Applications of artificial intelligence: Inferential sensors

What are virtual sensors?

Virtual sensors, called sometimes soft-sensors or eSensors or inferential sensors, are computer algorithms that generate values based on the machine-learning model of a physical process. Thanks to their efficiency and low cost, virtual sensors are attracting the close attention of all branches of manufacturing.

Companies are always striving to increase product quality and production efficiency, and to reduce costs. Environment pollution regulations further elevate their demands. Overall manufacturing performance heavily depends on quality and relevance of measurements coming from sensors. This innovative data-driven approach offers a new, effective way of obtaining accurate readings along the entire processing chain without installing expensive hardware. We build a mathematical model that learns every detail and every interaction of a given process on the basis of a large number of stored historical measurements. After a training process, this model is capable of calculating values for every single sensor at every stage of the production process without touching the physical world.

Applying Artificial Intelligence

Common cases of using infernal sensors:

• Backup physical sensors
• Forecasting future measurements
• Obtaining real-time reading in cases where physical sensors provide very low sampling frequency
• Replacing expensive physical sensors

Modern industrial facilities rely on a large number of sensors. The acquired data is recorded and stored for years in company databases. These databases are the starting point for any kind of machine-learning solution. Although real-world raw data generally contains a lot of impurities such as incorrect measurements from faulty sensors or simply missing entries, after the data is preprocessed and cleansed, it still allows us to build a model that will always provide accurate “measurements.”

Advanced virtual sensors that are supposed to be included in fully autonomous control processes are being equipped with self-diagnostic capabilities to evaluate their own reliability. BitRefine offers also maintenance services as the machine learning model needs to be retrained from time to time to fully reflect an actual state of the production line, such as physical wear of the equipment, which can lead to deviations.