Signal Interpretation in Two-Phase Fluid Dynamics Through Machine Learning and Evolutionary Computing
Bogdan Filipič, Iztok Žun, Matjaz Perpar · Industrial and Engineering Applications of Artificial Intelligence and Expert Systems · 2022
The paper shows how techniques of machine learning and evolutionary computing can assist in making human expertise in sensor data interpretation explicit and suitable for computer execution. The study refers to a specific task in two-phase fluid dynamics, i. e. the interpretation of probe signals detected in gas-liquid flow. Given a raw probe signal, the corresponding twostate signal needs to be constructed which denotes the presence of the two phases. Due to the lack of knowledge about the processes on a micro scale, no exact procedure exists for accomplishing this task. However, operators are capable of interpreting visually presented probe signals through experience and intuition. To imitate their performance, a prototype signal interpretation procedure was designed manually, and its parameters tuned with genetic algorithms. in an alternative approach, skill acquisition was performed automatically, using inductive machine learning. The induced signal interpretation procedures were tested successfully on air-water pipe flow under laboratory conditions.