Physics-Led Artificial Neural Network Prototyping in Solar Disinfectors

Introduction: Concentrated solar thermal energy in its best-known form is represented by compound parabolic concentrators (CPCs), the most exergetic and easy to implement of energy technologies, and able to replace water disinfection powered by fuels and electricity in emergency and isolated contexts. CPCs are undergoing incipient research combining physics and machine learning (artificial neural networks, ANNs). Methods: A diurnal CPC, and a novel compound parabolic condenser (CPCo, intended to be powered at night by photovoltaics), alongside their heating and cooling behavior, were simulated by ANNs. The simulation clarified features of desirable prototypes, without the expense of testing a prototype. Results: ANN models yielded predictions with an accuracy of 95% during the day and 88% at night (equivalent to 0.23 oC mean absolute error). Cycles involved in the disinfection process (e.g., solar intermittency and batch operation) were integrated in the simulation. Discussion: The CPCo and CPC accommodated a transparent absorber and slow cooling, respectively. Materials and geometry enabled some of the lowest levelized costs of any energy source. ANN diurnal-nocturnal simulation pointed to a hybrid solution to intermittency. Further modular prototyping should address environmental operation conditions, failures (decompression, insulation), and automation. Conclusion: The application studied here has bearing on most steam and heat for industrial processes (SHIP) that can use solar energy, thermal principles, and thermodynamic fundamentals (owing to, among others, Newton, Fourier, Boltzmann, and Stefan). As ANN accelerates prototyping under physical constraints, ANN simulations ought to be incorporated early in the design of SHIP systems.