What is an AI powered Soft-Sensor?

In process and manufacturing industries it is not always feasible to measure every variables of interest. In those scenarios soft sensors can be really helpful. So, Soft sensors, also known as virtual sensors, are models used to estimate the values of unknown process variables using the available measurements. Soft sensors are also used when physical sensors are very costly or real-time measurements are not possible.

Soft sensor models can be based on first principles utilizing the underlying fundamentals or physics or they can be data-driven. ML techniques can be very useful in developing data driven soft-sensors. For e.g. Feed Forward Neural Network (FFNN) is one of such techniques. Using FFNN soft sensor models can be developed to estimate the value of difficult to measure variable using the available measurement. As an example, suppose for a process "pressure", "temperature" and "flow-rates" are available, however "concentration" is what we want to estimate which is very difficult to measure in real-time. Now using FFNN model can be built to predict concentration from the variable data which are readily available.