High-precision smart calibration system for temperature sensors

• Published in: Sensors and actuators. A. Physical. 2019-10, Vol.297, p.111561, Article 111561
• Main Authors: Wei, Rongshan, Ouyang, Kui, Bao, Xiaotian, Gao, Xiong, Chen, Chuandong
• Format: Article
• Language: English
• Subjects: Algorithms; Calibration; High precision; Internet of Things; Interpolation; Linearity; MCU; Neural networks; Particle swarm optimization; Performance tests; Polynomials; PSO-BP neural network; Real time; Sensor calibration system; Sensors; Smart sensors; Studies; Temperature; Temperature sensors
• ISSN: 0924-4247; 1873-3069
• DOI: 10.1016/j.sna.2019.111561

An MCU-based sensor calibration system is proposed, which mainly employs particle swarm optimization (PSO)-back propagation (BP) neural network. The system firstly reads sensor data through I2C bus and then uses the BP neural network and PSO algorithm to automatically calibrate these data in real time. Sigmoid activation function was implemented via a piecewise polynomial fitting. The proposed calibration system achieves high precision and low hardware resource consumption. [Display omitted] •An MCU-based sensor calibration system employs PSO-BP neural network.•The system reads sensor data through I2C bus and then calibrates these data in real time.•Sigmoid activation function was implemented via a piecewise polynomial fitting.•The system achieves high precision with low hardware resource consumption. High precision and smart sensors make up an indispensable data entry for the Internet of Things technology. Nonetheless, conventional calibration algorithms mainly implemented on the software, such as least squares, polynomial fitting, and interpolation, exhibit limited calibration accuracy that does not reflect a real-time measurement of the sensors. The problem can be resolved with an MCU-based sensor calibration system proposed herein, which mainly employs particle swarm optimization (PSO)-back propagation (BP) neural network. The system firstly reads sensor data through I2C bus and then uses the BP neural network and PSO algorithm to automatically calibrate these data in real time. Sigmoid activation function was implemented via a piecewise polynomial fitting to create a trade-off between hardware resource and precision. A performance test conducted on temperature sensors showed a maximum error of 0.16 °C within the measurement range of −40–100 °C with three times the standard deviation (3σ) error of ±0.23 °C and overall linearity of 0.1143% after the calibration system was added as compared to the significantly higher error of ±0.63 °C without the calibration.