Search

During the last several years, the development of combinatorial technology has enabled synthesis of a huge amount of chemical compounds in a short time. The large number of variables makes the direct human interpretation of data derived from combinatorial experimentation for high-throughput screening (HTS) very difficult. Artificial neural networks using multilayer perceptrons (MLP) have been successfully applied to the regression problems with various material data. In this work, MLP model was applied to HTS of ferroelectric materials including Bi 4− x La x Ti 3 O 12 (BLT) and Bi 4− x Ce x Ti 3 O 12 (BCT). The model using MLP was made to predict the ferroelectric properties of whole feasible experimental conditions. Once a neural network model with high accuracy and good generalization performance was established, we could predict the expected optimal reaction conditions with the best characteristics. The highest gradient value obtained using MLP model is higher than the maximum value found from experiments, thereby accelerating the discovery of the optimal compositions and post-annealing time of BCT and BLT.

This study considers a scheduling algorithm and mathematical models for a real polyvinyl chloride (PVC) plant. According to general rules of the PVC plant, the basic PVC scheduling model is formulated as mixed integer linear programming (MILP). This model includes production, inventory management, packing, and shipment processes. The basic PVC scheduling model is customized for two PVC production processes with different characteristics, resulting in two detailed PVC scheduling models. An optimization algorithm has been developed to solve these scheduling models. This algorithm is a hybrid algorithm combining the genetic algorithm for integer programming (IP) and the interior point method for linear programming (LP). A heuristic technique has also been developed to reduce the number of solving LPs. Finally, the optimized results of the models were analyzed.

An integrated scheduling system with demand forecasting has been developed. The SARIMA (seasonal autoregressive integrated moving average) model is automated to forecast the demand. In order to reduce the cost and instability of production scheduling, a safety inventory determination is also automated additionally. A scheduling system based on the mathematical model and the efficient optimization approach has been proposed. The mathematical model has been proposed for scheduling of the entire process of a PVC (Poly Vinyl Chloride) plant, including production, inventory, packing and shipment. Additionally, a more efficient optimization approach has been proposed. For validation purposes, a real PVC process was tested. The optimization result shows that the proposed mathematical model and optimization approach resulted in an increase of at least 4.5% in profits, obtaining good results in a reasonable amount of time; approximately 5 minutes by a 3.0 GHz PC.

Kang, Sookil Clerc, Frederic Farrusseng, David Mirodatos, Claude Woo, Seong Ihl Park, Sunwon 1st International Combinatorial Catalysis Symposium (ICCS) JUL 20-22, 2008 Daejeon, SOUTH KOREA; The issues of heterogeneous catalyst optimization are presented in the framework of high throughput iterative screening. To be efficient, the optimization procedures should consider the limitations of the facilities in terms of screening capacities, experimentation costs, and experimental noise. The issues of algorithm reliability are also addressed. Based on the simulation results, this work highlights the most important features of the evolutionary strategies (ES) that lead to successful optimizations. We show that the monitoring of the population diversity during the optimization is a key parameter. Finally, we provide some best practice recommendations for experimentalists who are not experts in metaheuristic methods and who are willing to apply ES for material library designs.

During the last several years, the development of combinatorial chemistry has enabled synthesis of a huge amount of chemical compounds in a short time. Therefore HTS (High-throughput screening) is required for dealing with the enormous amount of data. But human intervention (trial & error method) in data mining of experimental results lowers the efficiency of HTS. So self-organizing neural networks that rapidly and accurately transact experimental results are needed for the improvement in HTS performance. The self-organizing algorithms that were previously developed have randomness which causes unrealiability of algorithms which means different trials give quite different performances. However, in the proposed algorithm, randomness of neural networks is effectively eliminated by the construction of the optimized neural networks structure with hidden-neuron and hidden-layer addition. So this algorithm always matches the complexity of the model to that of the problem very well. As a result, this algorithm can find a near-optimal network which is compact and shows good generalization performance without human intervention.