ENBIS-11 in Coimbra

4 – 8 September 2011 Abstract submission: 1 January – 25 June 2011

My abstracts


The following abstracts have been accepted for this event:

  • Multivariate linear regression models for heterogeneous liquid-liquid reactions in microstructured reactors

    Authors: C.M.S.G. Baptista1, T. Mendes1, C. Dias1, M. S. Reis1, P. Löb2
    Affiliation: 1Chemical Engineering Department, Faculty of Sciences and Technology-University of Coimbra, 3030-790 Coimbra, Portugal 2Institut für Mikrotechnik Main
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Multivariate Linear Regression , liquid-liquid , microstructured , reactors
    Submitted at 1-Jun-2011 10:37 by Tânia Mendes
    Accepted (view paper)
    5-Sep-2011 17:20 Multivariate linear regression models for heterogeneous liquid-liquid reactions in microstructured reactors
    Many bulk chemical processes are multiphase, often liquid-liquid, and in these heterogeneous systems several complex mechanisms for mixing, mass and heat transfer and chemical reaction take place simultaneously, affecting the reaction selectivity and yield. Mass transfer through the interfacial area and temperature control are known to influence the formation of by-products and microstructured reactors are being developed to improve mixing efficiency, while enhancing heat and mass transfer processes. Scale-up of these microreactors will certainly benefit from data collected in experiments and the modelling of these results, enabling to obtain valuable kinetic information.

    A broad range of experimental conditions were explored at labscale while processing an exothermic liquid-liquid reaction. After a preliminary analysis for screening the most influent operating conditions and colinearity, these experimental data were used to develop Multivariate Linear Regression (MLR) models capable of describing the formation of products as well as the two main by-products. Apart from reactor temperature, the independent explanatory variables were found to be inlet conditions or their nonlinear relationships. These include flow rates, concentration of main reactants or residence time, which can be set during operation in order to reach the target productivity and selectivity. To minimize the information required and facilitate its use, even during operation of an industrial plant, the models are based on the minimum number of variables still allowing for good prediction ability.

    The authors acknowledge the funding support from the 7th European Framework Program within the PILLS Project.
  • Bayesian Parameter Estimation of Kraft Pulping Process for Eucalyptus Globulus

    Authors: Maria Graça Videira Sousa Carvalho (1), Song Won Park (2)
    Affiliation: (1) - Chemical Engineering Dep., University of Coimbra, Portugal; (2) - Chemical Engineering Dep. Polytechnic School. University of Sao Paulo, Brazil
    Primary area of focus / application: Modelling
    Keywords: bayesian inference , kinetic model , kraft pulping , Monte Carlo method
    Submitted at 1-Jun-2011 15:05 by M. Graça V.S. Carvalho
    7-Sep-2011 12:00 Bayesian Parameter Estimation of Kraft Pulping Process for Eucalyptus Globulus
    The bayesian inference is a very good tool to make decisions under uncertainty. To build a kinetic model of a kraft pulping process, the bayesian approach is applied, discriminating the several models and estimating the model parameters by the Monte Carlo method. This approach is very efficient in describing kinetic equations for wood pulping. The kraft process is the most used one to produce cellulosic fibres as raw material for papermaking. Since there are different structures of lignin in wood, the kraft pulping process is a very complex treatment of this natural lignocellulosic material, coupling different stages of delignification besides the usual alkali concentration, temperature and time as independent variables. In this work the model and parameter estimation methods are described and the evaluation of its capabilities to build a kinetic model is discussed under a statistical perspective.
  • Comparative analysis of combination methods: accuracy and linear correlation between the errors of individual forecasts

    Authors: Vera Lúcia Milani Martins, Liane Werner, Diego de Castro Fettermann
    Affiliation: Industrial Engineering, UFRGS - Brazil
    Primary area of focus / application: Economics
    Keywords: Forecasting , Combining Forecasts , Accuracy , Correlated Errors
    Submitted at 2-Jun-2011 04:22 by Vera Lúcia Milani Martins
    5-Sep-2011 10:10 Comparative analysis of combination methods: accuracy and linear correlation between the errors of individual forecasts
    Combination of forecasts is considered a successful alternative to the techniques of individual forecasts. This format forecast incorporates features from different approaches with improve accuracy. In these forecasts, the errors obtained in the individual forecasts are considered by some authors as independent events. Other studies don’t reference the type of relationship between the errors of forecasts analyzed. Therefore, the aim of this study is to evaluate the relationship between the accuracy of forecasts combined and magnitude of linear correlations between the errors of individual forecasts. The method applied in this study used 50 industrial real series, which was randomly selected from M3-Competition study. The analysis proceeded with the following 3 steps: i) Individual modeling RNA and SARIMA, ii) Forecasting calculation by combination (mean, minimal variance with and without correlations between errors), iii) and the comparison of the combinations by measuring from accuracy. The results show that the combination by minimum variance that considers the null correlation show the highest frequency to selection according to the measures MAPE (60%), MSE (54%) and MAE (56%). It was observed that the method of minimum variance with not null correlation achieves 75% of the selections when the linear correlation value varies between 0.6 and 0.7.
  • Statistical Analysis of Xbar - chart performance by Using DOE

    Authors: Adler Yu, Maksimova O., Shper V.
    Primary area of focus / application: Education & Thinking
    Keywords: Xave-chart , special causes of variation , DOE , regression model
    Submitted at 2-Jun-2011 08:57 by Vladimir Shper
    Accepted (view paper)
    5-Sep-2011 17:25 Statistical Analysis of Xbar - chart performance by Using DOE
    This paper is a natural continuation of our last year presentation where we started to discuss an extended notion of special causes of variability and their impact on the construction and interpretation of Shewhart control charts. Here we analyze the performance of a very simple chart for averages under conditions when special causes change not only the parameters of the underlying distribution but its type as well. We use DOE for planning our study in order to make this search more systemic than one-factor experiment used traditionally.
    To this purpose we chose 9 factors responsible for different conditions of intervention into the process and realized the regular fraction factorial design of 2m3n with 36 runs for Xbar - chart. We simulated 10 parallel realizations in each run. The performance of a chart in each run was estimated by the power function (the probability of finding a special cause). On the base of this experiment we have built a regression model and investigated it. The most important result of our investigation is that an adequate regression model includes not only linear effects but pair interactions as well. This means that classical analysis of Shewhart charts based on one-factor experiments has limited value and should be specified. We hope that the results obtained will be useful for the development of SPC applications
  • Clients, Professional and Analyst Dimensions: Intra and Inter Relationships Analysis in the Decision-making

    Authors: Isabel F. Loureiro, Celina P. Leão, Pedro M. Arezes
    Affiliation: Engineering School of University of Minho, Guimarães, Portugal
    Primary area of focus / application: Business
    Keywords: Clients and Professional Profile , Decision-making , Ergonomic analysis , Multivariate Statistics , Weighting Table
    Submitted at 2-Jun-2011 16:09 by Celina Leao
    Accepted (view paper)
    7-Sep-2011 12:40 Clients, Professional and Analyst Dimensions: Intra and Inter Relationships Analysis in the Decision-making
    Considering the Clients, Professional and Analyst dimensions, the ETdA (Ergonomic Tri dimensional Analysis) matrix assembling leads to the weighting table helping the Analyst in the ergonomic intervention’ decision. A three level methodology is proposed: (1) descriptive analysis to allow the characterization and study of the different answers profile in Clients and Professional dimensions; (2) correlation between the different answer categories and the level 1 results; (3) ergonomic factors’ multivariate analysis. In the level 3, an inter and intra dimension analysis will be done. The main issue of the intra dimensions analysis is the ergonomic factors relevance and intensity study, that is, to understand how the different ETdA dimensions feel the ergonomic factors and, to measure the intensity of the ergonomic perception in each ETdA dimension. The inter dimension analysis, allows the understanding of the ETdA dimensions’ relationships importance helping the results weighting. The mechanisms that regulate the interaction between the ETdA dimensions will have a positive impact in the professional workplace (commercial areas with free circulation of people) and consequently in clients’ general opinion on those areas contributing for the success of management strategies.
  • Boosting Approach for Monitoring Non-Parametric Multivariate Control Charts

    Authors: Dr. Abbas Saghaei, Sepide Sahebi, Samaneh Maranlou
    Affiliation: Science and Research Branch, Islamic Azad University, Tehran, Iran
    Primary area of focus / application: Process
    Keywords: Multivariate Control Chart , Non-parametric control Chart , Data mining , Boosting
    Submitted at 4-Jun-2011 07:39 by sepide sahebi
    5-Sep-2011 17:15 Boosting Approach for Monitoring Non-Parametric Multivariate Control Charts
    Multivariate control charts are being applied to solve many statistical process control (SPC) problems. Hotelling T^2 is a well performed technique, but when the quality characteristics are not distributed normally it is not applicable. So in this situation the non-parametric multivariate control charts are required to deploy. Recently using data mining approaches in SPC is of interest to researchers. In this paper boosting approach to monitor multivariate non-parametric control chart based on a boosting method is presented. Boosting technique has been used to identify in-control and out-of-control condition, and the results of proposed method where achieved by using Monte Carlo simulation where compared with parametric multivariate control charts.