A bibliometric analysis on the relevancies of artificial neural networks (ANN) techniques in offshore engineering

Abstract

Recently, the use of artificial neural networks (ANN) in the offshore exploration and production industry to optimize decision-making and reduce costs and non-productive time have been increasing. Despite this trend, there have been only 11 reviews of ANN in offshore engineering published on the Scopus database. Therefore, this article aims to provide an update on the relevance of ANN in offshore engineering over the past 18 years (2005–2023) through a bibliometric analysis using Excel and VOS Viewer software. This analysis highlights the yearly increase in publications related to ANN implementations in offshore engineering and identifies the most cited publications, citation network analysis, authors, keywords, journals, institutions, and leading countries. The objective of this bibliometric analysis is to assist subsequent research and collaboration in this field by shedding light on ANN’s potential and identifying areas for further application. The identified cluster area publications encompass a range of topics, including drilling systems and the assessment of pipes. Furthermore, the significant fourfold increase in publications since 2005 indicates a growing interest among researchers in adapting ANN for various applications within this field. This could lead to further advancements, innovations, and improved solutions to promote collaboration and knowledge-sharing among researchers in this domain.

Keywords:

• artificial neural network
• offshore
• bibliometric
• VOS Viewer

1. Introduction

To imitate the structure and/or functionality of biological neural networks, scientists have developed computational and mathematical models called ANN. They can be applied to identify patterns in data or to model intricate connections between outputs as well as inputs. The foundations of ANN were extensively covered in the literature (Basheer & Hajmeer, 2000). It offers a basic knowledge of ANN and addresses the questions of when and why these computational tools are required, the inspiration for their creation, as well as their relationship to biological systems and other modeling techniques. Other than that, it also discusses the variety of ANN types and learning rules, design considerations, associated computations, implementation to real-world problems, as well as benefits and drawbacks. ANN study has its roots in the investigation by (McCulloch & Pitts, 1943), which addressed the distinction between nets equivalent and equivalent nets in the narrow sense as well as the precise use and significance of temporal research of nervous activity. The theory also pertains to mathematical biophysics by providing a tool for the rigorous symbolic treatment of known nets as well as a simple method of creating hypothetical nets with the necessary properties. Recently, numerous fields of engineering and science, including offshore engineering, manufacturing, electronics, automotive, aerospace, etc., have extensively employed ANN to model human behaviours (Rafiq et al., 2001).

In the oil and gas sector, complex issues have been overcome using ANN. Considering neural networks are designed to address complex issues that are not resolved using standard modeling tools owing to the problem’s complexity, they should not be employed to solve mathematical problems or problems that may be described analytically. Besides, referring to a number of case histories and papers in the offshore engineering literature, the ANN application in offshore engineering may be classified into four categories starting from exploration, drilling, production, and reservoir (Alkinani et al., 2019). Firstly, some instances of ANN implementations in exploration, which mostly include seismic data ANN applications. Utilising seismic data, (Aminzadeh & DeGroot, 2005) applied it to object detection and employed ANN to find numerous seismic objects. In order to identify the line pattern of the direct wave and the hyperbola pattern of the reflection wave in a seismogram, where K. Y. Huang et al. (2006) have implement using ANN. Employing a concrete instance from the Permian (Ross, 2017) proved how ANN may enhance the clarity and resolution of seismic data in tight sand with limited porosity, high bulk density, as well as velocity. For a very long time, drilling engineering has employed other ANN literature. In a different work, (Lind & Kabirova, 2014) investigated the prediction of drilling troubles utilising an ANN and a database of drilling parameters. The result agrees with outcomes of prior research by Elkatatny et al (Elkatatny et al., 2016), (Abdelgawad et al., 2018), and (Al-Azani et al., 2018), which performed experimental investigations on drilling fluid rheological properties as well as practising ANN to estimate drilling fluid rheological properties. Moreover, (Hoffimann et al., 2018). studied the drilling reports’ sentence classifications. According to their analysis, ANNs were utilised to create a method for automatically categorising sentences from drilling reports into three tables: Events, Symptoms, as well as Actions, using data from 303 wells.

Additionally, ANN has been employed in several petroleum production engineering applications. For example, (Nieto et al., 2018). extended this work to completion optimisation, which uses the ANN to strengthen the completion and preserve the parent well in the British Columbian Montney Formation. Nevertheless, there are several ANN implementations in reservoir engineering. Moreover, (Rashidi et al., 2018). present the application of elastic modulus along with adopting the ANN to compare the limestone formations’ dynamic as well as static moduli. Additionally, cite two Iranian formations, Asmari and Sarvak, as instances together with (Rashidi & Asadi, 2018) investigated the pore pressure estimation by applying the ANN to estimate formation pore pressure from drilling data.

Although several articles have been published about the application of ANN in offshore engineering, they have narrowed their attention to certain ANN method features. Additionally, 11 review papers regarding the utilisation of ANN in offshore engineering have only been published per the Scopus database. This document was created to present a summary of journals, the most productive nations, and institutions in the world in offshore engineering publications to assist subsequent research and collaboration, as well as to emphasise the absence of research in the field. Furthermore, to make obtaining data for subsequent research easier, the most frequent keywords from earlier research on ANN in offshore engineering were included in the existing paper. It was revealed that main contributions of this study in identification of key research trends which can help identify the main research areas within the intersection of ANN techniques and offshore engineering. It can highlight the specific applications of ANN in offshore engineering, such as structural analysis, risk assessment, operational optimization, or condition monitoring. This identification can provide a comprehensive overview of the field and its current focus. VOS Viewer software allows for the visualization of co-authorship and collaboration networks. By analysing author affiliations and collaboration patterns, the analysis can reveal clusters of researchers or institutions working on ANN techniques in offshore engineering. This mapping can highlight key contributors, collaborations, and research networks within the field.

Furthermore, this analysis can provide insights into publication trends, such as the number of publications, growth rate, and citation impact. By examining citation patterns, it becomes possible to identify influential papers, highly cited authors, and prominent journals in the field of ANN techniques in offshore engineering. This assessment helps understand the knowledge dissemination and impact within the research community. Either than that VOS Viewer software facilitates the visualization of co-occurring keywords and their evolution over time. By analysing the titles, abstracts, or keywords of publications, the analysis can identify the main research themes and how they have evolved within the field. This visualization can help understand the progression of research topics, emerging trends, and potential shifts in focus over different time periods.

This analysis can identify research gaps and areas that have received less attention within the field. By analysing the network visualization and keyword analysis, it becomes possible to identify potential research directions or underexplored topics. This information can guide researchers and decision-makers in identifying areas for future investigation or innovation. Additionally, this report provided estimations for how ANN techniques might be employed in offshore engineering research trends and provided several suggestions for additional research. In-depth exploration and comprehensive analysis of all these aspects will be thoroughly discussed in the results and discussion section.

2. Bibliometric analysis methods

This review was established, and bibliometric analysis was applied to achieve its objectives. VOS Viewer software is a bibliometric analysis and bibliographic map expert software designed by (Waltman & van Eck, 2010). After that, VOS Viewer software as a bibliometric assessment tool and Excel are utilized to map the scientific plotting of the field of research. Then, the results are discussed with possible research gaps and research direction. It was possible to describe present study trends regarding the relevant ANN techniques in offshore engineering by employing the software. This also facilitated researchers with a precise understanding to aid them in subsequent study and collaboration.

According to Figure 1, five phases were established to carry out the objectives of the present review. Utilising the Scopus database’s basic search option and the keywords (“artificial neural network*” OR “ann”) AND (offshore OR “oil and gas”) with a defined year range of “2005–2023,” papers were gathered within the first phase. Document gathering was conducted on 12 January 2023 (after the specified date, any document is not a part of the existing research). Consequently, phase 2 research on 1259 papers were downloaded and acquired from the Scopus database. The Scopus website was also employed to manually evaluate the county’s publishing output based on phase 2 first corresponding author address. To emphasise the scarcity of research and determine the top 15 productive nations, journals, as well as institutions on the ANN in offshore engineering publications, the downloaded materials were exported to VOS Viewer software and Excel in phase 3. Phase 3 additionally identified the top 20 most-cited academic papers as well as the top 20 most-utilised keywords. Nevertheless, the most significant documents were gathered relying on document citation in order to determine the literary trends (only the documents above 30 citations were selected). Besides, VOS Viewer software appears with 131 documents cited above 30 times and classified them into 12 clusters, explained in detail in Section 5. Overall, in detail in Section 5 establish that ANNs have found numerous applications in offshore engineering due to their capability to analyse complex and large datasets, model non-linear relationships, and make predictions.

Figure 1. Bibliometric review structure methodologies chart.

Some relevant uses include structural health monitoring which ANNs can analyse sensor data from offshore structures to detect damage or anomalies, enabling proactive maintenance and preventing failures. Then, as wave and current predictions. ANNs can be trained on historical oceanographic data to forecast wave heights, currents, and other environmental conditions. These predictions are vital for offshore operations, including vessel navigation, safety, and optimizing renewable energy installations. In addition, ANNs can assist in modeling and predicting fluid flow behavior in offshore oil and gas reservoirs. They can optimize production strategies, estimate reservoir properties, and guide drilling decisions. Additionally, it should be mentioned that ANNs can analyze historical data on offshore accidents, incidents, and operational parameters to assess and predict risk levels. This helps in designing safety protocols, optimizing emergency response plans, and minimizing potential hazards. In the final stage, the acquired tables and bibliography maps underwent thorough analysis and were subsequently examined and deliberated upon.

3. Results and discussion

3.1. Overview of the ANN in offshore engineering historical and geographical trends

As per the Scopus database, 1259 documents have been published in ANN in offshore engineering from 2005 to 12 January 2023. Figures 2, 3 illustrate the yearly publication’s historical trends as well as the total number of citations from 2005 to 2023. Figure 3 demonstrates that throughout the 18 years from 2005 to 2010, the number of yearly publications increased significantly while consistently growing. The number of publications remained largely constant; it began to rise in 2011 and then increased significantly in 2020. The yearly publishing of ANN techniques relevant to offshore engineering has increased by 7768.75%, surpassing the average growth rate of 409%. Additionally, from 2005 to 2016, the publication’s yearly citations grew, with minor variations, until marginally declining in 2017. This is to be anticipated as it requires some time for academic papers to possess a discernible impact since it requires a while for other people to read new research, pay attention to them, and incorporate them as references in other works. As a result, recently published research often has fewer citations. The overall number of ANN offshore engineering research articles includes contributions from 75 different nations. The 15 most productive nations in the ANN for offshore engineering research were listed in Table 1 according to the number of publications. With 17.63% of all publications over the previous 18 years in the field of ANN on offshore engineering, China has maintained its top ranking. With a total publishing rate of 16.92% in the literature of ANN in offshore engineering, the United States placed in second place, followed by Iran, the United Kingdom, and India with 9.53 %, 7.86 %, and 6.20 %, respectively. Other than that, more than 20 publications each came from Saudi Arabia, Brazil, Canada, Malaysia, Russian Federation, Norway, Australia, South Korea, Nigeria, and Taiwan. According to a review of the Scopus database utilising VOS Viewer software, the top 15 nations accounted for more than 90% of all publications and more than 79% of all citations between 2005 and 2023. Figure 4 demonstrates that the application of ANN is receiving a lot of interest in offshore engineering research, particularly in China, Iran, and the USA, where the use of a wide variety of advanced technologies and ANN elements is expanding quickly. This happens as technology advances, and thus does the establishment of artificial intelligence (AI). Computer scientists and business leaders are both at the forefront of AI, generating services and products in large quantities to enable the ground-breaking technology available to everyone.

Figure 2. Total publications based on the Scopus database from 2005 to 2023.

Figure 3. Sum of citations based on the Scopus database from 2005 to 2023.

Figure 4. Distribution of Top ten leading countries from 2005 to 2023.

Table 1. Distribution of Top 15 leading countries from 2005 to 2023 as per VOS Viewer

Country	TP	TP (%)	TLS	TC	Continent
China	222	17.63%	96	1897	Asia
United States	213	16.92%	111	2883	North America
Iran	120	9.53%	101	2424	Asia
United Kingdom	99	7.86%	62	1443	Europe
India	78	6.20%	50	955	Asia
Saudi Arabia	67	5.32%	46	723	Asia
Brazil	62	4.92%	46	516	South America
Canada	58	4.61%	89	1366	North America
Malaysia	55	4.37%	56	468	Asia
Russian Federation	45	3.57%	18	293	Europe
Norway	40	3.18%	10	305	Europe
Australia	39	3.10%	58	769	Oceania
South Korea	35	2.78%	30	387	Asia
Nigeria	28	2.22%	14	296	Africa
Taiwan	24	1.91%	5	221	Asia

TP; Total publication, TC; Total citation, TLS; Total link strength.

In contrast, relying on the initial corresponding author address given by the Scopus database, the overall 1259 articles were manually evaluated (piece by piece) to determine how many publications were published annually for each nation. The challenge in estimating the precise number of publications for each nation every year using the VOS Viewer software as well as the Scopus website led to the conduct of this study. The dispersion of publications for the top five dominating counties from 2005 to 2023 is depicted in Tables 2 and 3. Additionally, Figure 5 illustrates the yearly publishing total as well as the proportionate contributions from the top five countries, which are China, the United States, Iran, the United Kingdom, as well as India. Table 2 indicates that China maintained a continuously growing trend in paper publications, considerably surpassing other nations in 2005 and 2006. It continued along this trajectory until 2017, when it began to decline its yearly publishing share. This is due to the fact that the geographic dispersion of research articles on the subject of ANN in offshore engineering is expanding as well. For instance, only nine countries made a contribution to the total of 16 papers published in 2005 (the United States contributed 4 papers, China and the United Kingdom each contributed 3, Australia and India contributed 2, and the remaining countries each contributed only 1 paper). In contrast, in 2022, there were 49 countries and 170 papers, with China being the most productive with 55 papers (or 32.35% share). The central government’s desire to invest in technology and science, as well as China’s growing academic labour market, are the key causes of this astounding increase in ANN-related research contributions from China (Barton et al., 2017). (Figures 6, 7) illustrate that the majority of ANN-related papers were released in 2022, with 55 articles accounting for 32.35% of all ANN-related publications in offshore engineering. Additionally, 2020 shares the second-highest number of articles with 39 publications and a share (10%) of all publications from (2005 to 2023). Last but not least, (Figures 6,7) demonstrate that from 2005 to 2010, there were fewer than 30 publications relating to ANN in offshore engineering, but that number began to rise significantly in 2011.

Figure 5. Top 15 leading journals in ANN for research in offshore engineering from 2005 to 2023.

Figure 6. No of publications of top five countries relying on the Scopus database from 2005 to 2023.

Figure 7. Shares (%) of publications of top five countries in ANN for research in offshore engineering relying on the Scopus database from 2005 to 2023.

Table 2. Top five most vital contributors’ countries with respect to their number of publications annually

Publications	Year	China	United States	Iran	United Kingdom	India
16	2005	3	4	1	3	2
17	2006	3	0	0	3	1
25	2007	3	5	4	0	1
29	2008	3	7	2	3	4
29	2009	4	4	4	0	2
25	2010	2	3	4	2	0
38	2011	4	6	9	3	3
51	2012	9	5	10	4	3
48	2013	7	8	7	6	3
65	2014	4	11	7	5	5
53	2015	5	8	11	8	5
49	2016	5	9	8	4	0
64	2017	0	10	9	0	6
93	2018	17	19	7	6	5
136	2019	18	30	7	12	7
182	2020	39	31	7	13	5
156	2021	36	27	11	15	11
170	2022	55	25	12	9	11
16	2023	5	1	0	3	4
TOTAL		222	213	120	99	78

Table 3. Shares (%) of Top five countries in ANN for research with regard to offshore engineering from 2005 to 2023

Year	China	United States	Iran	United Kingdom	India
2005	18.75	25.00	6.25	18.75	12.50
2006	17.65	0.00	0.00	17.65	5.88
2007	12.00	20.00	16.00	0.00	4.00
2008	10.34	24.14	6.90	10.34	13.79
2009	13.79	13.79	13.79	0.00	6.90
2010	8.00	12.00	16.00	8.00	0.00
2011	10.53	15.79	23.68	7.89	7.89
2012	17.65	9.80	19.61	7.84	5.88
2013	14.58	16.67	14.58	12.50	6.25
2014	6.15	16.92	10.77	7.69	7.69
2015	9.43	15.09	20.75	15.09	9.43
2016	10.20	18.37	16.33	8.16	0.00
2017	0.00	15.63	14.06	0.00	9.38
2018	18.28	20.43	7.53	6.45	5.38
2019	13.24	22.06	5.15	8.82	5.15
2020	21.43	17.03	3.85	7.14	2.75
2021	23.08	17.31	7.05	9.62	7.05
2022	32.35	14.71	7.06	5.29	6.47
2023	38.46	7.69	0.00	23.08	30.77

3.2. Key journals of ANN in offshore engineering-related research

The gathered 1259 papers were published in 160 journals that were indexed in Scopus, demonstrating both the diversity of the publishing distribution and the depth of interest in the study field of offshore engineering’s application of ANN. This section examined 15 journals that, over the previous 18 years, have published the most ANN for research in offshore engineering (from 2005 to 2023). A ranking of such journals’ total publications, total citations, as well as their percentage of total publications and total citations on ANN in offshore engineering in the total publications of the journal, are shown in Table 4. The Journal of Petroleum Science and Engineering, which published 5.48% of all articles on ANN for study in offshore engineering and received 8.29% of all citations over the last 18 years, topped the list of all journals (from 2005 to 2023). Additionally, ANN-related publications compensate for 4.77% of all papers released by Ocean Engineering and receive 10.04% of all citations. Furthermore, the Proceedings of The International Offshore and Polar Engineering Conference, which ranks third in terms of publications, accounts for 3.18% of all publications and has received 0.59% of all citations. The Journal of Petroleum Science and Engineering publishes several collections of articles on diverse topics across the year. Since the journal’s establishment in 1987 till the date, around 44,739 citations have been accumulated, based on the Scopus database. In light of the fact that ANN-related papers make up 5.48% of all journals of petroleum engineering and science publications, it is reasonable to say that the journal is a storehouse for state-of-the-art study on general offshore engineering, of which ANN is merely one subset.

Table 4. Top 15 leading journals in ANN for research in offshore engineering from 2005 to 2023

Journal	TP	TP (%)	TC	TC (%)
Journal Of Petroleum Science And Engineering	69	5.48%	1180	8.29%
Ocean Engineering	60	4.77%	1429	10.04%
Proceedings Of The International Offshore And Polar Engineering Conference	40	3.18%	84	0.59%
Proceedings Of The International Conference On Offshore Mechanics And Arctic Engineering OMAE	34	2.70%	187	1.31%
Applied Ocean Research	15	1.19%	458	3.22%
Computers And Geosciences	13	1.03%	422	2.96%
Energies	13	1.03%	129	0.91%
Geophysics	13	1.03%	551	3.87%
Proceedings Of The Annual Offshore Technology Conference	13	1.03%	107	0.75%
Advances In Intelligent Systems And Computing	12	0.95%	8	0.06%
Interpretation	11	0.87%	306	2.15%
IEEE Transactions On Geoscience And Remote Sensing	10	0.79%	369	2.59%
SPE Middle East Oil And Gas Show And Conference Meos Proceedings	10	0.79%	116	0.82%
Energy	9	0.71%	213	1.50%
Iop Conference Series Earth And Environmental Science	9	0.71%	11	0.08%

TP; Total publication, TC; Total citation, TLS; Total link strength.

3.3. Key institutions of ANN in offshore engineering-related research

The top 15 institutional contributors to ANN in offshore engineering research for the previous 18 years (from 2005 to 2023) are included in Table 5, ranked by the number of publications on ANN in offshore engineering. They primarily originate from the top 15 industrialised nations. King Fahd University of Petroleum and Minerals in Saudi Arabia is the biggest source of ANN in offshore engineering-related research among the 160 universities covered in this evaluation, with the greatest publications share of (4.29%) as well as the total citations share of (4.80%) from 2005 to 2023. Moreover, Universiti Teknologi PETRONAS in Malaysia is the second-best provider, with a publication share of 2.78% as well as a citation share of 1.26%. However, the citation record is strong for top leading nations and institutions, which include England, Australia, Netherlands, China, as well as the USA. Other institutions from the top 15 contribute with greater than 10 publications, respectively.

Table 5. Top 15 leading institutions of ANN in offshore engineering-related research from 2005 to 2023

Institution	TP	TC	Country	TP (%)	TC (%)
King Fahd University of Petroleum and Minerals	54	683	Saudi Arabia	4.29%	4.80%
Universiti Teknologi PETRONAS	35	180	Malaysia	2.78%	1.26%
China University of Petroleum East China	29	278	China	2.30%	1.95%
China University of Petroleum-Beijing	22	75	China	1.75%	0.52%
Islamic Azad University	22	438	Iran	1.75%	3.08%
Universidade Federal do Rio de Janeiro	21	200	Brazil	1.67%	1.41%
The University of Texas at Austin	18	313	United States	1.43%	2.20%
Petroleum University of Technology	17	429	Iran	1.35%	3.01%
Amirkabir University of Technology	15	254	Iran	1.19%	1.78%
University of Tehran	15	309	Iran	1.19%	2.17%
University of Strathclyde	15	275	United Kingdom	1.19%	1.93%
Indian Institute of Technology Bombay	13	206	India	1.03%	1.45%
Ocean University of China	13	115	China	1.03%	0.81%
Universidade de São Paulo	12	27	Brazil	0.95%	0.19%
Memorial University of Newfoundland	12	301	Canada	0.95%	2.11%

3.4. Keywords’ characteristics

The total number of keywords provided by the 1259 articles that were chosen is 160, or 0.13 keywords per article. One hundred sixty keywords were divided into two groups in order to prevent the issue of duplication, and Table 6 presents the relevant frequencies. The following 121 keywords show up lower (50 times), making up 75.63% of the total 160 keywords; the remaining 39 keywords showed up over 50 times, showing a variety of ANN research studies. In addition, the most frequently utilised keyword was ‘‘neural networks” (724 times), followed by ‘‘artificial neural network” (557 times) as the second top keyword utilised. Other keywords that appeared above 100 times comprise ‘‘offshore oil well production” (345 times), ‘“forecasting” (313 times), ‘‘gas industry” (181 times), ‘‘machine learning” (181 times), ‘“prediction” (123 times), ‘‘artificial neural networks” (120 times), ‘‘petroleum reservoirs” (101 times) and, lastly the ones that are left have been used below (100 times).

Table 6. Top 20 used keywords and their frequencies

Keywords	Frequency	%	TLS
Neural Networks	724	57.51%	8545
Artificial Neural Network	557	44.24%	7066
Offshore Oil Well Production	345	27.40%	4427
Forecasting	313	24.86%	4538
Gas Industry	181	14.38%	2480
Machine Learning	181	14.38%	2564
Prediction	123	9.77%	1927
Artificial Neural Networks	120	9.53%	1339
Petroleum Reservoirs	101	8.02%	1487
Artificial Intelligence	99	7.86%	1376
Petroleum Reservoir Engineering	97	7.70%	1494
Support Vector Machines	84	6.67%	1336
Oil And Gas Industry	80	6.35%	990
Well Logging	80	6.35%	1285
Backpropagation	77	6.12%	1049
Petroleum Industry	77	6.12%	1096
Algorithm	76	6.04%	1224
Neural Network	76	6.04%	896
Mean Square Error	75	5.96%	1154
Optimisation	74	5.88%	1037

3.5. Citation network analysis

The most prominent research with over 30 citations (131 documents) was chosen from the collection of documents and categorised into 12 clusters, as shown in Table 9 and depicted in Figure 8, to categorise the publications. Authors labelled each cluster in accordance with its content after classifying each cluster utilising VOS Viewer. Moreover, 18 papers in cluster 1 were released to investigate and evaluate various elements of drilling system operations and design. Numerous topics and points of view have been addressed in the articles, covering the most popular approaches, including generic algorithms with intriguing hybrid developments, expert systems, fuzzy logic, as well as ANN. Furthermore, it has been used for complex system identification, selection, optimisation, forecasting, as well as control. Generally, regardless of model comparisons or test field trials, the outcomes of the application of ANNs, generic algorithms (GA), fuzzy tools, as well as support vector machines (SVM) in drilling evaluation have been very promising (Abbas, Bashik, et al., 2019; Abbas, Rushdi, et al., 2019; Ahmed et al., 2019; Al-Baiyat & Heinze, 2012; Anemangely et al., 2018; Ashhab, 2008; Bello et al., 2015, 2016; Bhattacharya et al., 2019; Harish et al., 2015; Jahanbakhshi, Keshavarzi, & Jafarnezhad, 2012; Jahanbakhshi et al., 2012; Mohamadian et al., 2021; Neto et al., 2013; Orrù et al., 2020; Shi et al., 2016). Documents related to permeability and chemical compound prediction were collected under cluster 2. The permeability and chemical compounds were usually assessed in 17 documents. Aside from its capacity to deal with nonlinearity, the primary benefits of an ANN approach over conventional techniques in the search for complex interrelationships between geophysical/geological processes and properties that there is no a priori option of the underlying mathematical model are essential. Very less work is needed to identify the sensitivity of the model to the selected input variables. Additionally, to establish precise and straightforward procedures that are simpler than current techniques, less complex, and involve fewer calculations to calculate the units of chemical compounds (Alkinani et al., 2019; Castro et al., 2014; Ghiasi et al., 2014; Ghiasi-Freez et al., 2012; Guéguen et al., 2016; Huang et al., 1996; Jamshidian et al., 2015; Kamari et al., 2014; Layouni et al., 2017).

Figure 8. Illustration of clusters of the publications as the most significant studies with more than 30 citations.

Table 9. Classification of the most significant publications with more than 30 citations based on the Scopus database from 2005 to 2023

Cluster	Classification
Cluster 1	Drilling system design and operations
Cluster 2	Permeability and chemical compounds prediction
Cluster 3	Prediction and evaluation of ultimate strength on the model uncertainty
Cluster 4	Petrophysical parameters data prediction
Cluster 5	Estimation of the breakthrough time of water coning
Cluster 6	Wave parameters prediction
Cluster 7	Remote sensing
Cluster 8	The capability of the ANN in multi-machine
Cluster 9	Wind power forecasting
Cluster 10	The estimation of the oil recovery factor (RF)
Cluster 11	Water quality application
Cluster 12	Prediction of the condition pipe

On the other hand, research (14 papers) that addressed the estimation and assessment of final intensity on model uncertainty were grouped under cluster 3. Documents in this cluster focused on improving ANN structure, which was instantly influenced by the quantity of training data available, the number of neurons in the hidden layer, the activation function selected, the training iterations, the randomisation procedure, as well as the identification of additional operational parameters (Al-Anazi & Gates, 2012; Casey et al., 2019; Durodola et al., 2017; Hassanvand et al., 2018; Mohamadi-Baghmolaei et al., 2015; Pu & Mesbahi, 2006). Petrophysical parameter data prediction occurred in cluster 4 as a frequent problem of data for the characterisation of hydrocarbon reservoirs. Thirteen papers were acquired under this cluster, including various petrophysical, geological, including seismic data that are ambiguous as well as intrinsically rife with uncertainty (Aleardi & Ciabarri, 2016; Kadkhodaie-Ilkhchi et al., 2009; Onalo et al., 2018; Veeken et al., 2009). In cluster 5, substantial research (13 papers) relating to the use of various models and methodologies to assess and predict the breakthrough time of water coning was gathered. The investigations included a wide range of topics, including fuzzy logic that was modified using a hybrid Kalman filter and genetic algorithm, ANN techniques, the estimation of oil and gas-oil ratio production, as well as other topics (Halali et al., 2016; Khamehchi et al., 2014; Panja et al., 2018; Shaik et al., 2020).

The wave parameters prediction research (12 papers) that sought to determine and predict the average wave period, significant wave heights, as well as wind speed at the coastal location were included in the 6th cluster executing the formulae that may be applied in applications where wave and meteorological data are present. Additionally, this cluster establishes that wave parameter hindcasting is essential for several applications in offshore and coastal engineering (Günaydin, 2008; Jain et al., 2011; Kumar et al., 2017; Mahjoobi et al., 2008, 2008; Mandal & Prabaharan, 2006). Cluster 7 consisted of 11 research pertinent to remote sensing (Cunha et al., 2020; Pham et al., 2019; Pochet et al., 2019; Zheng et al., 2019). The capability of the ANN in multi-machine has been discussed in cluster 8 (10 documents). This cluster presented stability improvement, introduced a new term like the mobility index, and, together with a few others (Bolandi et al., 2017; Elkatatny et al., 2018; Wang & Thi, 2013; Patil et al., 2011, 2012).

The wind power forecasting was collected in cluster 9 (8 documents). The studies present the design of offshore foundations first discussed, followed by some uncomplicated design calculations for sizing foundations and structures that are suitable for the wind-turbine challenge. Typically, wind power time series exhibit intricate dynamics because of nonlinearities in wind physics and the manner that power is transformed in wind farms (Hong & Satriani, 2020; Kisvari et al., 2021; Lu et al., 2018; Shafiee et al., 2016; Lin & Liu, 2020). In other instances, this cluster offers a method for using local variables (such as direction and wind speed) to describe a variety of these impacts employing statistical models. The 10th cluster collected the estimation of oil recovery factor (RF) studies (6 documents) that provide complete knowledge and understanding of non-technical and technical aspects concerning the nature of the reservoir, available technology, as well as economic conditions, including other means (Ahmed et al., 2017; Alsaihati et al., 2020; Mahmoud et al., 2019). As for cluster 11 (5 documents), it has been classified as a water quality application. When the time series structure is defined by nonlinear behaviours and autoregressive, a comparison of watershed environmental water quality management as well as predictive models for wastewater process control has been performed under this cluster (Cheng et al., 2008; Dellana & West, 2009). Cluster 12 consisted of only four pieces of research pertinent to the prediction of the condition pipe (El-Abbasy et al., 2014; Senouci et al., 2014).

3.6. Qualitative content analysis

The top three most referenced reviews and a total of 20 top-cited publications from the previous 18 years, from 2005 to January 2023, were utilised as the basis for the content analysis to provide a complete perspective of the study area of ANN in offshore engineering. The 1259 documents that were chosen from the Scopus dataset were initially divided into two categories: articles (748 articles, or 59.41%), as well as review papers (11 reviews, or 0.87%). Reviews consistently obtained more citations when documents were compared since, they clarified and summarised findings from recent literature on the related topics.

Table 7 includes the 11 reviews on reliability conducted over the last 18 years (2005–2023). The top three most referenced reviews out of the 11 reviews were identified manually. As per their citation, the following authors and years generated the top three reviews: Agwu et al. (2018) with 84 citations, Agwu et al. (2018) with 33 citations, and Soomro A.A. (2022) with 20 citations.

Table 7. Review documents authors and their citation on ANN in offshore engineering from 2005 to 2023

Authors	Years	TC
Agwu O.E., Akpabio J.U., Alabi S.B., Dosunmu A.	2018	84
Agwu O.E., Akpabio J.U., Alabi S.B., Dosunmu A.	2018	33
Soomro A.A., Mokhtar A.A., Kurnia J.C., Lashari N., Lu H., Sambo C.	2022	20
Solanki P., Baldaniya D., Jogani D., Chaudhary B., Shah M., Kshirsagar A.	2022	14
Uddin M.A., Jameel M., Razak H.A.	2015	7
Sulaimon A.A., Teng L.L.	2020	6
Christiansen N.H., Voie P.E.T., Winther O., Høgsberg J.	2016	5
García Márquez F.P., Peinado Gonzalo A.	2022	1
Liang B., Liu J., You J., Jia J., Pan Y., Jeong H.	2023	0
Okenyi V., Bodaghi M., Mansfield N., Afazov S., Siegkas P.	2022	0
Rogers J., Badger B.A.	2012	0

Table 7 demonstrates that the latest three reviews, which were published in 2023 and 2022, seemed to have no citations. Nevertheless, for 2012, it is possible that other researchers have not yet discovered the same field or another reason due to the research that has been done connected to the subject area. Furthermore, Section 6.1 will address the top 20 cited articles on the relevancies of ANN techniques in various of study deeply.

3.6.1. Most cited articles in ANN for research in offshore engineering-related research from 2005 to January 2023

In their studies, Zou & Shi (Zou & Shi, 2016) clarified that in addition to these supervised learning methods, certain unsupervised methods are also employed in this stage, including the contour analysis method as well as the shape analysis method. Table 8 displays the most cited articles associated with the topic of ANN in offshore engineering. The features in our system are adaptively learnt from spaceborne optical image data rather than being manually generated. During the ship candidate identification stage, all ship target candidates are highlighted while the unsuitable backgrounds are suppressed, utilising three convolutional layers as well as three nonlinear mapping layers. On the contrary, research by Li et al. (2018) established that a deep convolutional neural network (CNN) termed hierarchical selective filtering (HSF) may be used to recognise multiscale ship targets in optical remote sensing images. Secondly, a deep multiscale feature is embedded in a network for suggesting ship detection regions utilising an HSF layer. Four unsupervised learning techniques, including two supervised learning techniques comprising several of the most significant multi-attribute facies classification tools, were contrasted and contrasted by (Zhao et al., 2015). However, ANN and support vector machine (SVM) specifically label the output clusters to the intended facies by establishing a precise relationship between the input data vectors as well as a subset of user-labelled input training data vectors.

Table 8. Top 20 most cited publications in ANN for research in offshore engineering-related research from 2005 to January 2023

Authors	Citations	Cites Per Year
Z. Zou, Z. Shi	202	28.86
T. Zhao, V. Jayaram, A. Roy, K.J. Marfurt	189	27
Z. Huang, J. Shimeld, M. Williamson, J. Katsube	148	74
V. Das, A. Pollack, U. Wollner, T. Mukerji	146	36.5
Q. Li, L. Mou, Q. Liu, Y. Wang, X.X. Zhu	131	26.2
J. Mahjoobi, E. Adeli Mosabbeb	131	9.36
P. Flores, A. Tapia, G. Tapia	131	7.28
K. Shojaei	123	15.38
M.A. Ahmadi, M. Ebadi, S.M. Hosseini	123	13.67
M.S. El-Abbasy, A. Senouci, T. Zayed, F. Mirahadi, L. Parvizsedghy	118	13.11
A.F. Al-Anazi, I.D. Gates	111	10.09
E.A. El-Sebakhy	110	7.86
J. Mahjoobi, A. Etemad-Shahidi, M.H. Kazeminezhad	110	7.33
S. Mandal, N. Prabaharan	105	6.18
K. Shojaei, M. Dolatshahi	98	16.33
M. Mullen, R. Roundtree, B. Barree	98	6.13
A. Kisvari, Z. Lin, X. Liu	94	47
R. Parichatprecha, P. Nimityongskul	94	6.71
S.A. Dellana, D. West	90	6.43
U. Iturrarán-Viveros, J.O. Parra	86	9.56

The primary benefits of an ANN approach over conventional methods, aside from its capacity to deal with nonlinearity, are that no a priori choice of the underlying mathematical model is required, as well as the tiny effort is necessary to ascertain the sensitivity of the model to the selected input variables, according to (Huang et al., 1996), which was researched in searching for complex interrelationships among geophysical/geological processes and properties. The time needed to train the network is the biggest disadvantage of back propagation artificial neural network BP-ANN modeling. Moreover, (Al-Anazi & Gates, 2012) discovered discrepancies that imply a comparison between a multilayer perceptron network and the sensitivity of the SVM-based prediction of porosity and permeability to sample size and empirical loss functions. Even though it has been shown that for linear regression problems with small sizes and high dimensions, the prediction accuracy of the SVM using the e-insensitivity loss function surpasses that of the conventional least-squares and least-modulus methods, it would be useful to examine the effects of various loss functions on the prediction accuracy of nonlinear SVM regression with RBF Gaussian and sigmoid kernel functions. Permeability and porosity are two vital reservoir properties that have to do with how much fluid is held in a reservoir and how well it can flow, as per research by Iturrarán (Iturrarán-Viveros & Parra, 2014). The neural network technique effectively distinguishes an extremely permeable zone that correlates to high water production in the aquifer as a consequence of employing ANN to make an estimation. According to research by (Das et al., 2019), the absence of a substantial training dataset with labelled examples limits the application of neural networks to geophysical issues. A benefit of utilising a neural network for impedance inversion is that when the network weights are determined through training on a training dataset, the prediction of impedance only needs one input seismic data-trace.

In a different research, (Mahjoobi & Adeli Mosabbeb, 2009) investigated significant wave height and suggested an alternate method built on regression and classification trees for significant wave height prediction, showing a fresh method for predicting significant wave height relying on SVM. According to study findings by (Mahjoobi et al., 2008), neuro-fuzzy computing methods and fuzzy inference systems are also applied in the prediction of wave parameters. Research by (Mandal & Prabaharan, 2006) contrasts this work, which discusses the utilisation of neural network analysis in wave forecasting and is performed at substantial wave height with a 3-h lead period. Recurrent networks forecast more accurately than older neural network techniques, according to the findings.

It is crucial to consider the heating of the air that results from radiation, according to (Flores et al., 2005), who also noted that wind is a result of solar radiation. The reactive and active power generation policies may be set to achieve the maximum economic benefits for both producers and utilities, provided that the ANN is employed to forecast the wind speed at a wind farm. According to simulation findings, the innovative data-driven technique given by (Kisvari & Lin et al., 2021) may obtain a high degree of accuracy while incurring less computing costs when it comes to wind power forecasting. A leader-follower formation control approach was put out by (Shojaei, 2015) for a 3-degree-of-freedom (DOF) model of underactuated autonomous underwater vehicles. As per their study, the suggested controller can efficiently handle the formation tracking problem in the existence of actuator restrictions and model uncertainties. Conversely, (Shojaei & Dolatshahi, 2016) examined several tracking controllers for underactuated autonomous underwater vehicles (AUVs) using a 3-degree-of-freedom (DOF) model and found that neural networks and adaptive control methods accommodate all model uncertainties, including environmental disturbances.

To calculate the water coning breakthrough time in fractured reservoirs, a novel kind of smart technique known as the least square SVM was employed (Ahmadi et al., 2014). Note that significant attempts were made to offer a least square SVM, a low parameter technique that is resistant to the over-fitting problem. The breakthrough time of water coning was also calculated using further methods involving ANN and fuzzy logic calibrated using a mix of Kalman filter and genetic algorithm. A regression and ANN model were built by El-Abbasy & Senouci et al (El-Abbasy et al., 2014) in different research to forecast the potential forms of failure for oil and gas pipelines. While both strategies’ performances were determined to be comparable, the ANN methodology outperformed fuzzy logic in terms of output. In order to test the SVR modeling scheme’s potential as a fresh framework for forecasting the pressure-volume-temperature (PVT) properties of crude oil systems, El-Sebakhy (2009) conducted a study on the three separate published data sets.

In order to remove the mechanical rock properties as a variable in fracture stimulation treatment design and history matching, a work by Mullen et al. (2007) presents a composite model to give a solid rock property solution either with or without sonic log data. According to (Parichatprecha & Nimityongskul, 2009), there has not been any study done to date on employing neural networks to estimate high performance concrete (HPC) durability. The suggested ANN model is employed to examine the impact of mix proportion variables on the HPC durability. The findings demonstrate the models’ reliability and accuracy, as well as the efficiency of utilising ANNs to effectively estimate the permeability of chloride ions in a variety of HPC ingredients. A study by (Dellana & West, 2009) that is conducting a pilot study that investigates these challenges by contrasting the forecasting accuracy of a conventional linear autoregressive integrated moving average (ARIMA) model with a nonlinear, dynamic neural network model builds on the effectiveness of using ANN in offshore engineering. In this research employing artificial data sets, the time-delay neural network (TDNN) was equivalent to linear ARIMA in all other single-period prediction data sets. It was more accurate compared to linear ARIMA for a single-period prediction in four of the eight data sets.

3.7. An insight into the practical implementation of ANN techniques in offshore engineering publications

In this study, the status and trends of application ANN techniques in offshore engineering were analyzed based on the databases of Scopus, Excels and Bibliometric methods. From the visual networks, some valuable information can be obtained. Providing insight into the practical implementation of ANN in the offshore industry involves understanding how ANN is used to solve real-world problems. This can involve understanding the specific challenges faced by engineers in the industry. In Section 5, it has been previously elucidated that the techniques and application of Artificial Neural Networks (ANNs) within the domain of offshore engineering have been thoroughly expounded upon, relying on the extensively referenced scholarly work accomplished by prior researchers. ANN have found practical applications in various domains, including offshore engineering. In the context of offshore engineering publications, ANN have been utilized to address a range of challenges and enhance decision-making processes. In this study some of an insight into the practical implementation of ANN techniques in offshore engineering publications that can be concluded:

1. Predictive Modeling: ANNs are employed in offshore engineering to develop predictive models for various parameters and phenomena. For example, they have been used to predict wave heights, wind speeds, current patterns, and other environmental factors. By training ANNs on historical data, these models can provide valuable insights for optimizing offshore operations, designing structures, and improving safety.

2. Structural Health Monitoring: ANN techniques have been utilized for structural health monitoring of offshore structures such as platforms and pipelines. By analysing sensor data, ANNs can detect and predict potential failures, fatigue, corrosion, and other structural issues. This information helps in scheduling maintenance and inspection activities, ensuring the integrity and safety of offshore assets.

3. Reservoir Characterization: In offshore oil and gas exploration, ANNs have been employed for reservoir characterization. They assist in interpreting seismic data, well logs, and production data to estimate reservoir properties such as porosity, permeability, and fluid saturation. ANNs help in identifying potential hydrocarbon reserves and optimizing drilling and production strategies.

4. Risk Assessment and Decision Support: ANNs contribute to risk assessment and decision-making processes in offshore engineering. By analysing historical data and incorporating various factors, including environmental conditions, equipment performance, and human factors, ANNs can provide insights into potential risks and support decision-making for offshore operations, emergency response, and safety protocols.

5. Autonomous Systems: With the rise of autonomous systems in offshore engineering, ANNs play a crucial role in enabling intelligent decision-making and control. ANNs are utilized in autonomous underwater vehicles (AUVs), remotely operated vehicles (ROVs), and unmanned aerial vehicles (UAVs) to perform tasks such as seabed mapping, pipeline inspection, and environmental monitoring. ANNs enable these systems to adapt to changing conditions, detect anomalies, and make real-time decisions.

6. Data Integration and Fusion: Offshore engineering involves multiple data sources from sensors, satellite imagery, and numerical models. ANNs facilitate data integration and fusion by combining information from different sources to generate more accurate and reliable predictions. They can handle large and complex datasets, extract patterns, and provide valuable insights for operational optimization.

7. Optimization of Offshore Operations: ANNs have been applied to optimize offshore operations in terms of production efficiency, energy consumption, and cost reduction. By analysing historical data and considering various operational parameters, ANNs can identify optimal operational settings, maintenance schedules, and production strategies, leading to improved performance and resource utilization.

8. ANN techniques in offshore engineering publications demonstrate the wide range of applications and benefits that ANNs bring to the field. From predictive modeling to risk assessment, ANNs have the potential to revolutionize offshore engineering practices and contribute to safer, more efficient, and sustainable offshore operations.

3.8. Future directions of ANN techniques in offshore engineering publications

The identification of offshore engineering issues has been approached in several earlier research in a confined method. Lately, scholars have come to the realisation that the majority of offshore engineering issues are connected, either indirectly or directly. As a result, they have been forced to create new methodologies for contemporary studies and to introduce the relevance of ANN in offshore engineering, which involves and assesses the majority of offshore engineering issues. A total of 1259 papers, with an increment ratio of 7768.75% and an average rise of 409%, have been published between 1 January 2005, and 12 January 2023, according to the Scopus database. A consistent and significant growth in citations with an average rise of 4334% followed the growth in publishing. One of the factors that led the authors of the present study to forecast a rise in publications on ANN in offshore engineering in the following five years was the rapid pace of publication and citation growth.

Besides, this study has demonstrated a scarce of research (via clusters classification) in the ANN techniques application in offshore engineering aspects of studying the petrophysical parameters data prediction, the capability of the ANN in multi-machine, the estimation of oil recovery factor (RF), water quality application, and other aspects related to the prediction of the condition pipe. On the other hand, most of the applications of ANN techniques in offshore engineering publications were focused on wave parameters prediction, drilling system design and operations, and remote sensing. In the present research, cluster categorisation has made clear the necessity for additional research in the ANN approaches stated in offshore engineering. This indicates a predicted rise in publishing in these areas over the next five years.

4. Conclusion and recommendations

ANNs are an excellent tool for solving issues that are challenging to represent analytically. ANNs have been employed in several offshore applications and have demonstrated a respectable level of accuracy. The conclusion was reached after reading a considerable number of publications concerning the utilisation of ANNs in offshore engineering. This field has access to a vast amount of historical data, which may be utilised to forecast future events and aid in improved decision-making. Since there are so many uncertainties regarding the future, making estimates is never easy. ANNs may be employed to anticipate the future or provide accurate real-time predictions so that decision-makers can plan beforehand to address difficulties. Numerous restrictions have been placed on this study, one of which is the incapacity to address the usage of ANN in offshore engineering. The purpose of the present study is to enlighten the application of ANN in offshore engineering and demonstrate the need for more research in this area. Additionally, the present study has evaluated a number of sections that define the ANN principle and gather sufficient data.

The current review forecasts a rise in publications pertaining to the relevance of ANN in offshore engineering and in various other aspects predicated on the existing number of Scopus database publications and citations, as well as the cluster classification carried out by the Excel and VOS Viewer software. On the contrary, this study determined a set of useful recommendations for encouraging sustainable development and growth. The present review has acknowledged the significance of giving thorough regard to the application of ANN, which will play a significant and impactful role in obtaining high-quality and effective alternatives for certain offshore engineering problems that are challenging to solve analytically with ultimate satisfaction in the long run. It is important to note that while ANNs have demonstrated success in various domains, their implementation in offshore engineering requires careful consideration of data quality, network design, training methodologies, and domain-specific expertise to ensure reliable and accurate results.

Nomenclature

AI	=	Artificial intelligence
ANN	=	Artificial neural network
AUV	=	Autonomous underwater vehicle
BP-ANN	=	Back propagation artificial neural network
CNN	=	Convolutional neural network
DOF	=	Degree of freedom
GA	=	Generic algorithms
HPC	=	High performance concrete
HSF	=	Hierarchical selective filtering
PVT	=	Pressure volume temperature
RBF	=	Radial basis function
RF	=	Recovery factor
SVM	=	Support vector machines
TC	=	Total citation
TDNN	=	Time delays neural network
TLS	=	Total link strength
TP	=	Total publication
VOS	=	Visualization of similarities

Acknowledgments

The authors would like to thank the School of Civil Engineering, College of Engineering, Universiti Teknologi MARA (UiTM) for supporting this research.

Disclosure statement

The authors affirm that they have no established financial or interpersonal conflicts that may have been perceived to have influenced the research presented in this study.

Additional information

Notes on contributors

Mohd Hisbany Mohd Hashim

Mohd Hisbany MohdHashim (Corresponding Author) was born in Lenggeng, Negeri Sembilan, Malaysia in 1967. He obtained his BSc Degree in Civil Engineering from The University of Alabama, Tuscaloosa, Alabama, USA in 1992. He obtained a Master of Engineering (Civil-Structure) degree and Ph.D in Civil Engineering (on “Durability and Performance of Carbon Fibre Reinforced Polymer-Concrete Bonding System Under Tropical Climates”) from Universiti Teknologi Malaysia (UTM), Skudai, Johor, Malaysia in 2005 and 2010 respectively. He is currently an Associate Professor at the School of Civil Engineering, College of Engineering, Univesiti Teknologi MARA (UiTM), Shah Alam, Selangor, Malaysia. His research interest includes structural strengthening and repair, fiber reinforced concrete, wind engineering (street canyon), and offshore engineering (burst pressure pipelines and marine engineering (hull structures), all of which related to FRP and steel fibers. He has published numerous papers related to these research interests as well supervising postgraduates’ students.