1. Introduction
Global economic and political strategies focus on carbon neutrality objectives. Under the Paris Agreement, Canada committed to reducing its greenhouse gas (GHG) emissions by 30% below 2005 levels by 2030 and a driving target of net-zero emissions by 2050 [1]. At the provincial level, Quebec has committed to reducing its GHG emissions by 37.5% compared to 1990 levels by 2030, to decreasing by 40% its consumption of petroleum-based products, and to meeting carbon neutrality by 2050 [2]. In this context, it is necessary to gradually transition to cleaner energy sources in which LPG plays a defining role as a transition fuel toward the complete implementation of renewable energies [3]. Hence, despite the increasing demand for electric and fuel-cell electric vehicles, replacing the existing internal combustion engine fleet demands significant investments for all the required infrastructure [4,5]. From this perspective, LPG, as a low-carbon fuel, can be applied to the existing spark ignition type of internal combustion engine vehicles with no substantial part changes [6,7]. The relevance of this fuel is more than just transport [8,9] but also in applications such as domestic cooking [10,11] and the agricultural and industrial sectors [12,13].
While LPG offers numerous benefits, its usage is not without its safety considerations, particularly regarding its flammability. Hence, LPG-related incidents will be discussed in this review. Bariha et al. [14] accordingly reviewed the studies until 2016 related to failure and other LPG and natural gas vessel hazards during handling, storage, and transport. In this review, the authors focused on propane and LPG due to their relevance in the energy transition. They discussed the improvements to date in the existing models for the various hazardous scenarios, seeking to identify the most critical gaps and research priorities that could enhance the LPG and liquefied natural gas (LNG) risk assessment. Nevertheless, the consequences of LPG-related accidents are enormous despite their low likelihood [15]. Moreover, city expansion worldwide also increases the energy demand and could lead to high risks during the storage and transportation of fuels. Consequently, a secure energy supply will contribute to the city’s sustainable development and ensure the safety of its inhabitants [16]. Furthermore, investigations on LPG continue as tools and software advance, improving the accuracy of predictions for LPG-related incidents to prevent accidents, facilitate rapid emergency responses, and minimize casualties. As highlighted by Huffman et al. [17] in their bibliometric study of the transport risk of LPG and LNG, there is a significant amount of safety-related experience with LPG compared to LNG, especially in consequence analysis. Thus, within the framework of the current energy transition, utilizing the existing LPG safety-related experience, infrastructure for storage [18], transportation [19], and final applications [20,21] emerges as a viable option for implementing and entering the market for renewable fuels. Accordingly, this study aims to provide a comprehensive review of the recent progress in the studies on the consequences of LPG-related incidents. The article also discusses the current LPG-trending applications and the emergence of biopropane as a promising alternative as a direct drop-in fuel toward a replacement of liquified petroleum gases. It notably focuses on the CFD advancement works and addresses the rise of machine learning-coupled calculations to generate reliable data to manage and predict LPG accidents but also to benefit the new and clean fuel technologies that will fuel humanity.
2. Characteristics and Properties of Liquified Petroleum Gas
LPG is a gas product of petroleum refining or natural gas extraction, mainly consisting of propane, butane, and other light hydrocarbons. Thus, natural gas or other gases are purified to remove water and sulfur impurities. Further, it is converted to syngas via steam methane reforming [22]. In Canada, 85% of the propane comes from natural gas and is produced in plants located in Alberta, British Columbia, and Saskatchewan. The remaining 15% is derived from refining petroleum in refineries in all provinces except Manitoba, Nova Scotia, and Prince Edward Island [23]. The LPG composition can vary in a wide range from pure propane passing through different propane/butane ratios to pure butane, depending on the location [23]. However, in the United States and Canada, LPG consists primarily of propane [24,25,26]. Therefore, Grade 1 fuel in Canada must contain at least 90% propane for use in internal combustion engines and general industrial/commercial applications [27].
Whether at the refinery or from natural gas processing, LPG contains contaminants, primarily water and sulfur compounds, including hydrogen sulfide, carbonyl sulfide, and elemental sulfur. According to the type and concentration of contaminants within LPG, one or more purification processes, including adsorptive [28,29] and absorptive purification [30], are employed to meet LPG quality specifications. Herein, the international trade of LPG is governed by two primary standards established by the International Organization for Standardization (ISO). ISO 8216-3 [31] is crucial in classifying petroleum products, including light distillates. At the same time, ISO 9162:2013 [32] specifies the characteristics of propane and butane. Additionally, several national standards regulate LPG trade and usage across different regions. For instance, in Germany, the DIN 51622:2020-09 standard [33] sets guidelines for LPG, while the United Kingdom follows the BS 4250:2014 standard [32]. In the United States, the ASTM D 1835-22 outlines the characteristics of liquified petroleum gases intended for various purposes, including heating and engine fuel in domestic, commercial, and industrial settings [34]; the EN 589:2024 specifies the requirements and test methods tailored explicitly for LPG automotive fuels in Europe [35]. Meanwhile, the CAN/CGSB-3.14-2023 standard governs the specifications and usage in Canada [27]. Table 1 compares the properties and requirements of some of the LPG standards mentioned.
Regarding its storage, LPG can be kept either above or below ground. For the first storage type, the LPG could be held in pressure, refrigerated, and semi-refrigerated storage tanks, considering that the boiling point of LPG approximately varies from −44 °C to −0.5 °C [36] and that the vapor pressure (ps) of its main components, propane and butane, at 25 °C is 951 kPa and 243 kPa, respectively [37]. Regarding underground storage, LPG can be stored in salt caverns or caverns that withstand its vapor pressure [38]. As for transportation, it is possible to move LPG through pipelines, waterways, roads, and railways [39,40,41]. Figure 1 illustrates the LPG production process and its final applications, which are discussed in the following section.
Regarding its properties, LPG is an odorless and colorless gas that is flammable and heavier than air. Under room conditions, LPG is gaseous, but it can be liquified and maintained under pressure to facilitate its storage and transportation [42]. The main properties of LPG are presented in Table 2, along with a comparative assessment of LPG, gasoline, and diesel, which highlights the cleaner combustion characteristics of LPG.
3. Applications
LPG has a lower environmental impact than other fossil-based fuels and is similar to LNG [47]. For instance, Bicer and Dincer [48] conducted a comparative environmental impact assessment of alternative and conventional fueled vehicles, using a cradle-to-grave approach and considering global warming and other environmental impact categories. The study found that in terms of global warming potential 500a, vehicles fueled by LPG produced 0.225 500a kg CO2 eq, which is lower than gasoline vehicles and comparable to Compressed Natural Gas (CNG) vehicles with 0.27 and 0.205 kg CO2 eq, respectively. The current and emerging LPG applications are disclosed below.
3.1. Transportation
Within the framework of global carbon neutrality targets, the adoption of electric and hydrogen vehicles is lower than expected [49]. Accordingly, governments are driving policies to employ LPG instead of gasoline or diesel fuels in vehicles with internal combustion engines [50,51]. As shown in Figure 2, motor gasoline is expected to remain the primary transportation fuel in 2040. In countries belonging to the Organization for Economic Co-operation and Development (OECD), the primary alternative-fueled vehicles are electric and plug-in hybrid electric vehicles, followed by those fueled by LPG. In contrast, natural gas is the main alternative in non-OECD countries [52]. Therefore, the LPG conversion technology has been adjusted to accommodate the development of gasoline engines from the first to the sixth generation. This adjustment has led to improvements in thermal efficiency, fuel consumption, and emission control, enabling the conversion of conventional gasoline engines for LPG use [53]. Recent studies on modern internal combustion engine designs—such as enhanced spray behavior, better combustion control, and optimized fuel injection—suggest that LPG remains compatible with these ongoing innovations. Its clean-burning characteristics and adaptability to direct injection systems allow it to benefit from many of the same improvements applied to traditional fuels [54,55].
Furthermore, LPG-fueled vehicles can utilize the cooling potential from the phase change in the vaporizer, thereby improving the air conditioning system [56]. LPG could also be blended with standard liquid fuels to improve performance and decrease emissions in spark ignition [57] and compression ignition engines [58,59]. Moreover, advanced LPG blends with clean alternative fuels, such as dimethyl ether (DME) [60], methanol [61], and hydrogen [62,63], aim to contribute to these fuel industries to establish alternatives towards carbon neutrality.
LPG is also a feasible fuel for facilitating the transition to zero-carbon ships [8,47]. In line with this, Kim et al. [64] conducted a study on the use of LPG in small marine vessels in South Korea. The study found that in a simulation-based comparison, exhaust emissions from LPG engines were significantly lower, with reductions of 11.35% and 4.59% in CO2 emissions and substantial decreases of 92.55% and 98.58% in soot emissions, compared to gasoline and diesel engines, respectively.
3.2. Heating and Cooking
LPG supports the clean energy transition by contributing to reduced pollutant emissions. The LPG has particle pollution from fine particulates (PM2.5) of 5.4 mg/MJ, whereas wood, kerosene, and charcoal exhibit PM2.5 values of 116.6, 25.3, and 84.4, respectively [65]. Furthermore, LPG is widely utilized, particularly in regions near the equator, across key energy end-use sectors such as residential [66,67], agriculture [68], commercial sector [69], and industry [70,71].
3.3. Fuel Cells
Solid oxide fuel cells (SOFCs) are attractive energy conversion systems that can employ hydrocarbons as fuels due to their high operating temperature. Among the hydrocarbon fuels, LPG is a promising option due to its availability, high energy density, low cost, and ease of storage [72]. When hydrocarbons serve as the energy source for SOFCs, an expensive and external reformer is typically employed to break down larger molecules into smaller ones, such as H2, H2O, CO, and CO2. Alternatively, using an internal reformer or directly using hydrocarbons as a SOFC fuel are compelling alternatives. Furthermore, the direct use of propane as the SOFC fuel is a trending alternative [73,74].
3.4. Refrigeration
In addition to the well-known applications of LPG, this fuel has increased its relevance as a natural refrigerant since it has a very low saturation temperature and low global warming potential and absorbs latent heat from surroundings, thus lowering the temperature [75,76]. Notably, researchers are focused on improving LPG-based refrigeration systems by using graphene, among other nanoparticles, to improve the coefficient of performance (COP) [77]. Furthermore, the carbon dioxide and propane mixture is a promising replacement for refrigerants in systems requiring low and ultra-low temperatures [78]. Other applications include LPG utilization as feedstock in the petrochemical industry for light olefins [79,80], aromatics [81], and hydrogen [82] production.
4. Biopropane Perspectives
Despite the advantages of LPG over other fossil fuels, there is a critical need to develop sustainable and renewable biofuels to address the depletion of petroleum-derived products and their impact on climate change. The eventual commercialization of bio-propane will reduce demand for natural gas and oil while maintaining the use of this clean-burning fuel [83]. Biopropane has a long-term potential to reduce air pollution and mitigate carbon emissions from challenging-to-decarbonize industries, such as transportation and rural heating [84].
Biopropane can be produced through various processes, with one notable example being the hydrotreating of vegetable oils. The hydrogenated vegetable oil (HVO) process allows the obtainment of renewable diesel as the main product and biopropane as a byproduct [85,86]. The Finland-headquartered Neste has patented the NEXBTL™ process to produce HVO, which is a mature and already commercial-scale manufacturing process [87]. The production of bio-propane through the gasification and synthesis of cellulosic organic wastes is in a demonstration state. Other pathways for biopropane production are glycerine dehydrogenation [88], bio-derived n-butanol hydrothermal conversion [89], sugar fermentation [90], waste digestion [91], and microbial biorefinery using amino acids such as fatty acid photo decarboxylase [92]. Furthermore, the authors highlight using Power-to-X as an emerging technology to produce renewable propane by hydrogenating captured carbon dioxide with green hydrogen [93]. In closing, despite the bio-LPG production (200 thousand tons/year) being far from those of LPG produced (300 million tons/year), it presents to the propane industry a route to decarbonization, giving a bio-alternative for consumers and as a potential competitor for governmental tax credits [86].
5. Highlighted Accidents Involving LPG Storage and Transportation
As mentioned earlier, accidents involving LPG have a low frequency but high consequences for life, the environment, and property. Those events have various contributing factors, ranging from design conception to technical issues and human errors [94]. Part of the accident-prevention approach involves learning from previous accidents, which entails identifying the cause and sequence of events, aiming to avoid similar accidents and limit damages to enhance process safety [95]. In this regard, implementing standards [96,97] and best practices [98] guide process safety management and accident prevention. Depending on the specific process requirements, best practices related to this article discussion are found in several documents such as the API 2510, Design and Construction of LPG Installations [99], the Risk Management Program Guidance for Propane Storage Facilities (40 CFR Part 68) [100], and the Process Safety Management for Storage Facilities [101]. Regarding Canada, propane storage and handling are governed by the CSA B149.2:20 code [102]. Furthermore, CSA Z-767:24 Standard [103] aims to identify the performance requirements for industries handling hazardous materials that plan to implement or have already implemented a process safety management (PSM) system [104]. Table 3 highlights some of the LPG-related accidents with high consequences, many of which have been used as input data or reference scenarios for validating consequence modeling approaches. For example, the Wenling LPG explosion has been studied using modeling tools to evaluate dispersion and the consequences of the explosion, as discussed in the following sections. Likewise, industrial accident databases such as the Major Accident Reporting System (MARS, later renamed eMARS after going online) [105], the Major Hazard Incident Data Service (MHIDAS) [106], and Analysis, Research, and Information on Accidents (ARIA) [107], among others, are available and focused on the lessons learned from accidents involving hazardous materials such as LPG. In addition to informing regulations and operational practices, these documented cases reveal recurring patterns of failure that reinforce the need for advanced predictive methodologies—such as consequence analysis—to enable rigorous assessment of accident scenarios and support evidence-based risk management strategies.
6. Risk Analysis
As mentioned earlier, the low pollution and price, among other properties of LPG, have increased interest in and the consequent use of this fuel. However, like other fuels, LPG is highly dangerous if not handled, stored, and transported correctly. Figure 3 shows a qualitative comparison of hydrogen, ammonia, methane, and propane in terms of the Risk Index of Explosion (ERI), Lower Flammability Limit (LFL), Upper Flammability Limit (UFL), Minimum Ignition Energy (MIE), and Recommended Exposure Limit (REL). Accordingly, hydrogen’s high risks are related to ERI, UFL, and MIE, which means that hydrogen has a high probability and severity of combustion and explosion compared to the other discussed fuels. On the other hand, ammonia exhibits a higher risk of toxic damage to the human body caused by leakage during an accident [108]. In this context, the safety-related experience related to LPG storage and handling [17,109,110,111,112] will contribute to the safe production, storage, and distribution of new fuels on a large scale [113]. In this context, the risks associated with LPG have been extensively studied through accident investigations, experimental research, and validated consequence modeling, all of which are supported by mature safety codes and regulatory frameworks. These well-established elements, examined in the following sections, provide a reference framework to inform risk analysis and guide the safe deployment of emerging fuels such as hydrogen and ammonia.
The flammable nature of propane is its main hazard. Thus, it can cause injuries due to burns and thermal radiation, and suffocation may occur if LPG displaces air, resulting in a decrease in oxygen concentration [44]. Propane in storage could leak and lead to vapor cloud explosions (VCE) or boiling liquid expanding vapor explosions (BLEVEs). These explosions cause massive damage to humans, the environment, and infrastructures due to the accompanying fires and blasts [114], see Table 4. A BLEVE is the worst-case scenario for an LPG-related accident. This physical explosion may occur when a container with a pressure-liquified gas ruptures and its content is suddenly released. This results in an explosive mixture of vapor and boiling liquid. If the BLEVE involves a flammable substance, such as LPG, four main dangers can arise: Fire, thermal radiation from the fireball, pressure due to the blast, and flying projectiles hazards. Among them, the projectiles could cause the most significant consequences. It is worth noting that the container size has a significant impact on the likelihood of a BLEVE occurring. Thus, small tanks could BLEVE faster since they have thinner walls than larger ones, leading to faster heating [115]. Comparatively, a 400 L tank could rupture in 5 to 7 min, a larger 4000 L one may fail in 5 to 7 min, while a 40,000-L tank on a truck would take 8 to 12 min to rupture. In an ignited vapor cloud event, a fireball could be immediately formed. Accordingly, for a 400-L tank, the radius of the fireball is around 18 m, and for a 4000-L tank, that radius doubles to about 36 m. Moreover, for a 40,000 L tank, the fireball radius could achieve 81 m. Notably, the heat is radiated in all directions, and the minimum observation distance is 90, 150, and 320 m for 400, 4000, and 40,000 L tanks, respectively. On the other hand, if the released vapor cloud is not ignited immediately, an explosion can occur due to delayed ignition. The resulting blast is unpredictable, with highly probable catastrophic effects, thus highlighting the importance of thorough safety measures [116,117].
To accurately assess all the major hazards and associated risks, specialized software is available to assist governments and industries in complying with safety regulations and corporate best practices. Software such as PHAST, EFFECTS, ALOHA, SAFETI, and RISKCURVES, among others, play pivotal roles in these attempts. These consequence modelling tools employ various computational approaches, each offering benefits and limitations that depend on the complexity of the scenario and the level of detail required. PHAST from DNV [118] and EFFECTS from Gexcon AS [119] display superior accuracy with robust and validated models for complex scenarios involving hazardous materials such as LPG release and dispersion. ALOHA [120,121], from the US EPA, is also a valuable tool for consequence analysis and emergency planning; however, it is less accurate than the software mentioned above due to its simplified assumptions.
On the other hand, SAFETI (DNV) [122] and RISKCURVES (Gexcon AS) [123] significantly contribute to risk assessment practices by providing systematic methodologies for Quantitative Risk Analysis (QRA) in compelling hazard management. These tools integrate statistical accident frequencies with consequence modeling outputs, enabling structured risk quantification and prioritization of mitigation measures. Furthermore, the CFD technique allows the improvement of existing models and the assessment of new hazardous scenarios [124]. CFD methods provide high-resolution spatial and temporal simulation capabilities, which are particularly valuable in urban or complex environments, although they typically require extensive input data and significant computational effort. Thus, consequence prediction is a critical stage in risk assessment for estimating the effects of fire, explosion, and dispersion scenarios on surrounding communities (see Figure 4). Accordingly, it also brings a baseline for mitigation planning and process design improvement. As discussed in the following sections, one of the primary challenges for both CFD and emerging data-driven tools, such as machine learning, is the limited availability of large-scale experimental data for validation, particularly in rare or extreme scenarios. Nevertheless, recent studies have addressed these limitations through hybrid modeling approaches and validation against real-world accidents. Machine learning, in particular, has emerged as a complementary strategy to traditional simulations, enabling faster predictions by learning from available data, reducing computational demands, and supporting rapid risk estimation in emergency scenarios involving hazardous materials such as LPG [125,126].
6.1. Experimental Research and Consequence Prediction Methods
6.1.1. Experimental Assessment
Different studies have focused on the experimental investigation of LPG-related fires and explosions, aiming to understand their mechanism for establishing prevention measures. Some previous experimental studies on LPG safety are reviewed in Table 5.
Table 3Highlighted accidents involving LPG storage and transportation.
| Place | Year | Cause | Fatalities | Reference |
|---|---|---|---|---|
| Feyzin, France | 1966 | The operational failure of the plant operator caused an LPG (propane) leak. | 18 | [127] |
| Mexico City, Mexico | 1984 | LPG (propane/butane mixture) leakage followed by ignition caused several explosions—domino effect. | >500 | [128] |
| Rio de Janeiro, Brazil | 1972 | The operator lost control in a draining operation on an LPG sphere, leading to a BLEVE. | 38 | [129] |
| Sainte-Élizabeth-de-Warwick, QC, Canada | 1993 | An LPG (propane) tank near a barn was involved in a violent fire and ruptured by BLEVE. | 4 | [130] |
| Visakhapatnam, India | 1997 | LPG leakage in a storage vessel caused a flammable vapor cloud followed by ignition and explosion. | >60 | [131] |
| Bucheon, Korea | 1998 | LPG (propane/butane mixture) leakage during the discharge process from the tank into subterranean storage. | 1 | [132] |
| Toronto, Canada | 2008 | Propane release at a transfer facility. An unknown ignition source led to a VCE and BLEVEs. | 2 | [133] |
| Viareggio, Italy | 2009 | The derailment of a train carrying LPG (propane) caused a release. The gas cloud formed led to a flash fire. | 31 | [134] |
| Chiba, Japan | 2011 | Earthquakes led to LPG (propane) vessels collapsing, resulting in a BLEVE. | [135] | |
| Kannur, India | 2012 | A truck tanker overturned, producing an LPG leak and a large vapor cloud that ignited, leading to a BLEVE. | 20 | [136] |
| Linyi, China | 2017 | An LPG tanker leaked during unloading, leading to a significant explosion and fire. | 10 | [137] |
| Wenling, China | 2020 | An LPG (propane/butane mixture) tank truck overturned while transiting at high speed on an expressway ramp. The tank collided with a concrete guardrail and exploded. | 20 | [138] |
LPG-derived consequence scenarios.
| Fires/Explosions | Definition | Graphical Description | References |
|---|---|---|---|
| Pool fire | Combustion of a substance that evaporates from a layer formed by a liquid fuel pool. A pool fire exhibits high flame temperature and heat flux to its surroundings. | | [139] |
| Jet fire | Results from a liquid, vapor, or gas discharge into a free space from an orifice. The momentum of the discharge induces the mixture of the discharged material with the atmosphere. | | [140] |
| Fireball | It is a fire that burns sufficiently rapidly for the burning mass to rise into the air as a cloud or a ball. It occurs if a flammable liquid, gas, or dust cloud abruptly releases and has limited mixing with air before ignition. | | [141] |
| Flash fire | The combustion of the flammable vapor or gas is mixed with air, and the flame propagates through that mixture with no overpressure generation. | | [142] |
| Vapor cloud explosion (VCE) | Results from igniting a cloud of flammable vapor or gas in which flame velocities are sufficiently high to produce a pressure wave. | | [143] |
| Boiling liquid expanding vapor explosion (BLEVE) | A BLEVE could be described as the explosive release of expanding vapor and boiling liquid when a catastrophic failure occurs in a pressure vessel holding a pressure-liquified gas. | | [144] |
6.1.2. Empirical Modeling
Theoretically based empirical correlations, such as those from the Netherlands Organization (TNO), Baker-Strehlow-Tang (BST), and the TNT equivalent methods, are the most commonly used for predicting the consequences of explosions. These approaches are sometimes inaccurate, as they often include experimentally adjusted factors that can lead to overestimating the calculated values. However, these methods have an easy implementation and do not require specialized software, high computational consumption, or deep programming knowledge compared to the integral methods, which will be discussed later [145,146].
Lv et al. [147] experimentally studied the evolution of propane jet flame geometrical profiles and air entrainment characteristics. The authors quantified the air entrainment coefficients using an integral model derived from conservation equations. Furthermore, they proposed a characteristic-length scale that represents the competition between the jet momentum component in the horizontal direction and thermal buoyancy in the vertical direction, thereby normalizing the flame geometrical profile parameters. This opens the door to using the developed models in simulators such as Fire Dynamics Simulator (FDS) or FireFoam.
Birk [133] presented a theoretical method for calculating the required overpressure for condensation cloud formation due to a BLEVE or VCE explosion. Air temperature and relative humidity are required as input parameters for the model. The Sunrise explosion accident is used as a validation example, utilizing video records from the incident. Accordingly, the explosion energy and expected damages at other distances can be calculated using the overpressure and distance data, which can be helpful for accident analysis.
Kraft et al. [148] developed simple and dimensionless equations for directly estimating distances considering the occurrence of a BLEVE event. The model correlated design or operation variables with safety distances as output. The input variables included the substance (i.e., propane), vessel volume, target vulnerability, explosion temperature, and the Jakob Number. This number could be designed as a hazard index related to BLEVE events. The aim of the work is also to facilitate the quantitative risk analysis of complex events. Furthermore, Elizaryev et al. [149] performed a comparative analysis of three mathematical models for assessing thermal radiation at BLEVE at a propane storage tank. The compared models, designated as MM1, MM2, and MM3, are those proposed by M.W. Roberts, the project GOST 12.3.047-2012 of the National Standard of the Russian Federation, and Dhurandher et al., respectively [150,151]. Unlike the other models, MM1 considers the time-dependent nature of a fireball’s thermal radiation. On the other hand, the surface radiant heat flux employed by MM2 is 350 kW/m2, whereas that calculated from MM3 is 387 kW/m2. Therefore, the models were compared at different times of fireball existence, according to MM1, in which the surface radiant heat flux corresponded to those of the MM2 and MM3 models (t = 4.16 s and t = 3.15 s, respectively). The results showed that the height of the center of the fireball resulting from MM1 is significantly lower, at 59.22 m and 68.33 m, compared to MM2 and MM3, respectively, at the corresponding times. Accordingly, the authors concluded that the energy potential with the BLEVE effect on the tank estimated by MM1 is incommensurable compared to MM2 and MM3.
Table 5Previous experimental studies on LPG safety.
| Incident | Objective | Experiment Details | Studied Variables | Main Conclusions | Reference |
|---|---|---|---|---|---|
| Jet fire | Study on the horizontal jet flame that impinges a vertical plate. | Plate (Q235 low-carbon steel) dimensions: 1 m × 1 m × 5 mm, thermal conductivity: 53.6 W/(mK). Nozzle (stainless steel): inner diameters of 2.0 mm, 3.0 mm, and 4.2 mm. Spacings between nozzle exit and plate: 0.20, 0.25, 0.30, 0.35 and 0.40 m. | Effect of nozzle exit velocity, exit diameter, and exit-plate spacing on the horizontally impinging jet fire. | A new correlation coupling the turbulent Karlovitz stretch factor and the ratio of nozzle exit diameter to exit-plate spacing was developed for the flame extension area of both horizontally and vertically impinging jet fire. It is noted that the temperature profile holds a big difference in the upward and downward directions along the vertical plate. | [152] |
| LPG tank under fire | Study the consequences of an LPG vehicle tank failure under fire conditions. | Ten fire tests on toroidal LPG vehicle tanks with no safety devices were conducted. | Tank filling level, fragmentation distance, and the radius of the danger zone. | All the tested tanks failed with a BLEVE within a t < 5 min after ignition, accompanied by a fireball, a near-field blast wave, and enormous fragmentation, leading to a high risk to rescue services when an LPG tank is affected by a fire. | [153] |
| BLEVE | Study the first moments (early milliseconds) of small-scale BLEVE in propane vessels | A small-scale apparatus was constructed to record detailed images of the failure process and measure overpressures near the vessel. The apparatus is an aluminum tube with D = 50 mm and L = 300 mm. | Failure pressures: from 10 to 33 bar. Measurement of (i) properties: temperature and pressure and (ii) consequences: blast overpressure, loud imaging, and shock around the vessel, among others. | The observation revealed the presence of a Mach shock at the vessel at the early stage of the opening. The results also demonstrated that the lead shock is generated before and gone before the liquid starts boiling, which indicates that the vapor expansion is primarily responsible for the first shock overpressure. | [144] |
| BLEVE | Analysis of the peak overpressure from the lead shock produced by a BLEVE using a new method: spherical shock theory. | A small-scale apparatus was constructed to record detailed images of the failure process and measure overpressures near the vessel (R/Dtube = 0.175/0.050 = 3.5): Aluminum tube with D = 50 mm and L = 300 mm. | Overpressure in the near-field (distance from the BLEVE center to the target, measured in the range within ten times the diameter of the BLEVE vessel); high-speed images. | Using shock tube overpressure prediction and spherical shock propagation model, a model based on the vapor phase properties at failure and a spherical shock propagation model was developed to predict BLEVEs overpressures in the near-field. | [154] |
| BLEVE | Study the ground force effect of BLEVEs: their impact on bridges and other infrastructure. | A small-scale apparatus was constructed to record detailed images of the failure process and measure overpressures near the vessel (R/Dtube = 0.175/0.050 = 3.5): Aluminum tube with D = 50.8 mm and L = 300 mm. | The failure pressure Pfail (from 11.7 bar to 32.7 bar); liquid fill level φliq (from 0 to 87%); the weakened length through machining at the top of the tube, Lc (from 50 mm to 150 mm). | The liquid fill ratio and the length of the debilitated vessel govern the magnitude of the ground force, which, jointly with the impulse, were linearly related to the rupture pressure and liquid fill ratio. | [155] |
| Deflagration-to-detonation transitions (DDTs) | Prediction of DDTs at large scales in congested areas. | Different tests on DDTs using propane in a test rig of 50,000 ft3 (1500 m3) gross volume were performed by SRI International and Gexcon. | Variables: levels of congestion, confinement, and gas concentrations. The flame speed and overpressure measurements were used for validation. | The congested area plays a significant role in the occurrence probability of DDTs. Also, the effects of detonating clouds are more critical than previously thought, which should be considered in future plant layout assessments. | [156] |
Reinders et al. [157] developed a lumped equilibrium tank model to study the pressure and temperature of LPG increase in a thermally coated pressure vessel exposed to a surrounding fire outbreak from a BLEVE. The results were validated against experimental tests in which three m3 tanks were filled to 50% capacity. The tests were carried out by the TNO research organization in the Netherlands and the Bundesanstalt für Materialforschung und-prüfung
Furthermore, Hemmatian et al. [159] comprehensively compared BLEVE overpressure predictions. The authors compared the empirically calculated data based on theory with experimental data. They concluded that the overpressures predicted using theoretical-based empirical models (i.e., TNO [160] and Roberts [150]) were conservative due to assumptions regarding isentropic expansion, isothermal expansion, ideal gas behavior, and constant volume energy. In contrast, the predicted overpressures closely aligned with experimental data when real gas behavior and adiabatic irreversible expansion (RAIE) were considered (i.e., Casal & Salla [161] and Planas-Chuchi et al. [162]). Recently, Laamarti et al. [163] developed a predictive correlation that estimated the ground loading from small-scale propane BLEVEs. The correlation had the vessel size, burst pressure, fill level, and weakened length as inputs, and the outputs were peak ground force, force duration, and impulse. The results revealed a significant influence of the liquid phase on the ground load force and its duration. Thus, vessels close to 95% full of liquid led to BLEVE incidents with the highest ground loads. The authors emphasized that correlation validation should be conducted on larger scales.
Sellami et al. [164] studied the validation of the Sedov–Taylor blast wave model for a BLEVE. The proposed model allowed the estimation of the overpressure effect of a BLEVE phenomenon by determining physical quantities characterizing the blast wave evolution, such as radius and velocity. The results were validated with BLEVE experiments, and they are more suitable for large-scale (5.6 m3 cylindrical tanks containing 2 tons of LPG, heated until a pressure of 15.2 bar) than medium-scale explosions (2000 L tanks, containing between 109.5 and 538 kg of LPG, with failure pressures between 15 to 19 bar). Afterward, the authors embedded the proposed model within the quantitative risk analysis (QRA) of a gas-processing unit. The authors used conventional tools such as the Hazard and Operability Analysis (HAZOP) and the Failure Modes and Effects Analysis (FMEA), which disclosed that the LPG accumulator is a critical system. Thus, it is essential to enhance its safety by optimizing existing safety barriers and implementing new ones to prevent BLEVE catastrophic consequences. Furthermore, incorporating the QRA–Sedov–Taylor approach enabled the derivation of realistic predictions for these BLEVE consequences (i.e., overpressure and thermal radiation) [165].
6.1.3. Integral Methods
Researchers have been working on developing more accurate mathematical models for consequence analysis, which is crucial for improved risk assessment and decision-making in safety management. In this context, integral models such as those employed in software such as ALOHA, EFFECTS, and PHAST are lumped-parameter ones developed typically for one- or two-dimensional consequence predictions [166].
In that direction, Bariha et al. [167] studied the risks of LPG transfer operation in a bottling plant. The effects of flammable vapor dispersion, jet fire, pool fire, and fireballs were investigated using ALOHA and PHAST (unspecified versions and models). Using the same input data, the authors found differences in results, such as the maximum damage distance for a thermal intensity of 37.5 kW/m2, resulting in 24.16 and 10 m for PHAST and ALOHA, respectively. The results revealed that the thermal radiation, estimated between 4–40 kW/m2 for the studied bottling plant, could affect the nearby population. Furthermore, using ALOHA, Beheshti et al. [168] conducted a study on the impact of LPG leakage from domestic cylinders. The study involved leaking LPG from a valve with a 1-inch hole in cylinders of varying sizes (26, 60, 78, and 107 L). In a BLEVE scenario, it was observed that the thermal radiation at distances of 39, 48, and 53 m reached about 10 kW/m2, posing a risk of causing fatalities to individuals within that range. At greater distances of 55, 68, and 74 m, the resulting thermal radiation was measured at 5 kW/m2, which could potentially lead to second-degree burns. This study emphasizes the importance of developing an emergency response plan in the event of catastrophic scenarios involving domestic LPG cylinders. Another study employing ALOHA [169] was devoted to modeling the consequences of a gas leakage and explosion fire from an LPG (100% propane) tank within a filling station. Thus, the thermal radiation from a BLEVE is approximately 10 kW/m2 up to 184 m from the explosion and 5 kW/m2 at distances of 260 m from the tank. On the other hand, comparable results were obtained using the same software by Kukfisz et al. [170]. They investigated the potential safe distance between public buildings and LPG (100% propane) filling stations, which consist of two gas storage tanks and a dispenser. Three different BLEVE scenarios were studied, with a tank filled at 20%, 42.5%, and 85% capacity. According to the results, for BLEVE scenarios with short-term radiation emission, the limit value of 12.6 kW/m2 is reached for distances of up to 149 m if the LPG storage tank is filled with 85% at the time of the explosion.
Witlox et al. [118] performed an overview of the verification and validation of consequence modeling for accidental releases of toxic or flammable chemicals, including propane. This verification of different models, such as flammable effect, dispersion, pool spreading/evaporation, and discharge models, provided an extensive experimental database to validate the consequence models in the hazard assessment package PHAST and the risk analysis package Safeti (V.8.1). Malviya and Rushaid [171] performed the consequence analysis of a BLEVE scenario of an LPG sphere tank using the PHAST software (V.6.5, unspecified model) and compared it to numerical calculations. According to the PHAST results, the exposure duration to a fireball is 27.57 s. It decreased to 24.41 s using numerical calculations. Furthermore, the thermal radiation PHAST results differed significantly as the receptor distance increased. Thus, for a receptor distance of 100.12 m, the calculated thermal radiation is 58.36 kW/m2 and 62.33 kW/m2 for PHAST and mathematical calculations, respectively. However, for 330.36 m, the calculated thermal radiation is 43.56 kW/m2 and 50.37 kW/m2 for past and mathematical calculations, respectively. On the other hand, Alfatesh and Biak [172] analyzed the BLEVE caused by a loss of containment of an LPG road tanker using PHAST software (V.8.11) and compared the results with those from a mathematical model. The results evidenced that the analysis performed with PHAST resulted in a larger radius covered, except for the worst thermal radiation value, >37.5 kW/m2, with a receptor distance of <170 m and <175 m, for PHAST and the mathematical model, respectively. Meanwhile, for thermal radiation of 12.5 kW/m2, the distances are 295 m and 273 m for PHAST and the mathematical model, respectively. Moreover, Lyu et al. [138] studied and simulated the LPG tank truck accident in Wenling, China, using EFFECTS and ALOHA software (unspecified versions). EFFECTS was employed to simulate the probable distribution of the tank debris after the BLEVE and the evaporation and spreading of the LPG. In contrast, ALOHA software and the TNO multi-energy method were employed to simulate the VCE. The results showed that the values given by the software for the distance of the tank residues are significantly lower than the actual distance (400 m). Furthermore, the authors noted the difficulty in predicting explosions in confined environments using simple leakage-dispersion models.
Additionally, MacNguyen [112] employed PHAST software (V.8.71) to compare the hazards of hydrogen with those of standard hydrocarbon fuels, including LPG. Among other results, the authors showed that the thermal reach distance for a pool fire from a 6-in. bounded release at storage conditions is lower for hydrogen than for LPG and LNG. For example, for a thermal radiation level of 12 kW/m2, the downwind distance for LPG and LNG is 42 m, and for hydrogen, it is 20 m. The authors stated that the study will provide the industry risk stakeholders with helpful guidance toward transitioning from LPG and LNG to hydrogen fuels.
6.1.4. Computational Fluid Dynamics (CFD)
An advantage of CFD-based consequence analysis, which involves conducting numerical simulations of dispersion, explosion, and fire, is that this method can thoroughly consider obstacles and terrain, leading to more accurate results than the previously mentioned methods. The mainstream software packages for CFD simulations in safety processes are ANSYS Fluent and CFX, GEXCON FLACS, and open-sourced OpenFOAM, among others [124]. It is essential to note that the reliability and accuracy of CFD models depend heavily on the validation of experimental data. In this respect, the modeling capabilities of CFD remain limited, though substantial progress has been made recently [173]. Table 6 highlights selected CFD studies on LPG consequences analysis.
LPG has a high risk of pool fires during transportation, storage, and applications. Therefore, Yi et al. [174] analyzed the hazards of LPG in large-scale pool fires using the CFD ANSYS Fluent software (V.19.2). After comparison with benchmark experimental data, the authors demonstrated that the employed CFD model accurately predicted the radiation of large LPG pool fires, and safe separation distances between LPG facilities and surroundings were estimated. Afterward, Yi et al. [139] also analyzed how pool diameter and wind velocity could influence the configuration of LPG pool fires using an experimentally validated CFD model developed in the same software (ANSYS Fluent 2019 R2). The authors proposed more accurate correlations that provide more accurate predictions of flame height and tilt angle compared to the reported empirical models. Scarponi et al. [175] used a CFD modeling approach to simulate and predict the behavior of an LPG pressurized tank exposed to a partially engulfing pool fire using ANSYS Fluent software (V.18.2). The comparison of the pressure build-up, wall temperature predictions, and lading temperature with the results of experimental fire tests verified the model as a reliable tool for analyzing pressure tanks under complex fire conditions.
Mashhadimoslem et al. [176] studied and simulated vertical propane jet fire through the k-epsilon (k-ɛ), realizable k-epsilon (RNG k-ɛ), baseline (BSL), baseline Raynolds Stress (BSL RS), and Shear Stress Transport (SST) models using the ANSYS CFX software (V.15.0). The results showed that the SST turbulence model is the most suitable, with an average error of 4.7% for a jet fire simulation. Vijayan et al. [177] studied the main geometrical features of jet flames using the CFD code FDS for regular and reduced ambient pressure conditions, and it was validated using results from previous works. The authors found that low-pressure conditions could lead to higher jet flame areas, and the flame length mainly depends on the exit momentum.
A VCE, followed by a fire, represents one of the most dangerous and high-consequence events in chemical facilities involving flammable materials. Hu et al. [178] investigated the LPG vapor cloud explosion hazards using CFD models through the FLACS software (V.10.7). Accordingly, in an ignition event, a nonuniform LPG cloud can result in a turbulent flame impacting the diluted explosive cloud at high speed, resulting in more severe consequences. Should LPG leakage exceed 300 s and the aperture widen beyond 10 mm, the hazard zone (Pc > 2 kPa) experiences a sharp escalation, thereby emerging as the pivotal factor in controlling explosion hazards. The LPG safe reserve demonstrates minimal variation at low protection distances (Rp) and critical overpressures (Pc). However, it shows an exponential increase when Rp and Pc surpass 200 m and 20 kPa, respectively.
Furthermore, Wang et al. [179] studied the dynamic process and damage evaluation of the Wenling, China, accident in 2020 (see Table 3) through scene reconstruction in FLACS (unspecified version). The authors found that the expansion of the LPG explosion around the expressway moved along the spaces between the obstacles. The overpressure was sufficient to cause considerable damage to the surrounding structures. Additionally, a detailed damage analysis of coupling characteristics resulting from an LPG tank trailer explosion is proposed, which will be useful for future blast-resistant structure design. Wang et al. [180] employed FLACS software (unspecified version) to simulate the 2018 LPG tank truck accident in Bologna, Italy. The authors found that the continuous dispersion from the LPG tank trailer triggered the vapor cloud explosion. The proposed work involved a failure analysis and classification of accidents involving LPG tank trailers, focusing on the explosive destruction and damage mechanisms that contributed to hazard identification and emergency response. Recently, Dhiman et al. [181] evaluated the performance of the PDRFoam tool (unspecified version- an open-source variant of OpenFOAM) on the vapor cloud explosions of propane. The evaluation of results was evaluated against a database of medium-scale experiments. The results showed that the flame propagation is well-modeled. However, over-predictions for peak pressure were found. The authors noted that PDRFoam is a computationally cost-effective solver for CFD explosion modeling, and its open-source nature opens up possibilities for further modeling.
On the other hand, Wang et al. [182] performed a hazard analysis on a characteristic fireball of an LPG riad tanker BLEVE using the FDS code from NIST (unspecified version). The authors found that the bigger the fuel mass, the larger the longitudinal diameter of the fireball, as well as the fireball’s diameter and aspect ratio. Accordingly, the fireball gradually changes from momentum-driven to buoyancy-driven as the mass of fuel increases. The results showed that the simulation data agrees well with the experimental data and could help us understand the evolution and hazards of BLEVE accidents. On the other hand, Sellami et al. [183] also proposed a CFD approach to evaluate the thermal effects of a BLEVE in the gas processing plant using the FDS CFD code FDS from NIST (V.6.5.3). The approach considered a sensitivity analysis for choosing appropriate models to fire reactive flows. The procedure was validated through empirical correlations and large-scale data from the literature, demonstrating good agreement. However, the authors noted the importance of the liquid-gas transition and the container disintegration to model a BLEVE process entirely.
Weerheijm et al. [184] performed a quantitative risk analysis of tunnel LPG explosions. The authors simplified 3D-CFD codes to a 1-dimensional numerical gas explosion model developed specifically for tunnel conditions. The results showed that in the case of an instantaneous LPG release, the cloud is initially above the upper flammability limit (UFL). Depending on the initial amount of LPG released, the cloud will be mainly fuel-rich by the time it reaches the end of the tunnel or may fall within the flammability limits before the tunnel ends. Li [185] also studied the hazards of alternative fuel vehicles, including those working with LPG, using a one-dimensional CFD program that was verified with previous tests. An analysis of the peak overpressure in case of an explosion in a tunnel showed that for gas tank rupture and BLEVE, the peak overpressure decreases rapidly along the tunnel and is reliable for users more than 100 m away from the tank. On the contrary, a cloud explosion would be most severe and intolerable for tunnel users. These studies contribute to the safety design of vehicles and tunnels. Moreover, Cheng et al. [186] investigated the ground vibrations caused by an LPG BLEVE inside a road tunnel using FLACS and LS-DYNA software (unspecified versions) to simulate the BLEVE process and the response of arched tunnel and surroundings, respectively. The authors found that the cover depth of the tunnel, the rock type, and the porosity are significantly more influential than the tunnel lining concrete grade in the LPG BLEVE-induced ground vibrations.
Shelke et al. [187] performed CFD analysis for two hydrocarbon BLEVE fireballs using the FireFOAM CFD code (unspecified version). The authors estimated the FireFOAM capabilities to simulate the fireball diameter, lifetime, flame dynamics, and structure and compared the results with BAM experiments and an aircraft. The authors concluded that the calculated fireball diameter and lifting height are in good agreement with available video records. Abdel-Jawad [188] proposed a hybrid method (analytical-numerical) using Prugh’s method as a feasible alternative for rapid solution times for BLEVEs modeling without accuracy loss, using the exploCFD code (V.7.0). The explosion source was first calculated analytically regarding pressure and temperature, then fed into the exploCFD code to estimate the blast wave propagation in a 2D domain. The initial rupture, catastrophic failure, and subsequent ignition were modeled as occurring instantaneously at the beginning of the simulation. Compared to the experiments [189], the simulation results showed that for a sensor location 10 m away from the side of the tank, an overpressure overestimation by exploCFD was about 20%. Consequently, in further work [190] the same author improved the method by coupling it with the confinement-specific correlation (CSC) gas explosion model, demonstrating exceptional agreement with previous experiments for peak pressure, impulse, and the time history of the first peak parameters. Compared to prior work, for a sensor location 10 m away from the side of the tank, the exploCFD prediction for the maximum overpressure closely matches the experimental results [189]. Based on the author’s findings, there is a variation of a few hundred Pascals. Davidy [191], on the other hand, proposed an algorithm using large eddy simulation (LES), calculation of the convective and radiative heat fluxes, thermochemical and heat transfer analysis of the vessel coating, and the calculation of the time required to evaporate the liquefied propane using FDS simulator (V.5.0) and COMSOL Multiphysics (V.4.3b). The author proposed a new and flexible tool for analyzing pressure vessels exposed to jet fire. Future work with other hydrocarbon fuels, as well as BLEVEs caused by pool fires, is also possible.
Scarponi et al. [192] developed a two-dimensional CFD-based model using ANSYS Fluent (V.18.2.0) to observe the pressure build-up of LPG tanks exposed to fire until BLEVE occurrence. The model enabled the study of the effects of free convection and thermal boundary phenomena, thermal stratification in both the liquid and vapor phases, and liquid thermal expansion. The authors noted that the development model could support studies predicting vessel failure under fire explosion. Subsequently, Scarponi et al. [193] extended the work by developing a 3D model in ANSYS Fluent (V.18.2.0) to simulate the pressure build-up in equipment with saturated liquids exposed to fire. The comparison between 2D and 3D model results showed that the 2D models are reliable in modeling fluid behavior when homogeneous heat exposure conditions exist. However, 3D models overcome limitations such as temperature profiles and recirculation patterns. The same authors [194] recently performed a parametric analysis to investigate the behavior of a 1.9m3 bullet LPG tank in partial engulfment fire scenarios using ANSYS Fluent (V.18.2.0). Different cases were studied by varying the position of the zone exposed to fire and the filling degree of the tank. It was demonstrated that the engulfment mode and the filling degree strongly influence the tank’s pressurization rate, energy accumulation, and high-temperature mechanical wakening.
Furthermore, Li and Hao [195,196] employed the CFD-based software FLACS (V.10.7) to predict the BLEVE blast wave in open space and hindered environments. Accordingly, the authors found that the liquid flashing from the LPG tanks is slower than vapor expansion in generating shock waves. The maximum peak pressures from the models were mainly attributed to vapor expansion. Regarding shock waves in obstructed environments, obtaining simulation-based correlations for pressure predictions that can predict explosion loads is feasible. A good agreement was achieved between the CFD results and medium- to large-scale experimental data. In addition, Ma et al. [197] performed numerical simulations using the k-ɛ and eddy-dissipation models of FLUENT (unspecified version) to analyze the mass and heat transfer in an LPG leak from a vessel. The simulation results agreed well with the experimental observations, validating the phase transition during the LPG boiling process. This provided information on the temperature, pressure, and thermal radiation distribution of the explosion flow field.
Wang et al. [198] also studied the BLEVE loading on rigid structures using FLACS (V.22.LR2). Previous large-scale experiments were used to calibrate the numerical model. The authors analyzed the diffraction and clearing effects on the reflected waves concerning the dimensions of the structure. Consequently, reflection coefficient charts were developed to predict the reflected BLEVE overpressure on rigid structures. In their subsequent work, the authors [199] investigated the influence of structural stiffness and BLEVE wave duration on the reflected BLEVE pressure on the structure. The interaction of BLEVE waves with flexible structures was simulated using ANSYS Fluent (V.16.0) for the blast wave propagation coupled with ANSYS Mechanical (V.APDL.16.0) for structural analysis. The results showed that during the action of a BLEVE wave, the more flexible the structure, the smaller the peak reflected pressure and the longer the duration of the action on the structure. The results can be used to predict explosion loads when designing structures against BLEVEs.
On the other hand, Yuan et al. [200] studied, through numerical simulation using FLACS (V.10.7), the LPG gas explosion accident in Beijing on 15 March 2022 (an example of an accident in a high blockage environment). The results showed that the explosion of 4.0–4.1 m3 of LPG can destroy several walls, with a maximum indoor overpressure of 1.130 bar. Additionally, the explosion resulted in a high-pressure injury area of 10 m and a high-temperature injury area of 14 m outdoors. Also, if structures have the same spatial direction as the propagation direction of the initial wave, they will suffer less damage. Doors and windows constrain the normal venting process of the explosion reaction fluid in civil buildings, leading to severe outdoor consequences. Accordingly, reasonable pressure relief areas will contribute to preventing and mitigating disasters from explosion accidents. Furthermore, Kang et al. [41] proposed a CFD-based simulation methodology using satellite maps along the FLACS software (V.10.7) to visualize the consequences of the failure of an LPG tanker leak explosion, aiming to provide support for LPG accident prevention. Recently, Lyu et al. [201] studied the predictive ability of the model for LPG gas dispersion using the data from the LPG tank truck accident in Wenling, China, and using the commercial CFD code ANSYS Fluent (V.14.0). The geometric model and grid were established using geographic information system (GIS) data from Google Maps of the accident site. The authors discussed the multicomponent liquid flash and liquid pool evaporation and spreading. The authors found that one of their simulations was consistent with the distribution of accident damage. The proposed methodology could guide accident consequence assessment and emergency response measurement development. Furthermore, the authors encourage studying and validating the discussed model for other liquid gasses, such as liquefied natural gas, chlorine, and ammonia. These recent case-based studies reaffirm the relevance of CFD as a valuable tool for simulating real-world LPG incidents, thereby enhancing the accuracy of consequence prediction and supporting the development of emergency response strategies in complex environments.
Table 6Highlighted CFD studies on LPG-related consequences modeling.
| Incident | Objective | CFD Code | Models | Scenario | Main Conclusions | Reference |
|---|---|---|---|---|---|---|
| Pool fire | Heat radiation from large LPG pool fires | ANSYS Fluent | k-ɛ model; radiation: P-1 model; non-premixed combustion model; surface emissive power (SEP) model; | Three different pool diameters of 12.9, 14.9, and 16.9 m with atmosphere temperatures of 309, 306, and 312 K, mass burning rates of 29.087, 47.328, 44.426 kg/s, and wind velocity of 3, 2.5, 0 m/s, respectively. | The employed CFD model (compared to experimental data) accurately predicted the radiation of large LPG pool fires. Safe separation distances between LPG facilities and surroundings were estimated. | [174] |
| Pool fire | Flame height and flame tilt as functions of pool diameter and wind velocity. | ANSYS Fluent | k-ɛ model; radiation: P-1 model; non-premixed combustion model; | Air velocities of 0, 0.5, 2.5, and 3 m/s were selected for each of the following pool fires with diameter, ambient temperature, and average mass burning rate of (i): 10.4 m, 306 K, and 8.406 kg/s (ii): 12.9 m, 309 K, 12.932 kg/s, (iii): 14.9 m, 306 K, and 17.254 kg/s, and (iv) 16.9 m, 312 K, and 22.196 kg/s. | Higher horizontal wind velocity → stronger convection effect in the horizontal direction, reducing the flame height and increasing the flame tilt angle. | [139] |
| Jet fire | Compare different turbulence models for vertical propane jet fire simulation. | Home code coupled with ANSYS CFX | Turbulence models: k-ε, SST, BSL, BSL RS, and RNG k-ε; EDC combustion model; Monte Carlo radiation model. | Computational domain: Cylinder with L = 10 m and D = 2 m. Fuel inlet to simulate the vertical jet fire: Nozzle with D = 12.75 mm. Temperature range: 1500–1700 K. Fuel rates: 148.41 m/s (0.03 kg/s) and 252.75 m/s (0.19 kg/s). | The SST turbulence model is the most suitable, with an average error of 4.7% for a jet fire simulation | [176] |
| Jet fire | Flame geometry of horizontal turbulent jet fires in reduced pressures | FDS | LES turbulence model; EDC combustion model. | The domain size is 8 m × 8 m × 7 m. The release square nozzle (20 mm × 20 mm) is provided in the Y-Z plane, 1.3 m above the ground. Five ambient pressure conditions, ranging between 0.6 and 1 atm, and eight jet fuel exit velocities, between 27.5 and 125 m/s, were studied. | Low-pressure conditions could lead to higher jet flame areas, and the flame length mainly depends on the exit momentum | [177] |
| Vapor Cloud Esplosion | Study the hazard evolution of considerable LPG leakage and vapor cloud explosion | FLACS | RANS turbulence model | The model size of the LPG plant was 450 × 300 m, including six oil tanks, six refrigerated tanks, and four pressure tanks. The LPG vapor cloud with stoichiometric concentration was ignited at the top of the refrigerated tank; the initial temperature was 301 K. | LPG expansion about the ground occurs along the gaps between structural congestion. Continuous large-scale congestions, such as walls, significantly enhance the LPG expansion and concentration accumulation. | [178] |
| Vapor cloud Explosion | Evaluate the PDR approach (PDRFoam) to model the effect of small-scale obstacles: pipes on flame propagation and explosion overpressure. | OpenFOAM | κ-ϵ—PDR-based modifications for turbulence; laminar flamelet combustion model | Pipe geometry: 3 m in length, 1 m in breadth, and 1 m in height. The pitch is 0.33 m vertically and horizontally, and the obstacle diameter (D) is 0.01 m. The fuel-air mixture was filled in the domain with an equivalence ratio of unity. | The flame propagation is well modeled considering the database of vapor cloud explosion experiments. However, over-predictions for peak pressure were found. | [181] |
| BLEVE fireball | Analyze the consequences of the fireball from an LPG tanker BLEVE accident. | FDS | LES | A 200 m3 cube was selected as the domain. The sides and top of the domain behaved as an open atmosphere. Initial pressure: atmospheric pressure; initial oxygen mass fraction: 0.232; initial temperature: 20 °C. | The bigger the fuel mass, the larger the longitudinal diameter of the fireball, as well as the fireball’s diameter and aspect ratio. Fireball gradually changes from momentum-driven to buoyancy-driven as the mass of fuel increases. | [182] |
| BLEVE fireball | Evaluate the BLEVE thermal effects on a gas processing plant. | FDS | LES turbulence model; EDC combustion model; thermal radiation model | Plant: open calculation domain of 300 m × 300 m × 360 m. Relative humidity: 40%, ambient temperature: 20 °C. Accumulator: Temperature 40 °C; Pressure: 14.5 barg; Volume: 50 m3. | The procedure showed good agreement with experimental data. However, the authors noted the importance of the liquid-gas transition and the container disintegration in modeling a BLEVE process. | [183] |
| LPG BLEVE in a tunnel | Investigate the rock vibrations induced by an LPG BLEVE inside a tunnel. | FLACS and ANSYS LS-DYNA | κ-ϵ model; ALE algorithm | Model domain: 30 m length, 30 m width, 60 m height. LPG tank: volume 20 m3; diameter 2.4 m. The tank center is located at the center of the upper arc of the curved-wall-arched tunnel. | The cover depth of the tunnel, the rock type, and the porosity are significantly more influential than the tunnel lining concrete grade in the LPG BLEVE-induced ground vibrations. | [186] |
| BLEVE | Blast wave prediction of large-scale BLEVE in the open space. | FLACS | RANS turbulence model | A 2000 L propane tank is modeled as a 2.6 m long cuboid. Stretched grids start at 4 m from the edge of the core grid domain. | The liquid flashing from the LPG tanks is slower than vapor expansion when generating shock waves. The maximum peak pressures from models were mainly attributed to vapor expansion. | [195] |
| BLEVE | Blast wave prediction of medium to large-scale BLEVE in an obstructed environment. | FLACS | RANS turbulence model | Two domains were simulated with dimensions of 65 × 40 × 30 m3 and 230 × 60 × 40 m3. The obstacles in the following parametric study are located on the ground and at least 5 m away from the BLEVE source. | Only the vapor expansion from BLEVE was considered to simulate the shock waves in an obstructed environment. Good agreement was obtained between the CFD results and medium-large-scale experimental data. Simulation-based correlations for pressure predictions that could predict explosion loads were proposed. | [196] |
| BLEVE-VCE Wenling accident | Simulation of gas dispersion resulting from the accidental instantaneous release of LPG and reconstructing actual accident processes. | Ansys Fluent | RANS turbulence model; SIMPLE algorithm | CFD simulation domain: A high-resolution terrain geometric model of the region around the accident site with an area of 3.83 km2 and height of 300 m (ICEM software), containing buildings, viaducts, ramps, and trees. GIS data from Google Maps at a horizontal resolution of 10 m. | The LPG instantaneous-release model was constructed considering the multicomponent liquid flash and liquid pool evaporation and spreading. The droplets in the release source, terrains, and obstacles significantly affected the vapor cloud’s dispersion behavior and extension distance. | [201] |
6.1.5. Application of Machine Learning
Machine learning is increasingly important for analyzing consequences and assessing risks. It provides advanced predictive capabilities and improved accuracy in identifying potential hazards and their consequences by analyzing large datasets, facilitating assessment, and saving time for rapid emergency response [202,203,204]. Therefore, machine learning has been proposed to enhance the accuracy and efficiency of BLEVE pressure prediction. Thus, Hemmatian et al. [205] estimated the mechanical energy of propane and butane BLEVEs. The dataset, derived from real gas behavior and the adiabatic irreversible expansion method, was used to train the Artificial Neural Networks (ANN). A Bayesian regularization was the three-layer backpropagation training algorithm. Temperature and vessel filling degree at failure have been considered input parameters (plus vessel volume), and the ANN model estimated the BLEVE blast energy as output data, validated with experimental values.
In this context, Gabhane and Kanidarapu [206] analyzed worse-case scenarios for an LPG terminal. Firstly, the authors performed a Hazard Identification (HAZID) study to obtain the worst-case failure scenario data. They used them in the ALOHA software to generate safety distance measures, considering the effects of flammable vapor clouds, thermal radiation, and overpressure blast waves. The resulting dataset was divided for training, validation, and testing of the ANN. The results from the ANN demonstrated high accuracy in safety distance prediction, providing valuable insights for informed risk management decisions.
On the other hand, accurately simulating CFD model simulations of BLEVE events, including their wave propagation and effects on the surroundings, is time- and resource-consuming. Machine learning arises as a promising option for relieving those CFD drawbacks. In machine learning approaches, training data are required to train the model to map from input variables to selected target outputs [166]. Accordingly, Li et al. [207] developed CFD techniques that utilized an ANN model to predict the impact of blast-loading generated from BLEVEs. The authors found that the application of the ANN model significantly decreased processing times from hours to seconds. Furthermore, the results from the trained ANN showed an error of only 1.3% compared to CFD simulations based on actual experiments. Similarly, Li et al. [208] improved the BLEVE pressure predictions from CFD models through the ANN model. Despite the reliable estimation of BLEVE consequences inside a tunnel, this work lacked rigorous validation compared to previous work. Although ANNs enable expedited calculations related to the consequences of BLEVEs, their use could be made more straightforward. In this regard, Wang et al. [209] employed a surrogate multi-layer perceptron (MLP) model to improve the ANN training efficiency and accuracy. The results were verified based on the experimental data available in the literature. Likewise, Li et al. [210] conducted a comparative study on the most effective machine learning model for predicting the consequences of BLEVEs. The authors found that between the MLP, ResNet, Transformer, and lightGBM, the transformer neural network achieved the best results for BLEVE pressure prediction in an open space. In this sense, Li et al. [211] employed the same machine learning approach to predict the consequences of propane and butane BLEVEs on structures (rigid obstacles). This study opened the door for using Transformer as a surrogate model for CFD BLEVE consequences analysis for structural response predictions. The same authors [212] recently proposed using Graph Neural Networks (GNNs) to predict overpressure from a BLEVE. The authors found that the GNNs exhibited higher temporal resolution in pressure-time history prediction compared to the exclusive CFD simulations.
7. Conclusions and Outlook
LPG is increasingly positioned as a link to clean fuels in the current energy transition. LPG has environmental advantages over other fossil fuels. Moreover, its well-developed infrastructure in storage, transportation, and end-use applications can be utilized, sometimes without modification, for the production of single-use or blended clean fuels, allowing them to enter the market while strengthening their supply chain.
However, being such a widely used fuel, there are still accidents related to LPG, which, although they have a very low probability of occurrence, they have very high consequences for humans, the environment, and infrastructure. It is precisely the study of the consequences of possible scenarios caused by incidents at different stages of the LPG supply chain that allows us to learn lessons and make corrections to avoid new accidents, minimize their consequences, or respond rapidly to an emergency involving this fuel.
This article provided a comprehensive review of LPG applications in the transportation and energy sectors, as well as the prospects for biopropane. Especially on the topic of safety, this review also presented the risks associated with LPG handling and the study of the consequences of incidents related to this fuel. This work addresses a gap in existing reviews by connecting the development of safety modeling tools—ranging from empirical methods to machine learning approaches—with the larger context of how liquefied petroleum gas (LPG) contributes to the energy transition. Unlike previous studies that have focused solely on LPG applications or modeling techniques, this review highlights the interplay between these elements.
Experimental testing provides insight into the phenomena within LPG-related fires and explosions, serving as feed data for the discussed models. However, experimentation at a large scale is dangerous and expensive. Therefore, efforts have been focused on developing predictive models, including empirical, integral, and CFD models. Each model can be used depending on the profoundness of estimation required. Empirical models are used for initial screening or as a preliminary assessment of a given study. In addition to experimental results, they allow the validation of integral or CFD modeling results. This structured synthesis also clarifies the role of each modeling strategy within the current challenges of consequence prediction and supports hybrid approaches in risk assessment.
Integral-based models are valuable tools for predicting the consequences of potential accidents resulting from LPG release, considering factors such as weather conditions and the scenario type under study, and providing relatively easy and fast estimations. Also, integral-based tools are widely employed in companies and the competent authorities for emergency action plans. However, using CFD techniques, a real-time assessment that considers geographic and structural information to estimate the consequences of an incident is only possible. Depending on the modeling level of detail, it could take several hours of calculation, limiting their application in emergency responses.
The Wenling, China accident served as a valuable analysis tool and source of information. It has been studied using various tools for consequence analysis, ranging from simple models to those with high geometric and analytical complexity levels. Therefore, hybrid approaches that combine integral tools with CFD methods are an attractive alternative to reduce calculation times. Moreover, using CFD methods combined with machine learning enables a reduction in time from hours to seconds, accompanied by a slight increase in error. At the same time, using machine learning with CFD is the future of achieving detailed estimations of the consequences of accidents in complex geography or structures in very short times, which could be used in emergency plans. However, this strategy needs to be explored more profoundly and requires validation.
Moreover, it is here, considering the boom and the urgent need for implementing clean fuels, where the studies carried out with LPG can bridge the gap in analyzing the consequences of fuels such as hydrogen, DME, and ammonia. Human experience in the handling, storage, transport, and applications of LPG and fuels such as LNG can contribute to filling the gap that these new fuels present in terms of infrastructure and development and, in general, data for the validation of the models that are being developed to study the consequences of probable accidents due to their current and future implementation. This review offers a knowledge base for LPG safety and a scalable modeling framework that can be adapted to cleaner fuels, where experimental data are limited and a rapid, high-resolution risk assessment is urgently needed.
C.A.-S.: Conceptualization, and Writing—original draft. J.-P.L.: Review and editing. G.S.: Conceptualization, review and editing. B.R.d.V.: Conceptualization, review and editing. All authors have read and agreed to the published version of the manuscript.
The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding authors.
Carolina Ardila-Suarez is deeply thankful to the postdoctoral fellowship Claire-Deschênes of the Faculté de genie from the Université de Sherbrooke for the support Icons made by various authors from
The authors declare no conflicts of interest.
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Figure 1 LPG: from production to final applications.
Figure 2 2015–2040 projection of vehicle sales for OECD and non-OECD countries [
Figure 3 Qualitative comparison of ammonia, hydrogen, methane, and propane risks. Adapted from [
Figure 4 Study of LPG consequences scenarios.
Comparison of the key properties and requirements for LPG fuel as specified in ISO 9162:2013 [
| Component | ISO 9162:2013 Commercial | ASTM D 1835-Special-Duty Propane (USA) | CAN/CGSB-3.14-2023 Grade 1 (Canada) |
|---|---|---|---|
| Propane | - | - | 90% min. by volume |
| Butane | 7.5% max. %(molar) | 2.5% max. by volume | 2.5% max. by volume |
| Sulfur | 50 mg/kg max | 123 mg/kg max | 123 mg/kg max |
| Evaporation residue | 60 mg/kg max | 0.05 mL max per 100 ml | 0.05 mL max per 100 mL |
| Vapor pressure | 1550 kPa max at 40 °C | 1435 kPa max at 37.8 °C | 1435 kPa max at 37.8 °C |
Physicochemical and combustion-related properties of the main components of LPG compared to gasoline and diesel fuels [
| Component | Propane | Propene | n-Butane | Gasoline | Diesel |
|---|---|---|---|---|---|
| Boiling point @ 101.3 kPa (°C) | −42.1 | −47.7 | −0.5 | 30–220 | 160–380 |
| Vapor pressure @ 37.8 °C (kPa) | 1310 | 1561 | 356 | ~64 | ~2 |
| Density @ 15 °C (kg/m3) | 506.0 * | 520.4 * | 583.0 * | ~730 | ~840 |
| Gross calorific value @ 25 °C (kJ/kg) | 50,014 | 48,954 | 49,155 | ~44,300 | ~45,500 |
| Lower Flammability Limit, LFL (% vol. in air) | 2.3 | 2.2 | 1.9 | 1.4 | 0.7 |
| Upper Flammability Limit, UFL (% vol. in air) | 9.5 | 9.6 | 8.5 | 7.6 | 5.0 |
* Liquid density (@saturated pressure).
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
Countries worldwide are focused on the objective of zero emissions by 2050. However, the accelerated implementation of clean technologies has had some drawbacks, remarkably those related to safety issues. Liquefied petroleum gas (LPG) emerges as a transition fuel in this context, considering the following two aspects. First, LPG is a fuel that has environmental advantages compared to other fossil fuels, so the extension of coverage as a replacement fuel is a key factor. Second, LPG has a well-developed storage and transportation infrastructure that can be used, sometimes without modifications, for clean fuels, helping their implementation. Therefore, the safety analysis and the study of the consequences related to the hazards of LPG is a current subject that contributes, through all the tools reviewed in this article, to not only reduce the risks of this fuel but also to connect with the safety issues of clean fuels. This review article provides a comprehensive overview through consequence modeling tools, highlighting computational fluid dynamics (CFD) and machine learning to pave the way for the full implementation of clean fuels that will power the future of humanity.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details
; Rego de Vasconcelos Bruna