(469d) Integrating Graph Theory and Machine Learning to Design Reliable Wastewater Treatment Plants

Existing process synthesis and design approaches, such as superstructure optimization, are only as good as the user-defined technologies (process units), streams, and connections. The definition of these elements relies closely on heuristics, the experience of the design engineer, as well as reasonable initial guesses and bounds for the unknown variables [1], [2], [3]. It is expected that the final design resulting from these approaches will have adequate performance in terms of a predefined objective function. However, complex systems usually have multiple objectives and constraints that need to be satisfied while withstanding uncertainties. This creates additional difficulties in realistic problem formulation and existing solvers fail to converge to a global optimum. Thus, to demonstrate the utility of the integrated Graph Theory (GT) and Machine Learning (ML) approach, ‘Design of Cost-effective, Sustainable, and Reliable Wastewater Treatment plants’ is selected as the problem of interest. Wherein, cost-effective implies minimum capital, operating, and maintenance costs [4]; sustainable implies maintaining ecological balance to support the well-being of current and future generations [5]; and reliable implies the minimum failure probability [6].

To this end, we have developed the Two-Layer Process Synthesis (TLPS) algorithm, which contains solvers for linear or nonlinear equations and efficiently handles the combinatorial part of the design problem. Thus, it will allow for timely convergence and optimal solution of complex and realistic process synthesis problems generating reliable optimal paths or flowsheets. The risk of failure of the associated infrastructure’s elements is integrated into the TLPS algorithm through the application of machine learning classification algorithms on historical asset inventory, operating datasets, and weather forecasts to determine the failure probability. The ML-integrated TLPS algorithm is part of the enhanced process design framework, which will serve a two-fold purpose; (i) determination of structurally feasible, economically scalable, and sustainable paths for complex wastewater treatment (WWT) systems, and (ii) prediction of associated infrastructure reliability and resilience for proactive maintenance and management [7]. The TLPS resorts to the graph-theoretic approach known as the P-graph framework [8] for the identification of the n-best structures considering reliability and cost. Thus, a set of alternative designs is identified that provide insightful information about the system. The stakeholders will be able to redesign and retrofit existing facilities and plan maintenance schedules for safe and timely repairs with minimal investment. This will reduce operational downtime and lag between planning, adjustments, and implementation. Comparative assessments for WWT networks employing multiple modeling, optimization, and solution approaches will eliminate drawbacks and provide insights into non-intuitive solutions for enhancing process efficiency and maximum resilience.

References:

[1] H. Yeomans and I. E. Grossmann, “A systematic modeling framework of superstructure optimization in process synthesis,” Comput. Chem. Eng., vol. 23, no. 6, pp. 709–731, Jun. 1999, doi: 10.1016/S0098-1354(99)00003-4.

[2] J. D. Chea, A. L. Lehr, J. P. Stengel, M. J. Savelski, C. S. Slater, and K. M. Yenkie, “Evaluation of Solvent Recovery Options for Economic Feasibility through a Superstructure-Based Optimization Framework,” Ind. Eng. Chem. Res., vol. 59, no. 13, pp. 5931–5944, Apr. 2020, doi: 10.1021/acs.iecr.9b06725.

[3] W. Wu, K. M. Yenkie, and C. T. Maravelias, “A superstructure-based framework for bio-separation network synthesis,” Comput. Chem. Eng., vol. 96, pp. 1–17, Jan. 2017, doi: 10.1016/j.compchemeng.2016.10.007.

[4] K. M. Yenkie, “Integrating the three E’s in wastewater treatment: efficient design, economic viability, and environmental sustainability,” Curr. Opin. Chem. Eng., vol. 26, pp. 131–138, Dec. 2019, doi: 10.1016/j.coche.2019.09.002.

[5] U. Nations, “Sustainability,” United Nations. Accessed: Jul. 25, 2023. [Online]. Available: https://www.un.org/en/academic-impact/sustainability

[6] Z. Kovács, Á. Orosz, and F. Friedler, “Synthesis algorithms for the reliability analysis of processing systems,” Cent. Eur. J. Oper. Res., vol. 27, pp. 573–595, 2019.

[7] Roshani E. and Filion Y. R., “Event-Based Approach to Optimize the Timing of Water Main Rehabilitation with Asset Management Strategies,” J. Water Resour. Plan. Manag., vol. 140, no. 6, p. 04014004, Jun. 2014, doi: 10.1061/(ASCE)WR.1943-5452.0000392.

[8] F. Friedler, Á. Orosz, and J. Pimentel Losada, P-graphs for Process Systems Engineering. Cham, Switzerland: Springer International Publishing, 2022. doi: 10.1007/978-3-030-92216-0.