Ingeniería de Sistemas y Automática

The Complex Process Visualization Group

The group at a glance

A recently constituted group in the Systems Engineering and Automatic Control Area of the University of Oviedo (ISA) is the Complex Process Visualization Group. By the moment, the group consists of an Associate Professor, an assistant professor and three research engineers. In addition, several Graduate Projects and Ph.D. Thesis are currently being conducted.

The group is aimed at developing methods to visualize the working condition or state of complex processes from the available measurements. The key idea, in which we focus most of our research effort is based on the development of a nonlinear projection of the process state vector onto a low dimensionality (typically 2D) space for visualization purposes. Under certain conditions it is possible to work out a "cartography" of this space, defining regions which correspond to different working conditions of the process. The problem is then reduced to determine visually on a graph the regions which the projection of the process state reaches at every moment.

A justification of this idea can be given under the scope of the theory of dynamic systems, by considering the process as a dynamic nonlinear system whose state vector describes specifically the situation of the process in a certain moment. From this point of view, the measurement, feature extraction and dimensionality reduction stages can be conceived as successive transformations from the process state space onto lower dimensionality spaces which finally lead to a space whose dimension is lower or equal to three.

Most of our research effort is focused on the dimensionality reduction stage. There exist a number of techniques for dimensionality reduction in the state of art which allow to describe the latent structure of process data by means of a few variables. Among these, the Self Organizing Map (SOM) stands out by its topology preserving and probability density function approximation properties and reveals itself as an excellent projection method to be used in complex process condition monitoring. Its main drawback lies on the discrete nature of the image space. This can make difficult or impossible to visualize trends, slow drifts, limit cycles etc. on the state of the process by looking at the projected trajectory.
 
 

With the aim to define a continuous projection, we propose an architecture based on the interpolation between the input space and the output space based on kernel regression. The resulting projection has globally the same properties of the SOM, but being at the same time a continuous projection onto the visualization space.
 
 

Main research lines

On the basis of this idea, our main research topics are listed below:

• The development of a continuous extension of the self organizing map and its implications in complex process monitoring tecniques. Our current approach is mainly based on the paradigm of radial basis functions. More specifically we propose kernel regression techniques (GRNN, generalized regression neural network) for their simplicity and statistical soundness.

• To develop new methods which exploit the information stored in the SOM and show it in an understandable way. We are specially working on the integration of other a priori knowledge sources about the process, such as fuzzy inference systems, process model equations (residuals, parity equations, etc.), or labeled data. These sources of knowledge can be used to produce new SOM planes which we have called fuzzy maps, residual maps and activation maps, which allow to define regions in the visualization space associated to subtle process conditions.

• The analysis of the projection model residuals, which allow to detect deviations of the process with respect to the model, showing also the variables which most deviate from their expected values. In the same line, we are also working in the construction of what we call residual planes, which give information about the confidence level of each region in the visualization space.
• The proposals of new visualization techniques such as continuous 3D visualization of the process data and process state trajectory geometry, based on MDS methods

• The development of portable expert system for fault finding & maintenance of rolling mill
• Diagnosis Assistance for Process Performance Dynamic Plant Condition and Periodic Quality Defects in the Cold Rolling Area
• Modelo de presets para skin-pass de la línea de galvanizado II basado en modelos híbridos
• Proyecto VIBRA
• Publications Of The Group

Main Research Groups