ENBIS-11 in Coimbra

4 – 8 September 2011 Abstract submission: 1 January – 25 June 2011

My abstracts

 

The following abstracts have been accepted for this event:

  • Performance comparison of parametric and nonparametric profile monitoring

    Authors: Bianca Maria Colosimo*, Marcela Meneses** and Quirico Semeraro*
    Affiliation: *Dipartimento di Meccanica, Politecnico di Milano, Italy, **Escuela de Ingenieria en Produccion Industrial, Instituto Tecnologico de Costa Rica.
    Primary area of focus / application: Process
    Keywords: profile monitoring , statistical process control (SPC) , quality , B-spline , functional data , nonparametric mixed-effects
    Submitted at 25-May-2011 14:29 by Marcela Meneses
    Accepted
    5-Sep-2011 10:10 Performance comparison of parametric and nonparametric profile monitoring
    Profile monitoring is a recent field of research in Statistical Process Control (SPC) literature, which is attracting the interests of many researchers. As a matter of fact, in many industrial applications (e.g., calibration studies, geometric specification and process signal monitoring), the aim is to check the stability of a profile, which represents the functional relationship between a dependent and one or more independent variables.
    Most of the approaches on profile monitoring are based on combining functional data modeling to multivariate control charting. To this end, many different approaches have been proposed but very few performance comparison studies have been presented in the literature in order to guide practitioners in selecting the best approach for the problem at hand.
    This paper tries to fill this gap by comparing two of the most promising approaches for complex profile monitoring with autocorrelated errors. In particular, the first approach is based on combining functional data modeling via B-Spline regression and autoregressive error models via ARMAX. The second method is a nonparametric profile monitoring approach which has been recently proposed in the literature (Qiu, Zou and Wang, 2010).
    A real case study dealing with density measurements along the particleboard thickness (usually referred to as Vertical Density Profile -VDP) is taken as reference throughout the paper. With reference to this real case, performances of the parametric and nonparametric approaches are compared. Additional features as flexibility, computational simplicity and interpretability are discussed, too.
  • Towards a systematization of profile analysis methods as a basis for the development of flexible and generalized data analysis framework

    Authors: Véronique M. Gomes Ana C. Pereira Pedro M. Saraiva Marco S. Reis
    Affiliation: University of Coimbra
    Primary area of focus / application: Modelling
    Keywords: Multiway methods , calibration , monitoring , profiles
    Submitted at 25-May-2011 18:23 by Marco P. Seabra dos Reis
    Accepted (view paper)
    5-Sep-2011 16:33 Towards a systematization of profile analysis methods as a basis for the development of flexible and generalized data analysis framework
    Data generated by current processes or collected in activities of product characterization, tend to present increasingly complex structures. The traditional arrangement of data in two-way arrays of variables vs time, variables vs production units, or similar, is not longer enough to accommodate the structures found in process and product quality data, nor are the classical methods able to analyze them, in a effective and efficient way. For instance, batch processes typically generate 3-way data arrays (or even higher-order arrays, if spectral measurements or hyperspectral images, for instance, are collected over time), and products can be characterized by 1-way profiles (chromatograms, spectra, particle size distribution curves, etc.), 2-way profiles (GC-MS, HPLC-DAD, grey images), 3-way profiles (hyperspectral images, 3-way hyphenated instruments), etc. In this context, there is a current need to evolve to more general environments for data analysis frameworks, which are able to handle data with such different data structures while pursuing the analysis goals.
    In this context, we purpose in this paper: i) to interpret all these different data structures as different manifestations of we call, in an abstract sense, “profiles”; ii) a systematization of the profiles found in practice regarding the methods required to analyze them and the task in question, in order to find cross-dimensional patterns about how to analyze data more effectively, irrespectively of their intrinsic dimensionality, and, possibly, to develop a coherent ontology that facilitates the development of new integrated approaches for profiles analysis; iii) a comparison study of multiway methods for handling similar tasks (namely, calibration and process monitoring).
    As a result of this systematization, a general architecture of data analysis frameworks for handling profiles can begin to be developed, in terms of components required to accomplish each task, irrespectively of the dimensionality of the profile.
  • Statistical Process Control of a Multiple Stream Process with Varying Means

    Authors: Eugenio Kahn Epprecht Ítalo Parente de Barros
    Affiliation: Departamento de Engenharia Industrial, PUC-Rio, Rio de Janeiro, Brazil
    Primary area of focus / application: Process
    Keywords: group control charts , multiple stream processes , variable means , statistical process control , group charts
    Submitted at 26-May-2011 00:26 by Eugenio Epprecht
    Accepted
    6-Sep-2011 11:45 Statistical Process Control of a Multiple Stream Process with Varying Means
    This study shows the application of techniques of Statistical Process Control (SPC) in a cosmetics industry, in a situation in which conventional techniques are not applicable. The process to be controlled is composed of eight streams, which produce eight units of the product at a time. The process has the peculiarity that the means of the streams change in time, even in a condition of statistical control. The means not only wander but they change discontinuosly from run to run. The runs are short, preventing good estimation of the means because there is not enough data for a precise “Phase I” analysis before “Phase II” monitoring. The control schemes proposed in the literature hitherto for multiple-stream processes assume constant means, and streams with similar means and variance, and are therefore not applicable to this process. A new scheme was then developed for the statistical control of the process, which blends the principles of the “group charts” and of “acceptance control charts”. A review was also presented of some techniques of statistical control of multiple-stream processes, including traditional and more recent methods.
  • Focused Multivariate Analysis for Process Optimization

    Authors: Tania F.G. Guerreiro Cova, Jorge L.G.F.S. Costa Pereira, Alberto A.C.C. Pais
    Affiliation: Departamento de Química, Faculdade de Ciências e Tecnologia, Universidade de Coimbra, 3004-535 COIMBRA, PORTUGAL
    Primary area of focus / application: Process
    Keywords: Process , diagnosis , multivariate , analysis , optimization
    Submitted at 26-May-2011 16:36 by J. Costa Pereira
    Accepted
    6-Sep-2011 15:25 Focused Multivariate Analysis for Process Optimization
    Assessing process variables and estimating optimal conditions for manufacturing in the industrial plant is an extremely demanding task, essentially because of the huge number of factors impacting upon the production line. On the other hand, a large number of conditions varies independently, making them difficult to follow individually along time and establish their relationship with production results. Also, the manufacturing process cannot be directly transposed into the laboratory, and “scale-down/scale up” activities are expensive and time consuming, thus restricting the use of experimental design related-techniques.

    In this work, we show that multivariate data analysis is a powerful tool for industrial process diagnosis and subsequent optimization process. In fact, standard and state-of-the art methods stemming from chemometrics are easy to apply, and provide a deep insight in the process, allowing for subsequent identification of critical areas and sub-process, and providing a scaffold for modelling tasks.

    A case-study from the food industry is analysed in-depth, from the dual perspective of process diagnosis and focused optimization.
  • Integration of SPC / EPC in multivariate stationary and non-stationary processes

    Authors: Raquel D. Moita, Tiago J. Rato, Pedro M. Saraiva, Lino O. Santos and Marco S. Reis
    Affiliation: Department of Chemical Engineering, University of Coimbra, Portugal
    Primary area of focus / application: Process
    Keywords: SPC / EPC integration , Multivariate monitoring , Nonlinear model predictive control , PI controller
    Submitted at 26-May-2011 19:29 by Raquel Durana Moita
    Accepted (view paper)
    5-Sep-2011 10:30 Integration of SPC / EPC in multivariate stationary and non-stationary processes
    Several works have appeared in the literature regarding the integration of Engineering Process Control (EPC) and Statistical Process Control (SPC). These contributions can be conceived as belonging to two fundamentally distinct classes of integrative approaches: one in which the SPC mechanism supervises, without interfering, with an EPC controller, which is active all the time, and another in which the SPC mechanism acts as a trigger of the EPC control scheme.

    Regarding the first class of approaches, where both techniques act simultaneously, there are at least three possibilities to monitor the EPC controlled process: a) Monitoring the quality characteristic, using control charts to signal large deviations from the target; b) Monitoring the adjustable variables: large deviations in the quality characteristics will result in large adjustments made and thus these variables should also have information that can be used for process monitoring; c) Monitoring both the process inputs and outputs.

    In this paper, we consider the first class of approaches, in which the EPC control is active all the time, exemplified with two study cases, with characteristics that have not been addressed before in the literature. In the first case, a multivariate monitoring scheme is implemented in a chemical process together with existent PI feedback controllers. It consists of a continuous stirred tank reactor (CSTR), where a first-order endothermic and irreversible reaction takes place, which is also fitted with a heating jacket for thermal control. The second case considered refers to the multivariate monitoring of a lab-scale fed-batch Escherichia coli biomass production process, subject to a nonlinear model predictive controller (NMPC).
  • Process Multivariate Analysis: Relevant Indicators in Biodiesel Production

    Authors: Susana M.C. Lima Melo, Jorge L.G.F.S. Costa Pereira
    Affiliation: Departamento de Química, Faculdade de Ciências e Tecnologia, Universidade de Coimbra, 3004-535 COIMBRA, PORTUGAL
    Primary area of focus / application: Process
    Keywords: Industrial , process , diagnosis , biodiesel , multivariate , data , analysis
    Submitted at 27-May-2011 03:18 by J. Costa Pereira
    Accepted (view paper)
    5-Sep-2011 16:25 Process Multivariate Analysis: Relevant Indicators in Biodiesel Production
    Industrial process analysis is typically an extreme data analysis situation since a huge number of variables are needed to be considered simultaneously in order to bring some enlightenment into the process.

    Process segmentation can allays be used as an analytical approach to the system but this is usually a fallacious simplification since it do not attends in variable interaction along all the process.

    Other difficulty in the manufacturing process diagnosis arises from the impossibility to make a correct direct transposition between the plant and the laboratory, and “scale-down/scale-up” activities are imprecise, expensive and time consuming, leading to misinterpretation of the system. Also, process variables are restricted in variations in order to keep final product under quality specifications which is a painful situations for experimental design plans.

    For this purpose, our approach to overcome to this adverse situations in process diagnosis is to auscultate all industrial process, from raw material to product final characteristics, during a long time process evaluation, allowing standard variable fluctuation. System behaviour is posteriorly analysed via the use of some powerful multivariate data analysis applied to recording results data set.

    We will present an discuss a case-study at a Biodiesel line production.
    In this case, the transesterification process was fully studied using 144 process predictors and 66 oil characteristics as responses. Predictors sub-space was further characterized into raw material (15) and process variables (129) and each sub-space was explored in order to reveal a) interdependencies, b) relevant informative variables, c) redundant information, d) outliers and e) leverage.