Machine Learning Modeling in Industrial Processes for Visual Analysis

The use of visualization tools makes it easy to analyze the operation of an industrial process. Given the large number of variables involved in today’s industrial systems, it is necessary to use techniques that reduce both the size of the data and the number of samples. In addition, since industrial systems involve similar processes running in parallel, this information can be added to analyze the processes. This paper proposes the use of a variant of self-organizing maps, EnvSOM, which allows conditioning the projection of these maps based on a set of variables. This variant is applied to operational data from AQUASOL II pilot plant located at the Plataforma Solar de Almería (PSA), which consists of several flat-plate collector loops that share a common water distribution pipe. Projecting the data in a conditional manner visualization maps are generated based on differences that allow to determine the existing differences between the heating loops.