Aggregation of Risk Management and Non-Parametric Models to Rank Failure Modes of Radio Frequency Identification Systems

Featured Application

The method is applied to the RFID system but can be implemented to assess the risks of any other equipment or process.

Abstract

The failure mode causes and effects analysis (FMCEA) is a commonly used reliability approach. It identifies, predicts, and analyzes potential failure modes affecting the proper function of equipment or the process under study, along with their roots and consequences. FMCEA aims to evaluate and assess the risks resulting from their occurrence, intending to suggest corresponding repair, adjustment, and precautionary measures to be planned during the conception, instruction, or implementation stages. However, the FMCEA has been criticized in the literature for its many inherent shortcomings related to risk assessment and prioritization. Therefore, this study presents an enhanced FMCEA method to address the deficiencies of the traditional risk priority number (RPN) and improve the reliability of risk assessments and corrective actions. A data envelopment analysis (DEA), as a non-parametric method, is used to evaluate the efficiency of these failures by considering their fixing time and cost and deciding on their final priority ranks. Sub-failure modes and their interrelationships are also taken into account. The radio frequency identification (RFID) system was chosen as an example due to its core role in Industry 4.0 and the Internet of Things (IoT) to demonstrate the effectiveness and usefulness of the proposed method. A total of 67 failures related to both hardware and software parts, including the environmental impacts of this technology, have been disclosed. The results of the conventional and the suggested FMCEA methods are found to be considerably different, with ten failure modes classified as being the most efficient.

1. Introduction

A failure mode causes and effects analysis (FMCEA) is an analytical technique that combines technology and the knowledge of experts. It is used to identify the possible failure modes (FMs) of a product or a process and manages to eliminate their negative effects. The traditional FMCEA method prioritizes the FMs based on the decreasing values of their risk priority numbers (RPNs), which are equal to the product of the risk factors (occurrence, O; severity, S; and detection, D) [1].

However, the conventional FMCEA has been criticized in the literature due to several deficiencies, including (i) the assignment of equal weights to the risk factors in the RPN computation. Additionally, another drawback of the FMEA method is that (ii) attention is not paid to the time and cost spent to solve the FMs; sometimes, the time and cost spent to fix the riskiest FM can be used to fix many other different FMs in lower ranks. Another issue that may be faced while applying the FMCEA is (iii) that some FMs (parent failures) may have sub-failure modes (SFMs); those sub-failures are most likely to have different risk priorities with large gaps, which puts a big question mark on the overall rank of the parent FM. Moreover, (iv) the efficiency notion is not considered in the risk assessment in the conventional RPN method. Furthermore, (v) hidden or latent risks may remain underestimated or inaccurately estimated due to the possibility of various combinations of risk factors resulting in the same RPN value. Consequently, these undisclosed risks might eventually cause unforeseen errors or issues. Additional shortcomings associated with the FMCEA method, along with some of the methods suggested to address them, can be found in [2,3,4,5] and are summarized in Appendix A.

2. Literature Review

2.1. FMEA and MCDM Methods

A variety of studies and research have aimed to enhance the traditional FMEA by incorporating different multi-criteria and decision-making (MCDM) materials and methods [2,3,4,5]. For instance, the house of reliability (HoR) approach, as outlined in [6], was introduced to manage the domino effects of failure modes. This was achieved by segmenting severity into fundamental components and conducting a cost/worth analysis using quality functional deployment (QFD). Additionally, this method categorized risk factors into two levels and seven sub-risk factors, relying on subjective expert weighting. In another study [7], the TOPSIS (technique for order preference based on similarity to ideal solution) and DEMATEL (decision-making trial and evaluation laboratory) methods were integrated. Attention was focused on analyzing the prioritization of failure modes in FMEA, considering the direct and indirect relationships between failure modes and the causes of failure, and the subjective weights of the risk factors. Furthermore, the authors introduced the intuitionistic fuzzy hybrid TOPSIS approach in [8], wherein domain experts utilized linguistic terms to subjectively weigh risk factors, while their objective ordered weights were derived through the normal-distribution-based method; this approach incorporated risk indicators like cost, time, and maintenance. The analytic hierarchy process (AHP), DEMATEL, and VIKOR (a Serbian acronym for “Vise Kriterijumska Optimizacija I Kompromisno Resenje”, meaning multi-criteria optimization and compromise solution) were combined in [9], with subjective weightage provided directly by experts. The VIKOR method pinpointed risk gaps associated with failure modes across each risk factor. Meanwhile, the AHP approach, in conjunction with DEMATEL, offered a relation map between failures and their causes, aiding in the prioritization of failure modes. The failure modes were then divided into cause and effect groups. The objective weights of the risk factors were determined using the ambiguity measure (AMWRPN) in [10], while the subjective assessments were modeled as basic probability assignments (BPAs). Virtual reality (VR) technology was adopted for identifying potential hazards in [11]. The cloud model and preference ranking organization method for enrichment evaluation (PROMETHEE) were used in [12] to rank failure causes based on considerations such as downtime duration, spare part criticality, and safety-related risks. Furthermore, the degree of uncertainty and divergence degree were used to calculate the primary and secondary weights of decision-makers, respectively. The overall weight was determined using cloud model theory. Moreover, an artificial intelligence method used the fuzzy rule base (FRB) and grey relations theory (GRT) in [13] to address the incomplete information. A fusion of risk evaluation information using the TrFNs-WAIA (weighted arithmetic interaction averaging operator of the trapezoidal fuzzy numbers) operator based on Shapley Choquet, alongside the TODIM method, was formulated to derive the risk evaluation matrix. This was carried out by simulating the psychological behavior characteristics of the team members and considering the correlations among them, and the weights of risk indicators were calculated by considering their interaction [14]. Machine learning uses an artificial neural network (ANN) model and a fuzzy inference system (FIS) to assess the failures of busbars used in the electrical vehicle industry [15]. An integrated-approach WLSM–MOI–partial-ranking method that incorporated the imprecision approach with a nonlinear programming model for the calculation of FRPNs was presented in [16]. The subjective weights of the risk factors were assigned by the expert judgment method. A linear programming model used the fuzzy ordered weighted geometric averaging (FOWGA) operator to assign the objective weights of the risk factors and aggregate their fuzzy ratings for calculating fuzzy RPNs in [17]. The PRISM (partial risk map) methodology was employed to elucidate concealed risks using diverse aggregation functions and to visually represent these hidden risks through the PRISM pattern [18]. Likewise, The action priority (AP) method was utilized for the direct classification of failure modes into high risk (H), medium risk (M), and low risk (L). This approach prioritizes the consideration of severity (S) first, followed by occurrence (O), and finally, detection (D) [19]. The previously mentioned methods represent advancements in calculating the risk priority number (RPN), and they have gained extensive utilization across various industries.

2.2. FMEA and DEA Methods

Data envelopment analysis (DEA) is a non-parametric approach that employs linear programming (LP) models to measure efficiency and determine objective weights using observed data for multiple inputs and outputs [20]. In [21], objective risk factor weights derived by DEA models varied for each failure mode; the failure modes were ranked based on the fuzzy weighted geometric means (FWGMs) of their fuzzy ratings for risk indicators (fuzzy RPN). Subjective weights for these indicators were assigned using the expert judgment method and solved through α-level sets and linear programming models in addition to DEA [22]. In another study [23], the objective weights of risk factors were determined using the DEA method, and the significance of severity criteria was approached through weight restriction [23]. Furthermore, the exponential RPN (ERPN) and DEA model (CCR-AR) were integrated into the FMEA process in [24]. The authors in [25] developed and compared two models that combined FMEA and DEA: the slacks-based measure of the DEA (SBM-DEA) model and the data-driven approach (NDA). The latter was found to possess a relative advantage over the former. In [26], a DEA-FMEA methodology was presented, introducing two additional risk factors—cost and duration of treatment—as uncertain undesirable outputs. These were included alongside occurrence (O), severity (S), and detection (D), which were treated as inputs. The objective weights for these risk indicators were determined using the DEA technique. Interval values were incorporated into DEA in [27], which was then combined with grey relational analysis (GRA) to compute objective weights for risk factors and prioritize failure modes in an auto parts manufacturing process. The slacks-based data envelopment analysis (SBM-DEA) model, without outputs and using values obtained from the fuzzy inference system (FIS) and linguistic FMEA as inputs, was explained in [28]. In another fusion approach [29], AHP, fuzzy set theory, and DEA were combined, wherein experts in FMEA were assigned different weights using AHP. Moreover, the inputs for the DEA model comprised values of OSD and the reciprocal RPN, while the inverse of cost and time served as additional risk indicators set as outputs. Various types of information describing the heterogeneous features of risk factors were applied and clustered in [30], and a DEA model was utilized to rank failure modes using the values of risk factors as inputs. Additionally, in [31], fuzzy AHP, fuzzy VIKOR, and DEA (without input or output) models were integrated to determine the weights of each risk factor, where the relation of maximum to minimum weight was considered. Failure modes were obtained through the SHELL (software, hardware, environment, and liveware) model analysis in [32], where TOPSIS was used to determine the expert weights. The subjective and objective weight distribution of both experts and risk factors was computed by the DEA model using the risk factors as inputs. Meanwhile, the output was set to a constant of one; the cloud model was employed in order to cope with the fuzziness and randomness of the linguistic assessment information. Various other methods used in the literature to address the limitations of FMEA are summarized in Appendix B.

2.3. Method and Case Study Selection

Among the useful and effective methods stated in the previous section, the DEA approach was selected in this study to enforce and increase the efficacity of the traditional FMEA method, with the intention of covering the mentioned shortcomings: (i) The DEA model ensures the assignment of different weights to each risk factor to cope with the shortcomings. (ii) The time and cost required to solve each failure mode will be considered as additional risk indicators and will be used as parameters for the DEA model and deal with the shortage. (iii) Sub-failure modes and their interrelationships will be considered and synthetized using an aggregation procedure to handle the deficiency, and (iv) a combination of outputs and inputs that are different from those found in the literature will be used to address the efficiency limitation. The traditional RPN number has been used since the ranking of the failures will rely on the output of the efficiency measurement and not the RPN ranks. Other improved versions of RPN calculation can definitely be used, such as the exponential RPN, the partial risk map (PRISM), or the action priority (AP).

Why has RFID been selected as an illustrative example of the proposed method?

The fourth industrial revolution, or Industry 4.0, gains advantages every day. As a result of the Internet of Things and cloud technology, everything has become smart and linked together. Radio frequency identification technologies play a critical role in enabling the ease of these connections. Moreover, RFID helped to improve the efficiency and effectiveness of various processes, services, and products in different organizations. This technology became popular by the 1970s, and since then, we have been surrounded by it wherever we go. Every person has more than one RFID tag in their wallet, for example, a control access card, a passport, a bank card, an ID card, etc. This means that these small parts play crucial roles in our daily lives. Therefore, ensuring the proper functioning of this technology is of significant interest. For all of these reasons, this study used combined FMEA and DEA to reach an appropriate priority rank for failures of an RFID system, which helped to decrease the amount of time and money wasted, and increased the system’s efficiency, safety, and reliability.

The framework of this paper is presented in Figure 1. First, the modified FMEA is applied, followed by an aggregation procedure to combine the sub-failure modes of each parent failure mode. Thereafter, the DEA model used the outcomes of the aggregation procedure as inputs/outputs for the efficiency measurement; then, a final rank is assigned to each failure mode. The calculations are explained in Section 2.

3. Materials and Methods

3.1. FMEA Method

FMCEA is a reliability approach that eases the identification, prediction, and analysis of the potential FMs affecting the proper function of the equipment or the process being studied, as well as their causes and effects. This is in order to evaluate and assess the risks resulting from their occurrence and then to suggest the corresponding repairing, adjusting, and precautionary measures to be planned during the conception, construction, or implementation stages. The steps of the FMCEA are explained in detail in [33] and can be summarized as follows:

• Process selection and review;
• Generate ideas regarding possible ways a system or process might malfunction or experience faults;
• Enumerate the possible reasons behind each failure mode and outline the potential consequences associated with them;
• Prepare the FMCEA sheets and define the severity, occurrence, and detection scales;
• Allocate ratings for severity, occurrence, and detection levels corresponding to each failure mode;
• Consolidate the risk assessments contributed by the FMCEA team members;
• Compute the risk priority number (RPN) for every failure mode;
• Rank the failure modes based on their prioritization for action or intervention.

3.2. DEA Method

DEA is an empirical, comparative, and non-parametric method employed for the efficiency measurement. The fundamental roles of DEA are to contrast the efficiency of the decision-making unit under observation, $DM{U}_o$ (o = 1, 2, 3 … n, where n is the number of DMUs) to the efficiencies of all other $DM{U}_j$ (j = 1 … n) within the production possibility set (PPS) according to the resources they use (inputs X = ($x_{1j},x_{2j},\dots ,x_{mj})$) and the outcomes they produce (outputs Y = ($y_{1j},y_{2j},\dots ,y_{sj})$), and to estimate the production frontier (production function). The objective function of the DEA mathematical model maximizes the efficiency; it returns the optimal values of the weights assigned to each input and output under the principle that a given DMU is able to perform better if another DMU can carry out this function, assuming that the inputs and outputs are identical for all DMUs [34,35]. An input-oriented CCR model can be expressed as shown in Equation (1) and modeled in Figure 2.

$$\begin{array}{c}Max {\displaystyle {\displaystyle \sum }_{r=1}^s}{u}_r{y}_{ro}\hfill \\ S.t {\displaystyle {\displaystyle \sum }_{i=1}^m}{v}_i{x}_{io}=1\hfill \\ {{\displaystyle \sum }}_{r=1}^s{u}_r{y}_{rj}-{{\displaystyle \sum }}_{i=1}^m{v}_i{x}_{ij}\le 0 j=1,2,\dots ,n\hfill \\ u_r\ge 0 r=1,2,\dots ,s \hfill \\ v_i\ge 0 i=1,2,\dots ,m\hfill \end{array}$$

(1)

where s is the number of outputs and m is the number of inputs.

3.3. The Proposed FMEA–DEA Method

The conventional FMEA ranks the FMs in descending order of their risk priority number (RPN). For each FM, the RPN is equal to the product of its probability of occurrence (O), severity (S), and non-detection (D). Nevertheless, sometimes, solving a set of FMs with moderate RPNs would be more efficient than solving a single FM with a high RPN. This means that allocating the time and costs needed to solve one single FM with a high RPN to solving a group of FMs whose RPNs’ sum is equal to or larger than the former may be more reasonable. Consequently, it is recommended that the notions of cost and time be incorporated into the traditional FMEA and the risk prioritization method be revised. Furthermore, the causes and effects of the same failure mode are most likely to have different severity, occurrence, and detection rates. This makes it necessary to consider the scaled-down rates related to each cause separately, as well as the interrelationships between them. Since the causes of a given parent failure mode may have different ranks with some gaps, it is necessary to assign them a common or synthesized rank. Moreover, in some cases, the RPN number alone is not efficient in risk prioritization; this leaves us with a question regarding how to introduce the notion of efficiency to deal with this shortage.

The proposed solution in this paper will answer the question concerning which FMs should be fixed first or must be fixed prior if we want to consider the efficiency along with the risk associated with the causes and effects rates. Thus, in order to determine solving which FM is efficient, we are going to use a DEA model that allows us to minimize time, cost, O, S, and D by also considering the highest RPN. Additionally, causes and effects will be treated as sub-failure modes belonging to a given parent failure mode. Each of these sub-failure modes will have its own ratings, and they will be grouped in a later step in order to obtain their common rank. Therefore, in the first stage, a modified FMEA will be proposed; this will be followed by the application of the DEA approach in the second stage.

3.3.1. Stage 1: Modified FMEA

The suggested modified FMEA in this paper can be conducted according to the following steps:

• Step 1.1: Form the FMEA team; let T be the number of experts participating in the study.
• Step 1.2: Perform a detailed study of the system, the process, or the product being analyzed.
• Step 1.3: Identify and predict all possible failure modes (FMs) and their causes and effects, and treat the causes as sub-failure modes (SFMs).
  • Let the total number of failure modes be M, and the total number of sub-failure modes be N; it follows that each failure mode $F{M}_i$ has a number $n_i$ of sub-failure modes $SF{M}_{ij}$, where $i=1,2,3,\dots ,M$and $j=1,\dots ,n_i$. It can be clearly seen that

    $$N={{\displaystyle \sum }}_1^M{\mathrm{n}}_i$$

    (2)
• Step 1.4: Set the rating scale for each risk factor.
• Step 1.5: Prepare the FMEA tableau or sheet; there should be a field for the components, a second for the failures of those components, a third for the causes, and then a fourth for the effects. It must also include columns for risk factors ratings and two additional columns containing the expenses and time necessary to fix each cause of failure (or sub-failure mode).
• Step 1.6: Collect the data from the FMEA tables filled by the T experts.
  • Let $O_{ije}$, $S_{ije}$, $D_{ije}$, $Cos{t}_{ije}$, and $Tim{e}_{ije}$ be the occurrence, the severity, the detection, the cost, and the time rates of the sub-failure mode $SF{M}_{ij}$ assigned by the e^th expert, respectively, where $i=1,\dots ,M$; $j=1,\dots ,n_i \mathrm{and} e=1,\dots T$.
• Step 1.7: Assign a weight ${\lambda }_e$ to the e^th expert reflecting the diversity of the members’ background and experience.

  $${{\displaystyle \sum }}_{e=1}^T{\lambda }_e=1$$

  (3)
• Step 1.8: Aggregate the assessments and construct the $\left(N\times 5\right)$ judgment matrix as follows: for each sub-failure mode $SF{M}_{ij}$, the aggregated rates of the occurrence, severity, detection, cost, and time are $O_{ij}$, $S_{ij}$, $D_{ij}$, $Cos{t}_{ij}$, and $Tim{e}_{ij}$, respectively, and are calculated according to the following equations:

  $$\begin{gathered} O_{ij}={{\displaystyle \sum }}_{e=1}^T{\lambda }_e{O}_{ije}; S_{ij}={{\displaystyle \sum }}_{e=1}^T{\lambda }_e{S}_{ije}; D_{ij}={{\displaystyle \sum }}_{e=1}^T{\lambda }_e{D}_{ije}; \\ Cos{t}_{ij}={{\displaystyle \sum }}_{e=1}^T{\lambda }_{e}Cos{t}_{ije} ; Tim{e}_{ij}={{\displaystyle \sum }}_{e=1}^T{\lambda }_{e}Tim{e}_{ije} \end{gathered}$$

  (4)
• Step 1.9: Compute the RPNs for each sub-failure mode $SF{M}_{ij}$.

  $$RP{N}_{ij}=O_{ij}\times S_{ij}\times D_{ij} ;where i=1,\dots ,M and j=1,\dots ,n_i$$

  (5)
• Step 1.10: Assign a rank ${\mathrm{a}}_{ij}$ (${\mathrm{a}}_{ij}=1,2,\dots N)$for each $SF{M}_{ij}$ related to the $F{M}_i$based on the descending order of the $RP{N}_{ij}$ obtained in step 8.
• Step 1.11: Synthetize the procedure for the$n_i$ sub-failure modes$SF{M}_{ij}$ that belong to the same parent failure mode$F{M}_i$ in order to eliminate the rank gaps. Thus, the common weighted RPN for the SFMs belonging to the same parent FM will attempt to give more weight to the SFM with the highest $RP{N}_{ij}$ (or the smallest $a_{ij}$). On the other hand, solving an FM can be achieved only if all the related SFMs are solved; the total cost and time (aggregated cost and aggregated time) of each FM will be equal to the sum of the costs and times of its related SFMs. As a result, the aggregated occurrence $\left(AGG{O}_i\right)$, the aggregated severity $\left(AGG{S}_i\right)$, the aggregated detection $\left(AGG{D}_i\right)$, the aggregated cost $\left(AGGCOS{T}_i\right)$, and the aggregated time $\left(AGGTIM{E}_i\right)$ will be calculated as follows:

  $$\begin{gathered} AGG{O}_i=\frac{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}\times O_{ij}}{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}}; AGG{S}_i=\frac{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}\times S_{ij}}{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}}; AGG{D}_i=\frac{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}\times D_{ij}}{{{\displaystyle \sum }}_{j=1}^{ n_i}{a}_{ij}}; \\ AGGCOS{T}_i={{\displaystyle \sum }}_{j=1}^{ n_i}Cos{t}_{ij} ; AGGTIM{E}_i={{\displaystyle \sum }}_{j=1}^{ n_i}Tim{e}_{ij} \end{gathered}$$

  (6)
• Step 1.12: Construct the synthesized $\left(M\times 5\right)$ judgement matrix using the results obtained in step 10, and compute the aggregated RPN $\left(AGGRP{N}_i\right)$ for each $F{M}_i$.

  $$AGGRP{N}_i=AGG{O}_i\times AGG{S}_i\times AGG{D}_i$$

  (7)
• Step 1.13: Assign a rank ($b_i$) to each $F{M}_i$ according to the dropping order of the aggregated RPN ($AGGRP{N}_i$). All the sub-failure modes of a given parent failure mode will be assigned the same ($b_i$).

3.3.2. Stage 2: DEA Approach and Model Selection

We are interested in estimating the efficiency of each failure mode; consequently, the FMs will be considered as DMUs. Since the selection of the input and the output vectors for DEA models is of great importance and has a significant impact on the final results, it is good practice to make these decisions carefully. Starting from the assumption that, on the one hand, the inputs comprise what is in our hand now, and on the other hand, the outputs consist of what we will receive as a result, the RPN, O, S, and D would be suitable as inputs, while cost and time should be considered as outputs. This is because cost and time would not be spent if the corresponding RPN or FM was not selected for action. Moreover, the DEA model attempts to increase the outputs and decrease the inputs; in our case, we expect to consume a minimum amount of time and expenses to solve a maximum number of failure modes with relatively high RPNs. To reach this goal, it is recommended to set O, S, D, and the reciprocal RPN as inputs, while the inverse of the cost and time ought to be regarded as outputs. From standard DEA models, a CCR-input-oriented model has been selected to be applied in this study. This is because, for any FM, if the occurrence, detection, severity, and RPN increase, then its repairing cost and time will be proportionally increased. This indicates a constant return to scale assumption. The DEA model inputs/outputs and DMUs are modeled in Figure 3.

The steps of the second stage are as follows:

• Step 2.1: Compute the efficiency ${\theta }_i$ for each failure mode $F{M}_i$ according to model represented in Figure 3 and Equation (1).
• Step 2.2: Assign a rank ($C_i$) to each $F{M}_i$ according to the dropping order of their efficiencies ${\theta }_i$.
• Step 2.3: The $F{M}_i$ with equal efficiencies (i.e., Equal $C_i)$ will be ranked according to their rank ($b_i$) obtained in stage 1.
• Step 2.4: Assign a final rank ($d_i$) to each $F{M}_i$ and provide the FMEA–DEA report.

4. Case Study: Application of the Modified FMEA–DEA for the RFID System

4.1. An Overview of the RFID System

RFID is a wireless technology using the radio frequency electromagnetic (EM) field. The RFID system consists of the tag, reader, interface software, and host. The proper functionality of the RFID system is contingent upon the logical connection and compatibility of all of these components with each other. The components and the functioning of the RFID system can be found in [36,37,38] and are illustrated in Figure 4.

The collection and the identification of failure modes, their causes, and the effects, as well as current control processes, have been performed based on the team members’ brainstorming, interviews, and communications with many experts, along with the information retrieved from different websites and studies from the existing literature. The effects of RF radiation on human health were disclosed in [39]; meanwhile, the environmental impacts of the RFID and ICT technologies are discussed in [40,41,42]. A failure modes, mechanisms, and effects analysis (FMMEA) was used for testing the environment of qualification of the RFID system [43]; in the same context, the reliability of RFID tags was tested in humid [44] and harsh environmental applications [45] and under thermal storage [46,47]. Conversely, as stated in [48], faults may arise due to temporary antenna tearing, electromagnetic interferences, and optical inductions; the results showed that most vulnerability was related to implementation. Under the same scope, different errors between readers and RFID tags were mentioned in [49,50,51]. FMEA was applied to assess the analog part of the RFID system, presuming faultlessness in the digital segment, as indicated in [52]. Additionally, the FMEA methodology was utilized to scrutinize failure modes within the RFID middleware [53] and to investigate user misconfigurations [54]. Evaluations of RFID device performance within the blood supply chain were conducted, utilizing FMEA to identify departments necessitating reengineering [55]. A review of security concerns in RFID systems was presented in [56]. Furthermore, [57,58,59] outlined the issues and challenges encountered with RFID technology.

4.2. Data Collection and Failure Identification (Steps 1.1–1.3)

A risk assessment team of five groups of experts (not individuals) has been formed. Details regarding the leader of each group are presented in

Table 1. Expert Information.

	Group 1	Group 2	Group 3	Group 4	Group 5
Age	60	35	29	65	25
Experience	Professor at EMU, Northern Cyprus, for more than 30 years	Engineer at Lafarge, Casablanca, for more than 10 years	Mechatronic Engineer at Sews, Morocco, and part-time teacher at ENSA for more than 5 years	Electrical engineer and responsible for customer support at Siemens, Germany, for more than 40 years.	Electrical and electronic engineer assistant in a company for repairing electrical boards and researcher in Paris for more than 3 years.
Education	PhD and more in addition to hundreds or research papers.	Master of Informatic Engineering	Master of Mechatronic Engineering and Specialized Technician Diploma	Master of Electrical and Electronic Engineering	Master of Data Science and a PhD student in the Electrical and Electronics Department

.

Table 2. Failure Modes, Causes, and Effects of the RFID System.

FM Number (i)	RFID SYSTEM	Sub Systems	Failure Modes (FMs)	Number of SFMs (ni)	SFM Number (j)	Cause (s) of Failure (Sub-Failure Modes SFMs)	Effects
1	TAG	The power supply of the label	Inefficient conversion of DC energy from RF	2	1	The antenna provides insufficient power.	The tag is unable to turn on because of not receiving a correct feed; consequently, it does not send signals
					2	Threshold voltage alteration due to the combined diode and MOS.
2			Excess transformation of electrical power	1	1	Excessive energy emitted by the reader	The label is powered and communicates with the reader when it should not.
3		Integrated circuit and connections	Short-circuiting between bumps.	1	1	Adhesive becoming over-cured with age.	Incorrect or deteriorated operation of the tag.
4			Short-circuiting between the antenna pads.	1	1	Adhesive being under-cured due to high temperature.
5			Inadequate impedance among bumps.	1	1	Antenna corrosion occurring in low temperatures.
6			Insufficient impedance among the antenna pads.	1	1	Bumps corroding due to humidity.
7			An open connection between the bump and the antenna pad.	1	1	Filter particle corrosion resulting from temperature and humidity.
8			Excessive impedance between the bump and the antenna pads at the bonding interface.	1	1	Adhesive swelling caused by temperature fluctuations.
9			Partial fracture of the antenna lead, causing excessive impedance within the antenna.	1	1	Die lift because of humidity cycling
10			The antenna exhibits excessive impedance due to an open circuit.	1	1	Die disjunction from adhesive resulting from temperature and humidity fluctuations
11			The connection between the bump and the antenna pad is intermittent or discontinuous.	1	1	Adhesive detached from antenna because of ESD
12			There is a short circuit in the IC (integrated circuit).	1	1	Empty spaces in the adhesive caused by excessive bonding force.
13			Open in IC	1	1	Lack of compression for the filler because of insufficient bond force
14			Defective IC	2	1	Over bond temperature	Defective tag
					2	The electrical overstress (EOS) causes current leakage
15			Insufficient spacing between the antenna and the chip.	6	1	The bonding temperature is inadequate.	Possible non-detection of the tag
					2	The bonding time exceeds the required duration.
					3	Bonding time inferior to that required
					4	Mechanical crop
					5	Mechanical curving
					6	Mechanical squeezing
16		EEPROM	Insufficient memory	1	1	Limited or inadequate storage memory	Loss of the data
17			The EPC number in EPC memory bank of the tag is identical for two tags or more.	1	1	The electronic product code (EPC) is a randomly repeated number sequence utilized by the manufacturer.	The base station is unable to distinguish between two tags
18			Contactless cards are more susceptible compared to regular credit cards.	1	1	An appropriate reader can query the tag if it lacks proper security measures.	Ethical problems in addition to cloning and theft
19			Decryption of user data in the user memory bank.	1	1	Insufficient or inefficient tag encryption	Privacy issues along with tracking, blackmail, and coercion
20			Unauthorized access to tag data in the reserved memory bank.	1	1	Insufficient security measures in the access and lock passwords.	The tag can be hacked, copied, counterfeited
21	TAG	Positioning	Wrong tag orientation	1	1	Misalignment between the orientation of the tag’s antenna and the reader’s antenna.	The reader may fail to detect the tag, or it might detect it with errors.
22			Steep angle of the tag during interaction.	1	1	The tag’s front is not oriented towards the antenna.	Limited read range.
23		Application surface	Incompatibility between the surface material and tag type.	1	1	The antenna’s signal transmission and reception are highly sensitive to the material it is placed on.	Decreased read range, reduced read rate, or no reading capability.
24		Tag attachment	Detachment of the tag from the item.	1	1	Attachment damage caused by environmental factors like dust, water, UV light, chemicals, and temperature affecting the lifespan.	Inability to track the item.
25		Encasement	Failure to safeguard the tag IC and antenna.	1	1	Encasement damage in harsh environments such as exposure to water, extreme temperatures, or metallic surfaces.	Possible potential damage to the tag.
26	Reader	Reader API communications event management	Reader collision issue (reader–reader collision)	2	1	Interference: reader communicating with tags covered by another reader.	Simultaneous multiple readings of the tag by overlapping readers.
					2	Overlapping coverage areas of two readers.	Signal interference due to overlapping coverage areas of readers.
27			Tag collision issue (tag–tag collision)	1	1	Simultaneous data transmission attempts from multiple tags.	Reader’s inability to differentiate between signals.
28			Interference between multiple readers (reader–tag collision)	2	1	Inadequate distance between two readers.	Incorrect information
					2	Incorrect operating frequency settings for two adjacent readers.
29			Short reading distance	5	1	Incorrect reader frequency settings.	Tag is not read
					2	Mismatched label and antenna polarization.
					3	Label surface obstructed by other materials like metal.
					4	Unconnected RF cable between reader and antenna.
					5	RFID label’s specifications.
30			Cannot read card	6	1	Improperly connected serial or network cable.	The PC command is unable to be transmitted to the reader.
					2	Inadequately tightened RFID antenna SMA connector.	Missed reading
					3	Mismatch in ISO standards between the transponder and the reader.
					4	Size discrepancy between the tag and the receiver construction.
					5	Frequency mismatch between the tag and the reader.
					6	Damaged label.
31			Accidental reading of EPC (electronic product code) instead of TID (tag identification data).	1	1	Inadequate reader settings.	Incorrect ID, thus wrong information
32			Self-jammer interference.	1	1	Continuous wave signal sent to the tag.	Saturation of the receiver block leading to a decline in sensitivity.
33			The rectifier converts low-level sinusoidal voltage into DC voltage.	1	1	Mismatch in power link between the antenna and the rectifier input.	Incorrect power feed of the integrated circuit of the reader
34	The “between” tag and reader	Antenna modem	Failure to receive signals by the tag or reader.	4	1	Defective sensitivity detection of the tag or the reader.	Loss of information
					2	Electromagnetic field disturbances.
					3	External aggression affecting the antennas in the tag or the reader.
					4	Internal failure in the reader.
35			Inability to transmit signals to the tag or the reader.	2	1	Internal failure in the tag.
					2	Repetitive software attempts to communicate.
36			The transmission in continuous	3	1	Interior failure in the tag	Channel congestion or overload.
					2	Interior failure in the reader
					3	Repetitive trials to communicate by the software
37		Electromagnetic field	The electromagnetic waves emitted by the tag cannot reach the reader, and vice versa.	1	1	Unfavorable environmental conditions (metallic, liquid, etc.).	Loss of information
38	HOS	Computer middleware database, SD card	Hack	1	1	Inadequate or ineffective protective measures/tools.	Breaches in users’ privacy and security.
39			Software bugs	1	1	Ignorance, big data, slow system	Incorrect, duplicate and missed reading
40			Data transfer	1	1	Weak or faulty connection between the reader and the host (Wi-Fi/serial interface/Bluetooth, etc.).	The host fails to receive data and does not transmit information to the tag.
41			Virus attack	1	1	Inappropriate antivirus protection	Unauthorized access, compromised functionality, or system malfunction.
42	Health	RF waves	Thermal effects: can potentially harm the eye lens, causing damage, and heat cells and biological tissues. Possibly carcinogenic: There is a potential risk of causing cancer.	1	1	The distance between the reader and the human body is within 10 cm.	Elevated specific absorption rate (SAR) of the human head, surpassing 1.6, which is the maximum value permitted by the FCC in the US.
43	Green IT	heavy metals, silicon, aluminum, plastics: PVC or PA6	The soaring quantity of RFID tags is leading to a significant consumption of metals and toxic materials.	1	1	Chemical reactions occurring with the environment.	Pollution and use of rare metals

contains a compilation of identified failure modes along with their respective causes and effects.

As is seen in Table 1, a failure mode (FM) may contain sub-failure modes (SFMs) due to multiple causes.

4.3. Rating and Aggregating the Failure Modes (Steps 1.4–1.10)

Based on data provided by experts (Supplementary Material), the total number of the identified failure modes (FMs) is M = 43, with a total number of sub-failure modes (SFMs) of N = 67; the number of sub-failure modes $n_i$ related to each parent failure mode are depicted in Table 2; the rating scale for each sub-failure mode against each risk factor is from one to five. The number of experts participating in the risk assessment is T = 5; the teams consist of groups of experts and not individuals. Hence, there was insufficient evidence to highlight significant differences among FMEA team members regarding their assessment capabilities; therefore, we assumed an equal weight ${\lambda }_e=1/5 \mathrm{for} \mathrm{all} \mathrm{team} \mathrm{members}$. The outcomes of the aggregation of risk factor ratings are presented in detail in Appendix C.

Table 3. The gaps between the $SF{M}_{ij}$ of each parent $F{M}_i$.

FMi	1	14	15	26	28	29	30	34	35	36
Min ${\mathrm{SFM}}_{\mathrm{ij}}$ rank $(\mathit{min}{a}_{ij})$	9	27	20	46	37	4	7	6	3	11
Max ${\mathrm{SFM}}_{\mathrm{ij}}$ rank $(\mathit{max}{a}_{ij})$	23	36	64	60	39	66	62	38	8	40
Gap $(\mathit{max}{a}_{ij}-\mathit{min}{a}_{ij})$	14	9	44	34	2	62	55	32	5	29

presents the failure modes with sub-failure modes, as well as the ranking gaps among the latter.

The results presented in Table 3 show that there are large gaps between the minimum and the maximum rank of the sub-failure modes of each parent failure mode. In the next step, these gaps will be eliminated by applying a synthetization procedure.

4.4. Synthetization Procedure for the Sub-Failure Modes (Steps 1.11–1.13)

The outcomes of the modified FMEA and the sub-failure mode synthetization are summarized in

Table 4. The synthetization procedure outcomes.

FMi	AGGSi	AGGOi	AGGDi	AGGCOSTi (USD)	AGGTIMEi (Minutes)	AGGRPNi	Rank bi
1	3.528	2.64	2.764	0.50	90.00	25.7512	9
2	2	2	2.2	0.25	45.00	8.8	43
3	2.5	2.75	3	35.25	95.00	20.625	19
4	2.25	3	3.5	25.25	85.00	23.625	13
5	2.75	1.75	2.5	0.25	75.00	12.03125	40
6	2.5	2.25	2.75	25.25	85.00	15.46875	29
7	2.75	2	2.75	25.25	85.00	15.125	32
8	3.25	2.25	2.25	25.25	85.00	16.453	26
9	3.5	2.25	2.75	35.25	95.00	21.65	17
10	3.75	2.5	2.75	35.25	95.00	25.78125	8
11	3.75	1.75	2.25	25.25	85.00	14.76562	33
12	4	2	2.75	25.25	85.00	22	16
13	3.5	2.5	3	35.25	95.00	26.25	7
14	4.107	1.75	2.857	26.00	55.00	20.535	20
15	2.689	1.786	3.16	161.50	520.00	15.187	31
16	3.8	2	2	0.50	45.00	15.2	30
17	4	2	3	35.25	42.50	24	12
18	4.25	2.25	3.5	35.50	95.00	33.46875	2
19	4.25	2.25	3	35.50	95.00	28.6875	5
20	4.2	1.8	3	35.50	95.00	22.68	14
21	3.2	2.8	3	25.25	37.50	26.88	6
22	1.6	2.8	2.6	25.25	27.50	11.648	41
23	2.4	2.2	2.6	35.50	50.00	13.728	38
24	2.6	2.2	3	35.50	42.50	17.16	24
25	4.2	1.8	3.4	25.50	55.00	25.704	10
26	3.8	1.6	2.373584	110.00	180.00	14.43139	35
27	4	2	1.75	50.00	85.00	14	37
28	3.9026	2.097368	2.2	120.00	190.00	18.00756	23
29	2.866	2.02269	2.516312	176.25	260.00	14.59056	34
30	3.2530	1.5836	3.052244	161.50	340.00	15.72451	28
31	4.4	1.2	3	35.25	50.00	15.84	27
32	3.2	1.4	3	35.25	50.00	13.44	39
33	3.25	1.75	2.5	25.25	40.00	14.21875	36
34	4.4520	1.91780	2.638356	260.50	500.00	22.52677	15
35	4.1636	2.290909	3.309090	101.00	95.00	31.56380	3
36	3.6823	1.764705	3.229411	96.00	155.00	20.98558	18
37	4.2	3.2	2.2	27.50	55.00	29.568	4
38	5	1.5	3.25	40.00	85.00	24.375	11
39	3.2	2.6	2	25.25	55.00	16.64	25
40	4.4	3.2	2.6	30.00	32.50	36.608	1
41	4	2.5	2	45.00	55.00	20	22
42	3	1.8	3.8	37.50	95.00	20.52	21
43	2	2	2.4	10.00	265.00	9.6	42

.

As the results show, most of the SFM priorities changed when applying the synthetization procedure and considering the relationship between them to rank their parent FM. The synthetization yielded three categories of SFMs; the first category groups the SFMs with fewer priorities, and the traditional FMEA won in terms of ranks when synthetized; however, the consequences of these failures drastically affect the proper functioning of the system. For instance, the ${\mathrm{SFM}}_{29,4}$ (the sub-failure mode 4 of the parent failure mode 29) previously represented more than 1.49% of the failures, which means that it was not considered to be a very critical failure; however, in reality, the tag would not be detected by the reader if the reading distance was very short, leading to the system not functioning at all. The second category consists of the SFMs whose priority did not change or changed slightly for both methods. Finally, the third category is composed of the SFMs who lost ranks. Contrary to the SFMs of the first category, these failures became less important in terms of the risk they may cause in terms of degrading the operation of the system, but they did not lead to the malfunctioning of the RFID components. The occurrence is not very frequent, and this can be minimized by the application of simple preventive actions such as choosing the right tags and selecting the correct settings of the different parts for the right applications.

4.5. Measurement of the Efficiency of the FMs by DEA (Steps 2.1–2.4)

The outcomes of the modified FMEA and the synthetization procedure served as inputs/outputs of the DEA model explained in Section 3.3.1. The efficiencies of DMUs (FMs) were found using PIM DEA version 3.2; the application of the steps of stage two explained in Section 3.3.2 yielded the results in Appendix C, as summarized in Figure 5.

The ranks of the failure modes in the modified FMEA and the FMEA–DEA methods are contrasted in Figure 6. After the application of the DEA and the incorporation of time and cost, most FM ranks seemingly changed. Consequently, we obtained three categories of FMs: those with increased priority, those with decreased priority, and the few FMs that retained their position. For instance, $F{M}_2$ had minimal priority in the FMEA method, but when efficiency was incorporated together with cost and time, its priority rank switched from 43 to 9. This FM concerns the settings that can be adjusted in a short duration of time with the minimum dispensed, whilst not fixing it would lead to improper or even non-functioning of the system. At the same time, the failure mode $F{M}_{22}$, which is related to the positioning of the tag, was initially ranked last and moved into eighth by winning 33 ranks. Similarly, $F{M}_5$, caused by corrosion, won 33 ranks; this can be avoided by effectively choosing the tag. Meanwhile, the failure mode $F{M}_{34}$ moved down by 34 ranks, which was related to the age of the tag and the non-reception of the signals due to environmental disturbances. This FM occurs infrequently because the life cycle of the tags is usually sufficient for the destined application. Moreover, attempting to deal with the space between the tag and the reader would be very costly and is often uncontrollable.

Table 5. Suggested replacements of some FMs that lost ranks.

FMs with Increased Priority				FMs with Decreased Priority
Failure Modes	Sum of the Costs (USD)	Sum of the Time (Minutes)	Sum of the RPNs	Failure Mode	Cost (USD)	Time (Minutes)	RPN
31 and 17	70.5	92.5	39.84	35	101	95	31.563
16 and 33	25.75	85	29.418	19	35.5	95	28.687
1 and 24	36	132.5	42.911	36	96	155	20.985
2, 5 and 22	25.75	147.5	32.479	34	260.5	500	22.52

presents some of the FMs that became high-ranking against others that lost ranks, in addition to suggested replacements; obviously, other combinations are possible. As an example, we can use reduced time and cost to solve failure modes 2, 5, and 22 instead of $F{M}_{34}$. Furthermore, the summation of their RPNs is 32.479. This is more than the RPN of $F{M}_{34}$, which is equal to 22.52. Another example is $F{M}_{35}$, which is related to the signal transmission between the tag and the reader caused by wave disturbance. This dropped to the 28th rank by losing 25 positions with an RPN equal to 31.563; solving this FM requires USD 101 and 95 min. Thus, we can solve $F{M}_{31}$and $F{M}_{17}$ in less time (50 + 42.5 = 70.5 min) and at a cheaper cost (USD 35.25 + USD 35.25 = USD 70.5), with higher resulting RPN (15.84 + 24 = 39.84). This is because $F{M}_{31}$ is related to the reader setting, which is easy and important, and $F{M}_{17}$ occurs when different tags have the same EPC number and can be eliminated solely by quickly checking the tag numbers before using them in applications or projects.

In light of what has been stated previously, we can conclude that the most efficient FMs concern the transfer of data, source of energy, settings, and choice of tags (memory, humidity, life cycle, temperature), as well as the settling of the transponder and the base station.

5. Comparison and Discussion

5.1. Comparison of the Proposed Method and the Conventional FMEA

To demonstrate the efficacy of the suggested model, the conventional risk priority, the synthetized FMEA, and the FMEA–DEA have been applied in the given case study. Figure 7 contrasts the results of the conventional FMEA method and the proposed FMEA–DEA method in this study.

The rankings of sixty-seven sub-failure modes obtained through the three approaches applied in stages one and two are depicted in Figure 8. The percentage priorities are used instead of ranks in order to compare the three methods. Further elaboration on the computation of the priority percentages can be found in Appendix C. Based on the results in Figure 7 and Figure 8, observable distinctions exist among the risk priority rankings generated by the three FMEA methods. This disparity highlights that the conventional RPN, the synthesized FMEA, and the proposed model employ distinct mechanisms for establishing the risk priority ranking of failure modes.

From Figure 8, it can be seen that 21% of SFMs have similar priorities in both the conventional and FMEA–DEA methods, of which 12% had similar priorities in all three methods. Meanwhile, 79% of the SFMs had different priorities in the conventional FMEA and FMEA–DEA methods, of which 49% had similar priorities in the conventional and the synthetized FMEA methods but different priorities when DEA was applied; furthermore, 30% had different priorities in all the three methods. SFMs in category one maintained almost the same rank in all three methods. For example, $F{M}_{40,1}$ and $F{M}_{25,1}$, related to data transfer and encasement, respectively, are high-ranking. This is contrary to $F{M}_{6,1}$, $F{M}_{8,1}$, $F{M}_{26,1}$, $F{M}_{42,1}$, and $F{M}_{43,1}$, which are related to the surrounding conditions of the tags and also the environmental impacts of the RFID parts. Thus, solving the former failures will decrease the probability of associated problems. Other examples are $F{M}_{2,1}$ and $F{M}_{17,1}$ (category 3), which had a minor priority in both FMEA methods but became high-ranking when efficiency was considered. This is because the former is related to simple reader settings, and the latter is related to duplicate identifiers, and this can be solved in a short timeframe with low expenses. On the contrary, $F{M}_{18,1}$ and $F{M}_{35,1}$ (category 4) are very risky but inefficient; the risk is related to the exposed tags, while the inefficiency is related to the tag antenna failure, and both require considerable expenses and time to make the users aware of the importance of protecting their labels and to repair the infinite small circuit in the latter. SFMs in categories five and six were high-ranking in terms of risk in both FMEA methods and strongly decreased their priorities in terms of efficiency. For instance, $F{M}_{34,2}$ related to the electromagnetic field disturbance is risky but often not in control. Instead, we can decrease the distance between the tag and the reader and ensure that there is no obstacle such as liquid or metal, and the same time and cost can be spent solving $F{M}_{22,1},$which belongs to category three.

5.2. Comparison of the Proposed Method, the AP Method, and the PRISM Method

There are limited failure mode analyses within the RFID field, primarily focusing on either the tags [43] or the readers [54]. However, in a specific study [52], the authors delved into examining 12 failure modes associated with the tag, reader, and link layers. It is important to note that in these referenced studies, the use of FMEA was not aimed at ranking the failures but rather at identifying them. Consequently, the data provided could not be utilized for comparison purposes.

The findings from our study will be employed to utilize various methods outlined in the existing literature. These include the conventional FMEA, the partial risk map (PRISM) detailed in [18], and the application priority method referenced in [19] and elucidated in [60]. Figure 9 illustrates the results pertaining to the ranking and classification of the six methods.

From Figure 9, it is evident that the outcomes of the three distinct functions within the PRISM method exhibit close similarities but notably differ from both the conventional FMEA and the application priority (AP) method. Specifically, around 73% of failures were given higher priority in the PRISM method compared to conventional FMEA. Moreover, the AP categorizations differ from both the PRISM and conventional methods for approximately 80% of the failures.

The newly proposed FMEA–DEA method showcases disparities with the conventional FMEA in 80% of the failure modes (FMs) and diverges entirely from PRISM in all failure modes. In particular, approximately 73% of the failures were assigned a lower priority in the FMEA–DEA method compared to other methods. Additionally, the AP classification notably differs from the proposed FMEA–DEA method by approximately 67%.

In summary, it is evident that different methods yield distinct results. This was anticipated due to their individual mechanisms for evaluating and prioritizing risk across failure modes.

Table 6. The similarities among the compared methods.

	Conventional FMEA	PRISM	New FMEA–DEA	AP
Conventional FMEA	-	0%	20%	19%
PRISM		-	0%	20%
New FMEA–DEA			-	33%
AP				-

offers a summary of the percentage similarities between the outcomes of the compared methods.

5.3. Limitations of the Proposed Method

Although this method has its advantages, it also has some disadvantages. The limitations of this method can be summarized as follows:

(i)

Data collection and team forming can be critical;

(ii)

It considers equal weights for the risk assessors;

(iii)

Fuzzy logic or linguistic variables can be employed to ensure consistency among team members;

(iv)

Using an advanced calculation method for the RPN, such as PRISM, would be an improvement.

6. Conclusions

In this study, we incorporated the DEA method and cost and time notions for the purpose of evaluating the risk associated with failure modes in FMEA. Firstly, a detailed study of the RFID system was performed. This was carried out in order to shed light on its possible failures and their causes and effects by means of a modified FMEA that considers the relationships among the sub-failures and eliminates the ranking gaps between them. Secondly, the DEA method was employed to calculate the efficiency rates of each failure mode, as well as the coefficients of the risk factors and the risk indicators. The cost and time needed to fix them served as inputs/outputs for the DEA model.

To provide an illustration, the RFID showed that the final FM ranks resulting from the FMEA–DEA method are based on both risk and efficiency, and they are reliable and rational more often than those obtained by the traditional FMEA. Therefore, this method is effective and useful, and it may be generalized and used for assigning risk priority rates of failure modes in different applications, equipment, or processes.

In future research, attention should be directed towards six specific areas. Initially, this will include model rate failure mode risk factors utilizing five-point numeric scales. Given the intricacy of risk analysis problems, FMEA team members often encounter challenges in providing precise assessment information for failure modes concerning each risk factor. Consequently, future exploration could involve integrating the proposed FMEA with fuzzy theories like linguistic fuzzy sets. Secondly, within the proposed FMEA framework, the assumption is that the relative weights of risk assessors are equal. In some situations, the team members have varying backgrounds and knowledge. Therefore, it is crucial to devise efficient techniques for determining the suitable weights of experts, such as AHP or DEA, in future research. Third, other methods can be used to model the interrelationships between failure, causes, and effects, such as DEMATEL. Fourth, a variety of MCDM methods can be incorporated; different combinations can be obtained from Appendix B. Fifth, improved methods for RPN calculation can be used, such as PRISM instead of the traditional method, as presented in Appendix B. Sixth, apart from the RFID system, conducting additional case studies in the future will be beneficial to further validate the proposed risk priority model.