Search

[EN] A comparison among widely used multivariate latent variable-based techniques for supervised process fault diagnosis was carried out. In order to assess their overall performance several diagnosis criteria were proposed (C-1: most suspected fault assignment; C-2: threshold-based fault assignment). Additionally, it was evaluated i) how the size of the training set used to build the latent variable models affected the diagnosis ability of the methods under study, ii) how they behaved under new types of failures not included in the original list of fault candidates and iii) which of them were more suitable for either early or late diagnosis. To accomplish all these objectives, the approaches were tested in different scenarios. Two datasets were analysed: the first was generated by a Simulink-based model of a binary distillation column, while the second relates to a pasteurisation process performed in a laboratory-scale plant., This research work was partially supported by the Spanish Ministry of Economy and Competitiveness under the project DPI2017-82896-C2-1-R and Shell Global Solutions International B.V. (Amsterdam, The Netherlands).

[EN] A novel method and software system for rational handling of time series of multi-channel measurements is presented. This quantitative learning tool, the On-The-Fly Processing (OTFP), develops reduced-rank bilinear subspace models that summarise massive streams of multivariate responses, capturing the evolving covariation patterns among the many input variables over time and space. Thereby, a considerable data compression can be achieved without significant loss of useful systematic information. The underlying proprietary OTFP methodology is relatively fast and simple it is linear/bilinear and does not require a lot of raw data or huge cross-correlation matrices to be kept in memory. Unlike conventional compression methods, the approach allows the high-dimensional data stream to be graphically interpreted and quantitatively utilised in its compressed state. Unlike adaptive moving-window methods, it allows all past and recent time points to be reconstructed and displayed simultaneously. This new approach is applied to four different case-studies: (i) multi-channel Vis-NIR spectroscopy of the Belousov Zhabotinsky reaction, a complex, ill understood chemical process; (ii) quality control of oranges by hyperspectral imaging; (iii) environmental monitoring by airborne hyperspectral imaging; (iv) multi-sensor process analysis in the petrochemical industry. These examples demonstrate that the OTFP can automatically develop high-fidelity subspace data models, which simplify the storage/transmission and the interpretation of more or less continuous time series of high-dimensional measurements to the extent there are covariations among the measured variables., This research work was partially supported by the Spanish Ministry of Economy and Competitiveness under the project DPI2014-55276-C5-1R, Shell Global Solutions International B.V. (Amsterdam, The Netherlands), Idletechs AS (Trondheim, Norway), the Norwegian Research Council (Grant 223254) through the Centre of Autonomous Marine Operations and Systems (AMOS) at the Norwegian University of Science and Technology (Trondheim, Norway) and the Ministry of Education, Youth and Sports of the Czech Republic (CENAKVA project CZ.1.05/2.1.00/01.0024 and CENAKVA II project L01205 under the NPU I program). The authors want to acknowledge Prof. Bjorn Alsberg for providing the Vis-NIR equipment and the Laboratorio de Sistemas e Tecnologia Subaquatica of the University of Porto, the Hydrographic Institute of the Portuguese Navy and the University of the Azores for carrying out the REP15 exercise, during which the hyperspectral push broom image was collected.

[EN] This work describes an application of Multivariate Statistical Process Control to monitor soybean oil transesterification. For the development of multivariate control charts, near infrared spectra were acquired in-line during the evolution of ten batches under Normal Operating Conditions. They were then organized in a three-way array (batch × spectral variable × time). This structure was analysed by the two most commonly used approaches to develop batch monitoring schemes for handling such kind of data, referred to as Nomikos-MacGregor (NM) and Wold-Kettaneh-Friden-Holmberg (WKFH), respectively. To assess the performance of the approaches, eight test batches, during which specific interferences were induced, were manufactured. When applied for off-line monitoring, both NM and WKFH correctly pointed out such intentionally produced failures. On the other hand, concerning on-line monitoring, NM exhibited a better fault detection capability than WKFH. Contribution plots were found to highlight the spectral region mostly affected by the disturbances regardless of the modelling strategy resorted to., The authors would like to thank FACEPE/NUQAAPE, CNPq/INCTAA science funding programs for partial financial support. Research fellowships granted by the Brazilian agencies ANP/Petrobras and CNPq are also gratefully acknowledged. This research work was partially supported by the Spanish Ministry of Economy and Competitiveness under the project DPI2014-55276-C5-1R and Shell Global Solutions International B.V. (Amsterdam, The Netherlands).