ImXPAD S540 detector at D2AMΒΆ
This tutorial corresponds to the calibration the goniometer an ImXPAD detector composed of 8 stripes of 7 modules, many of which are defective, on a goniometer.
This detector is mounted on the goniometer 2theta arm at the D2AM beam-line, French CRG at the ESRF synchrotron.
The raw data files are available at: http://www.silx.org/pub/pyFAI/gonio/D2AM-15/
In [ ]:
%pylab nbagg
In [2]:
import os, time
start_time = time.time()
import fabio, pyFAI
print("Working with pyFAI version", pyFAI.version)
from pyFAI.goniometer import GeometryTransformation, GoniometerRefinement, Goniometer
from pyFAI.gui import jupyter
/mntdirect/_scisoft/users/jupyter/jupy35/lib/python3.5/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
Working with pyFAI version 0.15.0-beta4
In [3]:
#Download all data
from silx.resources import ExternalResources
#Comment out when outside ESRF
os.environ["http_proxy"] = "http://proxy.esrf.fr:3128"
downloader = ExternalResources("pyFAI", "http://www.silx.org/pub/pyFAI/testimages", "PYFAI_DATA")
all_files = downloader.getdir("LaB6_gonio_D2AM.tar.bz2")
print("List of files downloaded:")
for i in all_files:
print(" "+os.path.basename(i))
detector_file = [i for i in all_files if i.endswith("D5Geom-2018.h5")][0]
images = [i for i in all_files if i.endswith(".edf")]
npt_files = [i for i in all_files if i.endswith(".npt")]
List of files downloaded:
LaB6_gonio_D2AM
16Dec08D5_1777-rsz.npt
16Dec08D5_1791-rsz.npt
16Dec08D5_1729-rsz.edf
16Dec08D5_1763-rsz.npt
D5Geom-2018.h5
16Dec08D5_1725-rsz.edf
16Dec08D5_1728-rsz.npt
16Dec08D5_1784-rsz.npt
16Dec08D5_1763-rsz.edf
16Dec08D5_1728-rsz.edf
16Dec08D5_1735-rsz.npt
16Dec08D5_1727-rsz.npt
16Dec08D5_1729-rsz.npt
16Dec08D5_1770-rsz.edf
16Dec08D5_1784-rsz.edf
16Dec08D5_1730-rsz.npt
16Dec08D5_1742-rsz.npt
16Dec08D5_1770-rsz.npt
16Dec08D5_1725-rsz.npt
16Dec08D5_1726-rsz.npt
16Dec08D5_1777-rsz.edf
16Dec08D5_1756-rsz.edf
16Dec08D5_1735-rsz.edf
16Dec08D5_1730-rsz.edf
16Dec08D5_1742-rsz.edf
16Dec08D5_1791-rsz.edf
16Dec08D5_1749-rsz.npt
16Dec08D5_1726-rsz.edf
16Dec08D5_1756-rsz.npt
16Dec08D5_1727-rsz.edf
16Dec08D5_1749-rsz.edf
In [4]:
#Definition of the detector and deplay of an image and its mask:
d5 = pyFAI.detector_factory(detector_file)
print("Detector shape: ",d5.shape)
fimg = fabio.open(images[-1])
for k,v in fimg.header.items():
print(k, ": ", v)
f, ax = subplots(1, 2)
jupyter.display(d5.mask, label="mask", ax=ax[0])
jupyter.display(fimg.data, label=os.path.basename(fimg.filename), ax=ax[1])
Detector shape: (960, 578)
{
EDF_DataBlockID : 0.Image.Psd
EDF_BinarySize : 4439040
EDF_HeaderSize : 1536
ByteOrder : LowByteFirst
DataType : DoubleValue
Dim_1 : 578
Dim_2 : 960
Image : 0
HeaderID : EH:000000:000000:000000
Size : 4439040
VersionNumber : 1
Epoch : 1481327234.3919599056
det_sample_dist : 0
y_beam : 0
x_beam : 0
Lambda : 0.495938
offset : 0
count_time : 120
point_no : 24
scan_no : 906
preset : 0
col_end : 559
col_beg : 0
row_end : 959
row_beg : 0
counter_pos : 120 2689 156 15.3097 0 90.3692 15.3097 0 25 25 1749 5.65726e+09 0 0 0 174.925 0 0 15.3097
counter_mne : sec vct1 vct2 vct3 vct4 Imach pseudoC pfoil Emono Ecod img roi1 roi2 roi3 roi4 pico1 pico2 pico3 pico4
motor_pos : 24.0001 0.077502 89.9912 -89.9921 -0.0032 0.0022 57.1197 134.748 -32.9504 0.16656 -5 0.47558 -1.5 0 0 0 4.53604 0.1416 1.04 1.04022 1.04022 -4.4 -1.10211 -0.543725 -9.962 -14.038 -16.865 -7.195 24 -2.038 24.06 4.835 1
motor_mne : del eta chi phi nu mu keta kap kphi tsx tsy tsz rox roy tox toy mono inc1 courb courbb courbf omega khimono gamma su6 sd6 sf6 sb6 vg6 vo6 hg6 ho6 rien
suffix : .edf
prefix : 16Dec08D5_
dir : /users/opd02/raw
run : 1749
title : CCD Image
Data type cannot be displayed: application/javascript
Out[4]:
<matplotlib.axes._subplots.AxesSubplot at 0x7f92d326acc0>
In [5]:
# Define wavelength and create our "large" LaB6 calibrant
wavelength = 0.495938 * 1e-10
from pyFAI.calibrant import get_calibrant
LaB6 = get_calibrant("LaB6")
LaB6.wavelength = wavelength
print("2theta max: ", numpy.degrees(LaB6.get_2th()[-1]))
print("Number of reflections: ", len(LaB6.get_2th()))
2theta max: 179.17349767223402
Number of reflections: 236
In [6]:
#Use a few manually calibrated images:
npt_files.sort()
print("Number of hand-calibrated images :",len(npt_files))
Number of hand-calibrated images : 15
In [7]:
# Definition of the goniometer translation function:
# The detector rotates vertically, around the horizontal axis, i.e. rot2.
# Rotation both around axis 1 and axis 2 are allowed
goniotrans = GeometryTransformation(param_names = ["dist", "poni1", "poni2",
"rot1", "rot2", "rot3", "scale1", "scale2" ],
dist_expr="dist",
poni1_expr="poni1",
poni2_expr="poni2",
rot1_expr="scale1 * pos +rot1",
rot2_expr="scale2 * pos + rot2",
rot3_expr="rot3")
#Definition of the function reading the goniometer angle from the filename of the image.
def get_angle(metadata):
"""Takes the angle from the first motor position and returns the angle of the goniometer arm"""
return float(metadata["motor_pos"].split()[0])
print('filename', os.path.basename(fimg.filename), "angle:",get_angle(fimg.header))
filename 16Dec08D5_1749-rsz.edf angle: 24.0001
In [8]:
# Definition of the geometry refinement: the parameter order is the same as the param_names
rot3 = numpy.pi/2
scale1 = -numpy.pi/180
scale2 = 0
param = {"dist":0.5,
"poni1":0.05,
"poni2":0.05,
"rot1":0,
"rot2":0,
"rot3": rot3,
"scale1": scale1,
"scale2": scale2,
}
#Defines the bounds for some variables
bounds = {"dist": (0.2, 0.8),
"poni1": (0, 0.1),
"poni2": (0, 0.1),
"rot1": (-0.1, 0.1),
"rot2": (-0.1, 0.1),
"rot3": (rot3, rot3), #strict bounds on rot3
#"scale1": (scale1, scale1),
#"scale2": (scale2, scale2),
}
gonioref = GoniometerRefinement(param, #initial guess
bounds=bounds,
pos_function=get_angle,
trans_function=goniotrans,
detector=d5, wavelength=wavelength)
print("Empty refinement object:", gonioref)
#Let's populate the goniometer refinement object with all control point files:
for fn in npt_files[:]:
base = os.path.splitext(fn)[0]
fimg = fabio.open(base + ".edf")
basename = os.path.basename(base)
sg =gonioref.new_geometry(basename, image=fimg.data, metadata=fimg.header, control_points=fn, calibrant=LaB6)
print(basename, "Angle:", sg.get_position())
print("Filled refinement object:")
print(gonioref)
Empty refinement object: GoniometerRefinement with 0 geometries labeled: .
16Dec08D5_1725-rsz Angle: -0.003
16Dec08D5_1726-rsz Angle: 0.9998
16Dec08D5_1727-rsz Angle: 2.0
16Dec08D5_1728-rsz Angle: 2.9998
16Dec08D5_1729-rsz Angle: 4.0002
16Dec08D5_1730-rsz Angle: 4.9998
16Dec08D5_1735-rsz Angle: 10.0001
16Dec08D5_1742-rsz Angle: 16.9996
16Dec08D5_1749-rsz Angle: 24.0001
16Dec08D5_1756-rsz Angle: 30.9997
16Dec08D5_1763-rsz Angle: 37.9999
16Dec08D5_1770-rsz Angle: 44.9997
16Dec08D5_1777-rsz Angle: 52.0
16Dec08D5_1784-rsz Angle: 58.9995
16Dec08D5_1791-rsz Angle: 65.9999
Filled refinement object:
GoniometerRefinement with 15 geometries labeled: 16Dec08D5_1725-rsz, 16Dec08D5_1726-rsz, 16Dec08D5_1727-rsz, 16Dec08D5_1728-rsz, 16Dec08D5_1729-rsz, 16Dec08D5_1730-rsz, 16Dec08D5_1735-rsz, 16Dec08D5_1742-rsz, 16Dec08D5_1749-rsz, 16Dec08D5_1756-rsz, 16Dec08D5_1763-rsz, 16Dec08D5_1770-rsz, 16Dec08D5_1777-rsz, 16Dec08D5_1784-rsz, 16Dec08D5_1791-rsz.
In [9]:
# Initial refinement of the goniometer model with 5 dof
gonioref.refine2()
Cost function before refinement: 0.00030766121608429307
[ 0.5 0.05 0.05 0. 0. 1.57079633
-0.01745329 0. ]
fun: 1.9855225530234952e-08
jac: array([ 2.40916227e-08, 3.04897072e-07, 4.13640063e-07, -1.75538242e-07,
-9.57392364e-07, -1.12132525e-13, -5.24435840e-06, -7.14899229e-07])
message: 'Optimization terminated successfully.'
nfev: 178
nit: 17
njev: 17
status: 0
success: True
x: array([ 5.20464974e-01, 6.34779274e-02, 4.46077056e-02, 3.03313918e-03,
7.46909751e-03, 1.57079633e+00, -1.74719084e-02, -3.84262987e-04])
Cost function after refinement: 1.9855225530234952e-08
GonioParam(dist=0.5204649744819361, poni1=0.06347792740436164, poni2=0.04460770564596157, rot1=0.0030331391768589344, rot2=0.007469097514563215, rot3=1.5707963267948966, scale1=-0.017471908444435618, scale2=-0.00038426298745148534)
maxdelta on: dist (0) 0.5 --> 0.5204649744819361
Out[9]:
array([ 5.20464974e-01, 6.34779274e-02, 4.46077056e-02, 3.03313918e-03,
7.46909751e-03, 1.57079633e+00, -1.74719084e-02, -3.84262987e-04])
In [10]:
width=3
height=int(ceil(len(gonioref.single_geometries) / width))
fig,ax = subplots(height, width,figsize=(10, 25))
for idx, sg in enumerate(gonioref.single_geometries.values()):
sg.geometry_refinement.set_param(gonioref.get_ai(sg.get_position()).param)
jupyter.display(sg=sg, ax=ax[idx//width, idx%width])
Data type cannot be displayed: application/javascript
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
WARNING:matplotlib.legend:No handles with labels found to put in legend.
In [11]:
# Final pass of refinement with all constrains removed, very fine refinement
gonioref.bounds = None
gonioref.refine2("slsqp", eps=1e-13, maxiter=10000, ftol=1e-12)
Cost function before refinement: 1.9855225530234952e-08
[ 5.20464974e-01 6.34779274e-02 4.46077056e-02 3.03313918e-03
7.46909751e-03 1.57079633e+00 -1.74719084e-02 -3.84262987e-04]
fun: 1.9855093643189328e-08
jac: array([ 3.37489690e-09, 0.00000000e+00, 0.00000000e+00, 1.09849585e-08,
-8.71782191e-07, 7.70932331e-09, -3.46983076e-05, -5.59240269e-07])
message: 'Optimization terminated successfully.'
nfev: 15
nit: 1
njev: 1
status: 0
success: True
x: array([ 5.20464974e-01, 6.34779274e-02, 4.46077056e-02, 3.03313917e-03,
7.46909771e-03, 1.57079633e+00, -1.74719008e-02, -3.84262865e-04])
Cost function after refinement: 1.9855093643189328e-08
GonioParam(dist=0.5204649744811972, poni1=0.06347792740436164, poni2=0.04460770564596157, rot1=0.003033139174453971, rot2=0.007469097705424581, rot3=1.5707963267932088, scale1=-0.017471900847851926, scale2=-0.0003842628650156724)
maxdelta on: scale1 (6) -0.017471908444435618 --> -0.017471900847851926
Out[11]:
array([ 5.20464974e-01, 6.34779274e-02, 4.46077056e-02, 3.03313917e-03,
7.46909771e-03, 1.57079633e+00, -1.74719008e-02, -3.84262865e-04])
In [12]:
#Create a MultiGeometry integrator from the refined geometry:
angles = []
images = []
for sg in gonioref.single_geometries.values():
angles.append(sg.get_position())
images.append(sg.image)
multigeo = gonioref.get_mg(angles)
multigeo.radial_range=(0, 80)
print(multigeo)
MultiGeometry integrator with 15 geometries on (0, 80) radial range (2th_deg) and (-180, 180) azimuthal range (deg)
In [13]:
# Integrate the whole set of images in a single run:
res = multigeo.integrate1d(images, 10000)
jupyter.plot1d(res)
#Note the large number of peaks due to hot pixels ....
Data type cannot be displayed: application/javascript
Out[13]:
<matplotlib.axes._subplots.AxesSubplot at 0x7f92c91c99b0>
In [14]:
#Add hot pixels to the mask: pixel which are 15x more intense than the median in their ring.
thres = 15
old_mask = d5.mask.astype("bool", copy=True)
new_mask = d5.mask.astype("bool", copy=True)
for ai,img in zip(multigeo.ais,images):
b,a = ai.separate(img, 1000, restore_mask=0)
b[old_mask] = 0
b[b<0] = 0
print(sum(b>thres*a))
new_mask = numpy.logical_or(new_mask, (b>thres*a))
print(" Size of old mask", sum(old_mask),
" Size of new mask",sum(new_mask),
" Number of pixel discarded", sum(new_mask)-sum(old_mask))
4
4
4
4
4
4
4
4
4
4
4
3
3
3
3
Size of old mask 198678 Size of new mask 198683 Number of pixel discarded 5
In [15]:
# Update the mask
for ai in multigeo.ais:
ai.detector.mask = new_mask
# Integrate the whole set of images in a single run:
res2 = multigeo.integrate1d(images, 10000)
ax = jupyter.plot1d(res, label="Before hot-pixel removal")
ax.plot(*res2, label="After hot-pixel removal")
ax.legend()
Data type cannot be displayed: application/javascript
Out[15]:
<matplotlib.legend.Legend at 0x7f92c90a82e8>
In [16]:
# Integrate the whole set of images in 2D:
res2d = multigeo.integrate2d(images, 1000, 360)
jupyter.plot2d(res2d)
Data type cannot be displayed: application/javascript
Out[16]:
<matplotlib.axes._subplots.AxesSubplot at 0x7f92c906ed30>
In [17]:
print("Total execution time", time.time()-start_time)
Total execution time 105.95396304130554
In [ ]: