Next Article in Journal
Clustering of Pediatric Brain Tumors in Texas, 2000–2017
Previous Article in Journal
Imidacloprid and Its Bioactive Metabolite, Desnitro-Imidacloprid, Differentially Affect Ovarian Antral Follicle Growth, Morphology, and Hormone Synthesis In Vitro
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Threat and Risk Analysis-Based Neural Network for a Chemical Explosion (TRANCE) Model to Predict Hazards in Petroleum Refinery

by
Lalit Rajaramji Gabhane
and
NagamalleswaraRao Kanidarapu
*
School of Chemical Engineering, Vellore Institute of Technology, Vellore 632014, Tamil Nadu, India
*
Author to whom correspondence should be addressed.
Toxics 2023, 11(4), 350; https://doi.org/10.3390/toxics11040350
Submission received: 24 February 2023 / Revised: 24 March 2023 / Accepted: 5 April 2023 / Published: 7 April 2023
(This article belongs to the Section Exposome Analysis and Risk Assessment)

Abstract

:
Risk analysis and prediction is a primary monitoring strategy to identify abnormal events occurring in chemical processes. The accidental release of toxic gases may result in severe problems for people and the environment. Risk analysis of hazardous chemicals using consequence modeling is essential to improve the process reliability and safety of the refineries. In petroleum refineries: toluene, hydrogen, isooctane, kerosene, methanol, and naphtha are key process plants with toxic and flammable chemicals. The major process plants considered for risk assessment in the refinery are the gasoline hydrotreatment unit, crude distillation, aromatic recovery, continuous catalytic reformer, methyl–tert–butyl–ether, and kerosene merox units. Additionally, we propose a threat and risk analysis neural network for the chemical explosion (TRANCE) model for refinery incident scenarios. Significantly, 160 attributes were collected for the modeling on the basis of the significance of failure and hazardous chemical leaks in the refinery. Hazard analysis shows that the leakages of hydrogen and gasoline at the gasoline hydrotreatment unit, kerosene at the kerosene merox plant, and crude oil at crude-distillation units were areas of profound concern. The developed TRANCE model predicted the chemical explosion distance with an R2 accuracy value of 0.9994 and MSE of 679.5343.

1. Introduction

Rapid industrialization and the use of more advanced technology, materials, plants, and machinery have led to a significant jump in petroleum product consumption [1]. A petroleum refinery regularly handles many flammable and explosive products in large volumes in processing units and storage tanks. It is prone to major accidents such as fire, explosions, and the dispersion of toxic and flammable substances [2,3] due to the handling, processing, and storage of highly flammable substances and the operation of the facilities at elevated temperatures and pressure or cryogenic conditions [4]. These potential hazards become aggravated due to processing upsets, extreme physical conditions, and the accidental release of flammable hydrocarbons, causing major incidents [5]. The statistical data show that the frequency of fire and explosive hazards in petroleum refineries is increasing [6]. Equipment failure and the lack of a safety culture are the primary reasons for incidents. A catastrophic incident also leads to fire and explosions [7]. In a refinery complex, the gasoline hydrotreater unit (GTU), continuous catalytic reforming unit (CCR), methyl–tertiary–butyl–ether unit (MTBE), crude oil distillation unit (CDU), kerosene merox unit (KMU), and aromatic recovery unit (ARU) are the central functional units causing fire and explosive hazards [8]. Petroleum refineries have good safety regulations, standards, codes of practice, checklists, and operational protocols that provide multiple levels of protection against incidents [9]. However, despite that, the industry is still vulnerable to significant accidents such as fires, explosions, and the dispersion of toxic and flammable substances [10,11]. The Indian Oil Corporation Vadodara accident revealed the need to clearly understand the various types of accidents in oil installations and the importance of emergency operation centers [12]. There have been several global incidents of fires and explosions in the oil and gas industry, and there are various case studies and reports on accidents in this industry [13,14]. Emergencies arise due to causes such as explosions in a processing plant, accidents in storage facilities, during transportation, or due to technological failures, inappropriate or inadequate plant safety design, arson, malicious acts, human error, and natural calamities. Therefore, it is essential to investigate petroleum refineries’ worst possible safety hazards, evaluate their impact on people, property, plants, and society, and devise risk-control strategies [15].
Quantitative risk assessment in the processing industry is vital to protect workers and the environment [16]. Consequence analysis is a part of risk assessment, and its qualitative and quantitative methodologies help in estimating significant losses in the processing industry [17]. The accidental release of toxic gases, fire, explosions, and the dispersion of combustion products are some hazardous scenarios in processing industries [18]. Each scenario and its consequences could be calculated using modeling software such as CFD, ALOHA [19], WISER, and PHAST. Thermal radiation models estimate the amount of released heat due to explosions [20]. The results of consequence analysis are used to prepare emergency response plans [21]. Vapor cloud explosions in refineries and LPG plants are more severe in low-wind conditions as the vapor travels due to gravity [22].
Historical data availability is influenced by developing data-driven techniques in processing safety from fault identification to risk assessment. Past risk-assessment methods depended on steady-state statistical data to predict failure occurrence, causes, and consequences in the processing industry [23].
ML methods in risk assessment from corrosion cracking are applied with big data to integrate the Internet of Things [24]. Several soft-computing techniques and their application in safety evaluation were suggested for strategic decision-making in hazard management [25].
The critical decision-making parameter is effective for risk analysis measures regarding the worst-case chemical explosion failure in processing industries. Multimodal failure in industrial processes is difficult with a single solution at steady-state conditions. Therefore, several local data-driven strategies with different algorithms for the dynamic failure of processing components are used in pipeline incidents [26]. A multimodal machine-learning framework for process modeling with feature selection was implemented in the distillation column and achieved acceptable results. Potential hazardous chemicals in drinking water were predicted with two different supervised and unsupervised ML models. The prediction results were compared and had less than 5% false detection [27]. A multilayer perceptron neural network was developed with an accuracy of 84% for historical oil and gas pipeline data [28]. A comprehensive review of machine-learning approaches for oil and gas pipeline failure predictions was also conducted [29].
A new ANN framework, the operable adaptive sparse identification of systems (OASIS), was developed for chemical processing and successfully applied to reactor-separation risk analysis [30]. Risk assessment using the OASIS of an oil and refining plant was conducted from incidents on the basis of atmospheric and internal conditions and fault prediction.
An ANN framework using a Bayesian algorithm was proposed to predict faults of and assess the risk from corroded pipeline incidents [31]. Predictive analysis was performed on the LNG bunker price using a recurrent neural network, long short-term memory (LSTM), and gated recurrent methods. LSTM best suits highly volatile data in fault forecasting [32].
The literature shows that fire and explosion analyses were performed for storage tanks but not other critical process units of the refinery. In this context, the authors were motivated to conduct the present research on fire and explosion hazard analysis for the essential units of the refinery. The present study addresses the gap and focuses on analyzing worst-case fire and explosion scenarios in a petroleum refinery, which may occur during its operational life due to mechanical failures, sabotage, or natural calamities.
The present study aims to carry out the fire and explosion hazard analysis for critical plants of the refinery and to estimate the probable fire and explosion threats that will seriously impact people in the refinery, plant, equipment, and the surrounding community. The critical plants considered for the analysis are the gasoline hydrotreater unit (GTU), continuous catalytic reforming unit (CCR), methyl–tertiary–butyl–ether unit (MTBE), crude oil distillation unit (CDU), aromatic recovery unit (ARU), and kerosene merox unit (KMU). Additionally, different failure scenarios in the refinery plant and the measured threat zones (probably red, orange, and yellow) based on the distance traveled by the explosion of hazardous chemicals data were collected from NSCI (the National Safety Council India). The influenced parameters (wind speed, air temperature, and relative humidity) are considered to spread the exploded chemicals. This work aims to develop a TRANCE model for predicting threat- and risk-zone (red, orange, and yellow) distance with real-time industrial data when multimode failures occur at different process units in a refinery plant.

2. Materials and Methods

Industrial data from a large petroleum refinery complex in western India is used for the study. The data include process flow diagrams, process information, and regional meteorological data on the refinery’s location. The refinery is on around 500 acres of land surrounded by industrial estates, residential colonies, and other commercial establishments. This refinery has a present refining capacity of about 12 MMTPA. The main products produced in the refinery are naphtha, motor spirit, hexane, benzene, toluene, mineral turpentine oil, kerosene, jet fuel, diesel, light diesel oil, fuel oil, LPG, and bitumen. A systematic hazard-identification technique HAZID was followed in this study. The entire plants and pipelines in the refinery under this study are divided into several isolated sections. Each section was examined to identify the likely failure scenario regarding the leak size, the quantity of hazardous material contained in the isolatable section, and the estimated duration of the leak. Sixteen potential worst-case scenarios (seven each in summer daytime and winter night-time) were considered to perform risk analysis with the ALOHA hazard modeling tool and are given in Table 1.

TRANCE Topology

Soft-computing techniques via ANNs, FL (fuzzy logic), ML (machine learning), PR (probabilistic reasoning), and EA (evolutionary algorithms) have proved incredible efficiency for problem-solving in a variety of applications, specifically in risk assessments to support decision-making processes in an economical and time-saving manner. ANN is the most promising method to build a strong relationship between processes’ input and output features. ANN can be used in numerous applications such as system identification, modeling, process control, fault identification, and risk assessment. The intellectual task performance of ANN is similar to brain neurons by gathering and storing knowledge during the training stage. ANN is noteworthy in developing a surrogate process model of a real one. Perceptron artificial nodes could correlate and create a new response for non-linear and complex process data. A feed-forward multi-layered neural network is preferable for chemical process applications. A threat and risk artificial neural network for chemical explosion (TRANCE) incident structure was developed to transfer signals from input to output features. The performance capability of the TRANCE structure depends on located hidden layers/neurons between these input–output features based on activation and summation functions. The training algorithm and activation function are the key parameters that affect the TRANCE structure performance. The optimized TRANCE structure for refinery incidents is represented in Figure 1. This optimized structure has three input features, a hidden layer with five neurons of the sigmoid activation function, and an output layer with three neurons for three output features of the linear activation function. Levenberg–Marquardt (LM) algorithms are used to train the learning process of TRANCE for minimum error value. LM will develop a pattern by adjusting the weights and biases between input and output features instead of memorizing large datasets. LM is a trainlm back-propagation algorithm, which deals with large datasets with high convergence speed for ANN and optimizes the features iteratively until the predicted values reach an acceptable level. The specifications and overview of the proposed model information are provided in Table 2. The proposed network was applied to predict the safety distance of the refinery dataset generated from ALOHA. We examined the proposed framework capability with multi-failure scenarios over the single fault prediction done earlier. The worst-case scenarios identified from the HAZID and the affected weather conditions were applied to the ALOHA to generate safety distance measures. The effect of chemical explosion distance based on the weather conditions, air temperature, relative humidity, wind direction, wind speed, etc., were considered to measure the vicinity of hazardous chemicals. The chemical distances are classified as red, orange, and yellow zone depending on chemical concentration and the severe risk to humans and/or the environment. One hundred and sixty (160) different weather conditions were applied, and the safety distances were measured. The noise and error in datasets were clustered, and unnecessary data were removed. Datasets were divided for training, validation, and testing purposes as 70%, 15%, and 15% of the total, respectively. The training data were applied to the TRANCE model to train the model and to establish a correlation between the input and target signals. The input layer TRANCE model has three features (air temperature, relative humidity, and wind speed) that are significant effects of a chemical explosion. The output layer of TRANCE has three targets, namely red, orange, and yellow threat zones, for risk assessment. The validation datasets were used to evaluate the trained model and tune the mandatory parameters to attain the high-performance accuracy of the TRANCE model.

3. Process Description

The selected large-petroleum refinery is in western India. The Google map image of the refinery is given in Figure 2. The refinery is situated on around 500 acres of land, surrounded by industrial estates, residential colonies, and other commercial establishments. This refinery has a present refining capacity of around 12 MMTPA. The main products made in the refinery are naphtha, motor spirit, hexane, benzene, toluene, MTO, diesel, light diesel oil, fuel oil, kerosene, jet fuel, LSHS, LPG, and bitumen. The refinery process operations are illustrated in the block flow diagram in Figure 3. The major process plants of the refinery considered for this study are crude-distillation units (CDU-1 and 2), gasoline hydro treatment units (GTU), kerosene (KMU) units, continuous catalytic regeneration unit (CCR), and methyl–tert–butyl–ether (MTBE) unit.
The crude-distillation unit (CDU) is the first plant in the refinery. After desalting, the crude feedstock is preheated using recovered process heat. The feedstock then passes through a direct-fired crude-charge heater and later into the vertical distillation column just above the bottom at a pressure slightly higher than atmospheric pressure and temperatures of about 350–370 °C. Except for the heaviest fractions, all other fractions flash into vapor. As the hot vapor rises through the tower, its temperature decreases. Heavy fuels and asphalt residues are removed from the bottom. The major products, including lubricating oil, heating oil, kerosene, gasoline, and uncondensed gases, are drawn from the successively higher points of the distillation tower.
The topped crude withdrawn from the bottom of the CDU is heated to temperatures ranging from 370 to 420 °C in the vacuum-distillation unit (VDU). The heated topped crude is flashed into a multi-tray vacuum-distillation column which operates at an absolute pressure ranging from 350 to 1400 kg/m2. The catalytic-cracking unit (CCU) uses heat, pressure, and catalysts to convert heavy oils into lighter products, such as more valuable gasoline and distillate blending components. Gas oils from atmospheric distillation, vacuum distillation, coking, and de-asphalting processes are generally used as feedstocks. These feedstocks have a boiling point ranging from 340 to 540 °C.
The isomerization unit (ISOM) converts n-butane, n-pentane, and n-hexane into their respective iso-paraffins with extremely high-octane numbers. The isomerization unit is critical while converting the n-butane to isobutane and provides additional feedstock for the alkylation unit. Additionally, to convert the normal pentanes and hexanes into higher-branched isomers for gasoline blending. The isomerization process is similar to catalytic reforming, where-in the hydrocarbon molecules are rearranged. However, unlike catalytic reforming, isomerization only converts the normal paraffin into iso-paraffins. The naphtha hydrotreater unit (NHT) with hydrogen is used to desulfurize the naphtha fraction from the crude oil distillation and other units in the refinery. The gasoline hydrotreater unit (GTU) employs a catalytic chemical process to remove sulfur (S) from gasoline to reduce the sulfur dioxide (SO2) emissions that result from using gasoline while maintaining its octane value. GTU uses hydrogen to desulfurize some other distilled fractions removed from the crude oil distillation unit.
The refinery’s kerosene merox unit (KMU) desulfurizes the kerosene and jet fuel by oxidizing undesired mercaptans to organic disulfides. A fluid catalytic-cracking unit called FCCU converts the heavier, higher-boiling fractions from the crude distillation unit to lighter, lower-boiling-point fractions. In the hydrocracker unit (HCU), hydrogen converts heavier fractions from the crude oil and vacuum-distillation units into more lightweight products. The aromatic recovery unit (ARU) removes benzene and toluene from naphtha in the refinery.

4. Results and Discussions

Sixteen (eight each in daytime and night-time) hazardous leak scenarios at significant process units in the refinery were modeled to analyze the threat zones. The leakages considered are hydrogen gas at CCR and GTU, gasoline at GTU, methanol at MTBE, crude oil at CDU1, naphtha at CDU2, toluene at ARU, and kerosene leak at the KMU plant. The ALOHA modeling program output results for each leakage scenario for different process plants of the refinery are analyzed. The threat zones are categorized into red, orange, and yellow based on the leaked chemical concentrations. ALOHA’s default toxic level of concern values is considered to decide the type of threat zone in all scenarios. Stability class, wind speed, leak diameter, spill area, chemical release duration, and gas density are significant variables for the threat zones. Among them, the leak diameter plays a crucial role because chemical release quantity is a function of the leak diameter section.

4.1. Flammable Area of Vapour Cloud

The worst hazard levels due to the formation of a flammable area of vapor cloud during a summer day and winter night-time for process plants were plotted in Figure 4. The vapor cloud’s threat zones extend beyond refinery battery limits. For the hydrogen gas leak at GTU during the winter night, the vapor cloud spread up to 4000 m. The flammable vapor cloud for a gasoline leak at GTU spread up to 827 m, and for a crude oil leak at CDU1 spread to 1300 m during daytime summer. Results indicate that the winter night-time threat zones are slightly more significant than the summer daytime threat zones. Compared to the other process units, the hydrogen leak at GTU spreads more area, and there is a scope for vapor cloud explosion if the escaping gas ignites immediately.

4.2. Jet Fire from Thermal Radiation

Lethal thermal radiations for 37.5 kW/m2 intensity from a jet fire can be observed in Figure 5. The worst-case hazard results for the scenario of full-bore rupture (FBR) indicate that in case of hydrogen leak at the GTU process plant, jet fire from thermal radiation reached a maximum distance of 32 m during the winter night-time. These thermal radiations may cause 100% fatalities, significant damage to the plant, equipment, buildings, and property, and also causes first-degree burns.

4.3. Fireball Thermal Radiation

Fireballs radiations of 37.5 kW/m2 intensity for the process units are provided in Figure 6. This can cause the spread of fire by domino effect to other areas. Figure 6 shows the worst-case hazard zone with a maximum threat-zone distance of 83 m for the naphtha leak at CDU2.

4.4. Blast Force from Vapor Cloud Explosion

The overpressure blast effect due to the vapor cloud explosion will destroy buildings and cause severe injuries to the people residing at nearby places. Figure 7 shows that the impact of blast force will be severe during the winter night with a pressure of 8 psi magnitude that may reach up to 4300 m causing the destruction of buildings. Severe injuries will be caused to people within a distance of 4300 m in the downwind direction.
Compared to the hazards created by the other process plants, the catastrophic rupture of a large hydrogen pipeline at GTU is the most significant hazard in the refinery. It will significantly impact if the incident occurs during the winter night.

4.5. Performance of the TRANCE Model

The separated datasets with 112,24,24 for train, validation, and testing were iterated to achieve the minimum error value of the TRANCE model. In Figure 8, the learning rate of the proposed model was validated with MSE values of training, validation, and testing of TRANCE model datasets presented. Train, test, and validation curves show a minimum error of 282.82 at epoch 9 of total epochs 15. The curves have similar behavior from epoch 9 steadily without overfitting or underfitting. Thus, the TRANCE model for risk assessment was established successfully. The performance of the TRANCE model can also be evaluated from another essential parameter called the coefficient of determination (R2). The mathematical expressions for MSE and R2 are represented as follows [33]:
M S E = 1 N i = 1 N ( y T R A N C E , i y e x p e r i m e n t a l , i ) 2
R 2 = 1 i = 1 N ( y T R A N C E , i y e x p e r i m e n t a l , i ) i = 1 N ( y T R A N C E , i y m )
Here, y T R A N C E , i represents prediction values from the proposed model, y e x p e r i m e n t a l is true values or experimental target features and y m is the mean of all target values.
Error values between target and TRANCE outputs are mentioned in Figure 9, with the datasets on the vertical axis and the error on the horizontal axis. The error value displayed a linear change trend in testing, validation, and training datasets. It demonstrates that most of the data’s accumulated peak values are near zero and reveal the superior prediction strength of the TRANCE model.
A TRANCE-based model is developed by utilizing preprocessed industrial dataset of refinery plant failure scenarios. The dataset preprocessing is performed using Matlab 2022a with a computational time of 30 min on a 5 Core computer with 8 GB RAM. The data sets were trained and validated through this TRANCE model, and the error values are acceptable. Testing has been done to predict the threat-zone response from the TRANCE model for the new datasets, and the predicted values are collected. The obtained error results in all phases are presented in Table 3. In Figure 10, the accuracy of the TRANCE model in the training and validation phase is quite good, with the coefficient of determination values of 0.9998 and 0.998, and the error values (MSE) are 180.70 and 282.82, respectively, which is higher than the SVM-based risk assessment of rock bolts failure in coal mines [34].
The overall training, testing, and validation datasets are presented in Figure 11, and three zones are highlighted. This Figure is evidence of the high-precision performance of the TRANCE model in all three output features, orange and yellow zones, which are highlighted with respective colored circles. Similarly, Figure 11 represents the testing ability of the TRANCE model with high precision with the evidence majority of predicted values from all three threat zones lie on a linear regression line with an R2 value of 0.9994 and MSE value of 679.53. Data-driven model-based corrosion risk-assessment prediction from a gradient-boosting algorithm has less efficiency of R2 value of 0.98 compared with the TRANCE model [35].

4.6. Limitations of the TRANCE Model

A large modern refinery in India was considered for the pilot study. The rudimentary refining processes in petroleum refineries in the Indian context are similar. This study focuses on predicting the hazardous-release threat zones in case of release from various process units in the refinery. Therefore, the model will be helpful for other refineries as well. Industrial data from a large petroleum refinery complex in western India were used for the study. The failure scenarios data were used as input for the ALOHA. The TRANCE model dataset is generated from the weather conditions of the location and failure scenarios of the refinery together. In this context, authors can generate datasets by changing weather conditions in ALOHA based on their plant location. The same data can apply to this TRANCE model, but this proposed accuracy may vary. This model can change or modify the hyperparameters such as training algorithm, number of neurons, layer size, epoch size, and dataset splitting to get accurate predictions from the model.

5. Conclusions

The consequence modeling of a real-time petroleum refinery was carried out to analyze the fire and explosion hazards present at various key process plants and are considered for the analysis of worst-case fire and explosion scenarios. Hazard-modeling results indicated that the hydrogen leak at GTU resulted in a maximum threat-zone distance of 4000 m for a flammable area of vapor cloud at night-time. Jet fire for the same plant with thermal radiation intensity of 37.5 kW/m2 reached a distance of 32 m and damaged nearby pipelines or vessels. Fireball radiations of thermal radiation intensity 37.5 kW/m2 for the leak of naphtha at CDU2 reached 83 m distance, and it will cause the spread of fire by domino effect to other areas. Jet fire thermal radiation of intensity 12.5 kW/m2 reached a maximum of 81 m from the source of fire for the naphtha leak at the CDU2 process plant. This intensity of the radiation may result in fire ignition. The flammable vapor clouds of hazardous leaks can cross the battery limits of the refinery and cause a threat to the neighboring society.
The suggested TRANCE computing model has been validated with high accuracy of 99.98% to predict the distance traveled by the chemical explosion of refinery incidents. The prediction results show that the coefficient of determination values in training, validation, and testing are 0.9998, 0.9998, and 0.9994, respectively. The mean absolute error values of the TRANCE model are 180.7024, 282.8242, and 679.5343. These highly accurate results are evident in the reliability of the proposed framework for safety distance prediction compared to the past results. Refinery management can adopt this model to assess the safety distance from the hazardous chemical explosion based on the prior forecasted-atmosphere conditions from the weather department. The proposed model of this research helps predict the threat zones and prepare the appropriate emergency response plan for the refinery. The results will also help plant designers plan the safety distances while siting the plants and selecting the construction material for the refinery plant and equipment to avoid a domino effect due to thermal radiation impingement. It will provide useful information to the regulatory authorities to prescribe mandatory safety distances, insurers, and disaster management authorities to prepare emergency plans.

Author Contributions

L.R.G., Conceptualisation, methodology, validation, formal analysis, investigation, resources, writing—original draft preparation, visualization. N.K. is data curation and supervision. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data used in this work is incorporated in the paper itself.

Acknowledgments

Authors are grateful to NSCI and the Vellore Institute of Technology, Vellore, Tamil Nadu, INDIA, for their support.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Saloua, B.; Mounira, R.; Salah, M.M. Fire and Explosion Risks in Petrochemical Plant: Assessment, Modeling and Consequences Analysis. J. Fail. Anal. Prev. 2019, 19, 903–916. [Google Scholar] [CrossRef]
  2. Tauseef, S.M.; Abbasi, T.; Pompapathi, V.; Abbasi, S.A. Case Studies of 28 Major Accidents of Fires/Explosions in Storage Tank Farms in the Backdrop of Available Codes/Standards/Models for Safely Configuring Such Tank Farms. Process Saf. Environ. Prot. 2018, 120, 331–338. [Google Scholar] [CrossRef]
  3. Jayakumar, C.; Isac, S.; Prasad, D.M.R. Emergency Response Plan for Methane and Chlorine with Dispersion Modelling Using CAMEO. Int. J. Occup. Saf. Ergon. 2021, 28, 1802–1810. [Google Scholar] [CrossRef] [PubMed]
  4. Samia, C.; Hamzi, R.; Chebila, M. Contribution of the Lessons Learned from Oil Refining Accidents to the Industrial Risks Assessment. Manag. Environ. Qual. Int. J. 2018, 29, 643–665. [Google Scholar] [CrossRef]
  5. Freeman, R.A. CCPS Guidelines for Chemical Process Quantitative Risk Analysis. Plant/Oper. Prog. 1990, 9, 231–235. [Google Scholar] [CrossRef]
  6. Chettouh, S.; Hamzi, R.; Benaroua, K. Examination of Fire and Related Accidents in Skikda Oil Refinery for the Period 2002–2013. J. Loss Prev. Process Ind. 2016, 41, 186–193. [Google Scholar] [CrossRef]
  7. Dobashi, R. Fire and Explosion Disasters Occurred Due to the Great East Japan Earthquake (11 March 2011). J. Loss Prev. Process Ind. 2014, 31, 121–126. [Google Scholar] [CrossRef]
  8. Fahim, M.A.; Alsahhaf, T.A.; Elkilani, A. Safety in Petroleum Refineries. In Fundamentals of Petroleum Refining; Elsevier: Amsterdam, The Netherlands, 2010; pp. 357–376. [Google Scholar] [CrossRef]
  9. Bi, H.; Xie, X.; Wang, K.; Cao, Y.; Shao, H. A Risk Assessment Methodology of Aluminum Dust Explosion for Polishing Process Based on Laboratory Tests. Proc. Inst. Mech. Eng. O J. Risk Reliab. 2021, 235, 627–636. [Google Scholar] [CrossRef]
  10. Timbang, A.; Reddy Prasad, D.M.; Bin Azri, M.H.; Sofri, S. Addressing Process Safety Challenges in Downstream Industries in Brunei Darussalam. Int. J. Occup. Saf. Ergon. 2021, 28, 2284–2292. [Google Scholar] [CrossRef]
  11. Sofri, S.; Reddy Prasad, D.M.; Bin Azri, M.H.; Timbang, A. Analysis of Chemical Engineering Curriculum to Improve Process Safety Competency. Int. J. Occup. Saf. Ergon. 2022, 29, 642–650. [Google Scholar] [CrossRef]
  12. Gupta, K. Design of District Emergency Operations Centres, and the Case Study of Indian Oil Corporation Jaipur Depot Explosion. Int. J. Emerg. Manag. 2010, 7, 221–232. [Google Scholar] [CrossRef] [Green Version]
  13. Editors, G.; Jaromír Klemeš, J.; Yen Liew, P.; Shin Ho, W.; Shiun Lim, J.; Tarmimi Abu Bakar, H.; Han Siong, P.; Koy Yan, C.; Kidam, K.; Wijayanuddin Ali, M.; et al. Analysis of Main Accident Contributor according to Process Safety Management Elements Failure. Chem. Eng. Trans. 2017, 56, 991–996. [Google Scholar] [CrossRef]
  14. Sesseng, C.; Storesund, K.; Steen-Hansen, A. Analysis of 985 Fire Incidents Related to Oil-and Gas Production on the Norwegian Continental Shelf. In Safety and Reliability—Safe Societies in a Changing World, Proceedings of the 28th International European Safety and Reliability Conference—ESREL 2018, Trondheim, Norway, 17–21 June 2018; Taylor & Francis Group: London, UK, 2018; pp. 1847–1854. [Google Scholar] [CrossRef]
  15. Kriaa, S.; Bouissou, M.; Laarouchi, Y. A New Safety and Security Risk Analysis Framework for Industrial Control Systems. Proc. Inst. Mech. Eng. O J. Risk Reliab. 2019, 233, 151–174. [Google Scholar] [CrossRef]
  16. Tang, Y.; Jing, J.; Zhang, Z.; Yang, Y. A Quantitative Risk Analysis Method for the High Hazard Mechanical System in Petroleum and Petrochemical Industry. Energies 2017, 11, 14. [Google Scholar] [CrossRef] [Green Version]
  17. Yoon, H.; Park, J.; Lim, W.; Lee, K.; Choi, N.; Lee, C.; Moon, I. Integration of Qualitative and Quantitative Risk Assessment Methods for Gas Refinery Plants. Korean J. Chem. Eng. 2013, 30, 1368–1374. [Google Scholar] [CrossRef]
  18. Iskender, H. Risk Assessment for an Acetone Storage Tank in a Chemical Plant in Istanbul, Turkey: Simulation of Dangerous Scenarios. Process Saf. Prog. 2021, 40, 234–239. [Google Scholar] [CrossRef]
  19. Terzioglu, L.; Iskender, H. Modeling the Consequences of Gas Leakage and Explosion Fire in Liquefied Petroleum Gas Storage Tank in Istanbul Technical University, Maslak Campus. Process Saf. Prog. 2021, 40, 319–326. [Google Scholar] [CrossRef]
  20. Lobato, J.; Rodríguez, J.F.; Jiménez, C.; Llanos, J.; Nieto-Márquez, A.; Inarejos, A.M. Consequence Analysis of an Explosion by Simple Models: Texas Refinery Gasoline Explosion Case. Afinidad J. Chem. Eng. Theor. Appl. Chem. 2009, 66, 543. [Google Scholar] [CrossRef]
  21. Movahed, A.; Jahani, F.; Parvini, M.; Shakib, M. Consequence Analysis of Gas Condensate Leakage in a Gas Refinery to Develop an Emergency Response Plan. J. Occup. Hyg. Eng. Vol. 2019, 6, 1–8. [Google Scholar]
  22. Atkinson, G.; Cowpe, E.; Halliday, J.; Painter, D. A Review of Very Large Vapour Cloud Explosions: Cloud Formation and Explosion Severity. J. Loss Prev. Process Ind. 2017, 48, 367–375. [Google Scholar] [CrossRef]
  23. Choi, Y.; Bhadriaju, B.; Cho, H.; Lim, J.; Han, I.S.; Moon, I.; Kwon, J.S.I.; Kim, J. Data-Driven Modeling of Multimode Chemical Process: Validation with a Real-World Distillation Column. Chem. Eng. J. 2023, 457, 141025. [Google Scholar] [CrossRef]
  24. Alamri, A.H. Application of Machine Learning to Stress Corrosion Cracking Risk Assessment. Egypt. J. Pet. 2022, 31, 11–21. [Google Scholar] [CrossRef]
  25. Harirchian, E.; Aghakouchaki Hosseini, S.E.; Jadhav, K.; Kumari, V.; Rasulzade, S.; Işık, E.; Wasif, M.; Lahmer, T. A Review on Application of Soft Computing Techniques for the Rapid Visual Safety Evaluation and Damage Classification of Existing Buildings. J. Build. Eng. 2021, 43, 102536. [Google Scholar] [CrossRef]
  26. Kumari, P.; Wang, Q.; Khan, F.; Kwon, J.S.I. A Unified Causation Prediction Model for Aboveground Onshore Oil and Refined Product Pipeline Incidents Using Artificial Neural Network. Chem. Eng. Res. Des. 2022, 187, 529–540. [Google Scholar] [CrossRef]
  27. Samanipour, S.; Kaserzon, S.; Vijayasarathy, S.; Jiang, H.; Choi, P.; Reid, M.J.; Mueller, J.F.; Thomas, K.V. Machine Learning Combined with Non-Targeted LC-HRMS Analysis for a Risk Warning System of Chemical Hazards in Drinking Water: A Proof of Concept. Talanta 2019, 195, 426–432. [Google Scholar] [CrossRef]
  28. Elshaboury, N.; Al-Sakkaf, A.; Alfalah, G.; Abdelkader, E.M. Data-Driven Models for Forecasting Failure Modes in Oil and Gas Pipes. Processes 2022, 10, 400. [Google Scholar] [CrossRef]
  29. Zakikhani, K.; Nasiri, F.; Zayed, T. A Review of Failure Prediction Models for Oil and Gas Pipelines. J. Pipeline Syst. Eng. Pract. 2020, 11, 03119001. [Google Scholar] [CrossRef]
  30. Kumari, P.; Halim, S.Z.; Kwon, J.S.I.; Quddus, N. An Integrated Risk Prediction Model for Corrosion-Induced Pipeline Incidents Using Artificial Neural Network and Bayesian Analysis. Process Saf. Environ. Prot. 2022, 167, 34–44. [Google Scholar] [CrossRef]
  31. Bhadriraju, B.; Kwon, J.S.I.; Khan, F. OASIS-P: Operable Adaptive Sparse Identification of Systems for Fault Prognosis of Chemical Processes. J. Process Control 2021, 107, 114–126. [Google Scholar] [CrossRef]
  32. Kim, K.; Lim, S.; Lee, C.H.; Lee, W.J.; Jeon, H.; Jung, J.; Jung, D. Forecasting Liquefied Natural Gas Bunker Prices Using Artificial Neural Network for Procurement Management. J. Mar. Sci. Eng. 2022, 10, 1814. [Google Scholar] [CrossRef]
  33. Gollangi, R.; Nagamalleswara Rao, K. Energetic, Exergetic Analysis and Machine Learning of Methane Chlorination Process for Methyl Chloride Production. Energy Environ. 2022, 0958305X221109604. [Google Scholar] [CrossRef]
  34. Jiang, P.; Craig, P.; Crosky, A.; Maghrebi, M.; Canbulat, I.; Saydam, S. Risk Assessment of Failure of Rock Bolts in Underground Coal Mines Using Support Vector Machines. Appl. Stoch. Models Bus. Ind. 2018, 34, 293–304. [Google Scholar] [CrossRef]
  35. Ossai, C.I. A Data-Driven Machine Learning Approach for Corrosion Risk Assessment—A Comparative Study. Big Data Cogn. Comput. 2019, 3, 28. [Google Scholar] [CrossRef] [Green Version]
Figure 1. TRANCE framework development methodology.
Figure 1. TRANCE framework development methodology.
Toxics 11 00350 g001
Figure 2. Satellite image of a petroleum refinery.
Figure 2. Satellite image of a petroleum refinery.
Toxics 11 00350 g002
Figure 3. Block flow diagram of refinery processes.
Figure 3. Block flow diagram of refinery processes.
Toxics 11 00350 g003
Figure 4. Threat zones for a flammable area of the vapor cloud.
Figure 4. Threat zones for a flammable area of the vapor cloud.
Toxics 11 00350 g004
Figure 5. Threat zones for jet fire.
Figure 5. Threat zones for jet fire.
Toxics 11 00350 g005
Figure 6. Threat zones for fireball radiations.
Figure 6. Threat zones for fireball radiations.
Toxics 11 00350 g006
Figure 7. Threat zones for blast force from vapor cloud explosion.
Figure 7. Threat zones for blast force from vapor cloud explosion.
Toxics 11 00350 g007
Figure 8. Leaning rate curve for TRANCE model performance with training, validation, and testing datasets.
Figure 8. Leaning rate curve for TRANCE model performance with training, validation, and testing datasets.
Toxics 11 00350 g008
Figure 9. Error histogram plot of training, validation, and testing datasets.
Figure 9. Error histogram plot of training, validation, and testing datasets.
Toxics 11 00350 g009
Figure 10. The correlation coefficient for three threat zone prediction values from the TRANCE model in training and validation.
Figure 10. The correlation coefficient for three threat zone prediction values from the TRANCE model in training and validation.
Toxics 11 00350 g010
Figure 11. The correlation coefficient for three threat zone prediction values from the TRANCE model in testing and all together.
Figure 11. The correlation coefficient for three threat zone prediction values from the TRANCE model in testing and all together.
Toxics 11 00350 g011
Table 1. The list of worst-case hazardous scenarios.
Table 1. The list of worst-case hazardous scenarios.
SNo.Plant LocationWeather ConditionSituationLeak SizePipe DiaPipe Length
1GTU5DCatastrophic rupture of hydrogen pipeline in the summer daytimeFBR300 mm160 m
2GTU 1.5FCatastrophic rupture of hydrogen pipeline in the winter night-timeFBR300 mm160 m
3CCR 5DThe leak of hydrogen pipeline in the summer daytime150 mm450 mm137 m
4CCR s1.5 FThe leak of hydrogen pipeline in the winter night-time150 mm450 mm137 m
5GTU5 DThe leak of gasoline pipeline in the summer daytime150 mm450 mm250 m
6GTU1.5 FThe leak of gasoline pipeline in the winter night-time150 mm150 mm250 m
7KMU5 DCatastrophic rupture of kerosene pipeline in the summer daytimeFBR450 mm10 m
8KMU1.5 FCatastrophic rupture of kerosene pipeline in the winter night-timeFBR450 mm10 m
9MTBE5 DThe leak of methanol pipeline in the summer daytime150 mm150 mm30 m
10MTBE1.5 FThe leak of methanol pipeline in the winter night-time150 mm150 mm30 m
11CDU15 DThe leak of crude pipeline (IS2) in the summer daytimeFBR450 mm800 m
12CDU11.5 FThe leak of crude pipeline (IS2) in the winter night-timeFBR450 mm800 m
13CDU25DThe leak of naphtha pipeline (IS2) in the summer daytimeFBR450 mm30 m
14CDU21.5FThe leak of naphtha pipeline (IS2) in the winter night-timeFBR450 mm30 m
15ARU5 DThe leak of toluene in the summer daytime150 mm150 mm10 m
16ARU1.5 FThe leak of toluene in the winter night-time150 mm150 mm10 m
Table 2. Overview of the TRANCE model.
Table 2. Overview of the TRANCE model.
ParticularsSpecifications
Number of neurons in first and second hidden layer5, 3
Number of features in input and output3, 3
Training algorithmFeed-forward back propagation
Optimization algorithmTrainlm (Levenberg-Marquardt)
Activation functionSigmoid and Linear
Performance-evaluation functionMSE, R2
Minimum number of epochs15
Number of scenarios considered for risk assessment163
Table 3. Performance of the developed TRANCE model for three zones.
Table 3. Performance of the developed TRANCE model for three zones.
ObservationsMSER2
Training113180.70240.9998
Validation24282.82420.9998
Testing 24679.53430.9994
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.

Share and Cite

MDPI and ACS Style

Gabhane, L.R.; Kanidarapu, N. Threat and Risk Analysis-Based Neural Network for a Chemical Explosion (TRANCE) Model to Predict Hazards in Petroleum Refinery. Toxics 2023, 11, 350. https://doi.org/10.3390/toxics11040350

AMA Style

Gabhane LR, Kanidarapu N. Threat and Risk Analysis-Based Neural Network for a Chemical Explosion (TRANCE) Model to Predict Hazards in Petroleum Refinery. Toxics. 2023; 11(4):350. https://doi.org/10.3390/toxics11040350

Chicago/Turabian Style

Gabhane, Lalit Rajaramji, and NagamalleswaraRao Kanidarapu. 2023. "Threat and Risk Analysis-Based Neural Network for a Chemical Explosion (TRANCE) Model to Predict Hazards in Petroleum Refinery" Toxics 11, no. 4: 350. https://doi.org/10.3390/toxics11040350

APA Style

Gabhane, L. R., & Kanidarapu, N. (2023). Threat and Risk Analysis-Based Neural Network for a Chemical Explosion (TRANCE) Model to Predict Hazards in Petroleum Refinery. Toxics, 11(4), 350. https://doi.org/10.3390/toxics11040350

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop