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:

  • Control Chart ln(S2) to Control Process Variability

    Authors: José Gomes Requeijo, Ana Sofia Matos, Zulema Lopes Pereira
    Affiliation: UNIDEMI, Departamento de Engenharia Mecânica e Industrial, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, 2829-516, Caparica, Portu
    Primary area of focus / application: Process
    Keywords: Statistical Process Control , Dispersion Control Chart , Normal distribuition , ln(S2) statistic
    Submitted at 7-Jun-2011 17:27 by Ana Matos
    Accepted
    5-Sep-2011 17:00 Control Chart ln(S2) to Control Process Variability
    Statistical Process Control (SPC) uses a wide range of quality tools to provide organizations with processes that consistently satisfy their customer’s needs. The SPC methodology places special emphasis on control charts, through which process stability can be monitored.
    The traditional control charts developed by Walter Shewhart are based on certain assumptions, namely the presence of independent and identically distributed data following a Normal distribution. Whenever the sample size is equal or greater than 4 it is considered that, in most cases, the sample means may be modelled by a Normal distribuition. As regards the dispersion control accomplished through R or S charts, the assumption does not hold, since both range and standard deviation distributions are not Normal. However, classical R or S charts are still constructed considering the assumption of Normality, by establishing the control limits at three standard deviations of the corresponding distribution of the variable (R or S).
    To mitigate this weakened assumption, the authors suggest the use of the ln(S2) statistic. The ln(S2) statistical distribution is considered approximately Normal, provided that the individual quality data follows a Normal distribution. Control limits for ln(S2) control chart are presented, both for Phase 1 and Phase 2 of SPC. Additionaly, new factors specific to the ln(S2) control chart, which were developed by the authors, are presented assuming the same significance level of the traditional charts.
    Some examples are introduced, with different sample sizes, enabling the performance comparison between the ln(S2) chart and the R and S traditional charts.
  • Multivariate Control Charts based on Signal Processing Tools

    Authors: Prof. Song Won Park (1) Prof. Marco Paulo Seabra dos Reis (2) Prof. Gustavo Matheus de Almeida (3)
    Affiliation: (1) University of Sao Paulo, Brazil, (2) University of Coimbra, Portugal, (3) Federal University of Sao Joao del-Rei, Brazil
    Primary area of focus / application: Process
    Keywords: Fault detection and diagnosis , Multivariate statistical process control , Signal processing tools , Hidden Markov models , DAMADICS Benchmark
    Submitted at 8-Jun-2011 01:05 by Gustavo Almeida
    Accepted (view paper)
    5-Sep-2011 10:50 Multivariate Control Charts based on Signal Processing Tools
    Early fault detection and diagnosis is still a challenge in monitoring operations in industries worldwide. One reason for this, concerns the spatial overlapping among distinct fault classes, which implies that some abnormal situations may only be distinguished from others by taking into account the specific order of occurrence of the events. Most applications employ monitoring systems that make use of metrics based on residues, which are given by discrepancies between observed and estimated values. An alternative approach to such schemes is the adoption of signal processing tools, such as hidden Markov models (HMM). In this context, this work illustrates the use of the HMM method, which was already employed successfully in speech recognition applications, in order to carry out the monitoring tasks of both detection and diagnosis. In summary, the proposed methodology results in a multivariate control chart based upon HMM. We present its application to the case study of the DAMADICS benchmark, with the focus on the actuator system responsible for controlling the thin juice flow rate to the first stage of the evaporation station. Both abrupt and incipient faulty situations were investigated. Regarding the former type of faults, detection and diagnosis was done successfully, and in what concerns to the latter ones, they were carried out in a correct and progressive way. Some positive aspects to be pointed out in a control chart approach based on HMM, include: the familiarity arising from their similitude with classical control charts, the updated information about the system current state (tendency), the limitation of the search space of candidate faults, contributing to early system recovery, and the awareness of yet unknown faults.
  • Statistical Perspective of Model Selection for Industrial Chemical Recovery Boilers

    Authors: Prof. Song Won Park (1) Prof. Gustavo Matheus de Almeida (2)
    Affiliation: (1) University of Sao Paulo, Brazil, (2) Federal University of Sao Joao del-Rei, Brazil
    Primary area of focus / application: Modelling
    Keywords: Variable selection , Model selection , Steam generation , Industrial data analysis
    Submitted at 8-Jun-2011 01:17 by Gustavo Almeida
    Accepted
    7-Sep-2011 12:20 Statistical Perspective of Model Selection for Industrial Chemical Recovery Boilers
    Very large process historical databases are a common scenario in medium- and large-industries plants nowadays. Massive data repositories contains redundant, relevant and irrelevant variables to a goal in particular. Hence, the challenge is to select a subset of relevant variables according to the specific objective. Variable elimination has several benefits, e.g. economical, by reducing the cost and time of data collection, and physical, by making the data handling and the phenomena interpretation easier. Moreover, it improves the model performance which can be negatively affected by, e.g. the multicollinearity problem. In the first step, this study applies variable selection techniques, from classical ones such as stepwise regression, to genetic algorithms and principal components analysis, to obtain subsets of relevant process variables with respect to steam generation in a chemical recovery boiler. The case study uses a real database collected in a boiler belonging to a Brazilian Kraft pulping mill. It contains fourteen process variables and almost 3000 registers for each one of them. After selecting variable subsets of different sizes, the second part use them as inputs and analyze the results concerning model selection from a statistical perspective.
  • A Survey of Advances in Control Chart Methodology

    Authors: Moshe Pollak
    Affiliation: The Hebrew Universsity of Jerusalem
    Primary area of focus / application: Process
    Keywords: SPC , Cusum , Quality Control , Shiryaev-Roberts
    Submitted at 9-Jun-2011 08:00 by moshe pollak
    Accepted (view paper)
    7-Sep-2011 10:00 A Survey of Advances in Control Chart Methodology
    Classical control charts are Shewhart, Cusum, EWMA and variations thereof. Most of these methods require knowledge of baseline distributions/parameters, and in essence use a "representative" for the unknown post-change regime. Furthermore, for lack of an alternative, these methods have been applied to complicated situations that they are not designed for (such as applying a Shewhart chart to detect a change of regression). There is a price to be paid for ignorance of distributions an use of methods in situations for which they are not designed; usually there are discrepancies between nominal and actual ARL's and detection of a change may be slower than one could otherwise attain.

    We will survey a likelihood-ratio approach that guarantees efficiency as well as the accuracy of ARL's in complex scenarios. Examples will be given.
  • Risk-based Maintenance Policy selection by using Simulation Models and Design of Experiments

    Authors: Seyed Mojtaba Sajadi, Habib Mortazavi
    Affiliation: Industrial Engineering Department, Islamic Azad University Najafabad Branch,Najafabad ,Iran
    Primary area of focus / application: Education & Thinking
    Keywords: Risk-based Maintenance , Simulation , Design of Experiments , Maintenance Policies
    Submitted at 12-Jun-2011 08:45 by S.Mojtaba Sajjadi
    Accepted
    5-Sep-2011 16:55 Risk-based Maintenance Policy selection by using Simulation Models and Design of Experiments
    Maintenance is a complex process that is triggered by equipment failure or planned repair. This process requires planning, scheduling, control and deployment of maintenance resources to perform necessary maintenance activities. The simulation model replicates the existing operational policies and evaluates feasible policies to optimum system performance. The measures of performance are average server utilization and average waiting time. Also simulation and Design of Experiments (DOE) have been used to model different maintenance activities and polices such as CM, TBM and PM for one-machine, one-product manufacturing system to finding the best maintenance policy and comparison the different scenarios that minimize the total cost of maintenance and inventory. Each of these models focuses on an activity of the maintenance system.
  • Multivariate SPC of dynamic processes: application of DPCA-MD to the Tennessee Eastman process

    Authors: Tiago J. Rato; Marco S. Reis.
    Affiliation: University of Coimbra
    Primary area of focus / application: Process
    Keywords: Dynamic multivariate statistical process control , Principal component analysis , Dynamic Principal component analysis , Missing data imputation methods , Tennessee Eastman process
    Submitted at 14-Jun-2011 12:24 by Tiago Rato
    Accepted (view paper)
    5-Sep-2011 17:20 Multivariate SPC of dynamic processes: application of DPCA-MD to the Tennessee Eastman process
    The multivariate approaches commonly adopted to monitor industrial process are usually based on principal component analysis (PCA) [1, 2] and dynamic PCA (DPCA) [3]. The first method is able to deal with cross-correlation, while the second is employed when besides cross-correlation, auto-correlation is also present in data. However multivariate SPC using DPCA still leads to auto-correlated statistics. To handle this issue, we have developed a set of multivariate statistics based on the integration of DPCA and Missing Data (MD) methods. Such statistics, which are new, were applied to the Tennessee Eastman process [4], which is a representation of a typical industrial process presenting both cross-correlation dynamics and non-linearity. The results then obtained, are also comparable with others previously presented in literature (e.g. see reference [5]). From this study, we can conclude that the proposed monitoring statistics lead to the highest detection rates on 19 of the 21 process distribution studied, and showed to be superior, in a statistically significant sense, to their PCA and DPCA counterparts. The combined DPCA-MD statistics also presented lower auto-correlation, which improves their implementation and reliability.

    References:
    1. Jackson, J.E., Technometrics, 1959. 1(4): p. 359-377.
    2. Jackson, J.E. and G.S. Mudholkar, Technometrics, 1979. 21(3): p. 341-349.
    3. Ku, W., R.H. Storer, and C. Georgakis, Chemometrics and Intelligent Laboratory Systems, 1995. 30: p. 179-196.
    4. Downs, J.J. and E.F. Vogel, Computers and Chemical Engineering, 1993. 17(3): p. 245-255.
    5. Russell, E.L., L.H. Chiang, and R.D. Braatz, Chemometrics and Intelligent Laboratory Systems, 2000. 51(1): p. 81-93.