Shigeharu Kito

Neural network as a tool for catalyst development

Abstract This article introduces a novel information science technique, an artificial neural network, which will possibly be a powerful tool for catalyst development. As an example of synergistically generated catalytic functions, the neural network has been applied to the estimation of acid strengths of mixed oxides, and it has been shown that not only the interpolation but also the extrapolation of given acid strength data is possible within reasonable experimental error. The neural network also has been successfully applied to the estimation of catalytic performance, such as the catalytic activity of a series of lanthanide oxides in the oxidation of butane and the selectivities to various products in the oxidative dehydrogenation of ethylbenzene on a series of promoted SnO 2 catalysts. On the basis of these results, some remarks are given on its application to catalyst development.

An expert systems approach to computer-aided design of multi-component catalysts

Abstract INCAP-Muse, a prototype expert system for the design of multi-component catalysts, was developed. Started from the reactants and the products given in advance, INCAP-Muse identifies that best suited target reaction, and then selects the multi-component catalyst candidates for that reaction through various kinds of inference. When INCAP-Muse was applied to the catalyst design for the oxidative dehydrogenation of butene, a very satisfactory result was obtained.

2.17 Acid Strength of Binary Mixed Oxides: –Estimation by Neural Network and Experimental Verification

Abstract In order to demonstrate the feasibility of neural network system for the estimation of acid strength of mixed oxide, two series of mixed oxides were prepared and their acid strengths were compared with those predicted by the neural network. It was revealed that the acid strength of mixed oxide can be estimated by the system within a reasonable error.

Expert Systems Approach to Catalyst Design - Application and Experimental Verification

Abstract A prototype expert system for catalyst design, INCAP ( In tegration of C atalyst A ctivity P attern), was applied to the estimation of catalytic performances of a series of promoted SnO 2 catalysts in the oxidative dehydrogenation of ethylbenzene, and the estimated performances of eighteen promoted SnO 2 catalysts were compared with those measured experimentally by using a conventional flow reactor. The index to the overall performance adopted in the INCAP agreed fairly well with the experimental one, but the disagreement was significant for some catalysts. Then new indexes to the activity and selectivity were introduced and compared with those measured experimentally. In these cases, good correlations were obtained between them, which strongly demonstrates the feasibility of the expert systems approach to the design of catalysts.

Applicability of Neural Network to the Estimation of Acid Strength of Binary Mixed Oxides

The applicability of a neural network based on error back-propagation algorithms to the estimation of strength of synergistically generated acid sites in binary mixed oxides is discussed. As the results, the neural network is able to estimate such acid strength within experimental errors with the exceptions for some MgO -containing mixed oxides.

Knowledge-Base Approach to the Creation of Catalytic Reaction Mechanism

A prototype knowledge-based system was developed for the creation of catalytic reaction mechanism by effectively utilizing Prolog language for the representation of knowledge of the structure of chemical species, the elementary reactions, and so on. The feasibility of the system was demonstrated by taking the hydrogenation of carbon monoxide to acetic acid as an example.