Abstract:
In production systems, process planning and scheduling are the two core functions
that must be implemented before starting a production. The process planning generates a
suitable process plan for each part that indicates the sequences of machining operations of
the part, and the suitable machines and tools for that operations. This plan depends on the
available manufacturing resources and their operation time. The scheduling, then, specifies
the timetable of each operation from the process plans’ information. A significant conflict
during the scheduling process is that some determined resources are not valid or overloaded
because the process planning intentionally specifies the machines that make the least
manufacturing cost. Therefore, it is time-consuming to re-schedule or search for a feasible
solution that agree to the constraints of both process planning and scheduling.
Integrated process planning and scheduling is a trend of modern manufacturing
systems that would improve production in more flexible ways and overcome the
disadvantages of conventional procedures of process planning and scheduling. It is mostly
solved the integrated approach by using various metaheuristics algorithms. Hopfield
networks, in literature, show effective results in scheduling and can be implemented in
hardware level that has the computation powers and speed over the digital computations.
Otherwise, the analog computation has no iteration and obtain a real-time solution.
However, the Hopfield network is a local minimization method with slow convergence by
its gradient descent algorithm. Some modifications of Hopfield network should lead to a
fast convergence and global optimization problem.
This thesis presents two original and novel Hopfield neural networks architectures.
One of the proposed Hopfield neural networks modifies the standard Hopfield Network by
ii
properly incorporating the Broyden-Fletcher-Goldfarb-Shanno (BFGS) unconstrained
optimization algorithm. The second proposed Hopfield network is based on properly
incorporating the Nazareth unconstrained optimization algorithm in the standard Hopfield
network. One important issue to be investigated is the convergence speed of these two
novel Artificial Networks with respect to the convergence of the standard Artificial
Hopfield Network. The proposed model uses analog integrators to follow the traditional
algorithms of BFGS and Nazareth algorithms. The algorithms can be modeled by using
simple analog circuits, such as integrators, multipliers, and adders. The proposed
algorithms show a significant increase in the convergence speed of modified Hopfield
networks compared to the standard one. Also noise and adaptive learning parameter
technique are integrated to the modified Hopfield networks for global optimization. To test
and compare the effectiveness of the proposed Networks, they are implemented on the
solution of an integrated process planning and scheduling (IPPS) problem, which is a nondeterministic
polynomial time problem (NP-hard). Moreover, here, it is treated under a
novel formulation and to represent a modern manufacturing approach.
Description:
A Thesis Submitted to the Faculty of Graduate Studies and Research In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Industrial Systems Engineering, University of Regina. xii, 202 p.