Soledad Valero

Application of Artificial Neural Networks to Combinatorial Catalysis: Modeling and Predicting ODHE Catalysts

This paper shows how artificial neural networks are useful for modeling catalytic data from combinatorial catalysis and for predicting new potential catalyst compositions for the oxidative dehydrogenation of ethane (ODHE). The training and testing sets of data used for the neural network studies were obtained by means of a combinatorial approach search, which employs an evolutionary optimization strategy. Input and output variables of the neural network include the molar composition of thirteen different elements presented in the catalyst and five catalytic performances (C2H6 and O2 conversion, C2H4 yield, and C2H4, CO2, and CO selectivity). The fitting results indicate that neural networks can be useful in high-dimensional data management within combinatorial catalysis search procedures, since neural networks allow the ab initio evaluation of the reactivity of multicomponent catalysts.

MAS methodology for HMS

Developments in Holonic Manufacturing Systems (HMS) have been reported in three main areas: architectures, algorithms, and methodologies for HMS. Despite the advancements obtained in the first two areas the methodologies for HMS have not received great attention. To date, many of the developments in HMS have been conducted in an almost “empirical way”, without design methodology. There is a definite need to have methodologies for HMS that can assist the system designer at every development steps. This methodology should also provide clear and unambiguous analysis and design guidelines. To this end, in this work we present a Multi Agent Methodology for HMS analysis and design.

Neural networks for modelling of kinetic reaction data applicable to catalyst scale up and process control and optimisation in the frame of combinatorial catalysis

Abstract This work describes an application of artificial neural networks (ANNs) for modelling the kinetics of catalytic reactions using methods not based on fundamental knowledge. Thus, neural networks have been used to model the behaviour of different reactions under different reactor conditions. The modelling of catalytic reactions by neural networks has been demonstrated, and the influence of experimental error in input data has been estimated. In addition, a novel methodology for modelling catalytic data employing already-trained neural networks has been systematically studied by using experimental results from the catalytic hydroisomerization of different n -paraffins. It can be then expected that a new reaction system can be rapidly analysed with a small number of experiments if a library of well-trained neural networks which represent a series of different reaction networks is constructed.

JGOMAS: New Approach to AI Teaching

This paper presents a new environment for teaching practical work in AI subjects. The main purpose of this environment is to make AI techniques more appealing to students and to facilitate the use of the toolkits which are currently widely used in research and development. This new environment has a toolkit for developing and executing agents, called JADE-based game-oriented multiagent system (JGOMAS). The environment also has a dedicated website where students can access different documentation and information and interact with teachers. An actual case study of this environment applied to the practical work component of an advanced AI course is presented.

A FAST Method to Achieve Flexible Production Programming Systems

One of the main critical problems in manufacturing system domains is the production scheduling process, because an agile and reactive production planning and scheduling system is essential in manufacturing. The production scheduling process is a complex problem in which finding a suitable production scheduling can greatly increase the effectiveness of highly flexible production processes. Nevertheless, this high flexibility makes the production scheduling and acquisition of relevant data quite complicated. Therefore, there is a strong demand for a universal and flexible tool for production scheduling capable of increasing the utilization of resources and that supports a decision-making process for the selection of production orders. In this paper, a flexible and adaptive scheduling tool (FAST) to develop an adaptable, fault-tolerant, and scalable scheduling system for a manufacturing environment is presented. This approach is based on multiagent systems (MAS), which provide a natural way to solve problems in domains of this kind.

An argumentation framework for supporting agreements in agent societies applied to customer support

This work presents a system for customer support that integrates case-based reasoning functionalities with an argumentation framework for agent societies. This integration allows to automatically engage in agreement processes to decide the best solution to apply to solve an incidence that has been received in a call center. In this way, the quality of the response would be increased and the company running the call center can take advantage over its competitors in the market.

Goodness and lacks of MAS methodologies for manufacturing domains

Multi-agent system technology has achieved enough development level to be applied in complex problem domains, such as manufacturing systems. This work contributes to demonstrate this applicability, evaluating its goodness and lacks. Thus we have employed a production task scheduling problem in a ceramic tile factory as a real case study. This complex problem requires robust and flexible software applications.

Enhancing Smart-Home Environments using Magentix2

Abstract Multi-agent system paradigm has been envisioned as an appropriate solution for challenges in the area of smart-environments. Specifically, MAS add new capabilities such as adaption, reorganization, learning, coordination, etc. These features allow to deal with open issues in the context of smart-homes such as multi-occupancy, activity tracking or profiling activities and behaviors from multiple residents. In this paper, we present Magentix2 as a suitable MAS platform for the development of dynamic smart environments. Specifically, the use of Magentix2 ( http://gti-ia.upv.es/sma/tools/magentix2/index.php ) facilitates the management of the multiple occupancy in smart living spaces. Normative virtual organizations provide the possibility of defining a set of norms and organizational roles that facilitate the regulation and control of the actions that can be carried out by internal and external agents depending on their profile. Moreover, Magentix2 provides a tracing service to keep track of activities carried out in the system. We illustrate the applicability and benefits of Magentix2 in a set of scenarios in the context of smart-homes.

SoftComputing Techniques Applied to Catalytic Reactions

Soft computing techniques have been applied to model and optimize the kinetics of catalytic reactions. Genetic algorithms have been combined with already trained neural networks to obtain new catalytic samples with relevant quality values. Thus, a suitable model of a genetic algorithm has been designed and implemented for this problem. In order to improve the results offered by this genetic algorithm, different combinations of its parameters have been studied. Moreover, this soft computing approach has been applied to some industrial reactions. Therefore, specific neural networks trained for each kind of reaction and the genetic algorithm have been joined together, getting successful results.

Applying Genetic Algorithms in Chemical Engineering for Determining Zeolite Structures

Zeolites are crystalline materials widely used in many catalytic process in industry. Specifically, they have a major impact at petrochemicals, fine chemicals or gas separation. Thus, discovering new zeolites with specific properties is a high-impact target for the industry, due to their huge economical repercussions. New tools and techniques are needed to help in this task, because trial and error approaches prevail until now. In this work, we propose a new application of genetic algorithms for helping chemical engineers in the process of determining zeolite structures with specific properties. Our proposal takes advantage of some symmetry operation properties to improve the performance of the genetic algorithm. Furthermore, a suitable fitness function has been designed which evaluates all main features required for efficient zeolites.