Artificial Neural Network control of thermoelectrically-cooled microfluidics using computer vision based on IR thermography. (2nd February 2019)

• Record Type: Journal Article
• Authors: Rizkin, Benjamin A.; Popovich, Karina; Hartman, Ryan L.
• Is Part Of: Computers & chemical engineering. Volume 121(2019)
• Page Start: 584
• Page End: 593
• Publication Date: 2019-02-02
• Subjects: Thermal control -- Machine learning -- Neural networks -- Process control -- Autonomous microfluidics -- Process resiliency; ANN Artificial Neural network -- CAD Computer Aided Design -- CAM Computer Aided Modeling -- CNC Computer Numerical Control -- CGB Conjugate Gradient Backpropagation -- CGBPFRU Conjugate gradient backpropagation with Fletcher–Reeves updates -- CGBPPBR Conjugate gradient backpropagation with Powell–Beale restarts -- CGBPPRU Conjugate gradient backpropagation with Polak–Ribiére updates -- DLL Direct Link Library, compiled C/C++ code -- FPS Frames per Second -- IoT Internet of Things -- LM Levenberg–Marquardt (Damped Least Squares) with forward training -- NARMA Nonlinear Autoregressive Moving Average -- OSSBP One-step Secant backpropagation -- PID Proportional Integral Derivative control -- PWM Pulse Width Modulation -- q-NBP quasi-Newton backpropagation -- RBP Scaled Conjugate Gradient Backpropagation -- UART Universal Asynchronous Receiver–Transmitter -- UFPA Uncooled Focal Plane Array -- USB Universal Serial Bus -- VOx Vanadium Oxide thin film microbolometer; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2018.11.016
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: Artificial Neural Networks can be self-trained for pulsed-width modulation control. IR thermography delivered high-speed and high-accuracy data. Computer vision was used to segment an IR image into regions of interest. The system demonstrated resiliency to set point changes and disturbances. Improved temperature control and calculation speed was observed compared to PID. Abstract: High-speed and high-accuracy thermal control of reactors has always been of interest to chemical engineers. In this paper we present a new methodology for thermal control of a continuous-flow chemical reactor using non-contact IR thermography combined with computer vision and a predictive Artificial Neural Network. The system exhibits several key advantages over thermocouples and PID control including the ability to quantify and account for thermal diffusion in the system, to collect and process data very quickly and with high accuracy, to analyze the entire surface of the reactor, and to update its training based not only on the current thermal response, but also on external factors. We have constructed and validated such a system as well as shown improvements in its accuracy, rise time, settling time, set point tracking, and overshoot as compared to more traditional forms of thermal control, validating this as a possible approach for experimental and process control. Graphical abstract: Image, graphical abstract