ENBIS: European Network for Business and Industrial Statistics

My abstracts

The following abstracts have been accepted for this event:

• Online Deconvolution for Industrial Hyperspectral Imaging Systems
  
  Authors: Yingying Song (CRAN)
  Primary area of focus / application: Design and analysis of experiments
  Keywords: Hyperspectral image, Online deconvolution, LMS, ZA-LMS
  

  Submitted at 2-May-2018 02:32 by Yingying Song
  Accepted
  

  3-Sep-2018 15:50 Online Deconvolution for Industrial Hyperspectral Imaging Systems
  Hyperspectral imaging has received considerable attention in the last decade as it combines the power of digital imaging and spectroscopy. Every pixel in a hyperspectral image provides local spectral information about a scene of interest across a large number of contiguous bands. Several sensing techniques have been devised for hyperspectral imaging and can be categorized into four main groups: whiskbroom (point scan), pushbroom (line scan), tunable filter (wavelength scan), and snapshot. This paper is a first step towards the development of advanced online hyperspectral image processing methods required in industrial processes that aim at controlling and sorting input materials right after each line scanning. The aim of this paper is to address the fast online (sequential) deconvolution of hyperspectral images captured by pushbroom imaging systems. The proposed sequential deconvolution algorithm can be easily extended to whiskbroom systems.
  
  Pushbroom imaging systems make use of 2D sensors allowing to observe the scene line-by-line at each time instant. The stream of spatial-spectral arrays is stacked to form the hyperspectral image which is a 3D data cube. A hyperspectral image may suffer from spatial distortions resulting in a loss of spatial resolution. Assuming a constant acquisition velocity, any resolution improvement results in an increase of both blurring and noise level. Thus, the corresponding distortion can be modeled by linear invariant convolution.
  
  We introduce a least-mean-square (LMS) based framework accounting for the convolution kernel non-causality and including non-quadratic (zero attracting and piece-wise constant) regularization terms. This results in the so-called sliding block regularized LMS (SBR-LMS) which maintains a linear complexity compatible with real-time processing in industrial applications. The choice of these regularization terms is thus mainly motivated by the targeted application, namely, the inspection of objects put on a conveyor belt. At a given wavelength, the response of the conveyor after background removal is close to zero while that of the objects is supposed to be piecewise constant. A model for the algorithm mean and mean-square transient behavior is derived and the stability condition is studied.
  
  Experiments were conducted to validate the transient behavior model with different regularization parameter values. This model allows to assess the influence of each regularization parameter. It appears that the zero-attracting property results in a faster convergence to zero than that of the algorithm without any regularization. The first order derivative regularizer is favoring the reconstruction of piecewise constant objects along the spatial dimension by decreasing the difference between two adjacent rows. However, both zero-attracting and the first order derivative properties introduce a bias on the amplitudes. More experiments showed that there exist optimal values for the different regularization parameters. Experimental results on both simulated and real hyperspectral images proved that the proposed SBR-LMS outperforms standard approaches at low SNR scenarios which is the case corresponding to ultra-fast scanning with industrial imaging systems.
   
  Future works will focus on the automatic learning of hyperparameters. A joint online deconvolution and unmixing algorithm is also worth being studied. This is expected to yield a very low computational burden and accurate image restoration approach.
• Detection and Modelling of the Propagation Regimes in Fatigue Crack Propagation
  
  Authors: Florine Greciet (Safran Aircraft Engines, Université de Lorraine, IECL, Inria BIGS), Romain Azaïs (Université de Lorraine, IECL, Inria BIGS), Anne Gégout-Petit (Université de Lorraine, IECL, Inria BIGS)
  Primary area of focus / application: Modelling
  Keywords: Crack propagation, Modeling, Piecewise deterministic Markov Processes, EM algorithm, Trichotomic algorithm
  

  Submitted at 3-May-2018 11:35 by Florine Greciet
  Accepted
  

  5-Sep-2018 11:30 Detection and Modelling of the Propagation Regimes in Fatigue Crack Propagation
  In aeronautics the risks of engine parts failure are accentuated by the extreme environment in which these structures must operate. Each environment stress can cause a flaw in structure and the stress accumulation repeated on tens of thousands of hours of flight leads to the propagation of the crack. In the long term a crack can cause the break of the structure which can be fatal during a flight. It is essential to control this phenomenon to size the engine parts accordingly. In terms of sizing, the parts are designed to withstand during a defined period the stresses applied to them.
  
  Crack propagation modeling goes through the study of the fatigue crack growth rate law according to the stress intensity factor. The following three regimes are empirically observable through the propagation law. In regime I, referred to as the crack initiation region, crack propagation is a discontinuous process which is extremely slow at very low values of stress intensity factor. In regime II, a power-law relationship between crack growth rate and stress intensity factor range is observed. Regime III corresponds to a quick and unstable crack growth leading to rupture when the stress intensity factor tends to a critical value.
  
  To model the propagation law we propose to use Piecewise Deterministic Markov Processes (PDMPs) which are good candidates to model the three regimes mentioned before because it is a process governed by punctual random jumps governed by an ordinary differential equation between the jump times. It is important to note that times of transition between regimes are non observable on propagation curve because of these two following reasons: the crack propagation phenomenon is continuous and data are usually noisy.
  Therefore, the trajectory is hidden, that means times of transition between regimes, propagation parameters in each regime and the number of propagation regimes are unknown.
  The two methodologies that we propose aim at estimating all these quantities under the following assumptions:
  - In each regime the propagation is described by a polynomial function;
  - Regime changes preserve the propagation continuity;
  - Trajectories are observed through a Gaussian additive noise.
  
  The first methodology consists in assuming that the number of regimes is k and maximizing the likelihood with k regimes in two steps: the models parameters are estimated by an EM algorithm and we use an exhaustive search to find the times of transition. The number of regimes is determined a posteriori by a BIC criterion. The second approach is recursive and consists in estimating only one time of transition in the observed trajectory at each step. To overcome the invalidity of the model we maximize a penalized version of the likelihood by an EM algorithm (parameters estimation) through a trichotomic algorithm (estimation of transition times). When a time of transition is estimated, this procedure is repeated on the observed data after this time of transition. The procedure stops when the last estimated time of transition improves no more the BIC.
  
  Results of simulations obtained for these two methods will be presented.
• Sequential Design of Experiments to Estimate a Probability of Failure in a Multi-Fidelity Stochastic Simulator
  
  Authors: Séverine Demeyer (Laboratoire National de Métrologie et d'Essais), Rémi Stroh (Laboratoire National de Métrologie et d'Essais), Nicolas Fischer (Laboratoire National de Métrologie et d'Essais)
  Primary area of focus / application: Design and analysis of experiments
  Secondary area of focus / application: Metrology & measurement systems analysis
  Keywords: Multifidelity, Sequential design, Bayesian analysis, Gaussian process, Probability of failure
  

  Submitted at 4-May-2018 15:03 by Séverine Demeyer
  Accepted (view paper)
  

  4-Sep-2018 10:50 Sequential Design of Experiments to Estimate a Probability of Failure in a Multi-Fidelity Stochastic Simulator
  The use of computational codes has become common practice when experiments are not feasible or when their number is too parsimonious. The statistical modelling of numerical experiments with Gaussian processes yields a probabilistic decision framework to perform uncertainty quantification (UQ). Multifidelity simulators are considered here for their ability to further reduce computational costs of UQ studies by producing cheap low fidelity approximations of the system and costly high fidelity simulations. A sequential sampling strategy building on the stepwise uncertainty reduction (SUR) criterion is derived for multifidelity simulators (stochastic or deterministic) that compromises at each step between the cost of future observations and the expected reduction of uncertainty of the quantity of interest. The work is applied to a fire safety study to estimate the probability of failure of a smoke extraction system and its associated uncertainty.
• Learning the Uncertainty Propagation Rate in Remaining Useful Lifetime Prediction
  
  Authors: Yingjun Deng (Tianjin University), Alessandro Di Bucchianico (Eindhoven University of Technology), Mykola Pechenizkiy (Eindhoven University of Technology)
  Primary area of focus / application: Modelling
  Secondary area of focus / application: Reliability
  Keywords: Remaining useful lifetime (RUL), Recurrent neural network, Uncertainty propagation rate, Inverse Gaussian random variable
  

  Submitted at 6-May-2018 08:09 by Yingjun Deng
  Accepted (view paper)
  

  4-Sep-2018 12:00 Learning the Uncertainty Propagation Rate in Remaining Useful Lifetime Prediction
  In remaining useful lifetime (RUL) prediction with inspection data, it is a basic fact that long-term prediction has more uncertainty than short-term prediction. Assuming the RUL prediction to be an inverse Gaussian random variable, the ratio of its variance to its mean is considered to measure the uncertainty propagation rate (UPR) in the prediction.
  
  Specifically, in this paper we use a recurrent neural network (RNN) as the RUL predictor. Later the RNN and the UPR are jointly trained with inspection data and failure records via alternative minimization. The proposed algorithms are applied to RUL prediction in an industrial use case within the Horizon 2020 Prophesy project.
• Industry 4.0: How to Support Complex Glass Construction by Statistics
  
  Authors: Scheideler, Eva (OWL University of Applied Sciences), Ahlemeyer-Stubbe, Andrea (Data Mining and More)
  Primary area of focus / application: Design and analysis of experiments
  Secondary area of focus / application: Modelling
  Keywords: Metamodeling, Curved glazing, Design of Experiment, Prediction modeling
  

  Submitted at 7-May-2018 10:10 by Eva Scheideler
  Accepted
  

  5-Sep-2018 10:50 Industry 4.0: How to Support Complex Glass Construction by Statistics
  Large domestic and commercial buildings are typically fitted with extensive insulating glass units. The strength and resistance of these units must be designed to adhere to strict safety regulations. Meta models have been shown to be effective in predicting essential parameters related to safety aspects of glass. In ENBIS 16 we presented the first results regarding meta-models for predicting essential parameters regarding plane glass sheets. We have now extended the meta-model method to curved insulating glass units which have considerably greater complexity. The results are presented in a tool that is easy to use without expert knowledge. We show how the tool enables sales staff and architects to check quickly if chosen glass fulfills safety regulations at an early stage of planning. Such a planning tool saves time and money; the financial impact is large and because we can now deal with both plane and curved glass sheets, the increased flexibility leads to an even wider application and financial benefits.
• Sensitivity Analysis and Constrained Multiobjective Optimization of Computer Experiments: A Case Study of a Centrifugal Compressor Impeller
  
  Authors: Sonja Kuhnt (Dortmund University of Applied Sciences and Arts)
  Primary area of focus / application: Other: Computer experiments
  Keywords: Computer experiments, Kriging, Global optimization, Sensitivity analysis
  

  Submitted at 8-May-2018 10:46 by Sonja Kuhnt
  Accepted
  

  3-Sep-2018 14:20 Sensitivity Analysis and Constrained Multiobjective Optimization of Computer Experiments: A Case Study of a Centrifugal Compressor Impeller
  Computer simulations often replace real-life experiments in engineering and natural sciences nowadays. For the whole process chain of energy transformation, turbo machines play an important role. Centrifugal compressor impellers are a vital part of turbo machines. In our application, a specific impeller geometry is simulated over the entire operating range with regard to its efficiency. As the simulations are very time-consuming and complex, we built meta-models and base sensitivity analysis and optimization on them. Gaussian process models, better known as Kriging models, are a common choice.
  
  Sensitivity analysis investigates how the input variables contribute to the variation of the outcome. A popular method is the use of Sobol indices which quantify the importance of individual input variables or groups of them. We review the Sobol sensitivity indices and the more recently developed total interaction indices (TII) and show how to display them in so-called FANOVA graphs. Besides increasing the knowledge about the unknown black-box function, results from sensitivity analysis are useful within modeling and optimization.
  
  When searching for a global optimum, the well-known efficient global optimization (EGO) procedure considers prediction values as well as the model uncertainty. The compressor impeller application requires the optimization of multiple responses under additional constraints. We apply and compare possible extensions of the EGO algorithm.