Virtual Sensor Building
with PYTHIA Virtual Sensors

Virtual Sensors can bring significant benefits to your company. They can replace expensive or error-prone real sensors, minimize the need for lab measurements, or directly measure impossible or abstract parameters like quality or durability. Up to now, creating Virtual Sensors was hard and expensive. With PYTHIA Virtual Sensors, it can be done quick and simple.

Building Virtual Sensors

The conventional creation of virtual sensors is a time-consuming and expensive process. With PYTHIA Virtual Sensors, it can now be done much faster and simpler.

The construction of the underlying prediction model usually requires vast amounts of data and takes a lot of time. Data scientists have to synchronize and clean up the data. In close cooperation with domain experts, relationships between data have to be extracted following educated guesses. Causes of events have to be analyzed, and a prediction model has to be built. If neural networks are used as a basis, the correct design must first be found and then undergo extensive training.

With PYTHIA Virtual Sensors, this changes dramatically: The domain expert on its own can create virtual sensors quick and straightforward without the involvement of data scientists. PYTHIA Virtual Sensors takes care of both data cleaning and prediction model creation. Therefore, creating virtual sensors becomes many times faster, easier, and cheaper.

With PYTHIA Virtual Sensors, every company, no matter their level of experience, can create virtual sensors quick, simple, and fully automated.

How it works

Creating virtual sensors with PYTHIA Virtual Sensors is a quick and simple procedure that requires no data science expertise.

First, you provide the unfiltered, raw data of your processes (time-series data or tabular data, can be up to 1 TB). An in advance exclusion of individual signals or parts of the data is neither necessary nor recommended.

You then define which of the signals (parameters) within your data should be predicted (measured) by the virtual sensor. This can be a single signal or a combination of multiple signals.

PYTHIA Virtual Sensors then performs an automated data analysis and creates the sensor and a final report about the achieved quality of the sensor. This sensor can now easily be implemented on-cloud, on-sight, or on-edge.


quality Assurance – Process safety – waste reduction
Paper making – Measuring paper quality

During the continuous process of papermaking, paper quality is a critical process parameter. Unfortunately, a direct and online determination of this parameter is impossible since a lab measurement is required. Since it takes about one hour from sampling to the result, the operator is flying blind, so to speak, until the next lab value is available.

With PYTHIA Virtual Sensors, a virtual paper quality sensor can easily be created, which provides a live prediction of the quality. The operator can now follow paper quality’s progression continuously and live, which allows instant adjustments. This way, safety margins can be lowered, resulting in reduced material consumption and increased throughput.

Dead oil – Measuring oil viscosity

The dead oil viscosity is a critical process parameter that heavily depends on the type of oil. Even slight differences in composition can dramatically impact the viscosity, making it almost impossible to address this issue with classical black oil correlations.

With PYTHIA Virtual Sensors, a virtual viscosity sensor can be created live predicting the viscosity based on a model that was learned from the history of thousands of parameters.

Concrete – Measuring compressive strength

The concrete compressive strength is a critical process parameter that determines the quality of the concrete. It depends highly nonlinear on age and ingredients. Usually, it has to be determined in the laboratory at great expense using a crushing test on concrete cylinders, which can take up to a month and is prone to human error.

With PYTHIA Virtual Sensors, a virtual strength sensor can be created that determines the compressive strength live and on-sight from other well known and easily measurable parameters.

Mining – Measuring Silica during flotation

In a flotation plant, minerals are processed to concentrate the iron ore. A critical process parameter is the percentage of Silica, which is the impurity of the iron ore. This value can only be determined with one hour delay by a lab measurement.

With PYTHIA Virtual Sensors, a virtual Silica sensor can be created, which determines the iron ore’s impurity value from other well known and easily measurable parameters.

Based on this live prediction, the engineers can take corrective actions during the process to reduce impurity. Fewer impurities mean less ore goes to tailings where Silica is reduced in the ore concentrate. This reduces costs and effort and has a positive effect on the environment.

Liquid tank – Measuring temperature in the middle

A closed liquid tank has sensors to measure the temperature at the inlet and outlet. Besides, the temperature in the middle of the tank should be measured. This is difficult because the tank has to be opened, and a sensor has to be placed in the middle.

With the help of a virtual sensor, this needs to be done only once: With PYTHIA Virtual Sensors, a virtual temperature sensor can be created that predicts the temperature in the middle based on the inflow and outflow for the first tank. The same sensor can be used for all other tanks as well without the need to open them.

Sensor training on simulated data

The temperature in a nuclear fusion reactor cannot be measured directly because it is too hot. However, many other values can be measured, such as pressure, density, etc.

With a virtual temperature sensor, the temperature could be measured indirectly based on the other available parameters. To build that sensor, training data is required. This data can be obtained by a simulation of the nuclear reactor. The virtual sensor can then be created based on the simulation’s historical data and transferred to the real application.

These are only a few examples of PYTHIA Virtual Sensors applications. In fact, the potential industrial applications are countless.

General applicability

PYTHIA Virtual Sensors can be used in many industrial applications. If you are wondering whether it can be used in your particular case as well, focus on the following criteria:

  • There exists a critical process parameter whichs direct measurement is either impossible, expensive or prone to error.
  • There exists either time-series data, tabular data or both.
  • The available data contains enough information to predict the parameter (this can easily be checked in a pre-analysis)

If your conditions match those criteria, there is a good chance that PYTHIA Virtual Sensors can create a virtual sensor for this parameter.

Why PYTHIA? Find it out!