Developing ANFIS-FMEA model for assessment and prioritization of potential trouble factors in Iraqi building projects

1 Introduction

The construction sector is subject to notable challenges and is distinguished by elevated levels of risk and unpredictability. Numerous projects have encountered high cost and/or timeline fluctuations due to various challenges throughout the life cycle, ultimately rendering these projects unsuccessful.

Several researchers have attributed project difficulties and failures to ineffective management of trouble events, which can result in neglecting milestones and objectives. Effective project recovery management has emerged as a critical component of successful project management [1–3]. Therefore, developing innovative approaches for evaluating factors contributing to project difficulties and implementing project management strategies is significant. Many policymakers are currently seeking to establish an efficient and effective support system for managing trouble factors in construction projects within the industry to facilitate their recovery.

To mitigate the adverse effects of trouble factors in troubled projects, it is imperative first to conduct a thorough evaluation and identification of the necessary prerequisites for successful planning. Subsequently, implementing effective and efficient corrective and preventive measures is crucial. Therefore, enhancing awareness of the present circumstances is the optimal approach to making the decision effectively. The primary aspect of evaluating trouble factors is to analyze and recognize the current state of affairs [4]. Hence, meticulous planning is imperative to attain high reliability and availability objectives. The methodology employed in this research involves utilizing a priority ranking system for potential trouble factors based on specific decision criteria.

Several authors [5–9] have highlighted that the Failure Mode and Effects Analysis (FMEA) method is highly effective and may be considered the optimal approach in this field, the most prevalent and customary method for identifying the potential causes of project trouble. The conventional FMEA methodology has been widely employed in numerous industries and is a highly favored approach for reliability modeling. However, it has specific weaknesses and limitations and it suffers from certain shortcomings which were pointed out by several researchers [10–15], they have pointed out that traditional FMEA assumes that the three parameters (P,I,DCD) of FMEA have the same importance, Furthermore, it tends to produce exactly the same value of trouble priority number (TPN) despite a different set of (P,I,DCD) rating. Additionally, it needs to consider indirect relationships between parameters (P,I,DCD).

Many models have been presented in the literature to improve the effectiveness of FMEA and boost trust in the results in light of criticisms levelled against the FMEA by integrating the classic FMEA approach as an efficient tool with additional tools. There are three methodologies proposed to improve traditional FMEA in literature: The first methodology includes combining FMEA and “multi-criteria decision-making” (MCDM) methods such as “analytic hierarchy process” (AHP), “technique for order of preferences by similarity to ideal solution” (TOPSIS), “analytic network process” (ANP) and others. The second methodology uses fuzzy systems to enhance FMEA. The third methodology combines fuzzy systems with (MCDM) methods to improve and enhance FMEA.

Beiki Ashkezari et al. [16] proposed a method by integrating the FMEA as a practical method in reliability engineering and the MADM techniques through using AHP and TOPSIS as failure evaluation tools. The research results show that the suggested method helps the contractor to prioritize all potential failures and their failure modes in order to take corrective action and minimize the risk of a failure occurring, which leads to more effective implementation of the construction project and the fulfillment of the client’s requirements. Anugerah et al. [17] had used FMEA combined with the AHP to prioritize the sustainable related risk in the supply chains. The findings indicated that the most significant risks are an unsafe and hazardous working environment, natural disasters, and unreliable transportation networks. The accuracy of prioritizing the sustainable supply chain risk has been improved by the suggested modified method of risk management. Gupta and Gaikwad [18] used FMEA by combining with ANP to rank the most critical risk that may happen at a precast concrete plant. The findings indicate that six risks had the same “Risk Priority Number” (RPN) and the most critical risk occurring at the precast concrete plant is injury on the job site. Benbachir et al. [19] developed a practical and effective decision-making tool to ensure the sustainability of complex sewerage systems through prioritizing the failures and breakages of these systems. The proposed approach combines both FMEA and AHP methods. The AHP method was used to deal with complex decision-making problems.

After the Fuzzy-FMEA was presented, many researchers began to enhance this methodology in their research. Kamyab et al. [20] provided a methodology for assessing the safety performance in a machine-building company using the combination of FMEA and FIS. The safety management level, senior management level, and culture level were organizational elements that affected safety performance in this study. The findings of the current study indicated that the organization under study had a computed safety performance score of 75.3. Macura et al. [21] used Fuzzy Logic (Interval Type-2)-based FMEA to get a better understanding of the risk events that happen in railway infrastructure projects in the Republic of Serbia. The uncertainty in risk assessment is dealt with using Fuzzy Logic (Interval Type-2). Wang et al. [22] had combined the FMEA with the Fuzzy evaluation systems, the results show that the use of the benefits of fuzzy systems can compensate the FMEA method’s drawbacks and mitigate any losses brought by project risk, further improving the risk evaluation method, appropriate analysis, and control of the risks that may happen in the construction project. Tavakolan and Mohammadi [23] provided a combined method employing fuzzy systems and FMEA to evaluate the risks in construction projects and overcome the shortcomings of conventional FMEA in the construction sector. The findings show that the method of fuzzy comprehensive evaluation combines qualitative and quantitative analysis. It combines their benefits and enhances the risk assessment process. Cheng and Lu [24] proposed a method to assess the risk of pipe jacking construction by combining FMEA and fuzzy inference. This approach employed a three-step framework to represent the relationship between severity (S), occurrence (O), and detection (D) with the level of criticality of risk occurrences for the purpose of identifying and prioritizing the technical challenges. Sharma et al. [8] created an approach to support decisions for FMEA using fuzzy logic. This system included 384 fuzzy (if-then) rules, which contributed to making it simple to utilize for non-experienced beginners. Rule-based procedures are not always the best choice when comparing the various approaches because such methods require a wide variety of “if–then” situations. Because of these shortcomings, rule-based methods have drawn criticism

One of other available solutions to overcome these shortcomings is integrating FMEA method with fuzzy logic and MCDM approaches. Heidary Dahooie et al. [25] attempted to employ Stepwise Weight Assessment Ratio Analysis as the MCDM method and the fuzzy logic system to provide a better approach to the FMEA method. The findings show that the suggested method may produce logical and accurate priority rankings for occupational risk in the construction sector and is beneficial and flexible for dealing with complicated FMEA constraints. Boral et al. [26] suggested a novel integrated method for fuzzy-FMEA utilizing fuzzy “Multi-Attribute Ideal Real Comparative Analysis” (MAIRCA) and fuzzy FAHP. They initially used the FAHP approach to determine the relative importance of risk events, and then they introduced the adapted fuzzy MAIRCA (FMAIRCA) to evaluate potential risks. The findings of their study indicate that their improved FMAIRCA approach requires less computational effort and is capable of creating better decisions. Fattahi et al. [27] suggested an approach to enhance the FMEA method’s accuracy. The RPN for each failure has been replaced in their method by the “Fuzzy Weighted RPN”. The weights of three factors and failure modes were determined using the fuzzy-AHP and fuzzy “Multi Objective Optimization Based on Ratio Analysis” approaches. Jamshidi et al. [28] developed a comprehensive methodology which combines fuzzy FMEA with “Grey Relational Analysis” for analyzing risk in ERP projects by considering uncertainties. This approach suggested an organized method during the risk assessment process. In another study, Vahdani et al. [29] introduced a novel FMEA technique that used fuzzy logic and TOPSIS to enhance the risk assessment procedure. FMEA and the fuzzy TOPSIS approach were combined to create a new risk assessment method that addresses the shortcomings of the traditional method. The findings showed that implementing the suggested method helps in assessing the most significant risks of adopting the “Shortest geometric distance from the Positive Ideal Solution and the greatest geometric distance from the Negative Ideal Solution.”

Finally, authors [30–32] submitted a comprehensive review which included a thorough analysis of the research studies that had been published from 1998 to 2018 on traditional FMEA improvement methods. Findings show that the approach most frequently applied to improve FMEA was the fuzzy logic systems of all the reviewed articles, which is the most popular approach.

From this brief literature review, we conclude that our study will contribute as follows:

1. Developing a novel Adaptive (ANFIS-FMEA) model can be used in Iraqi building projects as an effective tool to support their top management to assess and prioritize potential trouble factors in a timely manner so that effective corrective measures can be made more successfully. The main difference between the suggested method is the integration of the FMEA approach and ANFIS technique which combines the advantages of “artificial neural networks” (ANN) and “fuzzy inference system” (FIS); this integration will contribute in a way by providing a more accurate trouble factors assessment.

2. In this work, the MCDM problem has been solved using the fuzzy AHP approach, where cost impact and time impact must be combined into one term.

3. This research seeks to improve the traditional FMEA by modifying its ranking system to account for the shortcomings of the conventional method of calculating the TPN and thus make it more useful as a decision support tool for use in construction industry recovery management by using ANFIS as it provides accelerated learning capacity and adaptive interpretation capabilities to model complex patterns.

2 FMEA

The FMEA was initially established as a structured design approach during the 1960s by the aerospace sector. Presently, it is widely recognized as a crucial instrument in reliability engineering. The FMEA has been established as a critical proactive measure that investigates, detects, and ranks the possible failure modes in a given system, process, service, or design. In addition, it can decrease failure modes through early identification [11,12].

The conventional FMEA involves assessing the RPN to establish the significance of critical failure modes. This process entails evaluating various factors, including the severity (S), likelihood of occurrence (O), and detection (D) associated with each failure mode. The RPN value obtained indicates the priority of the corresponding risk levels. The index is computed using equation (1), as demonstrated below [13,14].

The RPN is a numerical value that ranges from 1 to 1,000. It is determined by multiplying the Severity (S), likelihood of occurrence (O), and Detection (D) ratings of a failure mode. Components of a system that exhibit a high RPN are considered more crucial than those with lower values. The conventional FMEA methodology employs a numeric continuum spanning from 1 to 10 to denote the Severity (S), likelihood of occurrence (O), and Detection (D) factors. These terms are assigned values, and conclusions can be drawn based on those values [10,15].

The RPN based on the traditional FMEA of trouble factors is obtained by multiplying the parameters (O, S, and D) without considering their relative importance, and their weights are considered the same. In this study, we applied the FMEA approach to assess and identify the priorities of the trouble factors in Iraqi building projects with some modifications. The failure modes, Severity (S), occurrence (O), detectability (D), and RPN, therefore, were substituted by the terms “trouble factors,” “the amalgamated impact (I _T) of cost and time,” “the Probability of Occurrence (P),” “the degree of detection and control for potential trouble factors (DCD),” and “TPN.”

3 Adaptive neural fuzzy inference system (ANFIS)

The technique known as Adaptive Neural Fuzzy Inference System (ANFIS) is a fusion of the computational prowess of ANN and the advanced reasoning ability of a Fuzzy Inference System (FIS) [33]. Jang introduced this technique in 1993 [34], which has since been widely used. To clarify, the “Adaptive Neuro-Fuzzy Inference System” (ANFIS) amalgamates the advantages of Fuzzy Logic and Neural Networks, thereby constituting a suitable technique. The ANFIS employs Neural Network learning techniques to adjust the parameters of a Fuzzy Inference System (FIS) [34,35]. The ANFIS architecture can be obtained by incorporating the FIS into the adaptive network’s framework. The ANFIS is a data learning methodology that employs Fuzzy Logic to convert input data into a desired output. This is achieved through interconnected Neural Network processing elements and weighted information connections, which map numerical inputs to work [34,36].

The ANFIS model has garnered significant attention from scholars in various engineering and scientific disciplines due to its effective learning and reasoning abilities, particularly in light of the heightened uncertainties associated with construction projects. ANFIS has been identified as a potential alternative to be utilized, as evidenced by previous research [37–39]. This study employs ANFIS-FMEA’s learning ability to evaluate trouble factors in construction projects in Iraq. During the training phase, the settings of each node are adjusted to identify the governing rules that dictate the interactions between the input and output. The anticipated result is determined by an adaptive network that utilizes multi-layered ANN with adaptive nodes in a feed-forward manner. Using multiple types of membership functions (MFs) and selecting varying types and quantities of variables in constructing an Adaptive Neuro-Fuzzy Inference System (ANFIS) model can offer alternative options for comparative analysis.

4 The suggested approach for assessment and prioritization of the potential trouble factors in construction projects

In this research, the suggested methodology for the assessment and prioritization of the potential trouble factors in construction projects includes three phases as follows:

Phase 1 (Investigation and collecting required data): The current system used to assess and prioritize potential trouble factors in construction projects troubled projects are studied in this phase. Its specifications are investigated by collecting required data from these projects through a case study for many Iraqi construction projects and by open meetings of the managers of these projects. The database of the suggested model was collected from [40] Building projects.

Phase 2 (Identification of potential trouble factors): Potential trouble factors in these projects are identified and classified in this phase. Various methods and tools are used to determine the main factors of trouble & relevant important data in Building projects. Necessary information was collected through direct access to the relevant documents and other methods, such as questionnaires. The essential trouble factors of troubled projects were determined as described in Table 1.

Phase 3 (Assessment and prioritization of potential trouble factors): This Phase requires analyzing data collected for the potential trouble factors in Phase 2, assessing and finally prioritizing it. This phase includes the following steps:

Step 1: Preparing the three parameters of the traditional FMAE method. In this step, the parameters for each trouble factor [Impact index of trouble factors on objectives of projects (Cost & Time (I _T)), Probability of occurrence index of trouble factors (P), and Detection and Control Degree index for potential trouble factors (DCD [the more excellent value for (DCD) index indicates less ability to Detection and Control a troubling aspect]) are defined according to some recommendations made by PMI [40]. This attitude will help to prioritize the failure mods and their effects. The traditional scale of the factors with Linguistic description adopted for defining parameters of the FMEA is given in Table 2. Each trouble factor is evaluated on a numerical scale from (1) least likely of occurrence, most negligible impact, and most likely to detect trouble to (10) most likely of occurrence, most impact, and least detectable.

Step 2: In this work, the MCDM problem related to the parameter of impact (I) on the project objectives which is divided into two impacts: cost impact (I _C) and time impact (I _t) has been solved using the fuzzy AHP approach, where cost impact and time impact must be combined into one term (I _T). The combined impact of (cost–time, I _T) for any trouble factor will calculate by using Fuzzy analytic hierarchy process (FAHP) for any trouble factor will calculate by using FAHP.

The FAHP is a decision-making methodology incorporating fuzzy theory into the basic Analytic Hierarchy Process (AHP). It is a popular approach for addressing MCDM problems and has been widely utilized. The F AHP was introduced by Saaty [41] and is distinct from the AHP in its ability to manage trusted weights effectively [42]. Integrating the Analytic Hierarchy Process (AHP) methodology and fuzzy Logic provides increased adaptability in decision-making and evaluation. The FAHP is a decision-making methodology that emulates human cognitive processes when faced with imprecise and ambiguous data. The (FAHP) approach retains the fundamental attributes of the Analytic Hierarchy Process (AHP) methodology, enables the handling of quantitative and qualitative data, employs a hierarchical framework and pairwise comparisons, mitigates discordance, and yields weightings [43].

The steps and procedures applied in this method to determine the criteria weights are shown in the steps:

1. Preparing the data for the problem: This stage is represented by preparing the raw data for the problem, which is represented by calculating the relative importance of the severity of the impact on the cost and time of the project.

2. Fuzzy Pairwise Comparison: Construct pairwise comparison matrices among all the criteria in the dimensions of the hierarchy system (cost impact and time impact) to determine the more important of each two dimensions, as shown in Table 2. The triangular FAHP consists of three values which are presented in equation (2) but they are presented in equation (3) if the alternative is weaker [44].

1. Buckley [45] calculates each criterion’s geometric mean of fuzzy comparison values as shown in equation (4). Here, r ∼ i still represents triangular values.

   (4) r i ∼ = ∏ j = 1 n a ij 1 / n , i = 1 , 2 , 3 , … n ,

   where r i ∼ is the geometric mean of the fuzzy comparison values of criterion i to each criterion, a ij is a fuzzy value of the pairwise comparison of criterion i to criterion j, n is the number of criteria, and ∏ j = 1 n . is the pi-product of a ij from j = 1 to j = n.

2. The fuzzy weights of each criterion: [44] The fuzzy weights of each criterion can be found with equation (5)

   (5) w i ∼ = r i ∼ ⊗ ( r 1 ∼ ⊕ r 2 ∼ ⊕ … r n ∼ ) − 1 = ( l w i , mw i , uw i ) ,

   where the lw _i , mw _i , and uw _i stand for the fuzzy weight’s lower, middle, and upper values.

3. De-Fuzzily the weights of each criterion: the Centre of area method proposed by Chou and Chang [46], via applying equation (6)

   (6) M i = l w i + mw i + uw i 3 .

4. Normalization of the weights of each criterion: M _i is a non-fuzzy number. However, [44] it needs to be normalized by equation (7)

Table 3 describes the normalized weights of cost and time impact for each trouble factor based on the previous steps.

1. Calculating a combined impact of (cost–time, I _T) for each trouble factor by combining the weighted impact of cost and weighted impact of time as demonstrated in equation (8). The (I _T) values are used as inputs of the impact parameter which is one of the FMEA parameters.

Step 3: Assessment and prioritization of the identified trouble factors in building projects through developing the model based on using the (Neuro-Fuzzy) ANFIS technique which is characterized by effective and efficient learning. The present step involves the integration of ANN computational power and the advanced reasoning ability of a Fuzzy Inference System (FIS) to establish a precise framework that takes into account the weights of FMEA parameters and trouble factors in the assessment of trouble factors priority. The resulting output comprises each trouble event’s TPN. Subsequently, the trouble factors are prioritized following the computed TPN. Hence, decision-makers must prioritize high-ranking trouble factors that require prompt action in response to such factors. The responsible party for any trouble factors should respond commensurately with the relevant recovery strategy.

5 Development of the ANFIS model

The model’s database was obtained from a sample of 40 building projects. The ANFIS-FMAE model was created through MATLAB R2021a, provided by The Math Works, Inc. in Natick, USA, in 2021. The model comprises three input variables and one output variable. The study has identified three specific parameters of FMEA, namely P, I _T, and DCD, which have been utilized as input variables to forecast the TPN. The TPN is a metric to gauge the level of trouble in troubled projects and is considered the output variable in the models. The models’ input was initialized based on pre-calculated data extracted from troubled projects, following FMEA parameters. As mentioned, this pre-calculation was carried out in two phases as part of the suggested methodology.

The Neuro-Fuzzy Designer has been used to construct, design, train, and evaluate ANFIS-FMEA models within Sugeno-type fuzzy inference systems. These models are employed to assess and determine the prioritization of potential factors that may cause issues in building projects. The initial step involved randomly dividing the model’s data set into two sets. The initial data set comprised of 30 projects was utilized for training the model, while the other dataset consisting of 10 projects was employed for testing the model. The ANFIS-FMAE model was constructed and trained using a training data set. In contrast, a testing data set was utilized to assess the efficacy, dependability, and performance of the trained ANFIS-FMAE model in predicting the TPN and prioritizing the level of potential factors associated with troubled projects.

To build and choose the best proposed ANFIS-FMEA models that are characterized by the most suitable properties, a higher level of reliability, accuracy, and efficiency is required. Therefore, many models were built and trained several times until the desired accuracy level was reached, and the error is minimized by changing the properties of parameters and the architecture of the ANFIS model using different options of number and type (initial FIS for training) of MF for each input and output variable as well as other possibilities of division methods of the input and output data to generate several rules in developing and constructing phase of model and using different options of optimization methods of the train for each model throughout the model’s training phase.

Generally, in this work, each ANFIS model has three phases: constructing, training, and testing. Several options for the number and type of MFs and many options for the division of the input and output data to generate the rules are done in the constructing phase of models. Hence, for this task’s achievement, the options [(3 3 3), (4 4 4), and (5 5 5)] and the options (trimf, trapmf, guassmf, gauss2mf, pimf, dsigmf, and psigmf)] are used as number and type of (MFs) respectively to assist in developing the initial set of (MFs) for each input variables, while (constant and linear) are used as options of the type of (MFs) for the output variable. Also, division methods perform this model by extracting a set of rules. The methods of division are effective tools for categorizing the inputs that facilitate the training phase and determine the number of rules. For that, two options (subtractive clustering (SC) method and grid partitioning method) have been chosen as division methods of data. In the training phase, to choose the best optimization method for training models, two methods (pack-propagation and hybrid) are used as options for optimization methods to train models.

6 The results of trials to build two ANFIS models and comparison

Following multiple attempts to construct and train numerous ANFIS models utilizing distinct sets of parameter properties and architecture during the construction and training phases, as previously mentioned. Subsequently, the ANFIS models’ testing dataset was utilized to assess the performance and efficacy of each trained ANFIS model. This was accomplished by evaluating the root mean square error (RMSE) and the correlation coefficient (R) between the expected and actual output. The RMSE and the correlation coefficient (R) are utilized to evaluate the efficacy of ANFIS model training in identifying trouble factors and to furnish insights into the performance of correlations and the measuring of the degree of strong relationship between variables.

The study conducted a comparative analysis of the outcomes generated by the models. It determined that the model exhibiting the lowest RMSE and the highest correlation coefficient (R) is deemed the most optimal. The statistical parameters, namely the RMSE and the coefficient of correlation (R), were employed to evaluate and compare the performance of the models. These parameters were expressed mathematically as shown in equations (9) and (10), respectively [47,48]:

To choose the best model between many models that had been created, a comparison is done using a different set of the properties of parameters and the architecture of the ANFIS model, as shown below:

6.2 Using grid partitioning method

Grid partitioning is a technique to divide a given space into a grid-like structure, thereby preventing overlapping segments within the input space. This approach is documented in reference [50]. This approach involves partitioning the input data space into rectangular subspaces through an axis-parallel partitioning scheme. Additionally, each input is divided into identically shaped MFs. The number of fuzzy rules (if-then) can be expressed as M ⁿ , where n denotes the input dimension, and M represents the number of partitioned fuzzy subsets allocated for each input variable. In this stage, the grid partitioning method is selected as the data division technique. Multiple ANFIS models were constructed using various options for the number and type of MF for each input and output variable and different optimization methods. The objective was to identify the optimal model among the constructed models.

Figures 1–4 show the comparison between many models to identify the best model in building projects after using the grid partitioning method with different optimization methods of learning algorithm and several other types and number of MF for each input and output variable, and this was achieved after 1,000 epochs.

Figure 1

A comparison between many models according to training RMSE, testing RMSE and (R) after using the grid partitioning method with the optimization method (back-propagation algorithm) and the type of MFs to output variable (constant).

Figure 2

A comparison between many models according to training RMSE, testing RMSE and (R) after using the grid partitioning method with the optimization method (back-propagation algorithm) and the type of MFs to output variable (linear).

Figure 3

A comparison between many models according to training RMSE, testing RMSE and (R) after using the grid partitioning method with the optimization method (hybrid algorithm) and the type of MFs to output variable (constant).

Figure 4

A comparison between many models according to training RMSE, testing RMSE and (R) after using the grid partitioning method with the optimization method (hybrid algorithm) and the type of MFs to output variable (linear).

Finally, to obtain the best final characteristics of the parameters ANFIS model, the last comparison is made according to the lowest RMSE and largest R. Therefore, it can be concluded that the ANFISB68 model was the best. It has a great potential for assessment and ranking all identified potential trouble factors accurately in the building projects. The proposed model’s properties are shown in Table 5. Figure 5 presents plot of actual value vs. predicted TPN value of testing set of the proposed ANFIS- FMEA models.

Table 5

Specifications of the developed ANFIS-FMEA model in the building projects

Specifications of the developed ANFIS-FMEA model
Number of nodes	286
Number of linear parameters	125
Number of nonlinear parameters	45
Total number of parameters	170
Number of training data pairs	1,000
Number of fuzzy rules	125
The type of the MFs to input variable	Triangular MF
The number of the MFs to input variable	(5 5 5)
The type of MFs to output variable	Constant
The optimization method of training models	Hybrid learning algorithm
Training RMSE	0.10
Testing RMSE	0.03
R	1.0

Figure 5

Plot of actual value vs predicted TPN value of testing set of the proposed ANFIS-FMEA models.

7 Assessmethe Iraqi construction industry

The prioritization level of potential factors causing trouble in building projects was assessed and identified using the developed ANFIS-FMEA model based on the TPN. It is feasible to arrange TPN in descending order of priority after their scoring. The ranking results indicate that the most severe issues are situated at higher ranks, and then, the level of criticality decreases.

The factors identified as critical issues for these projects are presented in Table 6. The results of the comprehensive assessment show that the top five critical troubles in the building projects that should be given the highest priority were delayed payment to the contractor, change orders due to adding new works necessary to complete the work, change orders due to errors and design changes, force majeure (the bad security conditions in the country), and there are conflicts with other businesses as a result of lack of communication between government institutions, the highest ranking numbers (between first and fifth rankings of project).

Figure 6 illustrates a comparison of the ranking results of the potential trouble factors from traditional FMEA and ANFIS FMEA methods. The findings of the comparison with the conventional FMEA show that there were several differences occurred in ranking the trouble factors (X1, X6, X11, X12, X15, X16, X17, X18, X20, X21, X23, X24, X25, X27, X28, X29, X30, X31, and X32); also, the findings indicate that the highest five trouble factors are still X5, X9, X2, X8, and X3. The TPN based on the traditional FMEA of trouble factors was obtained by multiplying the parameters (P, I _T, and DCD) without considering their relative importance, and their weights are considered the same; despite some differences in ranking and, also, unlike the results of proposed method, the traditional TPN of trouble factor (X27) and (X28) had produce exactly the same value despite the different parameters (P, I _T, and DCD) of traditional FMEA. Fortunately, the suggested approach led to a more accurate computation of TPN, overcoming the defects of traditional FMEA, improving its effectiveness, and satisfying the major goals of the research through assessing and identifying the priorities of the trouble factors more appropriately and accurately.

Figure 6

A comparison of the ranking results of the potential trouble factors from traditional FMEA and ANFIS FMEA methods.

The results of current research show the effectiveness and high performance of the proposed method in terms of training and data analysis and the optimal setting necessary for better prediction of TPN. The performance of the ANFIS-FMEA model was evaluated using standard error measurements where the value of the RMSE was (0.03), which revealed the high ability of the proposed method to reduce (RMSE) between the expected value and the actual value and obtain very accurate prioritizing and ranking of trouble factors. The accuracy of the developed ANFIS-FMEA model was 100%. This makes ANFIS a great tool for simulating complicated processes, and it has an ability to learn, adaptability extensive parallelism with the fuzzy system which in the literature considers the best method used to improve traditional FMEA through tuning the parameters of a fuzzy inference system (FIS) by choosing the best fuzzy MFs, training these MFs and reduce fuzzy (IF-THEN) rules, thereby improving the performance of traditional FMEA and considering for relative importance of FMEA parameters, and this differs from the published literature [21,51]. Rashid and Al-Mhdawi [51] used only the traditional FMEA method for prioritizing risk factors in the Iraqi construction industry and assumed that the three parameters (P, I, DCD) of FMEA have the same importance, as the results show that some factors have different sets of (P, I, DCD) rating but it produced exactly the same value of TPN. Macura et al. [21] had developed a risk analysis model for railway infrastructure projects in the Republic of Serbia by using Fuzzy-FMEA. Developing and building a fuzzy model required a large number of fuzzy rules and choosing the best MFs which are the main problems for using the fuzzy systems alone; also, the results show that some factors have different sets of (P, I, DCD) rating but it produced exactly the same value of TPN. The proposed method by using ANFIS in our current research has successfully solved the main problems of the approaches used in the previous literature.

The disadvantage of ANFIS-FMEA model is reflected in the fact that it looks at individual trouble factors rather than their combination. Research can be evolved by developing ways to combine these factors. Moreover, in future, it would be interesting to increase the usability and reliability of the current proposed method with a large scale of data, and it is possible to generalize it to assess and identify the priorities of the trouble factors in infrastructures, roads, and bridges projects. Also, we hope that the current study results promote further research on selecting the appropriate response strategy for each trouble event based on TPN, which provides a guide to the types of response strategy and proper corrective actions that can be employed by practitioners with the aim of improving decision-making to mitigate or avoid the trouble factors. Broadly, ANFIS is good when the number of inputs is not more than five, but the more the inputs, the computational complexity of ANFIS-based model will be increased; therefore, as potential future directions and further research related to ANFIS performance improvement, we proposes combining the ANFIS with others methods such metaheuristic algorithm or particle swarm optimization, that may tune all the ANFIS parameters and improve its performance which can be compared with the current results.

8 Conclusion

The present study introduces a proactive and innovative methodology that exhibits promising potential for accurately assessing and ranking all potential trouble factors identified in building construction projects. The findings showed that the suggested method is capable of producing more accurate results and is effective in doing so. The outcomes derived from this methodology can aid the project team in effectively accomplishing a construction project by recognizing important problematic occurrences that necessitate thorough examination of underlying causes and prompt remedial measures, allowing ample time to address such critical issues. Therefore, implementing this methodology in subsequent endeavors can accomplish more successful undertakings. Additionally, it can be employed in assessing problematic factors and ranking other project types, such as infrastructure, road, and bridge projects.