Research Services in Manufacturing

courageouscellistAI and Robotics

Oct 29, 2013 (3 years and 10 months ago)

86 views

Research Services

in Manufacturing




MALTA

Exploiting ICT for the Maltese
Manufacturing Industry





Key Experts:

Dr Ernest Cachia



Dr John Abela



Dr Ing Saviour Zammit


Researchers:

Mr Armand Sciberras



Mr Joseph Bonello

ICT in Manufacturing


History of Manufacturing


Assembly line (Henry Ford)


Robots



Increasingly on the path of
innovation


To offer a broadened service
range
fulfilling
world
-
wide
customer requirements

ICT in Manufacturing


ICT potential


Fast


Accurate


Reliable


Updateable


Centralised Data


Assets remain in the company



Process 1


Automated
Optimisation of
Production Lines

Importance of Manufacturing


The Manufacturing Industry operates
in a very competitive market


Requires continuous innovation


Needs to provide added value services


Lower Costs


Higher Productivity


Faster product turnout


Better planning


Reduce Order
-
To
-
Shipping


time

Importance of Manufacturing


Efficient Scheduling and Planning
are main contributors to cost
reduction and efficiency


Helps the company allocate and
prepare resources to meet demand


ICT can add value by providing
low
-
cost tools that aid companies


Understand the scheduling
problems they face


Provide near
-
optimal schedules

Scheduling in Manufacturing


Scheduling refers to the problem of finding
an efficient sequence of utilisation of
human and machine resources


An efficient schedule reduces costs
through optimal allocation of resources


But it is a very difficult problem


Enumerating all possible schedules takes too
long to be feasible

Scheduling in Manufacturing


In Computer Science, scheduling problems
are well known


As examples of computationally complex
problems


As being notoriously difficult to find an
algorithm to solve them efficiently


Computationally complex
problems cannot be
solved in a “reasonable”
amount of time

Scheduling in Manufacturing


To determine an optimal
schedule, one must analyse the
performance of all possible
solutions of the problem


Enumerating all the possible
solutions takes a very long time


Also difficult is determining the
type of problem one has


Thus one is not sure how difficult
the problem is

Scheduling in Manufacturing


Operators have a number of rules
that help them find a schedule


Not always accurate and not
necessarily optimal


The project aims at helping
operators by


Helping them identify the problem


Help them identify the complexity of
the problem


Provide them with tools to generate a
schedule

Scheduling in Manufacturing


The project helps operators model
production lines using Graphs


The graph visually represents the
relationship between resources, jobs and
operations


Additional information
is captured to allow
the application to
detect the complexity
index of the problem

Scheduling in Manufacturing


The following types of problems can be
modelled:


Single Machine Problems: Bottleneck machine
Problems


Parallel Machine Problems: Jobs can be
processed on any of the different machines
available


Shop Problems: Machines
are dedicated to a
particular operation of a job
(Multi
-
Operation Model)

Scheduling in Manufacturing


The application is able to classify scheduling
problems in the following categories:


Polynomially

Solvable: These are the “easy”
problems i.e. one can obtain a schedule is a
reasonable timeframe


Pseudo
-
polynomially

Solvable: Problems that
are “easy” within certain bounds


NP
-
Hard: These are “hard”
problems, it is difficult to obtain
a schedule for these problems


Open: problems for which the
complexity status is not known

Scheduling in Manufacturing


Schedule generation depends on the type of
problem


Easy problems have an efficient algorithm


Hard problems may be approximated using
different techniques


Using heuristic techniques such as Genetic
Algorithms


Using mathematical
solutions such as
approximation
algorithms

Scheduling in Manufacturing


Generated schedules for hard
problems are near
-
optimal


Finding the optimal schedule is
not possible


Heuristic methods provide a
“good” trade
-
off between time to
generate the schedule and
optimality


Usually better than a human
-
generated schedule due to
computer’s speed in analysing
solutions


Scheduling in Manufacturing


A number of schedule generation
solutions are being investigated


A generic solver based on Genetic
Algorithms


Solvers based on mathematic
formulations


The proposed solution has a
plugin system aimed at


Extensibility to allow addition
of new algorithms and
solvers



Process 2


Automated QA
for Print Output

using Neural Networks

Quality Assurance in Manufacturing


Quality is a key factor in the selling
price of manufactured goods


Quality inspection is traditionally
done by human beings either:


At the end of the production line
on each and every item removing
defective products (slow and
expensive)


Inspecting samples from a
produced batch and assessing a
batch quality (defective products
are left in batch)

Motivation


Human inspection:


Not adequate for certain
inspection jobs


Not accurate enough


Slow and expensive


Not consistent and decisions
vary from one inspector to the
other



With recent advances in hardware and
development of AI techniques automating the
inspection process is more possible

Printing in an industrial setup


Printing is usually carried out
using:


Pad printing


Hot
-
foil stamping


Offset printing


Silk
-
screen printing


Printing Problems


All the printing techniques have their own
advantages and disadvantages and are
prone to particular defects


Changes
in
temperature


Sensitive to different
surfaces


Incorrect ink
mixtures


etc.


Preventing changes to these
variables is generally either
impossible or very expensive

Common Defects


Scratches and
c
racks



Missing or extra ink



Missing or incorrectly placed
components on PCBs



Smudges and misalignment
defects


Advantages of an automated system



Reliable 24x7



Accurate



Fast



Consistent



Cheap

Towards an automated system


The
project
aims at helping
operators
by


Guiding them in training and
setting up an automated
classifier system for
manufactured goods


Speed up the inspection
process of manufactured
goods


Increase the quality of
manufactured goods


Automated Optical Inspection

Image

capturing

Image correction,
background
removal, image
segmentation

Further image
processing and
feature extraction

Data Input to
Classifier and other
QA tests

Output from
Classifier


Image Processing


Alignment


Depending on the setup used, images
can be aligned to a
template



Background removal


Unwanted areas from the image can be
removed through masks and region
selection



Captured images can be broken down into
regions using intelligent algorithms
in order
to
be treated separately


Artificial Neural Networks


Mathematical models inspired
from the structure and functional
aspects of biological neural
networks


Made up of an interconnected
group of artificial neurons


Based on input, the structure
and connections of the
network change accordingly
to learn a particular task



Great for modelling complex relationships between
inputs and outputs

The WISARD Neural Network


At
training stages the
provided data
is extracted from the image and
stored in the
nodes



During live mode, the data is
extracted the same way from the
image and tested in the nodes



A similarity percentage is
calculated and the object can be
discarded if the value is under a
specific threshold



Other tests


Additional QA tests can be set to run in
parallel with the neural network



Label offset


Checking that a specified label printed on
the product is contained in a specified
boundary


Colour inclusion/exclusion


Checking for the presence of areas
defined by colour and size at specified
areas


Optical Character Recognition


Checking text at specified areas


Classification


All QA tests selected during
setup time are executed on
the captured image of the
product



All tests can either fail or
pass


The tool determines whether
the product should be
discarded or not




Thank you for

listening