Just-in-time semi-supervised soft sensor for quality prediction in industrial rubber mixers

Mixing is a ubiquitous operation in process engineering. It is not only used for combining materials, but also for promoting heat and mass transfer, increasing aeration, suspending solids, and modifying material structure. Measurement techniques have the potential to optimise industrial mixing processes and improve product quality by monitoring critical process parameters. Real-time sensing techniques that do not require manual material sampling are of particular interest owing to their automatic data acquisition capability. This makes them suitable for use in control systems for process automation and eventually as connected sensors in Industry 4.0. This review article focuses on these measurement techniques, defined as in- and on-line, along with their capability for implementation in industrial processes. The applications reviewed include liquid–liquid, gas–liquid, solid–liquid, solid–gas–liquid, in addition to solids blending. A technique selection section discussing the decision-making process when choosing a sensor and a summary table including the advantages, disadvantages, applications and limitations of each technique are provided. The article concludes by discussing the future of monitoring techniques for mixing processes.

A review of in-line and on-line measurement techniques to monitor industrial mixing processes

Quality in a manufacturing process implies that the performance characteristics of the product and the process itself are designed to meet specific objectives. Thus, accurate quality prediction plays a principal role in delivering high-quality products to further enhance competitiveness. In tubing extrusion, measuring of the inner and outer diameters is typically performed either manually or with ultrasonic or laser scanners. This paper shows how regression models can result useful to estimate both those physical quality indices in a tube extrusion process. A real-life data set obtained from a Mexican extrusion manufacturing company is used for the empirical analysis. Experimental results demonstrate that k nearest-neighbor and support vector regression methods (with a linear kernel and with a radial basis function) are especially suitable for predicting the inner and outer diameters of an extruded tube based on the evaluation of 15 extrusion and pulling process parameters.

/pdf/using-regression-models-for-predicting-the-product-quality-4lg8j8w5ap.pdf

Using regression models for predicting the product quality in a tubing extrusion process

In internal rubber-mixing processes, data-driven soft sensors have become increasingly important for providing online measurements for the Mooney viscosity information. Nevertheless, the prediction uncertainty of the model has rarely been explored. Additionally, traditional viscosity prediction models are based on single models and, thus, may not be appropriate for complex processes with multiple recipes and shifting operating conditions. To address both problems simultaneously, we propose a new ensemble Gaussian process regression (EGPR)-based modeling method. First, several local Gaussian process regression (GPR) models were built with the training samples in each subclass. Then, the prediction uncertainty was adopted to evaluate the probabilistic relationship between the new test sample and several local GPR models. Moreover, the prediction value and the prediction variance was generated automatically with Bayesian inference. The prediction results in an industrial rubber-mixing process show the superiority of EGPR in terms of prediction accuracy and reliability. © 2014 Wiley Periodicals, Inc. J. Appl. Polym. Sci. 2015, 132, 41432.

Real-time property prediction for an industrial rubber-mixing process with probabilistic ensemble Gaussian process regression models

Industrial Mooney viscosity prediction using fast semi-supervised empirical model

The current study shows how data mining and artificial intelligence techniques can be used to introduce improvements in the rubber extrusion production process. One of the keys for planning manufacturing values is prior knowledge of the properties of the material to be extruded. At present, such information is obtained through laboratory trials performed on samples taken off line after the elastomers have been manufactured, with the subsequent cost and delays. In view of these problems, the present study proposes a neural model capable of predicting the characteristics of rubber from the composition of the mixture and the mixing conditions, without having to wait for laboratory results, thus guaranteeing the traceability of the product in the process and the values according to their specific characteristics and also achieving a reduction in costs deriving from smaller amounts of discarded material during the performance of the tests, etc.

A neural network-based approach for optimising rubber extrusion lines

The most common goal of the factory owner is to achieve better quality in the final product by means of an improved process control. The significance and relevance of optimizing the existing control models is even greater in the open-loop control systems or in those governed by computational methods dependent on adjustable parameters. This talk reviews some typical industrial environments and focuses on some parts of them in order to show the real interest of these improvements. We will identify some difficulties in obtaining these improvements and show how the optimal control model for the manufacturing process can be obtained from data provided by sensors. We will also discuss some technical problems that are related to the main goal, and will identify some topics concerning outliers, density and topology. Also, we will show how these techniques can be applied as an instrumental toolbox in addressing some environmental problems.

Data mining and simulation processes as useful tools for industrial processes

This paper describes the outcomes of a study that the EDMANS(**) group has recently performed in a rubber extrusion process, focusing on the knowledge discovery phase previous to the system modeling. Some of the tools developed to satisfy the special needs of such a process are also presented: the CiTree algorithm for clustering subpopulations in massive databases and the PAELLA algorithm for outlier detection and data cleaning in non normal samples like those typically found in industrial processes. Finally, the results obtained by these data mining techniques when applied to a real rubber extrusion databases are shown.

http://www.wseas.us/e-library/conferences/2006prague/papers/514-209.pdf

F. Alba Elías

Papers

A neural network-based approach for optimising rubber extrusion lines

Data mining and simulation processes as useful tools for industrial processes

Knowledge discovery in rubber extrusion processes