Stochastic Evaluation of Disinfection Performance in Large-Scale, Open-Channel UV Photoreactors

The objective of the research described herein was to apply combinations of computational fluid dynamics and irradiance field models (aka, CFD-I models) to simulate the performance of full-scale ultraviolet (UV) disinfection systems, including variability. The central hypothesis of this research was that CFD-I models (which are essentially deterministic, by nature) can be applied via a non-deterministic (stochastic) approach to simulate process performance including variability, by allowing appropriate variations in input variables. Combined application of CFD simulations and ray tracing was employed to model the flow field (with the embedded particle tracking models) and fluence rate field, respectively, for the existing UV disinfection system at the Belmont WWTP in Indianapolis, IN, which comprises 7 channels with 384 high-output, low-pressure mercury amalgam UV lamps in each channel horizontally oriented, and two full-scale UV disinfection test channels, Aquaray® HiCAP, and LIT MLV-24A, which are designed for wastewater disinfection and water reuse applications, which use vertically oriented LP-HO mercury lamps. Ray tracing has been widely used for modeling optical (visible) lighting systems, which include visible light, LED, and solar light sources. Ray tracing simulates the spatial distribution of electromagnetic radiation from one or more sources by tracking a large number of rays that are assumed to emanate from the source(s) in a probabilistic manner. This approach allows rigorous simulation of optical behavior, including refraction, reflection, absorbance, and shadowing. By simulating the behavior of a large population of rays, it is possible to accurately describe the distribution of radiant energy within the system. In this study, fluence rate calculations were performed using commercial ray-tracing software (Photopia, LTI Optics, Westminster, CO). The practical application of ray-tracing was evaluated and validated through comparisons with experimental methods and other numerical models for small scale and large scale UV reactors. The ability of the ray tracing approach to accurately simulate fluence rate fields in UV photo-reactors was demonstrated, as were challenges and limitations related to ray tracing analysis implementation. For larger, more complex systems where physical measurements are more challenging, the implications of the simulation results are discussed. Monte Carlo simulations based on a CFD-I modeling approach were applied to predict variability in UV disinfection performance under the influence of the observed variability in the input parameters (including dose response behavior of E. coli and MS2, and initial bacterial count) for different reactors and operating conditions. The input parameters of the systems were measured using standardized experimental methods and on-site measuring devices. The results showed that variability in E. coli dose-response parameters (KA KB, c) and No plays an important role in the variability in predictions of effluent E. coli (N) and Reduction Equivalent Dose (RED). However, MS2 dose-response parameters caused insignificant variation in the predictions of RED. The stochastic CFD-I approach was used to simulate the behavior of the Belmont system for operating conditions that span the range of common use of this system. These simulations indicated that the likelihood of violating the regulatory permits for E. coli inactivation was essentially zero for most operating conditions. On the other hand, simulation of a hypothetical worst-case scenario at the Belmont UV system, based on peak flow rate, 60% lamp power and UV transmittance of 60%, and operating only one bank of lamps in each channel, resulted in a 16% probability that N will exceed permit limitations for E. coli . Sensitivity analysis between the remaining viable E. coli in the effluent stream (N) and the variable inputs, including UV dose-response parameters (DRP) and the initial concentration of E. coli (No), using Spearman’s rank test revealed the presence of associations between the variable output N and the variable inputs of the DRP and No; the degree and the ranks of this correlation appeared to be dependent on the operating conditions applied. Uncertainties in the CFD-I numerical simulations results and the number of simulated rays were also addressed in this work. Part of this study focused on developing interpolation and scaling methods that allows estimating unknown reactor UV dose distributions under various operating conditions with a reduced number of CFD-I simulations; this method was constructed based on the dependence relationships between UV doses and the operating parameters (i.e. flow rate, UVT, and lamp power). Lastly, a software tool was developed to facilitate programming of the stochastic Monte Carlo calculations for predicting the UV system performance (i.e., fractional inactivation, effluent microbial concentration, and RED) in accordance with the variability in dose-response parameters and No. The tool also includes a programmed algorithm which can be used to find optimum operating conditions for the UV disinfection process. This object-oriented program was constructed using MATLAB app design features. (Abstract shortened by ProQuest.