The proposed Risk Influential Factor Analysis (RIFA) is a proactive tool for evaluating the contributing RFs to an individual target risk or overall risk of a system. It is to assess the relative importance of different RFs in order to recognize RFs that are most in need of attention. Thereby, analyst concentrates on a risk/system, and performs the RIFA method. In Health, Safety and Environment (HSE) context, in most cases, a major accident is target risk. Major accident consists of an incident with severe impacts on human health, assets, and the environment (Abbassi et al., 2022). A major accident usually has low frequency and high consequences which are not well supported by conventional statistical methods due to data scarcity (Khakzad et al., 2014). Samples of major accidents are Macondo blowout and explosion, Chernobyl disaster, Coronaviruses contagion, etc.

The RIFA method employs three criteria to evaluate RFs. The criteria are a triple named “3C”, i.e. contributivity, controllability, and causality. Each criterion is rated in a 1–9 scale. Number 9 depicts the most critical case, and 1 stands for the least critical situation. As equation (1), the RIFA method defines a metric called Risk Factor Priority Number (RFPN), for each RF, by using weighted geometric mean of the three scores given by SME. Accordingly, c1, c2, and c3 are relative importance of the criteria given by SME, i.e. the higher c1/c2/c3 indicates the more important contributivity/controllability/causality. According to common practice, these weights are non-negative and normalized to add up to one (i.e. c1 + c2 + c3 = 1). RFPN ranges from 1 to 9. RFs with higher RFPN indicate greater concerns to the system, and require more attention. Obviously, on the one hand, RFPN function originates in the common formula of Weighted Product Model (WPM) that is an alternative evaluation method in MADM (Hatefi, 2025). On the other hand, it follows the FMEA idea which is computed its index by multiplying three criteria entitled: severity of consequences (S), possibility of occurrence (O), and ease of detection (D) (Dhalmahapatra et al., 2022; Chang et al., 2014). (1) RFPN = Contributivity c 1 × Controllability c 2 × Causality c 3 .

According to the process of the proposed methodology (explained in Section 2.3), ranks of the RFs obtained by the RIFA method need to be converted into the weights. For this purpose, a branch of the attribute weighting methods entitled “surrogate weighting” can be used. The relevant methods have been coded as I/+SW by Hatefi (2023c), i.e. Integrated and Surrogate Weights. The related methods take ranks of n items as inputs, and convert them to weights. Firstly, priorities of the items are determined. Next, ranks 1 , 2 , … , n are assigned to the item of the first-ranked, second-ranked … and last-ranked. After that, the weights are estimated using a straightforward formula. The estimated weights need to be in accordance with the weight space as equation (2), in which ω j is the weight of item by rank j. (2) { ( ω 1 , ω 2 , … , ω n ) | ω 1 ⩾ ω 2 ⩾ ⋯ ⩾ ω n , ∑ j = 1 n ω j = 1 , ω j ⩾ 0 } . The proposed Weight Estimation by Steady Trend (WEST) is a new method from I/+SW family. The WEST method is founded on the elementary weights estimated by arithmetic progression and geometric progression. In what follows, firstly a brief background on most famous I/+SW methods is stated, then platforms of the WEST method are explained in next sub-sections.

Figure 1 shows flowchart for implementing the RIFA-WEST methodology. The stages and steps are briefly described as follows:

Stage (A): Organize a panel of h SMEs ( k = 1 , … , h).

Stage (B): [Step I:] Assign a number between 1 and 5 to each SME. This number (denoted by d k ) shows level of expertise, on the basis of SME’s knowledge, experiences, skills and outlooks around the RFs. This number is subjectively determined. A very professional SME is assigned 5, and a tyro SME receives 1. [Step II:] Normalize d k by equation (12). (12) u k = d k ∑ i = 1 h d i , k = 1 , 2 , … , h .

Stage (C): [Step I:] Make a complete literature study, and draw up an initial RF list. For instance, published papers, books, related reports, and so on. [Step II:] SMEs benefit from the initial RF list to provide a final list of RFs. In fact, they are permitted to check RFs for recommending any corrective idea. Assume that N RFs ( RF 1 , RF 2 , … , RF n ) are finally identified.

The following stages (D) to (F) are taken by each individual SME:

Stage (D): [Step I:] Rate contributivity, controllability, and causality for each RF. [Step II:] Determine c1, c2, and c3. [Step III:] Calculate RFPN for each RF.

Step (E): Prioritize RFs in order of RFPNs, from the most important to the least important. Stage (F): Use the WEST rule to convert ranks to the weights. Let’s show the weights by w j k , i.e. the weight determined for RF j by SME number k.

Now, the individual results are combined through the following stages (G) and (H):

Stage (G): Combine the individual weights (received from h SMEs) by any valid method, e.g. by geometric mean formula. Anyway, as a recently suggested method, the proposed methodology advises to use the idea of Combined Compromise Solution method (COCOSO) (Yazdani et al., 2019). This combination is reached through six steps as equation (13) to equation (18). In equation (13) and equation (14), ω j k depicts weight of jth RF given by kth SME. (13) S j = ∑ k = 1 h u k ω j k , j = 1 , 2 , … , n , (14) p j = ∑ k = 1 h ω j k u k , j = 1 , 2 , … , n , (15) k j a = S j + p j ∑ i = 1 n S i + p i , j = 1 , 2 , … , n , (16) k j b = S j min i = 1 , … , n S i + p j min i = 1 , … , n p i , j = 1 , 2 , … , n , (17) k j c = S j + p j max i = 1 , . . , n S i + max i = 1 , . . , n p i , j = 1 , 2 , … , n , (18) FS j = 1 3 ( k j a + k j b + k j c ) + ( k j a × k j b × k j c ) 1 / 3 , j = 1 , 2 , … , n .

Stage (H): Prioritize RFs in order of the combined Final Score FS j .

Table 5 is to show process of the proposed methodology, ensure each step is clarified with input, output and how input to output transformation happens.

International Energy Agency (IEA), in 2018, published World Energy Outlook (WEO), which predicts an increment in world energy demand of about 25% by 2040. Fossil fuels, especially oil and gas, will continue to account for the majority of the supply to meet this growth in energy demand. That is why upstream oil and gas companies are still concentrating on enhancing the recovery factor from operational fields (Muggeridge et al., 2014). Upstream oil and gas industry consists of exploration and production of oil and gas fields. For a new oil field with no or short production history, the natural pressure in the reservoir is normally high enough for the oil or gas to flow to the surface. As long as the pressure in the reservoir remains high enough, the production can continue with natural pressure, i.e. primary recovery. The average oil recovery rate from mature oilfields around the world under normal pressure of fields, for most hydrocarbon reservoirs is somewhere between 20% and 40% (Muggeridge et al., 2014). In fact, normal recovery efficiency in Original Oil in Place (OOIP) is close to these numbers. This shows that there is a large potential for technologies to artificially increase the reservoir pressure (Abu-El-Ela et al., 2014). As a matter of fact, as production continues, the reservoir pressure starts dropping at a certain time when the fluids are depleted. When the pressure drops down to the level that continuing production is not economical, owners/analysts need to take a decision whether to abandon the wells or to apply certain artificial lift strategies to increase the reservoir pressure. There are two major strategies (Vora et al., 2021): Improved Oil Recovery (IOR) and Enhanced Oil Recovery (EOR). EOR strategies are used to recover mostly immobile oil that remains in the reservoir, while IOR strategies are employed to increase the recovery of mobile oil. Nowadays, various IOR/EOR technologies like water flooding; gas flooding, etc. are widely used in petroleum industry. One of these methods (the concentration point of the current paper), is known as Chemical EOR (CEOR). CEOR is an EOR technology in which a chemical combination is injected into the reservoir in order to increase sweep and/or displacement efficiency. CEOR processes can currently be considered as promising tertiary technologies for increasing oil recovery from depleted oil reservoirs (Flaaten, 2008). Nowadays, the number of CEOR programs has grown exponentially as the price of oil increases, and the cost of chemicals decreases. Typically, sandstone reservoirs are most frequently utilized for CEOR due to the higher permeability of sandstone compared to limestone reservoirs. There are many CEOR types (see Fig. 2).

Polymer flooding involves injecting polymer solution to decrease water mobility resulting in improved sweep efficiency. Surfactant flooding involves injecting surfactants that achieve low Interfacial Tension (IFT) with the displaced oil. The performed analyses have shown a chemical combination for injection is more productive. For instance, polymers are usually added to surfactants for mobility control, hence the name would be Surfactant-Polymer (SP) flood. SP flooding has advantage over the other EOR technologies because of its dual advantage of increasing both microscopic displacement efficiency and volumetric sweep efficiency (Flaaten, 2008). There are published reports around CEOR applications in real-world projects such as Daqing oilfield of China (Chang et al., 2006), St Joseph field in the offshore Malaysia (Du et al., 2011), and Angsi field located in the South China Sea (Othman et al., 2007). Figure 3 is a schematic of a typical deep-water application of CEOR showing the complex interaction of various aspects of the process. This complexity can influence subsurface efficiency, logistics, injection, production, and environmental aspects. Offshore production facilities such as the floating production, storage, and offloading vessel shown in this figure are often quite crowded, with little deck space available for the equipment and storage required for CEOR. The remote location of many offshore production facilities can make shipment and storage of chemicals difficult and expensive (Rany et al., 2011).

Processes in an oil and gas industry involve inherent risks, which should be kept under control for safe operation. IOR/EOR processes can have threats, e.g. adverse environmental impacts due to discharges to the marine environment and emissions to air (Vora et al., 2021). Among these solutions, CEOR is a complex process which exhibits a number of risks and uncertainties, thus technical risk analysis must be included in a CEOR plan. In fact, this operation is subject to many risks which may cause failures in operational and functional performance. Finally, it should be noted that a successful CEOR implementation depends on the success and the ability in addressing all the relevant risks upfront (Chai et al., 2011). In conclusion, in what follows, application of the proposed RIFA-WEST methodology for CEOR operation is presented. Notably, in all the assessment activities a panel of 12 SMEs has taken part.

The current case study is related to risk/RF ranking in IOR/EOR processes. Reviewing the relevant literature indicates that there are a few reported studies on risk ranking in IOR/EOR real-world situations. Dai et al. (2016) developed a multi-scale statistical model to apply CO2 accounting and risk analysis in an EOR environment at the Farnsworth Unit (FWU), Texas. Hatefi (2018) focused on the way of diagnosing and ranking RFs of a completed real-world gas injection project, i.e. a type of EOR methods. He employed a Nominal Group Technique (NGT) to establish a risk matrix structure, and to identify RFs. Vora et al. (2021) represented a review of potentially relevant Environmental/Ecological Risk Assessment (ERA) guidelines. They suggested an initial framework of an ERA model for understanding the environmental impacts from EOR solutions. Lee et al. (2021) developed a workflow for risk assessment and management, applied to the Southwest Regional Partnership on Carbon Sequestration (SWP) phase III demonstration project. This project is related to CO2-EOR, i.e. an oil production method in which oil recovery is enhanced by CO2 injection.

As mentioned before, CEOR can be technically complex with many RFs. Some key RFs are generic to CEOR development, such as the heterogeneity, chemical effectiveness, emulsion, issues in production of sales, specification oil, etc. However, there might be some RFs that are typical for a particular field, such as fractured injection, offshore environmental problems, large secondary gas cap, etc. In the current research we use the related literature (Flaaten, 2008; Rany et al., 2011; Peyro, 2018); then the CEOR RFs are identified and categorized as: •

General

A form (see Appendix B) was built and distributed among the 12 SMEs. Each of SMEs, individually assigned his/her relative importance about the RIFA criteria (contributivity, controllability, causality), and rated them. After that, the 26 RFs were ranked according to each SME, and then the WEST weights (in descending order: 0.0769, 0.0728, 0.0688, 0.0650, 0.0614, 0.0579, 0.0546, 0.0514, 0.0483, 0.0454, 0.0425, 0.0398, 0.0372, 0.0346, 0.0322, 0.0298, 0.0275, 0.0252, 0.0231, 0.0210, 0.0189, 0.0169, 0.0150, 0.0131, 0.0112, and 0.0094) were assigned to RFs. Table 6 displays the individual RF weights. For example, in view of SME1, RF19 (High temperature) is the first-ranked factor by weight 0.0769, whereas SME2 preferred to assign the biggest weight to RF22 (Chemical performance).

Having the individual weights, the COCOSO steps were performed as Table 7. For this purpose, the levels of expertise of 12 SMEs were adjusted as 3, 3, 3, 2, 3, 3, 3, 1, 1, 2, 2, and 2, respectively.

Final RF ranks are achieved based on final weights computed by the COCOSO process. Ranks represent that RF16 is the most critical factor, and RF26 is counted as the least important factor. The final ranking of the top ten RFs is determined as below: RF 16 > RF 9 > RF 23 > RF 15 > RF 21 > RF 14 > RF 25 > RF 10 > RF 13 > RF 22.

Robustness indicates the ability of an analytical method to maintain its outputs unaffected while slight variations are applied. To test the robustness of the proposed methodology, we discuss why the proposed methodology converts RFPNs to the WEST weights, and in fact why the methodology does not directly use RFPNs as the RF weights. To address the response, we use the case study data provided in the previous section.

In Table 8, regarding the case study discussed earlier, the RF ranks by the two methods (the WEST and RFPN) are shown. Firstly, this table represents that the RF ranks by the two methods are different. The rank vectors depict a 0.9921 Kendall’s coefficient of correlation as equation (19) (Kendall and Gibbons, 1990). In this equation, n is the number of RFs, and v is the number of vectors to be compared, thus herein v = 2. (19) t = 12 ∑ j = 1 n ( ( ∑ k = 1 v r k j ) − v ( n + 1 ) 2 ) 2 v 2 ( n 3 − n ) .

The preliminary reason for using the WEST weights instead of RFPNs is normalizing the outputs in such a way that for all SMEs, among 26 RFs, the most important RF receives a weight of 0.0769 (i.e. the highest weight by the WEST method), and the least important RF gets a weight of 0.0094 (i.e. the lowest weight by the WEST method). Conversely, if we use RFPNs, the range is so variable. As an example, SME1 gives RFPN = 5.1768 to the most important RF, whereas this number for SME2 is 7.6525.

A sensitivity analysis experiment is designed to justify the robustness of the proposed method. In this experiment, the robustness of the methods is verified by changing the input data. For this purpose, we performed N = 100 rounds of the computer analysis, so the following procedure was repeated 100 times:

(I) A parameter called “No. of changed points” denoted by x is randomly generated. The range of this parameter is [1 26] because there are 26 RFs in the case study.

(II) For each SME and for each criterion (contributivity, controllability, causality), a number of x input data are changed from the current value to a new value. In order to carry out this movement, two parameters are randomly generated, the former is the size of movement (a number between 1 and 8), and the latter is the sign of movement (+1 or −1). For example, if the current value is 6, and the parameters are 4 and −1, thereby the new value for replacement would be 6 − 1 ∗ ( 4 ) = 2. If the calculated value is less than 1 or more than 9, then these bounds are used instead.

(III) Calculate new ranks of RFs by the WEST weights and by RFPNs.

(IV) Calculate the Kendall’s correlation coefficients as the equation (19) for the two methods. For a given method, this coefficient is computed with regard to the initial RF ranks by SMEs and new RF ranks calculated in stage (III), i.e. the ranks before and after changing data in stage (II).

Table 9 represents the outputs of computer sensitivity analysis. The experiment was conducted with the use of a Visual Basic for application in the Excel programming language on a personal computer.

Even though Table 7 obviously displays the superiority of the WEST robustness over RFPN robustness in all the 100 rounds, we establish a hypothesis as the equation (20) to compare the Kendall WEST population mean and the Kendall RFPN population mean. (20) h 0 : μ KendallWEST − μ KendallRFPN = 0 , h 1 : μ KendallWEST − μ KendallRFPN > 0 .

Table 7 shows that the difference data (i.e. Kendall WEST minus Kendall RFPN) are paired. Indeed, there are two samples in which each datum in one sample is paired with one datum in another sample. Hence, we use the one-way paired t-student test. The t-student test statistic is calculated as 15.2620, and the critical range at 99.99% confidence level is t number greater than 3.165. Because 15.2620 > 3.165, we reject null hypothesis, and deduce that there is absolutely significant difference between the two populations. As a matter of fact, in the proposed methodology, the WEST weights significantly cause higher robustness in the results than that of directly using RFPNs.

The aim of the current part of the research is to address some features of the RIFA method compared to the classical FMEA method. Clearly, the two methods are different in application. The FMEA method is to prioritize risks, whereas the RIFA method is for prioritizing RFs. In spite of simplicity of the FMEA method, some FMEA drawbacks have been discussed by researchers (Ibarra et al., 2024; Dhalmahapatra et al., 2022; Ghoushchi et al., 2020). In what follows, some characteristics of the RIFA method are described from the view of the most important shortcomings of the classical FMEA method: •

In the FMEA method, the relative importance of the risk criteria (occurrence, severity, detection) is not considered (Keskin and Ozkan, 2009). Moreover, the calculation formula for RPN (multiplication of the three criteria) is questionable, such as the same RPN value may be achieved by different combinations of its criteria scores. Conversely, the RIFA method uses the three parameters (c1, c2, c3) as the relative importance of the RF criteria (contributivity, controllability, causality). By this strategy, different combinations of the criteria score result in different RFPNs.

As previously mentioned, the WEST method belongs to a family of attribute weighting methods entitled surrogate weighting by code I/+ SW (Hatefi, 2023c). There are many I/+ SW methods, some of them were reviewed in Section 2.2.1. A logical and acceptable approach to compare such methods is examining the match between the weights estimated by the methods and subjective weights achieved within real-world study-cases. Mean Absolute Difference (MAD) (equation (21)) is the index to measure the match. The method by lower MAD is better. (21) MAD = ∑ j = 1 n | ω j the method − ω j study − case | / n .

Fifteen real-life study-cases (see Table 10) were randomly derived from MADM papers. There was an attempt to have the cases from a variety of application fields, subjective or integrated weighting methods, and the number of attributes.

The WEST method was compared to six I/+SW methods, which are the EW, UV, RS, ROC, ROT, and ROL methods. Table 11 shows comparison results. In this table, for a given case, the first row stands for the MAD values, and the second row depicts ranks of the MAD values for the methods. To compare the methods, observing ranks received by the methods can be useful. Obviously, the WEST method, through all the instances, gets a rank between 1 to 4, in such a way that in only 2 of 15 the rank is 4. On the contrary, this measure of judgment shows very weak results for the EW, UV, ROT, even ROC methods. Let’s employ another index for making judgment. Based on the BORDA ranking rule, we calculate total ranks obtained from all the 15 instances to make judgement around the methods. This index and relevant overall ranks are shown at the two last rows in Table 11. This demonstrates relative comparability and good performance of the proposed WEST method over the other methods. From this point of view, the UV method acts as the worst performance. Considerably, the famous ROC method is placed at the fourth rank after the RS and ROL methods. In the table, the row entitled mean MAD confirms the above results.