# Building Virtual Sensors using Machine Learning

Virtual sensing techniques are useful to provide economical and feasible alternatives to impractical or costly physical measurements. A virtual sensing system can be implemented by creating a machine learning model able to extract information from other measurements and give, as an output, an estimate of the quantity of interest.

This approach promises cost savings because otherwise physical, costly, tasks are performed only virtualy.

Contents:

## Objectives

Virtual sensing tries to model our target variable so there is no need to measure it physically.

The challenge is to get the right variables for our model to predict the quantity of interest.

## Benefits

It allows to obtain meassures by using a machine learning model, reducing costs and work hours, as these meassures must usually be obtained in a laboratory.

## Approach

The way to virtual sensing is to create an approximation model that takes in some input variables related to out target and estimates the latter.

Neural networks can model the correct values of the target variable so we don't need to take physical meassures.

That allows to save costs and time on meassure's obtention.

## Data set

The data set contains measurements from our system or process.

It comprises state and performance variables.

State variables are those inputs that determine the system's performance and are not actionable by the company's technicians.

Some examples of state variables are:

• The pH of a water sample.
• The water salinity in a desalination plant.

Performance variables are the outputs of the system and depend on the state variables.

## Mathematical model

The model of a process is a mathematical description that adequately predicts the physical system's response to all anticipated inputs.

More specifically, it relates the performance variables with the state variables.

$$performance\_variables = function(state\_variables)$$

Neural networks are algorithms used to fit multi-dimensional and non-linear functions from data sets.

The inputs to the neural network include the state variables. The outputs from the neural network are the predicted performance variables of the system for that scenario.

## Conclusions

Virtual sensing makes companies save money and time since meassure taking becomes a software issue, which can improve planning and decission making.

Neural Designer uses machine learning to build predictive models that represent a broad range of variables associated with virtual sensing.