ENBIS-18 in Nancy

2 – 25 September 2018; Ecoles des Mines, Nancy (France) Abstract submission: 20 December 2017 – 4 June 2018

My abstracts


The following abstracts have been accepted for this event:

  • Effect of the Prior Distribution on the Mean in Control Charts

    Authors: Isabel Ramírez (Universidad Nacional de Colombia), Nelfi González (Universidad Nacional de Colombia)
    Primary area of focus / application: Quality
    Secondary area of focus / application: Quality
    Keywords: Control charts, ARL, Bayesian approach, Normal distribution
    Submitted at 25-Apr-2018 15:05 by Isabel Ramírez
    Accepted (view paper)
    3-Sep-2018 15:30 Effect of the Prior Distribution on the Mean in Control Charts
    In this work, control charts for the mean are constructed using a Bayesian approach. It is assumed that the quality characteristic to be controlled can be modeled by a normal distribution with known variance. The effect of the prior distribution for the mean is studied through a simulation study. Performance measures of the control charts are evaluated in phase II. A Bayesian conjugate model is established, therefore the prior distribution for the mean is normal. The posterior predictive distribution, which is normal, is used to establish the control limits. In the simulation study, the effects of sample sizes, distance of the prior mean in relation to the average of the calibration sample, and an indicator of how informative is the prior distribution of the population mean were studied. It is found that the false alarm rate can be exaggeratedly large if the prior is very informative, and in turn leads to a ARL-biased chart, that is, the maximum of the ARL is not given when the process is in control. In addition, when the size of the calibration samples and the future sample are small, there is a great influence of the prior specification on the power of the chart, particularly when the prior is very informative.
  • Prediction Uncertainty of Times Series and State Space Models

    Authors: Bjarne Bergquist (Luleå University of Technology), Peter Söderholm (Swedish Transport Administration and Luleå University of Technology)
    Primary area of focus / application: Reliability
    Keywords: Railway, Track geometry, Remaining useful life (RUL), Uncertainty, Kalman Filter, Time series, Prediction
    Submitted at 26-Apr-2018 09:35 by Bjarne Bergquist
    4-Sep-2018 09:20 Prediction Uncertainty of Times Series and State Space Models
    In this presentation, we discuss recursively updated time series predictions of the remaining useful life (RUL) of the railway track geometry. Geometrical properties of the track degrade, and RUL, in this case, means the degradation of the track geometry to a faulty state, requiring actions such as speed restrictions and track alignment.
    Track property sampling is irregular, and measurement errors differ between measurements. We compare models where we have assumed the same measurement error with models where we use the estimated precision of each measurement. The irregular sampling makes lumping of measurements into time intervals needed. We, therefore, analyse different interval selections for lumping of data. We also address needs for non-linear models and possibilities for seasonal component estimations.
  • Run-to-Run Control Based on Gaussian Bayesian Network in Semiconductor Manufacturing

    Authors: Wei-Ting Yang (École des Mines de Saint-Étienne), Jakey Blue (École des Mines de Saint-Étienne), Agnès Roussy (École des Mines de Saint-Étienne), Marco S. Reis (University of Coimbra), Jacques Pinaton (STMicroelectronics)
    Primary area of focus / application: Process
    Secondary area of focus / application: Mining
    Keywords: Run-to-Run (R2R), Fault Detection and Classification (FDC), Virtual Metrology (VM), Equipment signal, Gaussian Bayesian Network (GBN)
    Submitted at 26-Apr-2018 09:50 by Wei-Ting Yang
    3-Sep-2018 14:00 Run-to-Run Control Based on Gaussian Bayesian Network in Semiconductor Manufacturing
    Run-to-Run (R2R) control is the process regulating scheme commonly implemented in the semiconductor industry. Typically, key process parameters are regulated with respect to the measured quality features, e.g., the wafer thickness measurements. However, wafer quality can be affected by complex factors related to the equipment condition. In this study, the equipment condition is explicitly modeled based on the sensor signals and then integrated into the core of an R2R controller. The new R2R control scheme can reduce the process variability in a more effective way.

    For this aim, Gaussian Bayesian Network (GBN) is employed to analyze the implicit relationship not only between the control factors and process parameters but also among the process parameters and metrology. The cause-effect relationship between all the variables can be explicitly expressed in the form of a connected graph after applying GBN. Consequently, the variation of process parameters caused by the control factors can be estimated and the corresponding predicted metrology can be obtained simultaneously. In this regard, we are able to consider process control in a more global view.

    The effectiveness of this approach is demonstrated and validated via a practical case study, in collaboration with our industrial partner.
  • Insight into Aftermarket Automotive Sales, Factory Standards and Predicting Autopart Replacement

    Authors: Wayne Smith (Rain Data / Newcastle University), Shirley Coleman (Newcastle University), Jaume Bacardit (Newcastle University)
    Primary area of focus / application: Quality
    Keywords: Automotive aftermarket, Shewhart chart, Control limits, Return rates, Fuzzy matching, OEM standards
    Submitted at 26-Apr-2018 13:47 by Shirley Coleman
    Accepted (view paper)
    4-Sep-2018 10:10 Insight into Aftermarket Automotive Sales, Factory Standards and Predicting Autopart Replacement
    The automotive aftermarket sector collects large datasets on every aspect of buying car parts online via web hosted catalogues. These datasets contain a lot of information about the transactions of buying automotive parts (when, where, what) and details of the parts themselves. From this it is possible to derive patterns of buying habits, preferences, return rates and product attributes from the data. Using datasets collected from a software company (who provide both cataloguing and invoicing systems) this presentation will demonstrate three statistically applied techniques to this type of data that have provided both insight and benefit to business.

    Returns are an important consideration in any business that supplies goods. Returns can cost a company for many reasons, including, the cost of posting and identifying if a product is fit for re-sale. Here, the return rates of frequently bought car parts are plotted as a Shewhart chart, using a funnel plot with statistical limits to identify which return rates require investigation. This allows a method to prioritise attention to parts according to whether (and by how much) the return rates lie beyond the 3 standard deviation limits of the control chart.

    When referring to car parts, the OEM standard refers to the manufacturer of the original equipment - that is, the parts assembled and installed during the construction of a new vehicle. The aftermarket parts are those made by companies trying to match the OEM factory standard. There is considerable uncertainty about which parts match to which OEM numbers. The analysis presented here uses fuzzy matching to compare supplier parts to typical OEM factory standards. This process reveals those supplier parts which can be matched against other cars to which they weren’t originally intended, and in doing so highlights potential revenue for a supplier, in which they can sell their part against cars without modifying their product.

    Comparing invoices for replacement car parts and noting the mileage of the car at the time of replacement highlights the potential failure rate of non-serviceable car parts. Inspecting those parts that were replaced (along with the mileage at the point of replacement) and comparing these across different car models also demonstrates the reliability different car manufacturers. Function fitting of the part replacement per mileage aids in the process of identifying key points in a car’s history when failure may occur. This presentation explains this method in further detail.
  • Investigating the Influence of Different Powders on Coating Properties in an HVOF Spraying Process

    Authors: Eva-Christina Becker-Emden (Dortmund University of Applied Sciences and Arts)
    Primary area of focus / application: Modelling
    Secondary area of focus / application: Design and analysis of experiments
    Keywords: Thermal spraying, Protective coatings, Generalized linear models, Model selection, Factorial design, Blocking
    Submitted at 26-Apr-2018 14:03 by Eva-Christina Becker-Emden
    4-Sep-2018 12:20 Investigating the Influence of Different Powders on Coating Properties in an HVOF Spraying Process
    High velocity oxygen fuel (HVOF) spraying is a method to apply protective coatings on surfaces. It is of interest, how different powders influence properties of the coating such as layer thickness and hardness.

    To investigate differences between powders we run a factorial design with additional center points on two subsequent days, which is blocked and repeated for every powder. Taken into account are a WC-12Co powder of the type WOKA 3102 from Sulzer Metco, a WC-Co powder with small carbides, a WC-FeCrAl powder and a Cr3-C2 powder. We compare the obtained coating and particle properties for the different powders graphically. Additionally, we fit generalized linear models for some properties to gain insight into the influence of the process settings. Our model building is based on an all-subset selection using the Bayes Information Criterion as a selection criterion with the maximal model including all main effects. We consider different link functions under the assumption of gamma-distributed responses.

    We find that the WC-based powders provide a similar behaviour. In future, further experiments with the Cr3-C2 powder could be of interest, as its behaviour deviates from the WC-based powders.
  • Preserving Projections Properties when Regular Two-Level Designs are Blocked

    Authors: John Tyssedal (The Norwegian University of Science and Technology), Yngvild Hamre (The Norwegian University of Science and Technology)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Regular two-level designs, Blocking, Projective properties, Confounding
    Submitted at 27-Apr-2018 10:32 by John Tyssedal
    3-Sep-2018 15:10 Preserving Projections Properties when Regular Two-Level Designs are Blocked
    Box, Hunter and Hunter (2005) in their book “Statistics for Experimenters” mention projection properties as one justification for the use of two-level fractional factorials. But when fractionated regular two-level designs are blocked, using the standard way of blocking, their projective properties are dramatically worsen. Even a regular resolution V design that is of projectivity P=4, may become a P=1 design when blocked in two blocks. As a result, there are one or more two-factor interactions that will be completely confounded with the block-effect. In this presentation, we demonstrate a way of blocking two-level regular designs, such that their projective properties can be preserved at the expense of just a small decrease in efficiency. Thereby one can estimate most effects of normally interest even if the design is blocked. Designs with 16, 32 and 64 runs are considered.