ENBIS-12 in Ljubljana

9 – 13 September 2012 Abstract submission: 15 January – 10 May 2012

My abstracts

 

The following abstracts have been accepted for this event:

  • On the Construction of Control Charts for the Logarithmic Distribution

    Authors: Stelios Psarakis (Athens University of Economics and Business. Dept of Statistics), Elisabeth Topalidou (Athens University of Economics and Business. Dept of Statistics)
    Primary area of focus / application: Process
    Keywords: logarithmic distribution, control charts, Average Run Length, parameter estimation
    Submitted at 9-May-2012 10:48 by Stelios Psarakis
    Accepted (view paper)
    10-Sep-2012 16:45 On the Construction of Control Charts for the Logarithmic Distribution
    Logarithmic distribution is one of the discrete distributions with various applications some of which are in ecology, engineering, pharmacology, biochemistry, telecommunications and others. As a result of its variety of applications, it appears to be important that control charts to detect shifts in mean and variability should be constructed on the assumption that the distribution of the quality characteristic under study is the logarithmic distribution. Here we construct Shewhart-type and probability-type control charts for detecting parameter shifts for the logarithmic distribution and investigate their performances. We also deal with the construction of CUSUM procedures for the logarithmic distribution. The case of using estimate of the parameter of the logarithmic distribution in the construction of the control charts mentioned above is also discussed. Finally, the construction and the investigation of the performance of the proposed charts are illustrated using simulated and real data.
  • Profile Prediction from Online Data

    Authors: Marko Limbek (University of Ljubljana, Faculty of Economics)
    Primary area of focus / application: Mining
    Keywords: online portal user information, demography prediction, brute force modeling, automatic feature detection
    Submitted at 9-May-2012 12:07 by Marko Limbek
    Accepted (view paper)
    10-Sep-2012 15:55 Profile Prediction from Online Data
    The article describes analysing the data for ENBIS Challenge 2012. The goal was to predict demographic profile of online portal users based on what information they look for on the portal in order to be able to deliver tailored advertisment to them. A new model was built for each of the following demographic targets: age, job, sex, salary, household. Data consisted of 73730 records and 939 features. All models were built in the same way which proved to be the most succesful. First the most relevant features were chosen and then the eventual missing data and feature transformation was performed. Second principal component analysis was performed, further reducing the number of features and taking only the most relevant factors into account. Third a number od models was built on training data and best three models were chosen. The target was always a flag, so the models were all classifiers. Finally each of the targets was then predicted on test data and the results were quite satisfying, suggesting that models could be deployed on the online portal. The best predicted target was 'female sex'. What must be emphasized is that we did try to take into account different levels of portal access information, but the brute force method of automatic optimal feature detection and automatic modeling proved to be the more succesful.
  • A Bayesian Approach to Technical Performance Measures in Product Development Projects

    Authors: Pietro Tarantino (Tetra Pak packaging solutions), Leardi Carlo (Tetra Pak packaging solutions)
    Primary area of focus / application: Business
    Keywords: Technical Performance measures, Bayesian estimation, systems engineering, risk analysis
    Submitted at 9-May-2012 14:52 by Pietro Tarantino
    Accepted
    12-Sep-2012 10:25 A Bayesian Approach to Technical Performance Measures in Product Development Projects
    Technical Performance Measures (TPMs) refer to the subset (~6-10) of system & sub-system requirements that, if not met, put a product development project at cost, schedule or performance risk. They are tracked at key milestones, providing indications of technical solution progress, compliance to performance requirements and technical risk.  Practically speaking, for each TPM, at the beginning of the project, the analysis team develop a progress plan with expected value and tolerance bands. The current TPM value from physical test or simulation at the defined milestone is then assessed with respect to the expected value and the tolerances. If the parameter exceeds the tolerance, actions from management have to put in place. Methods for assessing technical risks associated to each TPM often do not consider the performance profile during the development process or the increase of test confidence level as technical maturity progresses.  This work aims at presenting an easy to use and interpret TPM monitoring tool developed with a Bayesian approach.  Examples from Reliability-type, Gaussian-type, Binomial-type and Poisson-type TPMs will be illustrated in theory and practice.  
  • A Quantitative Methodology for Designing Systems for Adaptability

    Authors: Pietro Tarantino (Tetra Pak packaging solutions), Carlo Leardi (Tetra Pak packaging solutions), Andrea Angelini (Tetra Pak packaging solutions), Roberts Nicolini (Tetra Pak packaging solutions)
    Primary area of focus / application: Other: Systems Engineering
    Keywords: Systems Engineering, Architecture Design, Systems Value, Simulation, Validation
    Submitted at 9-May-2012 14:55 by Pietro Tarantino
    Accepted
    10-Sep-2012 17:15 A Quantitative Methodology for Designing Systems for Adaptability
    Manufacturing industries, system products and customer services provide value through their ability to fulfil stakeholders’ needs and wants. These needs evolve over time and may diverge from an original system’s capabilities. Thus, a system’s value to its stakeholders diminishes over time. Some reasons for this decrease include growth in stakeholder wants and technological opportunities, which make an existing system, seem inadequate. Other reasons are growth in a system’s maintenance costs, due to effects such as depreciation and component obsolescence or in the environment, for example new rules and regulations and so forth. As a result, systems have to be periodically upgraded at substantial cost and disruption. Since complete replacement costs are often prohibitive, system adaptability is a valuable characteristic. The EU funded project AMISA, Architecting Manufacturing Industries and systems for Adaptability is based on two major financial theories: “Transaction cost” and “Financial Options” and on the Design Structure Matrix engineering tool. It is developing a generic quantitative and usable method and tool for architecting manufacturing systems and products for optimal adaptation to unforeseen future changes in stakeholder needs and technology development. This work aims at showing the state of the art of the DFA methodology development and some examples of the validation and simulation methodologies real-life pilot projects conducted by industry and SME.
  • Response Surface Design in Injection Moulding

    Authors: Magus Arnér (Tetra Pak Packaging Solutions)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Response surface, Design of Experiments, Injection moulding, Sampling strategy, Analysis of data, Nonlinear
    Submitted at 10-May-2012 11:26 by Magnus Arnér
    Accepted
    11-Sep-2012 16:50 Response Surface Design in Injection Moulding
    To solve a "frequent" failure mode we wanted to investigate the effect from tree different factors on the response. There was a suspicion about nonlinearities and therefore it was decided to perform a response surface design. The presentation will discuss the background of the problem, the set up of the experiment, the sampling strategy and the analysis of the data.
  • An Overview of Bayesian Networks for Dynamic System Analysis and Control

    Authors: Michael Ashcroft (Uppsala University)
    Primary area of focus / application: Modelling
    Keywords: Bayesian networks, stochastic modeling, assisted decision making, prediction, dynamic systems, control, machine learning
    Submitted at 10-May-2012 12:34 by Michael Ashcroft
    Accepted (view paper)
    12-Sep-2012 10:45 An Overview of Bayesian Networks for Dynamic System Analysis and Control
    Present an overview of Bayesian network based methods, focused on building towards an examination of their potential use in partially observable, dynamic system analysis and control. Topic will include:

    - Their (simplified) mathematical basis. An explanation of the differences between discrete, Gaussian and general networks and their comparative advantages.
    - Construction of model from domain knowledge or learnt from data. A discussion of the algorithms that permit the second and their guarantees.
    - Use of ensemble models.
    - Interpretation for system analysis, concentrating on the ability to identify redundant variables.
    - Use for predictive and decision theoretic purposes, their required inputs and their outputs.
    - Their use in modeling dynamic systems, and explanations of their relationship to Markov processes and the Kalman filter.