MSF MED
1.4 Possible scope for further research
Introduction and Literature Review 1.3.4 Summary
Extensive research has been carried out for the NLP optimization of MSF, RO and hybrid MSF-RO processes. Rigorous NLP formulations have been addressed in the literature for the optimization of MSF, RO and hybrid MSF-BR processes using deterministic methods (GRG and SQP methods). Few studies have targeted MINLP formulations and few authors have targeted the application of non-deterministic optimization algorithms (GA) for the optimal design of MSF, RO and hybrid MSF-RO processes. Further analysis of the literature approaches will be addressed in the next section i.e., possible scope for further research.
Introduction and Literature Review
has not been investigated by non-deterministic methods such as GA and compared with the results obtained for MSF-BR process configuration.
b) Only few studies targeted the comparative assessment of alternate RO processes. In one study, Zhu et al. (1997) confirmed that TRO-RSR process is the best among SRO and TRO-RSR processes. Maskan et al. (2000) confirmed that among TRO-RSR, TRO-RP, TRO-SRB-RR, TRO-SFRB, TRO-SFRR and SRO processes, maximum profit was obtained for TRO-RSR and TRO-SFRR processes. Marcovecchio et al.
(2005) opined that among SRO and TRO-SRB-RR processes, TRO-SRB-RR process provided lowest optimal cost. Similarly, Sarif et al. (2008) opined that TRO-SRB-RR process is optimal among alternate RO processes. Further, it is important to note that all literatures mentioned in this paragraph have adopted deterministic methods.
c) For RO process networks, optimization studies using non-deterministic methods (GA) were addressed only for SRO process by Guria et al. (2005) and Djebejian et al.
(2008). In this regard, it shall be noted that SRO process complexity is insignificant in comparison with the complexity of other RO processes with retentate or permeate reprocessing options. Thus, other complex RO process configurations involving either retentate or permeate reprocessing options were not investigated in the literature using non-deterministic methods.
d) Using deterministic optimization methods, very few studies have investigated the optimality of hybrid MSF-RO processes. Two studies (Helal et al. (2003) and Francois et al. (2006)) confirmed the optimality of hybrid 2 MSF-RO process configuration among alternative MSF-RO hybrid processes. The optimal design of hybrid MSF-RO processes has been investigated using Excel solver (GRG) and GAMS CONOPT (SQP) platforms.
Chapter 1 e) There has been only one study conducted by Abdul Rahim et al. (2010) for the optimization of hybrid 2 configuration using GA as non-deterministic method. In summary, a comparative assessment of alternate hybrid MSF-RO processes have not been addressed till date in the literature. In other words, alternate hybrid MSF-RO processes have not been studied using non-deterministic methods in the literature.
Even using deterministic optimization methods, studies were conducted for the optimality of alternate hybrid MSF-RO processes using very few hybrid configurations.
f) With respect to the ranking of alternate MSF, RO and hybrid MSF-RO processes based on minimal freshwater production cost, limited insights can be obtained from only two literatures. While Marcovecchio et al. (2005) inferred that TRO-SRB-RR is better than SRO processes, it did not infer upon the optimality of MSF and hybrid MSF-RO processes in conjunction with the RO processes. However, the later work of Marcovecchio et al. (2011) indicated that among MSF, RO and hybrid MSF-RO processes, the TRO-RSR process is the best. Further, these two literatures used deterministic optimization methods to rank the best process among available alternatives. Thus, limited insights can be gained from available literature with respect to the best configurations among MSF, RO and hybrid MSF-RO processes.
Considering the above critical summary of the literature data, the following aspects can be identified for possible scope for further research in this work:
a) Optimal design of MSF-BR, MSF-OT and MSF-M process configurations using various non-deterministic methods and their comparative assessment with respect to
Introduction and Literature Review
b) Optimal design of alternate RO process configurations with and without retentate and permeate recycle options using various non-deterministic methods and their comparative assessment with respect to solutions obtained from deterministic methods.
c) Identification of alternate hybrid MSF-RO processes based on the identified best processes among MSF and RO processes. The alternate hybrid processes are anticipated to have best performance among all possible hybrid MSF-RO process configurations.
d) Optimal design of identified alternate hybrid MSF-RO process configurations using various non-deterministic methods and their comparative assessment with respect to solutions obtained from deterministic methods.
e) Optimal water production cost based rank of best configurations among MSF, RO and hybrid MSF-RO processes using non-deterministic and deterministic optimization methods and their combinations.
Among various non-deterministic methods, differential evolution (DE), genetic algorithm (GA) and simulated annealing (SA) can be regarded to be prominent methods. Similarly, among deterministic methods, sequential quadratic programming (SQP) using conventional and multi-start approaches could be regarded to be very effective to achieve good quality optimal solutions. In this regard, it is important to note that the GRG method (Excel solver) is not effective to handle systems with large number of inequality constraints and hence could be in effective to solve hybrid MSF-RO processes, while it might be effective to solve MSF and RO process optimization sub-problems separately.
Further, it is important to note that while MINLP superstructure based process models could
Chapter 1 (Outer Approximation method), it is to be noted that these techniques adopt deterministic optimization methods to solve NLP sub-problem. Hence, these approaches do not rule out the possibility obtaining local solutions for superior NLP sub-problems (for which global solution exists and whose global solution is the best among alternate process configurations) and global solutions for inferior NLP sub-problems. In other words, MINLP formulation using deterministic approaches could not resolve the issue of identifying the best process configuration and its solution close to the global optimal domain. Therefore, it is apparent that MINLP process models with superstructure based approach needs to address non- deterministic methods (such as GA, DE, PSO etc.). However, the computational effort for the same would be significantly high. Instead, it is proposed in this work that initially NLP formulations shall be targeted using non-deterministic methods and using the solutions obtained for the NLP formulations, the MINLP formulation could be addressed in the near future using non-deterministic methods. In the Ph.D. thesis, only NLP optimization has been addressed so as to set benchmarks for furthering research in MINLP formulations for global optimization in the near future.
Among various deterministic and non-deterministic methods for NLP optimization studies, DE, SA, GA, SQP, MS-SQP have been selected in this work. These methods have been selected based on the following available options:
a) Differential evolution (DE) is an effective and efficient non-deterministic optimization method. DE algorithm can be coded in MATLAB programming environment.
b) MATLAB optimization toolbox has a wide variety of both deterministic and non- deterministic methods (GA, SA, SQP and multi-start SQP).
c) MATLAB inbuilt optimization toolbox solvers need to be investigated for their
Introduction and Literature Review
With this insights with respect to the MSF, RO and hybrid process and optimization techniques, the objectives of the thesis are being outlined next