HYBRID FLOWSHOP SCHEDULING UNDER UNCERTAINTY: ARCHITECTURE AND ALGORITHMS. APPLICATION TO THE CERAMIC TILE INDUSTRY.

In a time of global competence in which the lead times are increasingly more prominent as competitive element and where, in not few occasions, companies tend to offer a catalogue of extensive products and differentiated from the competence, multiple challenges exist that the Organizations should confront. Operations Management has the challenge to adapt the processes of Management of the Logistic and Productive Systems to the current needs.  In this process of change is habitual to leave from flexible small Productive Systems and to take the mass production approach in which is fundamental to employ the best "know-how" to try to obtain the most adequate performance of the available resources. The unfold of some good practices in the environment of the Production Scheduling can greatly help to improve the efficiency of the resources.  

Traditionally, Production Scheduling has been taken under a rather quantitative vision whose mission consisted of assigning, to arrange and to temporize the different works of the period based on the available resources.  Nevertheless, although still considering valid this approach, this thesis aims to emphasize that, in reality, the last goal of the techniques and methods developed during years within the environment of the Production Scheduling is not another than being applied within Decision Making Aid Systems. In this sense, the operative decisions taken in the Production Scheduler area should be connected in all the cases, at least, with its more direct decisional environment, as is that of the Production Planning.  

An in depth literary review within the extensive work carried out in more than 50 years of existence of what has been called as "Scheduling" shows the existence of a need that should be covered. It is a matter of reducing the distance among the problems presented by the scientific community and the ones that professionals want to solve.  The academic problems are characterized for being simplifications of the reality in which the productive systems, the working conditions, the organizing interrelations, or the own apparition of events are considered from a partial point of view, or even ignored.  For that reason in many cases the procedures and the techniques applied to the Production Scheduling in the organizations have not received the attention needed.  

One of the aspects to be considered to reduce the gap between the academic environment and the professional one is that of the Uncertainty.  Relatively few bibliography exists that consider the problem of the Production Scheduling under uncertain environments, and from the revised proposals, it has been decided to follow the model that proposes to divide the problem into two phases. The first phase, that receives the name of Predictive Scheduling, consists of establishing a Productive Schedule in a deterministic environment that serve of starting point. The second phase consists of modifying the Productive Schedule in process in the occurrence of an event that do it unfeasible, and it is called Reactive Scheduling.  The Predictive and Reactive Production Scheduling serve to generate Schedules from restrictions and objectives that can be different given since they are employed in different instants from the Productive period considered.  

With the objective of exploring paradigms that contributed a different focus to the traditional one relating to the use of techniques and tools to deal with the problem of Production Scheduling, in the current conditions, the possibilities of the Artificial Intelligence were explored, and in concrete the area related to the Intelligent Agents and Multi-agents Systems.  The paradigm of the Agents has been applied to diverse areas like Operations Management.  In the area mentioned its put in practice has been mainly focused on the environment of the Production Planning and of the Supply Chain Management.  This due to that the aid Agents theory to undertake aspects as the Business Integration, Organization Distributed, the Heterogeneous Environments, the Interoperability, the Dynamic Structures, or the Cooperation among others. From the need of including some of these aspects in the Production Programming ambit, it has been considered that the Multi-agents Systems can act as facilitators. 

Once the scientific contributions related to the environment of the "Scheduling" and the Multi-agents Systems have been analyzed, it has been identified that the objective of this thesis should be to introduce a Model that permit to Automate, by means of the employment of Multi-agents Systems, the Predictive and Reactive Production Scheduling, based on a Business Processes orientation that include Tools (Algorithms) for aiding to make decisions under Uncertainty, which are necessary to solve the functional/decisional vision that encloses the existence of events and its management, and that permit to take the first step toward a Dynamic Model. The objective set for the model is that of maintaining a rate of average productivity as high as possible in function of the available resources, having available the processes of Predictive and Reactive Scheduling objective functions that potentiate it. 

As a previous step to the construction of a proposal it has been carried out an important effort to correctly delimit its own validity. In this aspect, those characteristics of the problem desired to treat are analyzed, emphasizing its peculiarities and being been able to characterize as a Hybrid Flow Shop with Sequence Dependent Setup Times in conditions of Uncertainty.  The problem is mathematically modeled. 
  
In order to reach the proposed objective a Methodology, some Architectural elements, and some Tools that serve of guide and support are defined. These elements, at the same time, structure the elaboration of a proposal.  Through five steps defined within the methodology, and employing in each one of them the constructive elements and necessary tools, it is followed a path that is initiated with the elementary definition of the problem and that finishes with a test of operation for a particular case. 
 
With the two first steps of the methodology were determined the requests in which the proposal should be based to obtain its objective.  For the analysis of the requirements of the Predictive and Reactive Scheduling process and its relation with the Production Planning is employed the proposed named CIMOSA.  With the third step of the methodology, and in base to the identifying requests, it is undertaken the design of a Software Platform conceived as a Multi-agent System (IPSU-MAS) that facilitates the automation of the processes of Scheduling.  As design Methodology is employed INGENIAS, which is specific for developments based on Agents.
  
Before carrying out the fourth step, that consists of the implementation of IPSU-MAS, it is designed, developed, and analyzed a series of algorithms and procedures that necessarily should be included into the Software Platform.  If the platform is considered as a structure, these processes would be the motor that really enables the automation.  These are defined from a conceptual analysis and a practical contrast.  Among the most prominent algorithms develop it is necessary to emphasize those that permit the calculation of the Predictive Scheduling and of the Reactive Production Scheduling.  Although in smaller measure they also undertake those algorithms that regulate the coordination between the processes of Scheduling and Planning.

In order to be able to obtain a Predictive Production Schedule it is proposed the design and implementation of a new Genetic Algorithm called SMAGA.  The main difference among the proposal carried out in this thesis and the traditional Genetic Algorithms, is that in this case, a hybridization of the algorithm is carried out by combining the basic structure of this Metaheuristic with the characteristics of the Multi-agent Systems.  The objective is that the hybridization be produced so that the Agents take the role of "alive beings" and therefore, of individuals of the population.  With the introduction of the paradigm of Agents in this algorithm the proposal considers the existence of a Multi-agent System (SMAGA) inside another (IPSU-MAS).  After a proper parameter setting of the two variants of the SMAGA algorithms, the results are compared with the obtained by other proposals carried out for the same type of problems, turning out to be highly efficient.  For both the parameter setting and for the comparative analysis a known set of data is employed.  

In order to carry out the calculation of the Reactive Schedule, it is carried out a comparative analysis between a selected set of Heuristics and Metaheuristics, existing an own contribution (SSDS).  All the techniques have undertaken an extensive study under diverse circumstances, being able to obtain prominent conclusions regarding their behavior. 
Finally it is carried out, as part of the fifth step of the methodology, an application of the prototype IPSU-MAS, generated from the work carried out previously, to a real world company of the Ceramic Tile Industry.  It is considered that the companies of this sector are very adequate to carry out a practical experience due to the configuration of their manufacturing systems and competence level reached within their environment.  In the practical experience were compared the results obtained when employing the current methods of the Company Selected and when applying IPSU-MAS, in diverse settings.  The conclusion is that in all cases the results obtained with the IPSU-MAS are better, and many of them have reached values of productivity rate of between a 17% and a 29% better than the ones obtained with current techniques.