Development of a computational tool in Python language based on Principal Component Analysis, Self-Organizing Map and Support Vector Machines applied to process monitoring

Resumo

Industry is going through a new transformation, called Industry 4.0, in order to absorb recent advances
in technology as a way of meeting the constant need for efficient and automated processes. In the chemical industry
specifically, these techniques enable to understand the behavior of process guarantee his safety and good
performance. There are many methods for implementation of process monitoring system, between techniques of
unsupervised and supervised machine learning, such as the Principal Component Analysis (PCA), Self-Organizing
Map (SOM) and Support Vector Machines (SVM). In this context, in this paper was studied the development of a
computational tool for monitoring the process variables based on cited techniques in order to detect conditions that
can affect process performance and, consequently, product quality. The implementation was developed in Python
environment and it was applied in a generated fault data from data available in the literature for drying process in
an evaporator bed. The statistical metrics F1, accuracy and precision were used to evaluate the techniques and the
results indicated that the computational tool showed to be effective on application that was studied. Finally, the
tool presented advantages for being free, having an intuitive interface and can be easily used for process
monitoring.