noncontact laser scanner for surface reconstruction. These
geometry scans serve as the ground truth for the different test
objects used in this study and provide the basis for generating
the corresponding shape models needed for calibration algo-
rithms, as well as for comparing results from the structured
light sensor array.
Software for Simultaneous Hardware Control and Data
Analysis
For the fastest combination of data acquisition, trigger control,
and point cloud analysis feedback, all aspects were merged
into a common software interface built on a machine vision
library toolkit. During operation of the structured light sensors,
the control software is linked with the machine vision library,
enabling direct acquisition control and output data formatting
of the sensors. The acquired 3D data is analyzed immediately
after capture within the same software to display results. For
converting CAD geometries from CMM measurements or
shape models into formats compatible with the machine vision
library for shape-comparison analysis, a separate CAD software
tool was used.
Experimental Results and Discussion
In this section, experimental results are presented regarding the
WCS mapping and rapid 3D point cloud meshing from an array
of structured light sensors without requiring any part or sensor
articulation. The results show that the proposed automatic point
cloud statistical quality inspection algorithm can readily detect
when hairpin wire-form geometries exhibit deviations in the
expected 3D surface profile, which are indicative of wear in the
die-forming process. The future work of this research is also
summarized, specifically focusing on strategies to further reduce
the time required to perform a full 3D hairpin quality assess-
ment beyond what has already been achieved in this study.
Calibration Point Cloud Analysis for WCS Repeatability
To verify the procedure for repeatable WCS assignment across
all structured light sensors, the calibration artifact shown in
Figure 5 was first measured using the CMM system described
earlier. The CMM was used to measure the diameters of all
spheres and the positional center-point (x, y, z) coordinates
relative to the (0, 0, 0) coordinate defined on the calibration
block. This measurement was repeated 25 times on different
occasions, and the statistical average diameters and sphere
positions were used to generate representative CAD surface
models of the actual artifact assembly. The calibration artifact
was then placed within the field of view of all structured light
sensors, and 3D point cloud data was acquired from each
sensor. Surface model matching using a uniform best fit of
the CAD representation of the calibration artifact was per-
formed for each 3D point cloud, and the resulting positional
and angular alignment were recorded. It was determined that
a minimum of 50 scans from each sensor was required to
provide robust and reliable statistics for minimizing error in
determining the position and orientation of the surface model
within the point cloud data.
Once the orientation and position of the calibration
artifact had been determined in each sensor’s scene relative
to its coordinate system, a coordinate-system transformation
was performed for each sensor that placed the origin at the
position of the 10 mm diameter sphere closest to the (0, 0)
block markings. With the new common WCS applied to each
ME
|
ELECTRICVEHICLES
Figure 5. (a) CAD design of the calibration artifact for world coordinate
system (WCS) calibration of the sensor array. Sphere colors denote
sphere diameter in the design (b) image of the physical artifact
created for use in this study.
40
M AT E R I A L S E V A L U AT I O N • J A N U A R Y 2 0 2 6

sensor, a new set of point cloud scans (25 scans) was acquired
from each sensor, and the uniform best-fit surface model
matching was again applied to each scan to measure the arti-
fact’s position and orientation in the new coordinate system.
Because the new WCS matches the coordinate system of the
calibration artifact, the measured position and angular orienta-
tion are registered at the origin, with each of the (x, y, z) coor-
dinates exhibiting a measurement uncertainty of ±4 μm.
To verify WCS assignment repeatability, a measurement
system analysis (MSA) study was performed using a three-
sphere planar test object shown in Figure 6a, illustrating how
well point cloud data from four of the structured light sensors
mapped into the common WCS defined by the calibration
artifact. MSA is a structured procedure that has been applied
in manufacturing to assess the quality of measurement and
inspection systems. A Gage Repeatability and Reproducibility
(Gage R&R) Type 1 study (American Society for Quality Control
2003) was used to evaluate the repeatability of the integrated
inspection system by taking multiple structured light sensor
measurements of both the calibration artifact and the three-
sphere planar object on different days and positioned by dif-
ferent operators, thereby assessing the inherent variation of the
measurement process.
Once calibrated to the WCS, the three-sphere planar object
cloud data were fused, and sphere models were fit using a
uniform best fit. The diameters and (x, y, z) positions of each fit
sphere were recorded for each data sequence. These values were
then compared against the ground-truth measurements per-
formed by the CMM. Figure 6b shows an example subsampled
dataset of the three-sphere planar object acquired in uncali-
brated coordinate systems from each structured light sensor
plotted together. Figure 6c shows the same object after calibra-
tion to the common WCS. In both Figures 6b and 6c, colors are
used to distinguish point clouds from different sensors.
From the MSA study, it was determined that the diameter
measurements of all three 20 mm spheres (as confirmed by
CMM) displayed an average measurement uncertainty of
±10 μm from the nominal value. Positionally, the three spheres
showed relative distance measurement uncertainty of ±8 μm
with respect to one another. These values fall well within the
tolerance range expected for the 3D surface profile of the
hairpin wire-form geometry.
Wire-Form Point Cloud Analysis and Results
Figure 7a displays an example of the raw wire-form point cloud
data acquired from the structured light sensor array after per-
forming the calibration procedure to the common WCS. Colors
are used to distinguish point clouds from different sensors in
the image. Figure 7b shows an example of the final wire-form
point cloud after data fusion and subsampling from the struc-
tured light sensor array. A similar Gage R&R Type 1 study was
performed for a set of five production-relevant wire-forms
that had been measured by both the CMM system and the
structured light sensors. Comparable repeatability and reli-
ability results were observed for the hairpin shape however,
because the measurement for the hairpin involves a full 3D
surface profile comparison against CMM results, the nature of
the measurement differs. The CMM shape model undergoes
a uniform best fit to the fused wire-form point cloud, followed
by a closest-distance computation from each point in the cloud
to the CMM shape model. Across the set of hairpins evaluated,
all apex data points from the fused, subsampled point clouds
were found to be within ±50 μm of the CMM measurements
for the same hairpins held in the fixture shown in Figures 2d
and 4. This result provides strong confidence that the struc-
tured light sensor array is capable of reliably determining
whether the hairpin 3D shape geometry lies within the permis-
sible ±250 μm tolerance range.
With calibration established for the sensor array, the
objective of the study shifted toward rapid acquisition and
3D reconstruction of the hairpin wire-forms, enabling them
to be analyzed and visualized relative to the accepted toler-
ance band. To support this method, a production-relevant
gold-standard wire-form was repeatedly measured using both
the CMM system and the structured light sensor array to create
an average 3D reconstruction shape model for comparison
Figure 6. (a) Three-sphere planar test object illustrating how point cloud data from four of the structured light sensors is mapped to the common
WCS defined using the calibration artifact shown in Figure 5 (b) uncalibrated subsampled sphere point clouds from each sensor (c) subsampled
sphere point clouds after calibration to the common WCS. In Figures 6b and 6c, colors are used to distinguish point clouds from different sensors.
J A N U A R Y 2 0 2 6 • M AT E R I A L S E V A L U AT I O N 41