|
| |
New Plant Start-up - 9/97 to 6/98
Role: On-site Project Management / Manufacturing Management Consulting
 |
Client: Eisai, Inc.,
RTP
,
NC
|
|
Industry: Pharmaceutical (Drug Manufacturing) - FDA regulated |
|
ERP package: QAD - MFG/PRO - Win95/WinNT |
|
Team: RWI: 3 consultants - Client: 4 managers, 8 associates |
|
Cut-over date: In phases - 2/98 (Receiving) to 5/98 (AP) |
|
FDA Validation: ERP package configuration was validated on 2/98 |
Personal accomplishments
|
Managed ERP project and reported to Plant Controller |
|
Analyzed, designed and configured production business processes
using ERP software |
|
Planned entire project, managed software configuration process
that maps business processes into QAD configuration tables, lead most design
sessions |
|
Configured many package areas |
|
Trained many client associates |
|
Supported other IT projects including Bar Coding & FDA
Validation effort |
(top of
page)
New Plant Start-up - 1/96 to 3/97
Role: On-site Project Management / Process Engineering
|
Client: Paragon Trade Brands,
Gaffney
,
SC
|
|
Industry: Medical Devices - Level II - FDA regulated |
|
ERP package: Lilly Software - Visual Manufacturing - Win95/WinNT |
|
Team: RWI: 3 consultants - Client: 5 managers, 3 team resources,
10 associates |
|
Cut-over date: 10/96 |
Personal accomplishments
|
Managed production related IT projects and reported to VP of
Manufacturing |
|
Participated to definition of Concept of Operation document |
|
Developed Business Processes (formalized into Standard Operating
Procedures) for all production related activities (receiving, inventory
management, production planning & control, packaging, shipping and spare
parts inventory management) and their supporting documents (such as the Batch
History Record) |
|
Analyzed, designed and configured production business processes
using ERP software |
|
Trained associates and deployed procedures |
|
Developed a spare part inventory system (physical layout, part
numbering, management procedures, data loading) |
|
Developed comprehensive Inventory Control program to improve
data entry accuracy and timeliness |
(top of
page)
3/94 to 12/95
Role: Project Management / Technical Leadership
|
Project: Information System for scheduling and optimizing PCB
assembly |
|
Industry: Electronics - PCB Assembly |
Personal accomplishments
|
Managed research program on constrained optimization/scheduling
techniques - cooperation with
North Carolina
State
University
|
|
Managed development of distributed real-time SPC application |
|
Designed & developed many scheduling modules using
object-oriented database (C++) |
(top of
page)
6/97
Role: JAD session facilitation / Finite Capacity Scheduling Expertise
|
Project: Requirements Definition and RFP generation - duration:
2 weeks |
|
Industry: Optical Cable Manufacturing |
Personal accomplishments
|
Lead JAD session (over 20 people) to define project objectives,
critical success factors and requirements |
|
Developed Request For Proposal document (60 pages) ready to be
sent to selected (12) vendors
|
(top of
page)
1989
Role: Software Developer
|
Project: Qualification of castings to ensure successful machining
(value: elimination of waste) |
|
Industry: Heavy Machineries - Machining of castings |
Personal accomplishments
In this project I developed an optical probe for a 3-D
measurement system used to qualify casting parts prior to machining. The
technique consists of projecting parallel stripes onto a part with a stripe
projector. A regular optical camera is used to visualize the stripes. For a flat
surface, the stripes appear very regular (geometrical distortions in the stripe
projector and camera optics tend to bend the stripes slightly and vary their
spacing). Depression and elevation in the height of the surface of the part
being viewed change the spacing between the stripes, thereby modulating the
phase of the original stripe pattern. This is similar to radio frequency
modulation except that here the phase of the stripe pattern carries information
about the 3-D shape to be measured. The recovery of the shape information is
done by demodulating the stripe pattern as follows:
|
Compute the Fourier transform of the stripe pattern. |
|
Demodulate the signal around the center carrier frequency (the
frequency of the stripe pattern when the surface is flat). This is accomplished
by shifting the frequency band encoding the signal to a steady state level. |
|
Compute the inverse Fourier transform. |
|
Compute the phase of the signal and unwrap it into a continuous
profile to remove phase shifts between neighboring pixels due to bounded phase
values. |
|
Subtract the phase profile corresponding to the flat reference
surface. The resulting phase profile directly corresponds to the shape of the
part. |
|
Convert phase values to height values (this is done by first
calibrating the system over the entire measurement range: it associates height
differences for the reference surface with corresponding phase shifts). |
Note that this approach only generated relative height
information. Absolute height information can be extracted if one point in the
profile is known. This is accomplished by separately projecting a single stripe
and determining its absolute position through regular triangulation. A major
strength of the technique is that it is fairly insensitive to amplitude
variation in the signal. Additionally, the height information is available all
over the field of view.
Because of non linear distortions in the camera lens and the
stripe projector, the phase profiles for the flat reference surface are
calibrated at various heights. The measurement range is then limited by the
depth of focus of the combined camera/stripe projector pair. For a field of view
of about one inch. I achieved one thousandth of an inch height resolution
accuracy. About twenty-five stripes are visible in the field of view. The
measurement range is about one inch (the camera and the stripe projector are
mounted at the end of a robot arm).
This work resulted in patent #5,243,665: "Component Surface
Distortion Evaluation Apparatus and Method" - Sep 1993.
(top of
page)
1997
Role: Image Processing Software Developer
|
Project: Grading of bolts based on stamping information to ensure
lack of contamination (Value: elimination of below grade bolts) |
|
Industry: Military Equipment Production |
Personal accomplishments
Because suppliers have been providing FMC with bolt batches of
inconsistent grading (resulting in huge costs associated with sorting out below
grade bolts), it was decided to visually inspect bolts and classify them
according to their head markings (set of concentric bars with vendor mark). An
inspection technique was needed that would work on randomly oriented bolts. I
decided to use circular profiles to classify bolts. The technique I developed is
as follows:
|
Extract the gray scale image data along circular profiles (while
removing camera rectangular aspect ratio). Depending on which model in the
library is selected, a different circular profile may be used. |
|
Extract edges using the Laplacian of a Gaussian operator. |
|
Binarize edges (1 if edge is positive; otherwise 0). |
|
Compute the correlation function (for each orientation) between
the unknown and model bolts stored in the library (using their corresponding
binary profiles). |
|
Determination of the best match (highest correlation) in the
library. |
|
If the best match is good enough, assign bolt to appropriate bin;
otherwise assign bolt to reject bin. |
The binary representation was chosen for several reasons:
|
It carries most of the information contained in the gray scale
image. |
|
The correlation function of a binary function reduces to a XOR
function which can be implemented very efficiently. |
|
The correlation function of the sign of the Laplacian edge is
strongly peaked; this suggests that an efficient search technique can be used to
locate the correlation peak in the correlation function. |
I successfully tested this approach using a dozen bolts as part
of the acceptance test; I then transferred the technology to the engineering
group.
(top of
page)
1996
Role: Image Processing Software Developer
|
Project: Development of efficient visual recognition algorithms to
recognize aerial targets |
|
Industry: Military Equipment Production |
Personal accomplishments
As part of the target acquisition project, I implemented an
automatic target recognition module to identify infra-red signatures of airborne
targets. The input data consisted of global target silhouettes. The technique I
implemented used Fourier descriptors of the boundary. It consists of the
following steps:
|
Binarize the image data given by the infra-red sensor. |
|
Segment the image to extract blobs while removing noisy data. |
|
Extract blob silhouettes in the form of polygonal approximations.
A very fast technique was used where each pixel on the boundary is visited once.
The determination of whether a boundary point is a corner point is made by
reaching data stored in a lookup table that encodes possible pixel sequences
approximating vectors 10 pixels long or less. |
|
Sub-sample blob silhouettes into complex functions of the plane
(1024 samples). |
|
Compute Fourier transforms of complex functions. |
|
Normalize shape outlines in the Fourier domain (translation,
scale, rotation and origin) to a standard set of conditions. For example, the
shape’s center of gravity is translated to the left upper most pixel in the
image. |
|
Extract the 32 most significant complex coefficients for each
shape; typically: F(-16), …, F(-1), F(1), …, F(16). |
|
Measure the euclidian distance between each unknown shape and the
model shapes stored in the library using their normalized coefficients. |
|
For each unknown shape, select the best match (smallest distance)
as candidate target identification. |
The database consisted of more than 200 views of 2 airplanes
stored on a videodisk. Views for each airplane were given along similar rotation
axes with a 5 degree increment. The outlines were generated by a CAD modeling
package. Each view was normalized according to the technique previously
described, and was stored in the library. I obtained a 93% identification rate
(correct view within plus or minus 5 degrees).
(top of
page)
(Home)
|
