ENBIS-17 in Naples

9 – 14 September 2017; Naples (Italy) Abstract submission: 21 November 2016 – 10 May 2017

My abstracts


The following abstracts have been accepted for this event:

  • Predictive Maintenance of Smartmeter's Concentrators

    Authors: Jairo Cugliari (Université de Lyon), Stephane Bonneway (Université de Lyon), Pierre Achaicha (Université de Lyon)
    Primary area of focus / application: Other: Machine learning for energy management
    Keywords: Predictive maintenance, Functional data, Wavelets, Random forests
    Submitted at 6-Mar-2017 16:03 by Jairo Cugliari
    12-Sep-2017 09:00 Predictive Maintenance of Smartmeter's Concentrators
    Smart meters on electric systems allows a distant configuration and monitoring of the client needs through communication between the units and a supervisor. Predictive maintenance is a key element to ensure the appropriate functioning of the whole system.

    Other than the data registered from the client electricity consumption, the meters send different kinds of logs and events. We use events data from a early testing phase of smartmeters to construct a predictive model.

    Our approach uses the event data as time series or more generally functional data. Using wavelets we extract handy features from sets of events. Then, a random forest classifier is used to predict the health status of meters’ concentrators from the constructed features.
  • Statistical Monitoring of Probability Density Functions for Image Data

    Authors: Alessandra Menafoglio (Politecnico di Milano), Marco Grasso (Politecnico di Milano), Piercesare Secchi (Politecnico di Milano), Bianca Maria Colosimo (Politecnico di Milano)
    Primary area of focus / application: Other: Quality engineering applied to advanced manufacturing
    Keywords: Statistical Process Control, Image-based process monitoring, Functional data analysis, Constrained curves, Bayes space
    Submitted at 6-Mar-2017 17:51 by Alessandra Menafoglio
    11-Sep-2017 10:30 Statistical Monitoring of Probability Density Functions for Image Data
    In image-based statistical process control, the quality of monitored parts is often related to the overall distribution of relevant statistical descriptors rather than to the measured value for one single feature (e.g., their mean or variance). The latter distribution can be represented in the form of a probability density function (PDF) and then used for the purpose of controlling the stability of the process. Control charts for profile monitoring have been discussed in the literature on unconstrained functional data (Woodall et al., 2004 and Noorossana et al., 2012). Nonetheless, the latter lead to non-optimal results when applied to density functions, since they are not designed to properly account for the data constraints (positivity, integral to unity). In this communication, we present a profile monitoring approach based on the theory of Bayes Hilbert spaces (Egozcue et al., 2006, van den Boogaart et al, 2014), that allows to handle the data constraints through the use of a suitable geometry. Based on the simplicial functional principal components analysis (SFPCA) of Hron et al (2016), we effectively cope with the data dimensionality and constraints, and build an original control chart scheme for density data. Through simulation we show that our approach outperforms the competitor functional approaches. We illustrate the methodology through a real case study on the quality control of foamed materials production.
  • Harvest Time Prediction for Batch Processes

    Authors: Max Spooner (Technical University of Denmark), Murat Kulahci (Technical University of Denmark)
    Primary area of focus / application: Other: Reliability of Subjective Measurement Systems
    Keywords: Batch process, Statistical Process Control, Quality control, Bioprocess
    Submitted at 6-Mar-2017 18:16 by Max Spooner
    13-Sep-2017 10:50 Harvest Time Prediction for Batch Processes
    In many batch processes there is great variation in batch duration. Some batches progress faster than others due to variation in raw materials, physical conditions and operator behaviour. Product quality is affected by batch harvest time - a batch which is allowed to progress too long will differ to a batch that is harvested prematurely. Often the decision when to harvest depends on the judgement of an expert and requires immediate action. In this work we present a method for predicting the optimal harvest time at an early stage in the ongoing batch based on the data accumulated up to the present. The method is demonstrated using real data. Predictions are updated and gain accuracy as the batch progresses and additional data is accumulated. Besides the process measurements, the batch dynamics (acceleration/deceleration in progress) are incorporated through dynamic time warping. Different approaches to dealing with the high dimensions of batch data, such as variable selection methods and latent structure methods are contrasted. The proposed model allows the harvest time to be predicted at an early stage in the ongoing batch so that preparations can be made downstream and resources may be used accordingly.
  • Gaussian Processes for Adaptive Design: Learning Non-Stationarity

    Authors: Sébastien Marmin (IRSN), David Ginsbourger (Idiap), Jean Baccou (IRSN), Jacques Liandrat (Centrale Marseille)
    Primary area of focus / application: Other: ISBA session on Bayesian Optimization
    Keywords: Covariance kernels, Kriging, Sequential design, Computer experiments
    Submitted at 6-Mar-2017 19:17 by Sébastien Marmin
    12-Sep-2017 11:40 Gaussian Processes for Adaptive Design: Learning Non-Stationarity
    We consider the problem of approximating a function from scarce training samples. We focus on Gaussian process (GP) models, where the objective function is assumed to be a realization of a Gaussian random field. In this Bayesian approach, sequences of new evaluations and final predictions are driven by the assumed prior covariance kernel (through posterior distributions of the objective function) as well as by infill sampling criteria. In many test cases, functions exhibit heterogeneous variations depending on input space regions. It is then common to assume non-stationary prior covariance kernels. Here we define the so-called WaMI class of non-stationary covariance kernels dedicated to functions for which high variations occur along unknown non-canonical directions. Corresponding WaMI GPs generalize both Multiple Index GPs and tensorized warped GPs. We further propose targeted criteria relying on derivatives of the posterior GP. These criteria favour sampling in high variation regions, and we provide formulae for reducing their evaluation. Finally, GP models and criteria are compared on two test cases. On a first mechanical engineering application for nuclear safety, derivative-based criteria are shown to outperform usual variance-based criteria under a stationary model. Yet, prediction errors are even lower when combining variance-based criteria with WaMI GP. Treed Gaussian Processes are also compared on this and a second application from NASA, on which it first outperforms WaMI on small data set but is overcome by WaMI GP in sequential settings.
  • Additive Quantile Regression for Electricity Load Forecasting

    Authors: Matteo Fasiolo (University of Bristol), Simon N. Wood (University of Bristol), Yannig Goude (EDF R&D), Raphael Nedellec (EDF R&D)
    Primary area of focus / application: Other: Machine learning for energy management
    Keywords: Quantile regression, Generalized additive models, Electricity load forecasting, Smoothing splines
    Submitted at 6-Mar-2017 20:42 by Matteo Fasiolo
    Accepted (view paper)
    12-Sep-2017 09:20 Additive Quantile Regression for Electricity Load Forecasting
    Energy utilities need accurate electricity load predictions for the purpose of production planning. Generalized additive models are useful tools in this context, because they offer a compromise between flexibility and interpretability. However, it is sometimes difficult to find an adequate parametric model for the conditional distribution of the response. In addition, production planning might require estimates of only some conditional quantiles of electricity load, rather than an approximation to the whole distribution. Hence, for this application, quantile regression might represent an interesting alternative to distributional approaches. In this talk we will describe how the GAM framework can be used to fit additive quantile regression, in a stable and computationally efficient manner. We will also present an R package, called qgam, which can be used to fit quantile GAMS, and we will show how the proposed method's performance compares with that of alternative methods, in the context of electricity load forecasting.
  • Cost-Optimal Control Charts for Health-Care Data

    Authors: András Zempléni (Department of Probability Theory and Statistics, Eötvös Loránd University, Budapest), Balázs Dobi (Department of Probability Theory and Statistics, Eötvös Loránd University, Budapest)
    Primary area of focus / application: Modelling
    Secondary area of focus / application: Quality
    Keywords: Control chart, Economic, Markov chain, Medical data, Stationary distribution
    Submitted at 6-Mar-2017 21:52 by András Zempléni
    11-Sep-2017 16:00 Cost-Optimal Control Charts for Health-Care Data
    In an earlier paper Zempléni et al (2004) introduced a Markov chain-based method for optimal design of Shewhart-type control charts, based on the economic calculations, which originate from Duncan (1974).

    Control charts are traditionally used in industrial statistics. In this presentation we present a new approach, which is suitable for applications in the health care sector. Here most of the papers use standard process control charts for quality assurance (see e.g. Duclos et al., 2009). We adapt the Markov chain-based approach of Zempléni et al (2004) and develop a method, which suits for the real-life medical applications, where not only the "shift" (i.e. the degradation of the patient's health) can be random, but the sampling interval (i.e time between visits) and the effect of the treatment too. This means that we do not use the often-present assumption of perfect repair which is usually not realistic in case of medical treatments. The average cost of the optimal protocol, which consists of the average sampling frequency and control limits can be estimated by the stationary distribution of the Markov chain. We illustrate the approach by simulated data, based on real-life medical protocols and observations.

    A.J. Duncan (1974): Quality Control and Industrial Statistics (4th edn). Homewood: Illinois.
    A. Duclos, S. Touzet, P. Soardo, C. Colin, J. L. Peix and J. C. Lifante (2009): Quality monitoring in thyroid surgery using the Shewhart
    control chart, British Journal of Surgery, 96: 171–174.
    A. Zempléni, M. Véber, B. Duarte and P. Saraiva (2004): Control charts: a cost-optimization approach for processes
    with random shifts. Applied Stochastic Models in Business and Industry, 20: 185–200.