ENBIS-14 in Linz

21 – 25 September 2014; Johannes Kepler University, Linz, Austria Abstract submission: 23 January – 22 June 2014

My abstracts

 

The following abstracts have been accepted for this event:

  • Modeling of Non-Continuous Time-Pressure Curve Classes based on Pre-Volumes

    Authors: Astrid Ruck (Autoliv B.V.&Co. KG)
    Primary area of focus / application: Modelling
    Secondary area of focus / application: Design and analysis of experiments
    Keywords: Modeling, Non-continuous curve classes, Robustness, Effects
    Submitted at 20-Jun-2014 15:23 by Astrid Ruck
    Accepted
    22-Sep-2014 16:00 Modeling of Non-Continuous Time-Pressure Curve Classes Based on Pre-Volumes
    Pyrotechnical seatbelt pretensioners operate at high pressure levels. In order to avoid a bursting behavior, closed pressure systems are often equipped with a safety valve. To demonstrate product robustness burst tests are performed. Every single burst test is affected by a so-called pre-volume, which leads to a corresponding class of measured curves, i.e. pressure vs. time, which are not continuous at the bursting time.

    For the purpose of modeling the burst pressure and its corresponding time for a given pre-volume we transform the curves in a fixed environment of the bursting time. This approach can be applied to numerous curves affected by several other factors.
  • A Two-Step Method by Clustering and Forecasting a State Variable for Real Data Application

    Authors: Thomas Lehmann (Daimler AG), Joerg Keller (Daimler AG)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Quality
    Keywords: Forecasting, Clustering, Time series, Resampling, Automotive application
    Submitted at 20-Jun-2014 17:46 by Thomas Lehmann
    Accepted (view paper)
    22-Sep-2014 10:45 A Two-Step Method by Clustering and Forecasting a State Variable for Real Data Application
    Abstract. This paper presents a novel approach to perform improved forecasting of a state variable in real world data. Upstream identification of groups in data with methods of unsupervised learning should lead to increased quality of forecasting. Two thematically different attempts are presented. Finally, the group-specific forecasting is made by time series models.
    Evidence Accumulation and model-based clustering in terms of Growth Mixture Models are used as methods of clustering. The validation is done with Delete-1 Jackknife Resampling. Forecasting of detected groups is done through AR-/MA and ARIMA time series models, through Rollin Origin Cross Validation.
    For the experiment a real world data set from automotive sector is chosen. We explain the benefit of group specific over object specific forecasting for transient data.
    Two groups are derived, for which the decreasing state variable is predicted up to a given level on a given time frame.
  • When the Heart is Racing

    Authors: Bernard Francq (University of Glasgow), Suzanne Lloyd (University of Glasgow), Elaine Clark (University of Glasgow), Peter Macfarlane (University of Glasgow)
    Primary area of focus / application: Consulting
    Secondary area of focus / application: Modelling
    Keywords: Linear regression, Quantile regression, Predictive interval, ECG normalization
    Submitted at 22-Jun-2014 18:30 by Bernard Francq
    Accepted
    23-Sep-2014 10:35 When the Heart is Racing
    There has been little work done in assessing the effects of ethnic variation in ECG (electrocardiography) appearances while it is known that gender can affect the ECG results (e.g. QRS duration is known to be longer in white males compared to females). A study set out to address this shortcoming in a limited way by analysing ECGs from 4 different racial groups (more than 4000 patients composed of Caucasians, Chinese, Indians and Africans). Twelve lead ECGs were recorded in digital form on different machines. The digital data were returned to Glasgow and analysed using the same version of the Glasgow program. The goal of this study is also to define ‘normalized intervals’ where the patient’s result should ideally lie according to his/her age, race and gender, e.g. 96% predictive intervals based on the upper 98% limits and the lower 2% limits.

    Two statistical approaches were undertaken to assess the effect of race, age and gender on the ECG results and to define 96% normalized intervals.
    First, the well-known linear regression where the means are modelled assuming the normality of the data (with, if needed, a mathematical transformation undertaken prior to the analysis). The classical hyperbolic predictive intervals are not suitable in this context but straight predictive intervals can easily be computed from the predicted values of the linear regression (the predicted means).
    Second, the quantile regression where the desired quantile is directly modelled. This approach is obviously more flexible and does not assume the normality of the data.

    This presentation will compare both statistical approaches with simulations and the real data. The flexibility of the quantile regression will be discussed as well as the BLUP property of the linear regression. The ‘normalized intervals’ computed from both approaches will be compared to the observed quantiles (computed per age, race and gender) or moving quantiles with real data.
  • How to Regress and Predict in a Bland and Altman Plot?

    Authors: Bernard Francq (Université Catholique de Louvain)
    Primary area of focus / application: Metrology & measurement systems analysis
    Keywords: Measurement methods comparison, (Correlated)-errors-in-variables-regressions, Bland and Altman plot, Predictive interval, Tolerance interval
    Submitted at 22-Jun-2014 18:35 by Bernard Francq
    Accepted
    22-Sep-2014 16:00 How to Regress and Predict in a Bland and Altman Plot?
    Analytical laboratories continuously assess the uncertainties and reliability of their measurement systems so that their clients can take the right decision. This leads to the development of new measurement methods which should be more precise, less expensive but chiefly equivalent.

    To assess equivalence in method comparison studies, two main methodologies are presented separately in the literature. First, the approach based on errors-in-variables regressions that focuses on confidence intervals. Two devices are considered equivalent when they provide similar measures notwithstanding the random measurement errors.
    Second, the well-known Bland and Altman approach with its agreement intervals. Two devices are considered interchangeable when the differences in their measurements are not meaningful in practice.
    This presentation will explicit and compare both approaches. Tolerance intervals will be presented as better alternatives than agreement intervals and two new consistent regressions are proposed to compute predictive intervals in a Bland and Altman plot. The two methodologies will be reconciled and their similarities discussed with simulations and real data.

    It will be then concluded that errors-in-variables can (unfortunately) not be avoided in method comparison studies, although the Bland and Altman approach was, initially, proposed and applied to avert the complexity of this statistical method.
  • Multilevel Functional Principal Component Analysis of Façade Sound Insulation Data

    Authors: Raffaele Argiento (CNR-IMATI), Pier Giovanni Bissiri (CNR-IMATI), Antonio Pievatolo (CNR-IMATI), Chiara Scrosati (CNR-ITC)
    Primary area of focus / application: Modelling
    Keywords: Functional data analysis, Mixed effects models, Sound insulation data, Principal component analysis
    Submitted at 22-Jun-2014 20:06 by Raffaele Argiento
    Accepted
    22-Sep-2014 16:40 Multilevel Functional Principal Component Analysis of Façade Sound Insulation Data.
    In this work we analize data from a sound insulation of façades study, the experiment consisting of independent measurements executed several times by different operators on the same residential building. Mathematically, data can be seen as functions of an acustic parameter over the spectrum of the frequencies.

    In these studies, it is important to assess the within and between group variability in the measurements of façade sound insulation. Moreover, in the engineering literature it is known that the indices of sound insulation are more variable at low frequencies, compared to higher frequencies. Therefore, we employ a multilevel
    functional principal component analysis (FPCA, Di et al~2009) to decompose the
    functional variance both at the data and at the group level.

    Our method allows ranking the performance of the operators on the basis of their measurements' variability and their different performances at either low frequency (relative high variability) and high frequency (relative low variability) spectra.
  • Fuzzy Logic in the Assessment of Alternative Measurement Systems

    Authors: Magdalena Diering (Poznań University of Technology), Krzysztof Dyczkowski (Adam Mickiewicz University of Poznań), Agnieszka Kujawińska (Poznań University of Technology)
    Primary area of focus / application: Six Sigma
    Secondary area of focus / application: Process
    Keywords: Quality control, Alternative measurement system analysis, Cross tab method, Fuzzy logic
    Submitted at 22-Jun-2014 23:12 by Magdalena Diering
    Accepted (view paper)
    22-Sep-2014 16:40 Fuzzy Logic in the Assessment of Alternative Measurement Systems
    Quality control in manufacturing process means checking the consistency of the process or product with the internal or external customer requirements. Most often it is done by direct measurement or observation. The main objective of the quality control is to increase the chance that the product (process) is free from defects when passing it on further stages of the production process or on to use.
    Alternative control is a special case of quality control. It can be performed by measuring or checking and classifying the object (product) into one of a number of states (in the specific case – into one of two, for example: good/bad or OK/No OK).
    To assure that quality control of manufacturing is a reliable process and its outcomes are on accepted level, measurement system must be evaluated (variation of the measurement system should be known and accepted). There are many procedures to assess the capability and reliability of measurement system.
    The paper presents the directions of attribute measurement systems analysis (MSA) development. The study pointed out the possibility of using the fuzzy logic elements in this type of measurement systems. The work presents the basic methods and procedures used in the MSA studies, and also pointed out aspects of their integration with fuzzy logic tools. Case study is presented.