ENBIS Spring Meeting 2017

28 – 30 May 2017; Monastery of Schlägl in Upper Austria Abstract submission: 11 November 2016 – 5 March 2017

My abstracts


The following abstracts have been accepted for this event:

  • Simulation of Dynamic Customer Complaints

    Authors: Jens Michalski (Daimler AG)
    Primary area of focus / application: Reliability
    Keywords: Automotive engineering, Field prognosis, Reliability simulation, Warranty claims
    Submitted at 26-Feb-2017 19:13 by Jens Michalski
    30-May-2017 12:55 Simulation of Dynamic Customer Complaints
    The current and future product monitoring of technical components and systems in the automotive industry poses immense challenges to manufacturers and in particular suppliers, as well as extensive requirements for the participating business areas. In addition strongly differentiated market and customer requirements are related to the entire vehicle and all its vehicle derivatives. Furthermore a steadily increasing number of system and component variants from an increasing number of involved suppliers can be noticed.
    Since the demands on reliability and quality in the context of safety of the entire vehicle are among the most important requirements not only for the manufacturer but also for the customer, a targeted product monitoring (in front of the customer) must be achieved within the shorter development times. Particularly under the premise of the influencing variables in the field, there is a risk for failure mechanisms which were not detected in the development or which could not be eliminated early before the start of production. To ensure the detection of early and future errors in the field, maintain customer acceptance as well as adherence to legal frameworks, stochastic analyses are a general method to analyse/detect them. To implement these statistical methods, customer-specific data must be available regarding the failure mechanism of the component or system.
    While in mechatronic and electrical systems an immense increasing number of load data of the respective customers can be collected by means of on-board diagnostics, there is still a very high proportion of components and systems in which this extensive data basis cannot be generated (for example: interior or mechanical components). However, the given data base must be used as a representative sample of the customer behaviour to initiate targeted measures to improve reliability and safety in field and to determine the given field actions. Under specific circumstances this limited data information can lead to non-precise reliability prognosis using state of the art methods.
    In the context of this publication, the consideration of dynamic failure mode changes (for example: changes of customer complaints) is implemented directly inside a mathematical model to be able to estimate a more realistic complaint behaviour by the customer. Within the publication the needed data and the developed statistical approach to estimate future complaints are presented at the beginning. Afterwards the developed simulation model is concretized by means of a practical example.
    The focus of this example is as follows:
    -Explanation of the given data and the related failure mechanism (technical aspects and customer behavior/acceptance)
    -Calculation of the given data set and estimation of future complaints behavior (relative distribution)
    -Setting up the boundaries of the simulation model and explanation of the simulation itself to estimate the absolute amount of customer complaints
    -Results and discussion of the derived field actions
  • Predictive Maintenance with Latent Variable Models in Industry 4.0

    Authors: Alberto Ferrer (Multivariate Statistical Engineering Group, DEIOAC, Universitat Politècnica de València), Pedro Villalba-Torán (Multivariate Statistical Engineering Group, DEIOAC, Universitat Politècnica de València), Pedro Hernández-Ariznavarreta (Secure E-Solutions, Madrid)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Process
    Keywords: Predictive Maintenance (PdM), Condition-Based Maintenance (CBM), Principal Component Analysis (PCA), Partial Least Squares (PLS)
    Submitted at 26-Feb-2017 23:57 by Alberto J. Ferrer-Riquelme
    Accepted (view paper)
    29-May-2017 10:25 Predictive Maintenance with Latent Variable Models in Industry 4.0
    Predictive maintenance (PdM) is a maintenance strategy that switches from schedule-based or cycle-based maintenance to more cost-effective condition-based maintenance. PdM defines methods to predict or diagnose problems in a piece of equipment based on trending of test results. These methods use non-intrusive testing techniques to measure and compute equipment performance trends. Condition-based maintenance (CBM) is a methodology that combines predictive and preventive maintenance with real-time monitoring. PdM uses CBM systems to detect fault sources well in advance of failure, making maintenance a proactive process. CBM accurately detects the current state of mechanical systems and predicts the systems’ ability to perform without failure. This means that equipment maintenance is only undertaken if equipment condition analysis estimates a high failure probability. Success of equipment condition analysis depends on data availability and real-time analysis capability. In modern industries adopting the Industry 4.0 paradigm, the result of Industrial Internet of Things (IIoT) connecting intelligent physical entities to each other allows complex equipment units to have embedded sensors and special modules (agents) providing connection to the monitoring center. This is leading to the so-called Big Data problem in Industry 4.0. Big data exhibit high volume and correlation, rank deficiency, low signal-to-noise ratio, complex and changing structure, and missing values. Classic statistical techniques are not feasible for analyzing Big Data streams. In this talk we illustrate the potential of latent variable-based multivariate statistical methods to analyze Big Data streams and visualize extracted information in a way that is easily interpreted and that is useful for Time to Failure (TTF) prediction or Remaining Useful Life (RUL) estimation to be used for PdM. The methodology will be illustrated using aircraft engine run-to-failure data simulation [1].

    [1] Abhinav Saxena, Kai Goebel, Don Simon, and Neil Eklund. "Damage Propagation Modeling for Aircraft Engine Run-to-failure Simulation." 2008 International Conference on Prognostics and Health Management (2008)
  • Predictive Maintenance on Hydrocarbons Pumps: A Machine Learning Approach

    Authors: Constant Bridon (OCTO TECHNOLOGY), Veltin Dupont (OCTO TECHNOLOGY), Matthieu Lagacherie (OCTO TECHNOLOGY), David Campion (TOTAL), David Teixugueira (TOTAL), Laurent Castanie (TOTAL)
    Primary area of focus / application: Modelling
    Secondary area of focus / application: Process
    Keywords: Predictive maintenance, Machine learning, Sensor monitoring, Oil industry
    Submitted at 27-Feb-2017 10:21 by Constant Bridon
    30-May-2017 12:15 Predictive Maintenance on Hydrocarbons Pumps: A Machine Learning Approach
    The electrical submersible pump (ESP) is an efficient and reliable artificial-lift method for lifting moderate to high volumes of fluids from wellbores. The ESP is normally tubing hung from the wellhead. Due to this situation, downhole failures have strong impacts, especially in offshore environment: Lost production occur and pulling operations are costly as they request wells and drilling unit services immobilization.
    That's why it is critical to closely monitor ESP devices, and why they are equipped with a significant amount of sensors. The volume of generated data by those sensors is large and difficult to be handled by classical statistical approaches, especially in terms of generalization and industrialization. Indeed, the ESP’s local environment (through-flowing fluids, ESP type, etc.) is very specific to the pump and we still need to generate exploitable results on a reasonable time. Moreover, human monitoring is not a solution since it would be too expensive in trainin.
    Throughout this paper, we present an agnostic method, i.e not constrained to one type of ESP and its environment, that combines unsupervised and supervised learning techniques to predict failures.
    We designed an innovative method to integrate as much business side heuristics as possible. These refined data proved to be small signals with significant discriminant power in failure prediction. In order to generate those signals, we based our approach on an existing artifact already used by operatives, called a diagnostic matrix, also called FMEA (Failure Mode & Effect Analysis) providing scenarios explaining a failure types. This work is the basis of the evolution of curative maintenance toward predictive or even preventive maintenance for one of a Major Oil & Gas player.
  • A Monotonicity Preserving Transformation for Confidence Regions of Unsymmetric Multivariate Data

    Authors: Florian Sobieczky (SCCH Software Competence Center Hagenberg)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Reliability
    Keywords: Monotonicity preserving transformation, Sphericity, Predictive modelling, Error detection
    Submitted at 27-Feb-2017 15:36 by Florian Sobieczky
    29-May-2017 10:25 A Monotonicity Preserving Transformation for Confidence Regions of Unsymmetric Multivariate Data
    For the purpose of producing more accurate confidence regions for non-symmetric continuous multivariate unimodal data we introduce and validate a method which bijectively maps the sample space by a non-linear transformation that is preserving convexity of the pdf's contour-surfaces. Polygonal shaped confidence regions are defined using different polytopes: Generalizations of the octahedron, the tetrahedron and the triakis tetrahedron to d dimensions are used and serve as a skeleton in modeling the underlying distribution. Comparison with distributions of higher sphericity and different decay properties become possible. A simulation study from the line production environment comparing several predictive modelling techniques exemplifies the method's power in the case of different asymmetric distribution types.
  • Bayesian Replacement Policies for Rail Tracks

    Authors: Refik Soyer (The George Washington University)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Modelling
    Keywords: Semi-parametric models, Minimal repair, Decision analysis, Gamma process, MCMC
    Submitted at 1-Mar-2017 07:50 by Refik Soyer
    30-May-2017 11:35 Bayesian Replacement Policies for Rail Tracks
    In this talk we present a Bayesian decision theoretic approach for replacement strategies for systems that are subject to wear. In so doing, we consider a semi-parametric model to describe the failure characteristics of the system by specifying a nonparametric form for cumulative intensity function and by taking into account effect of covariates by a parametric form. Use of a gamma process prior for the cumulative intensity function complicates the Bayesian analysis when the updating is based on failure count data. We develop a Bayesian analysis of the model using Markov chain Monte Carlo (MCMC) methods and determine replacement strategies. Adoption of MCMC methods involves a data augmentation algorithm. We show the implementation of our approach using actual data.
  • A Statistical Model for Multivariate Variation Description of Wafer Acceptance Test Data for Reliable Chip Design

    Authors: Thomas Riebenbauer (Joanneum Research Forschungsgesellschaft mbH), Birgit Sponer (AMS AG), Peter Scheibelhofer (AMS AG)
    Primary area of focus / application: Quality
    Secondary area of focus / application: Process
    Keywords: Semiconductor industry, Corner model, Principal Component Analysis, Dimension reduction
    Submitted at 2-Mar-2017 12:27 by Thomas Riebenbauer
    29-May-2017 10:05 A Statistical Model for Multivariate Variation Description of Wafer Acceptance Test Data for Reliable Chip Design
    In semiconductor manufacturing, chip design and its reliability are sensitive to process variation. Physical parameters, which are modelled to form the basis of a new chip design, are influenced by manufacturing variability. Thus, it is of crucial interest to find adequate models to describe the physical device behavior for e.g. MOS (metal oxide semiconductor) transistors, resistors, capacitors or diodes that are needed for electrical engineering. So called corner models form a dense description of the parameter measurements and its variability. They define a high dimensional convex limit hull to enclose the variation of the physical parameters in their normal operating conditions. The limit hull is composed by a specific number of corners. This forms the basis for simulating new chip designs.
    In the present work, statistical dimension reduction techniques are applied to reduce the number of corners and thus to reduce the complexity and the corresponding simulation effort of a corner model as it is used at ams AG. It is shown that a constructed framework for model complexity reduction based on a statistically robust version of principal component analysis yields promising results.

    This work has received funding from ENIAC Joint undertaking under FP7 research (No.:621270) and FFG (No.:8437406).