The IoT seminar held in Tampere University of Applied Sciences (TAMK) inspired us to write about our Collo Liquids Analyzer, and how it relates to IoT and Artificial Intelligence, that is now more and more used in big data analysis. Collo® Liquids Analyzer demo was presented at the exhibition, showing water analysis and recognizing small amounts of oil or salt in water.
The IoT architectures in general include sensors, communication to cloud and data analysis. The analysis can consist of algorithms for big data and even AI-based algorithms.
Collo® Liquids Analyzer has versatile architectural possibilities. It can be used as an analyzing part in plant automation and connected to automation systems via standard interfaces. It can also act as an IoT system having 3G/4G communication to a cloud that runs extensive Collo analysis algorithms. It can even run in both modes at the same time; communication within plant automation and cloud connection.
In the IoT world there are wide range of sensors, like temperature, humidity, pressure and surface level sensors for example. Sensors are typically measuring one or two variables, and the updating interval is long, e.g. sending data every 15 minutes.
Collo® Liquids Analyzer has higher data communication standards. It is a resonator that creates an electromagnetic field in liquid. It measures liquid with several frequencies, so the amount of raw data is high, and also data transmission happens in short intervals.
The typical IoT communication mechanisms are 4G and new communication opportunities for low amount of data like LoRa and now emerging NB-IoT. Digita Ltd in Finland provides LoRa, and telecom operators will sell NB-IoT that is now in a piloting phase.
LoRa and NB-IoT are not very well suited for Collo’s need for high amount of data transmission. For Collo, 4G-communication is currently the best choice, and in the future 5G will emerge. LoRa or NB-IoT could be used for transmission of very highly analysed Collo data.
Big data analyses, like calculations of correlations, are typical for process diagnostics. Collo analysis software gets big amount of raw data describing the interaction of the liquid with electromagnetic fields of different frequencies. Collo analysis calculates dozens of features out of this data. Typically, one or two features can be used to describe the process behavior the customer is interested in.
For example; a customer wants to know when the mixing of ingredients is completed, Collo analyzer shows a couple of features, and when they stabilize, mixing is accomplished. In addition to direct customer needs, features can be used for discovering new phenomena in the process. For example, when a feature shows a regular cyclic behavior with a certain cycle time, it can be correlated with other process variables with similar cycle times, and thus causes for process disturbances can be found and corrected.
AI and machine learning provide new opportunities for data analysis. The major reasons for increased interest in machine learning are the increase of processing capacity, the advances in methods and the availability of open source software, like TheAno, PyTorch and Caffe2. Collo uses advanced methods in the analysis.
For example, many process variables are dependent on temperature, and customers are interested in knowing the values of the variables in a standardized temperature like in 20 degrees Celsius, not just in the measurement temperature. Collo uses neural network algorithms to make this standardization.
Future opportunities with AI
In the future, we see more opportunities for AI and machine learning in Collo. We collect for different liquids the data describing the interaction of the liquid with electromagnetic fields of different frequencies. We call this data the digital fingerprint of the liquid. With continuously growing data base of fingerprints and machine learning algorithms correlating customer data with our data base, we can help our customers make revolutions in their processes!
Author: Raimo Korhonen / ColloidTek
Read also: Why measuring just viscosity is not enough