(61w) Accelerating Process Design and Optimization with Novel Computational Tools

Since the global energy requirements are expected to increase and most of the world-wide energy demand is still supplied by fossil fuels, a search for a suitable transition fuel is underway. Amongst the conventional energy sources, natural gas (NG) is generally considered to be the cleanest and highest growing fossil fuel. Its liquefied form, LNG, is suitable for long-distance transport, and so it is one of the pillars of energy supply in many countries. However, NG liquefaction is an energy-intensive process, accounting for 30–40% of LNG cost. This imposes a pressure on cutting the NG liquefaction costs down. Moreover, because more and more attention is paid to sustainable development, environmental and safety considerations cannot be ignored.

Most of the published studies on simulation-based LNG process optimization focus solely on the main liquefaction process, where they usually aim at minimization of specific or total energy consumption (SEC/TEC), or exergy efficiency. Various feedstock quality, environmental impact, or process safety are usually not considered. While genetic algorithms (GA/NSGA-II) are among the prevalent optimization methods, the studies often use a very limited number of individuals and/or generations during the optimization as both simulation-based optimization using GA/NSGA-II and process safety evaluation are excessively time-consuming.

To tackle these problems, two novel computational tools have been developed – Parallel Genetic Algorithm Interface II (PAGAN-II) to greatly improve the optimization speed, and Hazard Detection Software II (HADES-II) to drastically decrease the necessary time for inherent safety assessment.

The problem of time-intensive optimization lies in great number of repetitive simulations required in simulation-based optimization. Furthermore, this repetition overwhelms the cache memory, which also gradually increases the optimization time. As a response, PAGAN-II, a Matlab-based algorithm compiled into a user-friendly graphical user interface (GUI), has been developed. As opposed to the traditional approach, PAGAN-II fully exploits the fact that Aspen Plus simulations can run asynchronously. To accelerate the computation, PAGAN-II creates a user-defined number of copies of the desired simulation. During the optimization, the entire population is sent to the optimization engine at once, the optimization engine assigns each simulation engine an individual, and the simulation engines start evaluating these individuals simultaneously. Once a simulation engine becomes idle, another individual is assigned to it. Thus, optimization time is greatly decreased. Additionally, PAGAN-II periodically reinitializes the simulation engines to diminish the effect of gradual optimization slow down.

The problem with safety evaluation lies in repetitive set of computations for each unit operation required for safety indices in inherent safety assessment. Several authors tried to create automated software solutions, however, a great number of input values needed to be entered manually. As full potential of the programming interface is far from exploited, we have created a fully autonomous Hazard Detection Software II (HADES-II). This Matlab-based computational tool gains full control over Aspen Plus simulations and automatically loads every crucial information directly from the simulation without the need to enter anything manually. Furthermore, HADES-II encompasses a self-learning internal component database which currently contains information on approx. 300 most used hazardous chemicals. Because different safety indices put different weights on different safety aspects and there is no consensus on which indices should be used, HADES-II calculates a range of indices out of which the user can choose: Dow’s Fire and Explosion (Dow FEI) and Chemical Exposure (Dow CEI) Indices, Comprehensive Inherent Safety Index (CISI), Process Route Index (PRI), Process Stream Index (PSI), and Hazard Identification and Ranking system (HIRA). Furthermore, HADES-II also calculates overall process safety indices based on safety polygons, where applicable. Finally, HADES-II can operate as individual GUI, and thus serve as a simple user-friendly safety assessment tool, or it can be called directly from any Matlab function, and thus be incorporated into single- or multi-objective optimization studies.

To demonstrate the capabilities of the proposed tools and in the spirit of the abovementioned, an extensive optimization and feasibility study of 5 MTPA propane-precooled mixed refrigerant (C3MR) LNG liquefaction process was done. In the study, 12 different types of NG were taken from published literature as feedstock and sorted into three categories:

• low-nitrogen feedstock (<1 mol. % nitrogen),
• mid-nitrogen feedstock (1–4 mol. % nitrogen),
• high-nitrogen feedstock (>4 mol. % nitrogen).

For each type of feedstock, the C3MR process was optimized to obtain minimal SEC while 19 process parameters were optimized. These parameters included: exhaust pressures of compressors and throttle valves, mixed-refrigerant and NG temperatures in the main cryogenic heat exchanger, and composition of mixed refrigerant. Optimization was carried out using PAGAN-II with 100 individuals over 100 generations and 12 parallel simulations.

Subsequently, a feasibility study of flash gas post-processing alternatives was carried out. In the base case (present in virtually all C3MR studies), LNG is reduced to storage pressure in a flash vessel, whereas the flash gas is not utilized further and is supposedly subjected to flaring. We studied five additional post-processing alternatives:

1. Utilization of the cooling potential of the flash gas in NG liquefaction and subsequent flaring.
2. Alternative 1 + recirculation of the flash gas (no flaring).
3. Alternative 1 + gas turbines as prime movers and flash gas as gas-turbine fuel.
4. Replacing the flash vessel with an integrated nitrogen rejection unit (NRU) venting stripped nitrogen to atmosphere.
5. Alternative 4 + gas turbines as prime movers and NRU side product as gas-turbine fuel.

The study was carried out with the following constraints: < 1 mol. % nitrogen in LNG, > 99.99 mol. % purity of vented nitrogen, < 15 mol. % nitrogen content in gas-turbine fuel. Each alternative was assessed from economic, environmental, and safety point of view. For economic evaluation, levelized main product unit cost was used. For environmental evaluation, direct and indirect carbon dioxide emissions were calculated. Safety assessment was done using Dow FEI and Dow CEI indices.

Optimization of the 12 feedstock alternatives for C3MR process resulted in a 14% average decrease in SEC. To demonstrate the performance of PAGAN-II, a performance test was carried out using 1 (standard approach), 2, 3, 4, 6, 8, and 12 parallel simulations. The performance was tested using a desktop computer with an AMD Ryzen 9 3900X 3.80 GHz 12-core processor and 48 GB RAM. For reproducibility, every test run was initiated with identical initial population. For time reasons, test runs were done with 50 individuals and 20 generations. Results of the performance test are shown in Figs. 1 and 2. Based on the performance test results, the optimization time was estimated as 468 hours (19.5 days) using the standard approach. However, the optimization of 12 alternatives using PAGAN-II lasted approx. 32 hours (1.3 days) which translates into approx. 1460% increase in relative computation rate. Furthermore, standard optimization of 12 process alternatives using a combination of 200 individuals over 200 generations (commonly used in studies) was estimated to take over 125 days, whereas the estimated time using PAGAN-II was less than 6 days (i.e., a factor of 21).

Results of the feasibility study are displayed in Fig. 3, where the position of points relative to the vertices of the hexagon of alternatives represents the optimality of the respective alternative for the respective feedstock. From the economics’ point of view, all low- and mid-nitrogen feedstocks favor Alternative 3, and all high-nitrogen feedstocks favor Alternative 5 – both alternatives employing gas turbines. However, from the environmental point of view, all feedstocks prefer alternatives which minimize on-site emissions, i.e., Alternatives 2 and 4. Unsurprisingly, safety assessment declared the base case as optimal for every feedstock, because it comprises the lowest number of equipment and the lowest amount of recirculated material. To choose the best alternative for each type of feedstock, each aspect was subjected to multi-criteria analysis using fuzzy non-dimensionalized distance from the respective optima. To sum up, Alternative 2 is suitable for low-nitrogen feedstocks with nitrogen content < 0.5 mol. %, Alternative 3 is suitable for low-nitrogen feedstocks with nitrogen content of 0.5–1.0 mol.% and for mid-nitrogen feedstocks, while Alternative 5 seems to be suitable for high-nitrogen feedstocks.

In such feasibility studies, safety assessment usually creates a bottleneck as manual or even somehow automated safety assessment for each alternative can take hours. However, safety assessment utilizing HADES-II took about 20 seconds per alternative and per feedstock, which resulted in overall time of ca. 16 minutes for all feedstocks and applicable process alternatives.

To summarize, developed computational tools express vast potential for use in chemical engineering practice which has been proven through extensive optimization and feasibility studies.