Overview / Usage

There is a great increase in the quantity, quality and accessibility of data, due to lowering costs of sensors, communication devices and data storage. More and more chemical plants are dependant on sensors to perform fine control, and newer processes tend to utilize multivariable control. That makes the problems of having a sensor failing much stronger. This project intends to create a program capable of create virtual signals to be sent to the control/monitoring system atenuating the problems of having a sensor failing, making process control and monitoring more robust

Methodology / Approach

We will develop a dataset simulating a chemical process. First a simple one, Van der Vusse reactor, later a more complex one: Tennessee Eastman process. We intend to use a denoising autoencoder to encode all information contained in the dataset into a single neural net. We will also compare the program with other techniques used like running a linear regression or using the mean signal as a new input. After we will use this new algorithm in a dynamic simulator to testo for stability

Technologies Used

Simulink
Python
Scipy
SKlearn
Keras