You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
497 lines
19 KiB
R
497 lines
19 KiB
R
# CHI.R
|
|
# Functions to read and manipulate data from the CHI760 potentiostat/galvanostat
|
|
# Taha Ahmed, Jan 2011 - Feb 2011
|
|
|
|
# CONTENTS
|
|
source("/home/taha/chepec/chetex/common/R/common.R")
|
|
# >>>> ocp2df
|
|
# >>>> chronocm2df
|
|
# >>>> chronoamp2df
|
|
# >>>> amperometry2df
|
|
# >>>> cv2df
|
|
# >>>> lsv2df
|
|
|
|
|
|
|
|
|
|
##################################################
|
|
################### ocp2df #######################
|
|
##################################################
|
|
ocp2df <- function(datafilename) {
|
|
# Function description: for recorded amperometric i-T curves
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
datafile <- file(datafilename, "r")
|
|
chifile <- readLines(datafile, n = -1) #read all lines of input file
|
|
close(datafile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+[e,]"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by either the letter 'e' or a comma.
|
|
# Note that backslashes are escaped due to the way R handles strings.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 2, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "time", "potential")
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Run\\sTime\\s\\(sec\\)")
|
|
names.attr <- c("RunTime")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
# attrow.idx should now contain only one matching row
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|
|
|
|
|
|
|
|
##################################################
|
|
################# chronocm2df ####################
|
|
##################################################
|
|
chronocm2df <- function(datafilename) {
|
|
# Function description: chronocoulometry data
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
datafile <- file(datafilename, "r")
|
|
chifile <- readLines(datafile, n = -1) #read all lines of input file
|
|
close(datafile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+[e,]"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by either the letter 'e' or a comma.
|
|
# Note that backslashes are escaped.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, step = factor(s),
|
|
matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 2, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "step", "time", "charge")
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Init\\sE\\s\\(V\\)",
|
|
"^Final\\sE\\s\\(V\\)",
|
|
"^Step\\s",
|
|
"^Pulse\\sWidth\\s\\(sec\\)",
|
|
"^Sample\\sInterval\\s\\(s\\)",
|
|
"^Quiet\\sTime\\s\\(sec\\)",
|
|
"^Sensitivity\\s\\(A/V\\)")
|
|
names.attr <- c("InitE",
|
|
"FinalE",
|
|
"Steps",
|
|
"PulseWidth",
|
|
"SamplingInterval",
|
|
"QuietTime",
|
|
"Sensitivity")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|
|
|
|
|
|
|
|
##################################################
|
|
################# chronoamp2df ###################
|
|
##################################################
|
|
chronoamp2df <- function(datafilename, wearea = 1) {
|
|
# Function description: chronoamperometry data
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
datafile <- file(datafilename, "r")
|
|
chifile <- readLines(datafile, n = -1) #read all lines of input file
|
|
close(datafile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+[e,]"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by either the letter 'e' or a comma.
|
|
# Note that backslashes are escaped.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, step = factor(s),
|
|
matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 2, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "step", "time", "current")
|
|
# Calculate current density
|
|
currentdensity <- ff$current / wearea
|
|
ff <- cbind(ff, currentdensity = currentdensity)
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Init\\sE\\s\\(V\\)",
|
|
"^High\\sE\\s\\(V\\)",
|
|
"^Low\\sE\\s\\(V\\)",
|
|
"^Init\\sP/N",
|
|
"^Step\\s",
|
|
"^Pulse\\sWidth\\s\\(sec\\)",
|
|
"^Sample\\sInterval\\s\\(s\\)",
|
|
"^Quiet\\sTime\\s\\(sec\\)",
|
|
"^Sensitivity\\s\\(A/V\\)")
|
|
names.attr <- c("InitE",
|
|
"HighE",
|
|
"LowE",
|
|
"InitPN",
|
|
"Steps",
|
|
"PulseWidth",
|
|
"SamplingInterval",
|
|
"QuietTime",
|
|
"Sensitivity")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|
|
|
|
|
|
|
|
##################################################
|
|
############### amperometry2df ###################
|
|
##################################################
|
|
amperometry2df <- function(datafilename, wearea = 1) {
|
|
# Function description: for recorded amperometric i-T curves
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
datafile <- file(datafilename, "r")
|
|
chifile <- readLines(datafile, n = -1) #read all lines of input file
|
|
close(datafile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+[e,]"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by either the letter 'e' or a comma.
|
|
# Note that backslashes are escaped due to the way R handles strings.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 2, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "time", "current")
|
|
# Calculate current density
|
|
currentdensity <- ff$current / wearea
|
|
ff <- cbind(ff, currentdensity = currentdensity)
|
|
# Calculate charge densities and differentials
|
|
charge.df <- It2charge(ff$currentdensity, ff$time)
|
|
ff <- cbind(ff, timediff = charge.df$timediff,
|
|
chargedensity = charge$charge,
|
|
sumchargedensity = charge$sumcharge,
|
|
dQdt = charge$dQdt)
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Init\\sE\\s\\(V\\)",
|
|
"^Sample\\sInterval\\s\\(s\\)",
|
|
"^Run\\sTime\\s\\(sec\\)",
|
|
"^Quiet\\sTime\\s\\(sec\\)",
|
|
"^Sensitivity\\s\\(A/V\\)")
|
|
names.attr <- c("InitE",
|
|
"SamplingInterval",
|
|
"RunTime",
|
|
"QuietTime",
|
|
"Sensitivity")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
# attrow.idx should now contain only one matching row
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|
|
|
|
|
|
|
|
##################################################
|
|
#################### cv2df #######################
|
|
##################################################
|
|
cv2df <- function(datafilename, wearea = 1) {
|
|
# Function description:
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
cvfile <- file(datafilename, "r")
|
|
chifile <- readLines(cvfile, n = -1) #read all lines of input file
|
|
close(cvfile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+,"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by a comma.
|
|
# Note that backslashes are escaped.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, segment = factor(s), cycle = factor(ceiling(s/2)),
|
|
matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 3, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "segment", "cycle", "potential", "current", "charge")
|
|
# Calculate current density
|
|
currentdensity <- ff$current / wearea
|
|
ff <- cbind(ff[, 1:5], currentdensity = currentdensity, charge = ff[, 6])
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Init\\sE\\s\\(V\\)",
|
|
"^High\\sE\\s\\(V\\)",
|
|
"^Low\\sE\\s\\(V\\)",
|
|
"^Init\\sP/N",
|
|
"^Scan\\sRate\\s\\(V/s\\)",
|
|
"^Segment\\s=",
|
|
"^Sample\\sInterval\\s\\(V\\)",
|
|
"^Quiet\\sTime\\s\\(sec\\)",
|
|
"^Sensitivity\\s\\(A/V\\)")
|
|
names.attr <- c("InitE",
|
|
"HighE",
|
|
"LowE",
|
|
"InitPN",
|
|
"ScanRate",
|
|
"Segments",
|
|
"SamplingInterval",
|
|
"QuietTime",
|
|
"Sensitivity")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|
|
|
|
|
|
|
|
|
|
##################################################
|
|
################### lsv2df #######################
|
|
##################################################
|
|
lsv2df <- function(datafilename, wearea = 1) {
|
|
# Function description:
|
|
# CH Instruments potentiostat records all data using standard SI units,
|
|
# so all potential values are in volts, currents are in amperes,
|
|
# charges in Coulombs, time in seconds, etc.
|
|
#
|
|
lsvfile <- file(datafilename, "r")
|
|
chifile <- readLines(lsvfile, n = -1) #read all lines of input file
|
|
close(lsvfile)
|
|
#
|
|
sampleid <- ProvideSampleId(datafilename)
|
|
#
|
|
rgxp.number <- "^\\-?\\d\\.\\d+,"
|
|
# regexp that matches a decimal number at the beginning of the line.
|
|
# Matches numbers with or without a negative sign (hyphen),
|
|
# followed by one digit before the decimal, a decimal point,
|
|
# and an arbitrary number of digits after the decimal point,
|
|
# immediately followed by a comma.
|
|
# Note that backslashes are escaped.
|
|
#
|
|
numrow.idx <- regexpr(rgxp.number, chifile)
|
|
# Save the match length attribute to another variable,
|
|
numrow.len <- attr(numrow.idx, "match.length")
|
|
# then scrap the attribute of the original variable.
|
|
attr(numrow.idx, "match.length") <- NULL
|
|
#
|
|
i <- seq(1, length(numrow.idx) - 1, 1)
|
|
j <- seq(2, length(numrow.idx), 1)
|
|
# Start indices of data ranges
|
|
starts <- which(numrow.idx[i] != 1 & numrow.idx[j] == 1) + 1
|
|
# End indices, except for the last
|
|
ends <- which(numrow.idx[i] == 1 & numrow.idx[j] != 1)
|
|
# Fix the last index of end indices
|
|
ends <- c(ends, length(numrow.idx))
|
|
#
|
|
ff <- data.frame(NULL)
|
|
for (s in 1:length(starts)) {
|
|
zz <- textConnection(chifile[starts[s]:ends[s]], "r")
|
|
ff <- rbind(ff,
|
|
data.frame(sampleid, segment = factor(s),
|
|
matrix(scan(zz, what = numeric(), sep = ","),
|
|
ncol = 3, byrow = T)))
|
|
close(zz)
|
|
}
|
|
names(ff) <- c("sampleid", "segment", "potential", "current", "charge")
|
|
# Calculate current density
|
|
currentdensity <- ff$current / wearea
|
|
ff <- cbind(ff[, 1:4], currentdensity = currentdensity, charge = ff[, 5])
|
|
#
|
|
### Collect attributes of this experiment
|
|
# These attributes are specific for each kind of experiment,
|
|
# be careful when adapting to other electrochemical data
|
|
rgxp.attr <- c("^Init\\sE\\s\\(V\\)",
|
|
"^Final\\sE\\s\\(V\\)",
|
|
"^Scan\\sRate\\s\\(V/s\\)",
|
|
"^Sample\\sInterval\\s\\(V\\)",
|
|
"^Quiet\\sTime\\s\\(sec\\)",
|
|
"^Sensitivity\\s\\(A/V\\)")
|
|
names.attr <- c("InitE",
|
|
"FinalE",
|
|
"ScanRate",
|
|
"SamplingInterval",
|
|
"QuietTime",
|
|
"Sensitivity")
|
|
for (n in 1:length(rgxp.attr)) {
|
|
attrow.idx <- regexpr(rgxp.attr[n], chifile)
|
|
attrow.len <- attr(attrow.idx, "match.length")
|
|
attr(attrow.idx, "match.length") <- NULL
|
|
attr(ff, names.attr[n]) <- strsplit(chifile[which(attrow.idx == 1)],
|
|
"\\s=\\s")[[1]][2]
|
|
}
|
|
#
|
|
return(ff)
|
|
}
|