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.

213 lines
8.9 KiB
Python

from struct import unpack
#from pylab import *
import sys
from array import array
# Updated 10/11/15 BMM : Added mesh.
# Updated 10/11/15 MD : Added dictionnary initialisation in mesh.
# Update 8/11/2019 MD : Improve performance (used array instead of unpack)
class rma:
def __init__(self):
self.file = 'not opened yet'
# May want to do some other things here later
def open(self, filename):
self.file = open(('%s.rma' % (filename)), 'rb')
self.header = self.file.read(1000).decode("utf-8")
self.type = self.header[0:10]
self.title = self.header[100:172]
self.geometry = self.header[200:300]
self.num_nodes = int(self.header[40:50])
self.num_elements = int(self.header[50:60])
if self.type == 'RMA11 ':
self.num_constits = int(self.header[60:70])
if len(self.header[80:90].strip()) == 0:
self.num_sedlayers = 0
else:
self.num_sedlayers = int(self.header[80:90])
self.constit_name = []
self.constit_name.append("NULL")
i = 1
print(self.num_constits)
print(self.header[300:1000])
while i <= self.num_constits:
# print self.header[300:400]
self.constit_name.append(self.header[300 + (i - 1) * 8:308 + (i - 1) * 8])
if self.header[300 + (i - 1) * 8:308 + (i - 1) * 8] == " SSED ":
self.constit_name.append(" BSHEAR")
self.constit_name.append("BedThick")
j = 1
print(self.num_sedlayers)
while j <= self.num_sedlayers:
self.constit_name.append(" L%dThick" % j)
j = j + 1
i = i + 1
# print i
# print self.num_constits
# print i
# if self.num_sedlayers > 0:
# self.num_constits=self.num_constits+2+self.num_sedlayers
# Header format in RMA2
# HEADER(1:10) ='RMA2 '
# HEADER(11:20)=DATEC = Date of run
# HEADER(21:30)=TIMEC = Time of run
# HEADER(31:40)=ZONEC = Time zone
# HEADER(41:50)=NP = Number of nodes
# HEADER(51:60)=NE = Number of elements
# HEADER(101:172)=TITLE = Title line
# HEADER(201:300)=FNAME = File name of binary geometry
# Header format in RMA11
# HEADER(1:10) - Model
# HEADER(11:20) - DATEC of model runtime
# HEADER(21:30) - TIMEC of model runtime
# HEADER(31:40) - ZONEC of model runtime
# HEADER(41:90) - '(5I10)' NP,NE,NQAL,IGS,LGS or '(5I10)' NP,NE,NQAL,ISEDS,NLAYT
# HEADER(101:172) - TITLE
# HEADER(301:540) - '(30A8)' CLABL
# HEADER(201:248) - FNAMD directory of the geometry
# HEADER(251:298) - FNAM2 filename of the geometry
def mesh(self, filename):
self.meshfile = open(('%s.rm1' % (filename)), 'r')
l = self.meshfile.readline()
while l[0:5] != ' 9999' and len(l) != 5:
l = self.meshfile.readline()
l = self.meshfile.readline()
self.node_e = {}
self.node_n = {}
self.node_z = {}
while l[0:10] != ' 9999' and len(l) != 10:
ll = l.split()
self.node_e[ll[0]] = ll[1]
self.node_n[ll[0]] = ll[2]
self.node_z[ll[0]] = ll[3]
l = self.meshfile.readline
def next(self):
# This reads the next timestep and populates the variables.
# Reset all of the variables
self.time = 0.0
self.year = 0
self.xvel = []
self.yvel = []
self.zvel = []
self.depth = []
self.elevation = []
self.temperature = []
self.salinity = []
self.sussed = []
# Add in an entry to fill position 0. Important since RMA files start numbering nodes from 1.
self.xvel.append(-999)
self.yvel.append(-999)
self.zvel.append(-999)
self.depth.append(-999)
self.elevation.append(-999)
self.temperature.append(-999)
self.salinity.append(-999)
self.sussed.append(-999)
if self.type == 'RMA2 ':
# WRITE(IRMAFM) TETT,NP,IYRR,((VEL(J,K),J=1,3),K=1,NP),(WSEL(J),J=1,NP),(VDOT(3,K),K=1,NP)
t = self.file.read(12)
if t:
a = unpack('fii', t)
self.time = a[0]
np = int(a[1])
self.year = a[2]
if (np != self.num_nodes):
print("Warning - NP (%d) on this timestep does not match header (%d)" % (np, self.num_nodes))
#b = unpack('%df' % 5 * np, self.file.read(20 * np))
b = array('f')
b.fromfile(self.file, 5 * np)
# b = np.fromfile(self.file, dtype=float,count = 5 * np)
tempA = [(x - 1) * 3 for x in range(1, np + 1)]
tempB = [np * 3 + (x - 1) for x in range(1, np + 1)]
self.xvel.extend([b[i] for i in tempA])
self.yvel.extend([b[i + 1] for i in tempA])
self.depth.extend([b[i + 2] for i in tempA])
self.elevation.extend([b[i] for i in tempB])
if self.type == 'RMA11 ':
self.constit = []
c = 0
# Start at zero to leave an empty array space. Constits numbered 1 .. num_constits
while c <= self.num_constits:
self.constit.append([])
# Put a null into the 0 position so that RMA node numbers are in the same numbered array position.
self.constit[c].append(-999)
c = c + 1
# READ(file1,END=100) TETT1,NQAL,NP,IYRR, ((VEL(K,J),J=1,NP),K=1,3), (wd(j),j=1,np), (wsel(j),j=1,np), ((TCON1(K,J),J=1,NP),K=1,NQAL-5)
t = self.file.read(16)
if t:
a = unpack('fiii', t)
self.time = a[0]
nqal = int(a[1])
np = int(a[2])
self.year = a[3]
if ((nqal - 5) != (self.num_constits)):
print("Warning - NQAL-5 (%d) on this timestep does not match header (%d)" % (
nqal - 5, self.num_constits))
if (np != self.num_nodes):
print("Warning - NP (%d) on this timestep does not match header (%d)" % (np, self.num_nodes))
b = array('f')
b.fromfile(self.file, nqal * np)
#b = unpack('%df' % nqal * np, self.file.read(4 * nqal * np)) #4 because single precision float type have 4 bytes
tempA = [(x - 1) * 3 for x in range(1, np + 1)]
tempB = [np * 3 + (x - 1) for x in range(1, np + 1)]
tempC = [np * 4 + (x - 1) for x in range(1, np + 1)]
self.xvel.extend([b[i] for i in tempA])
self.yvel.extend([b[i + 1] for i in tempA])
self.zvel.extend([b[i + 2] for i in tempA])
self.depth.extend([b[i] for i in tempB])
self.elevation.extend([b[i] for i in tempC])
for c in range(1, self.num_constits + 1):
(self.constit[c].extend([b[np * ((c - 1) + 5) + (x - 1)] for x in range(1, np + 1)]))
if self.type == 'RMA10 ':
# WRITE(IRMAFM) TETT,NP,NDF,NE,IYRR,((VSING(K,J),K=1,NDF),VVEL(J),WSLL(J),J=1,NP),(DFCT(J),J=1,NE),(VSING(7,J),J=1,NP)
# WRITE(IRMAFM) TETT,NP,IYRR,((VEL(J,K),J=1,3),K=1,NP),(WSEL(J),J=1,NP),(VDOT(3,K),K=1,NP)
t = self.file.read(20)
if t:
a = unpack('fiiii', t)
self.time = a[0]
np = a[1]
ndf = 6
ne = a[3]
self.year = a[4]
if (np != self.num_nodes):
print("Warning - NP1 (%d) on this timestep does not match header (%d)" % (np, self.num_nodes))
tempRead = np * (3 + ndf) + ne
b = array('f')
b.fromfile(self.file, tempRead)
# b = numpy.fromfile(self.file.read(4 * tempRead), dtype=float)
#b = unpack('%df' % tempRead, self.file.read(4 * tempRead))
i = 1
while i <= np:
self.xvel.append(b[0 + (i - 1) * 8])
self.yvel.append(b[1 + (i - 1) * 8])
self.depth.append(b[2 + (i - 1) * 8])
self.salinity.append(b[3 + (i - 1) * 8])
self.temperature.append(b[4 + (i - 1) * 8])
self.sussed.append(b[5 + (i - 1) * 8])
self.zvel.append(b[6 + (i - 1) * 8])
self.elevation.append(b[7 + (i - 1) * 8])
i = i + 1
if len(self.xvel) == 1:
# Note that this is a 1 now as we've filled the zero array position
return False
else:
return True