Using the DAKOTA Sensitivity Analysis and Optimization tool [
40], various analyses were carried out to understand system behavior considering both technical and economic constraints. DAKOTA is a comprehensive and freely available software developed by the Sandia National Laboratories, which implements a set of methods for optimization, parameter fitting, and sensitivity and uncertainty analysis. The case studies presented here focus on high salinity draw streams because PRO processes that use stream pairs with small salinity differences, such as seawater and freshwater, have little chance of being economically feasible.
3.1. Energetic and Economic Optima Occur at Different System Conditions
To arrive at this result—that energetic and economic optimum occurs at different system conditions—a sensitivity analysis was carried out using membrane sizes ranging from 20,000 to 200,000 m
2 and hydraulic pressure ranging from 10 to 100 bar. Flowrates of the draw and feed were scaled up in tandem with membrane sizes for each analysis. Membrane properties, stream properties, and simulation limits used in the setup of this study are shown in
Table 1.
The results reported in
Figure 4 are for 200,000 m
2 membrane area at equal volumetric flowrates of the draw and feed such that (
where:
For this membrane area, the data shown in
Figure 4 are for net power produced (
and the LCOE at various values of applied pressure on the draw solution side. The net power of the system can be defined as:
From
Figure 4, it can be observed that the maximum net power produced and the minimum LCOE occur at operating pressures that are significantly different from each other. The effect of the trade-off between the cost of generation and power production is clearly seen. At 200,000 m
2, the pressure that minimizes LCOE is approximately 25 bar, while the maximum power production is at 79 bar. While larger areas correspond to higher power generation, the combined effect of the increased cost of pumping required to deliver the draw solution through the system and the higher cost of the membrane due to its large area makes operating the system in maximum power production mode undesirable.
3.2. Optimal Economical Operating Pressure Range Decreases as Plant Size Increases
The effect of system size in terms of the feed flowrates was evaluated while fixing the draw flowrate and other process parameters.
Table 2 presents the data used in performing the simulation for this section. Membrane properties and stream properties remain the same as in
Table 1. The flowrate for the draw is fixed while that of the feed varies, and thus
varies accordingly in the range from 0.25 to 0.62.
Figure 5 shows the LCOE versus applied pressure for a membrane area of 20,000 m
2 and a draw solution flowrate of 0.25 m
3/s, while
Figure 6 presents the results for 200,000 m
2 membrane areas and a draw solution flowrate of 2.52 m
3/s. It can be observed from comparing the behavior of LCOE between the two different plant sizes that as the system size increases, the range of pressures for which LCOE can be minimized narrows. At 20,000 m
2 and for different values of
, the plot of LCOE (USD/kWh) against pressure (bar) in
Figure 5 shows that, between 43.75 and 66.25 bar, the LCOE reaches a minimum. The percentage change in the value of LCOE between these pressures ranges from 1.77% at
= 0.25 to 4.26% at
= 0.63. The change in LCOE within this pressure range is less than 5%, which indicates that the system can operate within this pressure range with little detrimental effect to plant techno-economics. For comparison, the two renewable energy technologies with the least value of LCOE (Hydro and Onshore Wind) differ from each other by about 11.3% [
41]. At 200,00 m
2 membrane area and 2.52 m
3/s draw solution flowrate, the range of pressure that minimizes the LCOE is between 21.25 and 33 bar, as shown in
Figure 6, providing a narrow pressure operation band of 11.75. This is because the combined effect of higher pumping costs and higher total membrane costs narrows the range of operating conditions for which LCOE might be minimized.
3.3. Effect of Membrane Intrinsic Properties on LCOE
Membrane intrinsic properties affect how much power can be produced via PRO for a specific configuration and this, in turn, affects the value of LCOE. An ideal membrane would be one that maximizes flux by increasing the value of water permeability and minimizes salt permeability [
42]. Although membrane properties such as water permeability (
A), salt permeability (
B), and structural parameter (
S) are usually assumed to be independent of pressure, Madsen et al. [
43] published findings showing that
B and
S both increase with pressure for HTI-CTA membranes. However, water permeability (
A) was reported to be independent of pressure. The results obtained by Madsen et al. for the effects of pressure on the salt permeability and structural parameter were fitted by the following linear relationship and used in the simulator so that these parameters are expressed as functions of applied pressure.
where
y represents
B or
S at a given pressure
P,
a is the initial value of salt permeability (
B’) or structural parameter (
S’) at an initial pressure
Po, and
b is the rate of increase of
B or
S versus pressure.
Table 3 presents the parameters and process conditions used in the problem setup to evaluate the effect of water permeability, salt permeability, and structural parameter on LCOE. The effect of water permeability on LCOE was studied by varying its value in the range of 0.2–10 L·m
−2·h
−1·bar
−1 while keeping other parameters at the values reported in
Table 3. The simulation results are presented in
Figure 7, and these results show a negligible decrease in LCOE beyond certain values of water permeability. For example, increasing the “
A” value from 0.2 to 0.6 L·m
−2·h
−1·bar
−1 at an operating pressure of 30 bar, results in a decrease in the LCOE by 25.35%. However, increasing “
A” from 2 to 5 L·m
−2·h
−1·bar
−1 and from 5 to 10 L·m
−2·h
−1·bar
−1 at the same pressure, only caused LCOE decline of 4.28% and 1.42%, respectively. This indicates that although a high value of water permeability is desirable for increasing water transport through the membranes, further increase beyond a certain value has little impact on the economics of the process.
Figure 8 shows the effect of salt permeability on LCOE by varying the initial value of
B at pressure = 10 bar,
B’, from 0.0 to 0.75 L·m
−2·h
−1, with the rate of increase of “
B” versus pressure and other parameters remained the same as reported in
Table 3. These results presented in
Figure 8 indicate that at any specific pressure, as the salt permeability increases, the LCOE also increases. This is not unexpected because higher values of salt permeability indicate that the membrane is more susceptible to reverse salt flux, which results in lowering the salinity gradient across the membrane and, thus, lowering the power produced for the same economic costs. At any specific value of salt permeability, the LCOE reaches a minimum but increases beyond a certain pressure. This is because, beyond optimal pressure where maximum power is produced, a further increase in pressure actually leads to a decrease in power production. This combined effect of increased pumping cost (due to increased pressure) but decreased power production causes a rise in the value of LCOE. Also, the LCOE increased at a faster rate with increasing salt permeability as the pressure deviated from the optimum value. For example, comparing the curves obtained at 0 and 0.75 L·m
−2·h
−1, when the pressure is increased from 40 bar to 60 bar, the LCOE increased by 16.15% for the membrane with 0 L·m
−2·h
−1 and 35.62% for that with 0.75 L·m
−2·h
−1, respectively.
The contribution of the structural parameter “
S” to costs was estimated by varying the initial value of “
S” at 10 bar pressure (
S’) in the range of 5.00 × 10
−5–1.00 × 10
−3, while fixing the rate of increase of
S with pressure and other parameters at the same values shown in
Table 3. The results obtained are shown in
Figure 9 and indicate that, at any specific value of pressure, the LCOE increases as the value of
S increases. Higher values of LCOE indicate less efficiency in power generation due to lowering water transport as a result of increasing
S caused by concentration polarization. For a specific value of a structural parameter, the LCOE reaches a minimum and then increases beyond a certain pressure. This trend is generally observed for Pressure–LCOE relationships, as increased pumping costs and decreased power production beyond optimum pressure result in the trend seen.
3.4. Optimization Studies
Optimization studies were performed to better understand plant conditions and dimensions that minimize LCOE, as well as compare the LCOE values obtained from PRO with other renewable energy technologies. First, we compare the results of optimization from the use of an ideal membrane and a real membrane with 100% efficiency of the mechanical components. Then, the assumption of perfectly efficient mechanical components is relaxed, and the LCOE is calculated using the results from the optimization study with real membranes. All three results are compared with other renewable energy technologies using global average values from the IRENA database [
41]. The optimization was carried out using the DAKOTA optimization tool [
40] with the LCOE as the objective function to be minimized. The flowrates of the draw and feed, applied pressure, and membrane area were the parameters varied. Sensitivity analysis (not shown) was used prior to optimization studies to obtain initial estimates for these parameters.
Table 4 shows the membrane properties for both ideal and real membranes, as well as process and stream conditions used in the problem set up.
The results obtained from the optimization study are shown in
Table 5. These values represent the optimum values of operating parameters that attained the lowest achievable LCOE for a single-stage PRO using the specified process conditions and stream salinities. The levelized cost of USD 0.0704/kWh was achieved for an ideal membrane, which is better than geothermal energy, offshore wind, and CSP, using the 2019 global average value reported by IRENA [
41]. Nonidealities in the membrane lead to a 92.7% increase in the required membrane area as well as a 78.5% increase in the value of LCOE. At this value of LCOE, PRO technology is only better than CSP [
41].
The International Renewable Energy Agency (IRENA) released the Renewable Energy power generation costs for the year 2019 [
41]. The reported results in
Figure 10 are used as a basis for comparing different renewable energy sources. The values reported are global averages.
Figure 10 shows that the LCOE of PRO with real membrane performance and considering equipment efficiencies is higher than that from other sources. However, it is comparable to both solar photovoltaics and concentrated solar power, considering their costs at the start of their mainstream production in 2010, as shown in
Figure 11. Solar PV found uses in satellite communication before becoming more mainstream for electricity generation for domestic use. This initial capital-intensive use enabled improvement in the technology, and external encouragement led to large-scale deployment that enabled some cost savings [
44]. In this same light, research has been conducted on the use of PRO for water flooding applications in oil reservoir pressure maintenance [
45]. Going further, we explore PRO multistage design configurations to determine if the extra energy production from multistage design justifies its costs.
3.5. Multistage Analysis
PRO and RO systems have many similarities, and it has been shown that multistage operation improves the energetic performance of a reverse osmosis process [
46,
47,
48]. Wei et al. [
46] showed that for a reverse osmosis process, the energetic benefits obtained from staged designs are confounded by increased costs of membrane area—although, with careful design, the energetic benefits could outweigh the costs. Some interesting multistage designs for PRO have been reported in the literature. Chung et al. [
49] classified all two-module multistage designs into two broad families—interstage pressure control and input exergy control. Interstage pressure control refers to designs with either a pump (for interstage pressurization) or a turbine (for interstage depressurization) between membrane modules. Pressure control is important in reducing entropy generation because it helps to achieve a more uniform osmotic driving force [
50,
51].
Input exergy control refers to a design philosophy that allows for modification to the draw and feed streams as they enter the process. This control method seeks to maintain the salinity gradient across the membrane in both stages. Altaee et al. [
18] and Li [
22] studied dual-stage systems PRO that can be classified as interstage pressure control. While Altaee et al. had a design with interstage pressurization, Li had interstage depressurization. He et al. [
52] introduced four designs for two-stage PRO systems, and each design is named depending on whether the draw or feed solution is split upon entering the process. Their design can be classified as input exergy control. Of the four designs, two are studied in this work; the first of them is the continuous draw differentiated feed (CDDF) design, as shown in
Figure 12, which has the draw stream moving continuously from one stage to the next while fresh un-concentrated feed is supplied to each stage. The second design is the differentiated draw continuous feed (DDCF), as shown in
Figure 13. In this configuration, fresh draw enters each stage while the feed flows unhindered from one stage to the next.
The optimization studies were first performed to minimize the LCOE, and the results from the single-stage optimization study using real membranes are set as the boundaries for the multistage system. A total membrane area of 137,255 m
2, applied pressure of 48.48 bar, and draw flowrate of 0.2853 m
3 were used for the dual-stage system. Optimization was conducted to determine the right distribution of membrane areas between stages, given that the total area of the membrane is constant. In addition, we found the optimal allocation of flowrates to each stage. The distribution of flowrates (either draw or feed, depending on the design configuration) between membrane stages is referred to as ‘split ratio’, and it sums up to 1. The volumetric flowrate of the feed was not limited to the value obtained from a single stage, but instead, it was allowed to vary within a range shown in
Table 6. This was done under the assumption that seawater is plentiful and would not be a limit to our design.
Table 6 shows the membrane properties, stream conditions, and optimization limits used in the set up of this run.
The results of the two-stage simulations are presented in
Table 7, and they show that carrying out multistage design with the aim of minimizing costs is detrimental even to the amount of energy that can be produced from the system. In both dual-stage designs, the total amount of power produced is less than that produced in a single stage. For the CDDF, the cost of production is the same as in the single stage, but lower energy production leads to higher LCOE. For the DDCF, the cost of energy produced is higher due to the presence of an added turbine. However, the energy produced is lower than that with the single-stage design; hence the LCOE is higher.
Another interesting observation about system design is the allocation of membrane area and flowrate of either feed or draw (depending on if the design is split feed or split draw) to each stage. In CDDF, 26.4% of the total membrane area is allocated to stage 1, while 31.0% of the total membrane area is allocated to membrane 1 in DDCF. These percentages are calculated by dividing the total membrane allocated to stage 1 by the total membrane area used for the optimization. This points to the fact that large-scale design of such systems would not have an equal distribution of membrane areas between stages to be effective. Similarly, 27.2% of the total feed into the system is fed into stage 1 in CDDF, while 29.7% of the total draw is fed into stage 1 in the DDCF configuration.
Studies were also carried out to maximize the net power produced by the plant. Doing this relaxes the constraints of economics and enables comparison between single-stage and multistage designs on the basis of energy available. This was first performed for a single-stage system using real membrane properties, as reported in
Table 4. Instead of the objective of minimizing LCOE as was earlier done, herein we set the objective to maximize net power produced. Comparing the results in
Table 7 for the single stage and the results presented in
Table 8, we see the trade-off one makes in designing a single-stage system to minimize costs as against maximizing power production. Net power produced was doubled in designing the system to optimize for power production against costs. However, this leads to a 34% increase in the levelized costs of electricity.
Similar analysis can be conducted for the multistage configurations. The results from the single-stage study are set as the boundaries for the multistage system. A total membrane area of 233,672 m
2, applied pressure of 69.47 bar, and draw flowrate of 0.5901 m
3 were used for the dual-stage system.
Table 9 presents information used in setting up the study to maximize the net power of multistage systems. Other membrane-specific and operating parameters were the same, as reported in
Table 6.
The results of this optimization are shown in
Table 10. Similar trends for area and volume distribution between stages are also observed here. Here, 33.3% of the total membrane area is allocated to stage 1 of the CDDF design, while 23.2% is allocated in the DDCF design. Additionally, 37.2% of the feed and 21.3% of the draw is allocated to stage 1 in CDDF and DDCF, respectively. In designing to maximize net power produced, the multistage designs produce more net power than the single stage; however, the levelized costs are also higher. CDDF design produces 1.7% more power than single stage and has a 5% higher levelized cost. The DDCF design produces 1.9% more net power than the single stage but has a 7.2% higher levelized cost. The simultaneous optimization of six variables (pressure, total membrane area, draw flowrate, area allocated to stage 1, split ratio, and feed flowrate) might lead to larger net powers, but this study was beyond the scope of this work.