULL, C1 = NULL, Cr = 20, Cf = 10, T0 = 0, Tc = 0.1, Tf = 0.1, Tr = 0.2, a = 0.5, b = 0.1, d1 = 1, d2 = 1, sided = "one")

Arguments

h

sampling interval. It can be a numeric vector or left undefined. See 'Details'

H

decision interval. It can be a numeric vector or left undefined. See 'Details'

n

sample size. It can be an integer vector or left undefined. See 'Details'

delta

shift in process mean in standard deviation units when assignable cause occurs (delta = |mu1 - mu0|/sigma), where sigma is the standard deviation of observations; mu0 is the in-control process mean; mu1 is the out-of-control process mean. Default value is 2.

lambda

we assume the in-control time follows a exponential distribution with mean 1/lambda. Default value is 0.05.

P0

profit per hour earned by the process operating in control. See 'Details'.

P1

profit per hour earned by the process operating out of control

C0

cost per hour due to nonconformities produced while the process is in control.

C1

cost per hour due to nonconformities produced while the process is out of control.(C1 > C0)

Cr

cost for searching and repairing the assignable cause, including any downtime.

Cf

cost per false alarm, including the cost of searching for the cause and the cost of downtime if production ceases during search.

T0

time to sample and chart one item.

Tc

expected time to discover the assignable cause.

Tf

expected search time when false alarm occurs.

Tr

expected time to repair the process.

a

fixed cost per sample.

b

cost per unit sampled.

d1

flag for whether production continues during searches (1-yes, 0-no). Default value is 1.

d2

flag for whether production continues during repairs (1-yes, 0-no). Default value is 1.

nlevels

30. It works only when contour.plot is TRUE.

sided

distinguish between one-, two-sided and Crosier's modified two-sided CUSUM scheme by choosing "one", "two", and "Crosier", respectively. See details in xcusum.arl

par

initial values for the parameters to be optimized over. It can be a vector of length 2 or 3. See 'Details'

contour.plot

a logical value indicating wether a contour plot should be drawn. Default is FALSE.

call.print

a logical value indicating whether the "call" should be printed on the contour plot. Default is TRUE

...

other arguments to be passed to optim function.

Value

The ecoCusum function returns an object of class "edcc", which is a list of elements optimum, cost.frame, FAR and ATS. optimum is a vector with the optimum parameters and the corresponding ECH value; cost.frame is a dataframe with the optimum parameters and the corresponding ECH values for all given n(if n is not specified, cost.frame won't be returned); FAR indicates the false alarm rate during the in-control time, which is calculated as lambda*(average number of false alarm); ATS indicates the average time to signal after the occurrence of an assignable cause, calculated as h*ARL2 - tau, where tau is the expected time of occurrence of the assignable cause given it occurs between the i-th and (i+1)st samples. The echCusum function returns the calculated ECH value only.

Details

When parameter par is specified, optimization algorithms would be used as default. par can be specified as: par = c(h, H) where h and H are the initial values of smapling interval and decision interval when n is specified; or par = c(h, H, n). Good inital values may lead to good optimum results.

When parameters h, H, n are all undefined, ecoCusum function will try to find the global optimum point to minimize the ECH (Expected Cost per Hour) using optimization algorithms (optim function), but in this case n would not be integer. It is usually helpful for the experimenter to find the region where the optimum point may exist quickly. When h and H are undefined but n is given as an integer vector, ecoCusum function will try to find the optimum point for each n value using optimization algorithms. When h, H and n are all given, ecoCusum function will use a "grid method" way to calculate the optimum point, that is ECH for all the combinations of the parameters will be calculated. The "grid method" way is much slower than using optimization algorithms, but it would be a good choice when optimization algorithms fail to work well.

There is strong numerical and theoretical evidence that for given L1, the value of L0 approaches its maximum when k(reference value) is chosen mid-way the between AQL and the RQL: $k = mu0 + 0.5*delta*sigma (Appl. Statist.(1974) 23, No. 3, p. 420). For this reason we treat k as a constant value and optimize n, h and H. For cost parameters either P0, P1 or C0, C1 is needed. If P0 and P1 are given, they will be used first, else C0 and C1 will be used. For economic design of the CUSUM chart, when d1 and d2 are both 1, only if the difference between P0 and P1 keeps the same, the results are identical. If the difference between C0 and C1 keeps the same, the optimum parameters are almost the same but the ECH(Expected Cost per Hour) values will change.

echCusum is used to calculate the ECH (Expected Cost per Hour) for one given design point.

References

Weicheng Zhu, Changsoon Park (2013), edcc: An R Package for the Economic Design of the Control Chart. Journal of Statistical Software, 52(9), 1-24. http://www.jstatsoft.org/v52/i09/

Lorenzen and Vance (1986). The economic design of control charts: a unified approach, Technometrics, 28. 3-10.

Chiu, W.K. (1974). The economic design of CUSUM charts for controlling normal means, Journal of the Royal Statistical Society. Series C (Applied Statistics), 23(3), 420-433.

See Also

ecoXbar, ecoEwma, xcusum.arl, optim, update.edcc, contour

Examples

#Chiu, W.K. (1974). Applied Statistics, 23, p427 Table3, row 1-2,14
## LINE 1
## global optimization to h, H and n, when lambda = 0.01, "Nelder-Mead" optimization algorithm doesn't work
#(y <- ecoCusum( P0=150,P1=50,Cr=30,d1=0,d2=0))
## we can try other algorithms:
(y1 <- ecoCusum( P0=150,P1=50,Cr=30,d1=0,d2=0,method="BFGS"))
# Based on the global optimum above, we specify the range of the
# parameters like this
(yy1 <- ecoCusum( h=seq(1.3,1.45,by=.01), H=seq(.5,0.6,by=.01),n=4:6,
P0=150,P1=50,Cr=30,d1=0,d2=0))
## LINE 2
(y2 <- ecoCusum( P0=150,P1=50,Cr=30,d1=0,d2=0,lambda=0.05))
(yy2 <- ecoCusum( h=seq(.6,0.7,by=.01), H=seq(.5,0.6,by=.01),n=3:6,
P0=150,P1=50,Cr=30,d1=0,d2=0,lambda=0.05))
contour(yy2)
## LINE 14
(y14 <- ecoCusum(n=30,P0=150,P1=50,Cr=30,delta=0.5,d1=0,d2=0,method="L-BFGS-B"))
(yy14 <- ecoCusum(h=seq(2.55,2.65,by=0.01),H=seq(0.3,0.4,by=0.01),
n=28:30,P0=150,P1=50,Cr=30,delta=0.5,d1=0,d2=0))
#Douglas (2009). Statistical quality control: a modern introduction, sixth edition, p470.
ecoCusum(lambda=.05,P0=110,P1=10,Cr=25,Cf=50,Tr=0,Tf=0,Tc=1,T0=.0167,a=1)
ecoCusum(h=seq(0.75,0.85,by=.01),H=seq(.55,0.65,by=.01),n=4:6,lambda=.05,
P0=110,P1=10,Cr=25,Cf=50,Tr=0,Tf=0,Tc=1,T0=.0167,a=1)