Machine Learning-Based Ambient Temperature Estimation Using Ultrasonic Sensor

Abstract

Temperature plays a vital role in determining the environmental conditions. Non-contact ultrasonic sensors use time of flight (ToF), which depends on the speed of sound in the measurement medium. The medium of propagation influences the speed of sound, and in the air medium, it is highly affected by temperature, humidity, and other gases present in the medium. Ambient temperature can be estimated using the speed of sound, time of flight, and the object’s distance from the ultrasonic sensor with proper compensation of humidity effect. The ultrasonic temperature measurement system determines the average temperature of the medium based on the changes in ultrasonic sound speed in the medium of travel. This paper proposes a non-contact ultrasonic sensor-based ambient temperature estimation system using two machine learning approaches: multiple linear regression (MLR) and support vector machine (SVM) regression. A low-cost 40 kHz ultrasonic transducer (HC-SR04) is used for the experiment to determine the temperature of the medium. The proposed ultrasonic sensor-based temperature estimation system is preferable in confined spaces such as room, boiler, tank, and other industrial applications where the temperature needs to be measured in a non-contact manner. To validate the proposed system’s accuracy, experiments are conducted in different environmental conditions with temperature ranging from 22 to 45∘ C and relative humidity ranging from 30 to 85%. Experimental results indicate that in the proposed measurement system, the temperature estimation error is bounded by ±0.4∘ C.