In this section, we present selected publication works of the OptimalCore partners, related to Supply Chain Planning and Optimization and other relevant academic fields.
Supply Chain Management in Greece, current and future trends
Laboratory of Quantitative Analysis, Logistics and Supply Chain Management, Department of Mechanical Engineering, Aristotle University of Thessaloniki, Greece
A. Panagiotopoulou
(Supervisors: E. Iakovou, D. Vlachos, Industrial Sponsors: A. Dimitriadis, T. Rafailidis)
Campaign planning for manufacturing supply chains
Laboratory of Quantitative Analysis, Logistics and Supply Chain Management, Department of Mechanical Engineering, Aristotle University of Thessaloniki, Greece
T. Rafailidis, E. Iakovou
Abstract:
In the process industry including pharmaceuticals, metals and chemicals, manufacturing supply chains are particularly long and complex. Moreover, the manufacturing operations are highly capacity intensive. In addition, companies must not only deal with the usual complexities of fluctuating demand from changing customer needs, competition, and economic pressures, but also with abrupt and unexpected changes that can upset carefully devised plans. In such an environment, Campaign Planning is pivotal for maintaining the efficiency of the manufacturing process. Consolidating the manufacturing demand by creating campaigns helps minimizing the total capacity setup time. Customer service and inventory levels need to be further taken into account while consolidating to achieve the desired supply chain management goals. In this work, we present the complexity of Campaign Planning in Manufacturing and further discuss ways to improve it. We demonstrate why the new dynamic way of manufacturing demand consolidation is superior to the traditional fixed one. Finally, we present best practices through a case study with a pharmaceutical customer and related technologies. We conclude by outlining the need for proactive decision-making by employing a quantitative what-if-scenario analysis to further mitigate the supply chain’s risk.
RTN-based rolling horizon algorithms for medium term scheduling of multipurpose plants
Computers & Chemical Engineering, Volume 21, Supplement 1, 20 May 1997, Pages S1061-S1066
A.D. Dimitriadis, N. Shah, C.C. Pantelides
Abstract:
One of the major advances in the scheduling of multipurpose plants over the past decade has been the increasing usage of rigorous mathematical programming approaches based on increasingly general problem representations such as the Resource-Task Network (RTN) framework recently proposed by Pantelides (1994). These approaches often require tile solution of large optimisation problems, the size of which increases with the length of the time horizon being considered. For medium-term scheduling problems typically considering time scales ranging from one month to a year, the solution via the direct application of the mathematical programming techniques mentioned above is very often impossible with the currently available computer hardware and solution methods. This paper presents three "rolling horizon" algorithms that are formally based on the rigorous aggregated RTN formulation presented by Wilkinson (1996).
Short term product distribution plan for multisite batch production, warehousing and distribution operations: solutions through Supply-Demand Network and Resource-Task Network optimisation approaches
Computer Aided Chemical Engineering, Volume 8, 2000, Pages 1171-1176
B.P. Das, N. Shah, A.D. Dimitriadis
Abstract:
An industrial production-distribution planning problem encompassing multisite batch processing plants, warehousing and distribution operations, containing a large number of process operational constraints has been investigated using both Supply-Demand Network and Resource-Task Network approaches. Both approaches have been able to solve the problem optimally and generated an integrated product distribution plan; but the distribution patterns are found to be different. Detailed analysis indicate that the Resource-Task Network approach offers a higher potential in solving varieties of integrated planning problems including complicated batch process operations.
Efficient modelling of partial resource equivalence in resource-task networks
Computers & Chemical Engineering, Volume 22, Supplement 1, 15 March 1998, Pages S563-S570
A.D. Dimitriadis, N. Shah, C.C. Pantelides
Abstract:
The Resource-Task Network process representation provides a conceptually simple, unified framework for the development of mathematical programming formulations for multipurpose plant scheduling and design. However, the considerations of complex production processes sometimes leads to very large RTNs and, consequently, mixed integer optimisation problems that are difficult or impossible to solve using currently available techniques. One way of reducing the size of the RTN representation of a process is to identify functional equivalences among subsets of the available resources, thereby allowing a more aggregate treatment of these resources. This paper provides the necessary theoretical basis for the exploitation of partial resource equivalence, which allows large RTNs to be reduced to smaller but completely equivalent ones. Such reductions are particularly significant in problems involving many sequence-dependent changeovers. Examples illustrating these ideas and their impact are presented.
Front : A novel method for training fuzzy neural networks with TSK rules
Control and intelligent systems, vol. 26, no3, 1998, pp. 92-101 (24 ref.)
A.D. Dimitriadis, J.B. Theocharis, G. Vachtsevanos
Abstract:
This paper proposes a novel algorithm, the FRONT (Fuzzy Random Optimization Network Training), for training fuzzy neural networks comprising Takagi-Sugeno-Kang (TSK) fuzzy rules. The FRONT algorithm is based on random optimization methods that have been successfully applied in training of multilayer neural networks. The parameter update is accomplished using Gaussian random vectors with a certain mean and variance values. The mean values are modified during training in favour of directions where a reduction of the objective function was detected during previous iterations. In this paper the search variances are allowed to vary, during training, following a prescribed reduction schedule. The variance reduction improves the overall performance of the training process, and is decided by monitoring the rate of change of the objective function. The FRONT qualities are investigated by a series of simulation experiments. These results demonstrate that FRONT is a robust training method that performs better than the back-propagation algorithm. Finally, a fuzzy neural network is trained by FRONT to predict the outputs of the gas furnace process, showing the validity of FRONT in modelling tasks.
