Using Virtual Sensors to Optimize Process Operation The Power of Virtual Sensors and Predictive Analytics Applications

Ivana Lukec, Ph.D.
Using Virtual Sensors to Optimize Process Operation

It is not the question anymore if predictive analytics can improve the plant operation - the only question left is "How much?"

These days everyone is talking about IIoT (Industrial Internet of Things) and big data, but do we really know what those are capable of doing when installed on real plants? Do we have a full picture of their practical benefits and potential improvements?  

Process history and real time data hold the huge power if they are exploited in the right way. They are essential for optimal operation of any plant. 

The information that is hidden in the process data reveals improvement potentials of any plant - but only if we look close enough: using the right knowledge and the right tools for the right set of process data. The three kings of predictive analytics: data, tools & process knowledge!

When you have those three, then optimized can really be anything - distillation columns, reactor performance, heat exchanger operation, furnace operation, emissions, economical indicators – the sky is the limit! And the creator's mind!


What are virtual sensors?

The key elements for using predictive analytics applications are virtual sensors. They are plant models devoted to the estimation of plant variables. They are known as inferential models, virtual sensors, or soft sensors.

Information derived from process data can be used for building mathematical models to monitor & control the process or to receive & predict valuable information.

Virtual sensors (soft sensors, inferentials, etc) are an estimation of any system variable or product quality developed by using mathematical models and data acquired from other available ones.

Virtual sensors are installed as a real time application that is able to perform an estimation, visualisation and even control of a process variable or variables.

They represent an alternative paradigm to traditional instruments and allow us both to customize the measuring facility capabilities to user application and to take control of the way measurement results are used and presented.
What is their most important value is that by using them, it is possible to predict important variables or product qualities that are related to process safety, product quality and economical aspects of any process. They represent the basis for real time process optimization.

Virtual sensors are a valuable tool in many different industrial fields of application, including refineries, chemical plants, cement kilns, power plants, pulp and paper industry, food processing, nuclear plants, urban and industrial pollution monitoring, just to give a few examples. They are used to solve a number of different problems such as measuring system back-up, what-if analysis, real-time prediction for plant control, optimization, sensor validation and fault diagnosis strategies.

Mathematical models of processes designed to estimate relevant process variables can help to reduce the need for measuring devices, improve system reliability, become a solid ground for process optimization and develop tight control policies.

Virtual Sensors offer a number of attractive properties:

  • they represent a low-cost alternative to expensive hardware devices, allowing the realization of more comprehensive monitoring networks;
  • they can work in parallel with hardware sensors, giving useful information for fault detection tasks, thus allowing the realization of more reliable processes;
  • they can easily be implemented on existing hardware (e.g. microcontrollers) and retuned when system parameters change; Soft Sensors using Industrial Applications
  • they allow real-time estimation of data, overcoming the time delays introduced by slow hardware sensors (e.g. gas chromatographs), thus improving the performance of the control strategies.

Real-world applications

So, what are the real-world experiences of using virtual sensors on process sections such as distillation columns, reactors, heaters? And what can be achieved with their application?

Optimizing debutanizer column

When looking into any distillation column, two variables are of key importance for the plant experts and are related to the top and bottom product quality. Operating distillation column without disturbance and on the limit of required top and bottom product quality means operating the column optimally and on the best economy. 

Real-time estimation of both the C5 (stabilized gasoline) content in the overheads of the debutanizer column and the C4 (butane) content in the bottom flow enables continuous monitoring of top and bottom product quality and is the basis for optimal column operation. Bottom and top product qualities and the consequence of the vapor-liquid equilibrium that takes place inside the column. The temperature profile of the column is the result of mass and heat transfer and as such defines product qualities. Therefore, by stabilizing temperature profile, we are able to stabilize distillation operation. By operating the column at the limit of both the bottom and top product qualities, we are able to run the column optimally while achieving the best possible economy. While having the real-time information about top and bottom product qualities, we have the information and ability to react and operate the column optimally.

Optimal operation of hydrotreatment reactor 

Reactors are representing the hearts of many units in chemical industry and are the key spots for optimization. Reliable online quality prediction of reactors chemical processes often encounters different challenges, including process nonlinearity, complex interactions between variables, and strict product quality requirements.

When it comes to refining units, hydrotreatment processes are very important in achieving the requirements of low-sulfur fuels. Hydrotreatment of gas oil fractions is commonly accomplished in the trickle-bed reactors where there are three phases: gas (mostly hydrogen), liquid (gas oil) and solid catalyst particles. 

Soft sensors have proven their predictability of sulfur content in the product and can work very efficiently in parallel to a hardware analyzer as a backup while their measurement becomes impossible due to instrument failure, maintenance or repair, or can be applied as the independent measurement.

Why are the virtual sensors important for reactor optimization? First of all, they enable continuous monitoring and prediction of a certain product quality an hour (sometimes even more) before this quality can be recorded with the hardware sensor. This gives enough time to take the necessary steps and keep the quality as close to the required one.

Any online optimization (such as APC) is not possible without the virtual sensor which  continuously and in real-time estimates important product quality.

The estimation becomes important as well to maintain reactor operation in optimal limits in order not to harm catalyst life more than necessary and keep it in the best performance as long as possible. 

Estimation of NOx emissions in large plants

We are witnesses of the climate change and necessary restrictive limits on emissions, which include NOx emissions. Sophisticated monitoring strategies for NOx emissions are now a matter of interest. 

Stationary sources, such as power plants and refineries, significantly contribute to the emission of nitrogen oxides by means of chimney fumes produced by the combustion of residuals deriving from a number of processes. Hence, industries are deeply interested in the development of measurement and/or estimation strategies for these pollutants.

NOx emissions are measured using an on line analyzer. To have comprehensive information about emitted pollutants, other chemicals are also monitored. Data produced by the analyzer are collected in a refinery database. Due to the harsh environment, the analyzer is frequently off-line for scheduled maintenance and during these periods mathematical models need to be used to estimate the NOx level.

Mathematical models are as well used to predict any disturbances and possible violations of required limits, which gives enough time to take adequate steps in order not to violate any restrictions.

Model development for virtual sensors

Models that are used for virtual sensors development can be based on black-box models or have a thermodynamic background. However, the plants of tomorrow will not possible without those kind of virtual sensors and predictive analytics. Models that combine process knowledge, process data and the right tools and experts who are able to combine those three will be of immense worth for optimal control of plants for the future.