Ultrasonic Sensor Modeling with Support Vector Regression

doi:10.3390/s25030678

2025-01-23

Dang, Duy Ngoc

Tri Minh

Rui Araújo

Khang Hoang Vinh Nguyen *

C. Le

DOI：10.3390/s25030678

原文链接

查看原文

摘要

En 中文

This study proposes a novel approach for predicting the output behaviors of the Pepperl+Fuchs 3RG6232-3JS00-PF ultrasonic sensor. The sensor, integrated into the Festo MPS-PA Didactic System, serves to monitor the water level in a tank, facilitating water extraction to bottles delivered via a conveyor belt. This modeling approach represents the initial phase in the creation of a digital twin of the physical sensor, providing the capability for users to observe the sensor's response and forecast its life cycle for maintenance objectives. This study utilizes the Festo MPS-PA Compact Didactic System and support vector regression (SVR) for data acquisition (DAQ), preprocessing, and model training with hyperparameter optimization. The objective of this modeling approach is to establish a digital framework for transition towards Industry 4.0. It holds the potential for creating a digital counterpart of the entire MPS-PA System when combining the proposed sensor modeling technique with computer-assisted design (CAD) software such as Siemens NX in the future. This would enable users to oversee the entire process in a three-dimensional visualization engine, such as Tecnomatix Plant Simulation. This research significantly contributes to the comprehension and application of digital twins in the realm of mechatronics and sensor systems technology. It also underscores the importance of digital twins in enhancing the efficiency and predictability of sensor systems. The method used in this paper involves predicting the rate of change (RoC) of the water level and then integrating this rate to estimate the actual water level, providing a robust approach for sensor data modeling and digital twin creation. The result shows a promising 6.99% error percentage.

Keyword:

support vector regression

digital twin

virtual sensor

kernel selection

hyperparameter optimization