Failure Modes Analysis Related to User Experience in Interactive System Design Through a Fuzzy Failure Mode and Effect Analysis-Based Hybrid Approach

Abstract

User experience (UX) is crucial for interactive system design. To improve UX, one method is to identify failure modes related to UX and then take action on the high-priority failure modes to decrease their negative impacts. For the UX of interactive system design, the failure modes under consideration are human errors or difficulties, and thus the risk factors concerning failure modes are subjective and even subconscious. Existing methods are not sufficient to deal with these issues. In this paper, a fuzzy failure mode and effect analysis (FMEA)-based hybrid approach is proposed to improve the UX of interactive system design. First, hierarchical task analysis (HTA) and systematic human error reduction and prediction approach (SHERPA) are combined to identify potential failure modes concerning UX. Subsequently, fuzzy linguistic variables are employed to assess the risk parameters of the failure modes, and the similarity aggregation method (SAM) is adopted to aggregate the fuzzy opinions. Then, on the basis of the aggregation results, fuzzy logic is adopted to compute the fuzzy risk priority numbers that can prioritize the failure modes. Finally, the failure modes with high priorities are considered for corrective actions. An in-vehicle information system was employed as a case study to illustrate the proposed approach. The findings indicate that, compared with other methods, our approach can provide more accurate results for prioritizing failure modes related to UX, and can successfully deal with the subjective and even subconscious nature of the risk factors associated with failure modes. This approach can be universally utilized to enhance the UX of interactive system design.

1. Introduction

Interactive systems are closely related to people’s lives. Designing interactive systems involves many different types of products, such as software systems, interactive products, and environments. The goal of designing interactive systems is to make them enjoyable to use, enable them to perform useful tasks, and improve the quality of life for those who use them. In other words, interactive systems should provide a good user experience (UX). UX involves developing high-quality interactive systems that fit with people and their ways of living [1].

Designers need to place users at the center of the design process and eliminate the failure modes that harm users the most. Generally, a failure mode in UX is an action that causes a user to stray from the path to successful completion. To enhance UX, a systematic approach to prioritizing the failure modes related to UX is needed. Several studies have been conducted on prioritizing failure modes related to UX. Nielsen [2] used severity to rank the failure modes, and the failure modes were rated as high, medium, or low. The failure modes were prioritized based on severity level. Besides severity, there are other factors affecting the risk of failure models. Rubin and Chisnell [3] used criticality, which was the combination of severity and probability, to rank the failure modes related to UX, and categorized the failure modes into seven levels. Travis and Hodgson [4] employed a decision tree that was constructed by using three questions (What is the impact of the failure mode? How many users are affected by the failure mode? Will users be bothered by the failure mode repeatedly?) to rank the failure modes and categorize them as low, medium, serious, and critical. These studies all focused on prioritizing failure modes related to UX; however, they did not systematically consider the risk factors that affect the priorities of failure modes.

Failure mode and effect analysis, commonly referred to as FMEA, is a meticulously systematic approach that is utilized for the identification and prevention of potential product or process failure modes before they even have a chance to occur [5]. FMEA is a proactive risk management tool that concentrates on preventing defects, enhancing safety, and boosting customer satisfaction [6]. Using FMEA, the risk priority number (RPN) that involves the severity, occurrence, and detection of failure modes can be determined, and the failure modes can be prioritized through the RPN. However, if two failure modes have the same RPN, it is not possible to distinguish between the prioritizations of the two failure modes [7]. For the UX of interactive system design, the failure modes under consideration are human errors or difficulties rather than mechanical or electronic component failure modes, and thus the risk factors including occurrence, severity, and detection are subjective and even subconscious, making them difficult to use exact numbers for estimation [8]. To deal with these situations, FMEA needs to be combined with other methods. Several studies have combined fuzzy logic with FMEA for the assessment of risk factors [9,10]. Compared with traditional FMEA, fuzzy logic-based FMEA has the following advantages: it allows the analyst to employ linguistic terms to evaluate the risk factors and can yield more accurate results [11]. Fuzzy logic-based FMEA has been studied in various fields, such as the aerospace industry, construction industry, and process industry [12,13]. Nevertheless, to the best of our knowledge, very few studies have been devoted to employing fuzzy logic-based FMEA to improve the UX of interactive system design.

When carrying out FMEA for the best results, a team is needed [14]. To aggregate the team members’ opinions concerning risk factors of failure modes, many studies used average value [15]. However, when there exist extreme values, the extreme values can be averaged, and the preference values of experts could be diminished. In other words, the consensus extent cannot be embodied, which could impact the validity of the evaluation result [16]. The similarity aggregation method (SAM) provides a systematic method to objectively consolidate individual opinions during group decision-making and can convert linguistic terms used for evaluation risk factors into fuzzy numbers [17]. Furthermore, through the aggregation steps, we can establish the judgment matrix and effectively derive consensus information for group decision problems [18].

Failure mode identification is crucial for improving the UX of interactive system design; however, FMEA does not provide a method to accomplish this. Hierarchical task analysis (HTA) is a commonly used task analysis method that can decompose tasks into clearly identifiable steps. This process helps to effectively identify potential failure modes [19,20]. In addition, the systematic human error reduction and prediction approach (SHERPA) employs a well-designed behavioral classification interconnected with an external failure mode taxonomy to discern potential failure modes related to human behavior [21]. SHERPA uses an error list methodology and claims to achieve the highest accuracy in error prediction [22]. Several studies have combined HTA with SHERPA for failure mode identification [23].

The main purpose of this paper is to improve UX in interactive system design. The primary contribution of this paper is the integration of various methods from different fields into a unified framework and its application to interactive system design, a field where it has not been previously utilized. In this framework, HTA and SHERPA are combined to detect the failure modes related to UX; moreover, fuzzy linguistic variables and SAM are integrated to analyze the failure modes. In addition, fuzzy logic is used to calculate the fuzzy risk priority number (FRPN), which is used to prioritize the failure modes. The proposed approach can accurately prioritize the failure modes related to UX, as well as effectively manage the subjective and even subconscious nature of risk factors associated with failure modes. In this paper, we hypothesize that the failure modes related to UX have different priorities, and our approach can identify and rank all the failure modes as well as provide corrective actions for the failure modes with high priorities. The remainder of this paper is arranged as follows. Section 2 gives a theoretical background review of UX in interactive system design, FMEA, and fuzzy set theory. Section 3 presents our proposed methodology. Section 4 demonstrates the procedures with a case study, and its results are discussed in Section 5. Finally, Section 6 summarizes the main conclusions.

2. Theoretical Background

2.1. UX in Interactive System Design

UX is defined as a person’s perceptions and responses that result from the use or anticipated use of a product, system, or service [24]. It is all about a user’s overall experience with a product, system, or service, encompassing emotional reactions, attitudes, ability to succeed efficiently, and other aspects. There exist two types of qualities related to UX [25]. One is pragmatic qualities, which are usability goals that relate to the tasks users must complete in order to reach their goals, such as suitability for the task, efficiency, and error tolerance. The other is hedonic qualities, which are experience goals related to the subjective impression concerning the overall interaction with the product, such as aesthetic impression, emotional appeal, and novelty [26].

UX is a crucial driving force for interactive system design [27]. Several researchers have proposed qualitative principles or rules that can be used to enhance the UX of an interactive system [28,29]. Measuring UX is focused on ensuring that the interactive system is appropriate, and thus it is at the heart of interactive system design. Various metrics related to measuring UX are subjective and even subconscious. Therefore, many studies have used fuzzy set theory for UX evaluation [30,31,32]. Generally, a failure mode in UX is an action that causes a user to stray from the path to successful completion. Common types of failure modes concerning UX include behaviors that prevent task completion, taking an incorrect sequence of actions, misinterpreting some pieces of content, and not seeing something that should be noticed [33]. Some of the changes made to solve certain failure modes may introduce new failure modes. Therefore, identifying and removing the biggest failure modes is important for improving the UX of an interactive system.

2.2. FMEA

The objective of FMEA is to identify all potential failure modes of a process or product. A failure mode occurs when the product does not perform as it should or when it malfunctions in some way. Even the simplest products have many opportunities for failure modes. In traditional FMEA, the risk of a failure mode and its effects are measured through three factors that are severity, occurrence, and detection [34]. Severity refers to the effect of the failure mode when it takes place; occurrence refers to the probability of the failure mode occurring; detection refers to the probability of the failure mode detecting. The RPN of a failure mode can be obtained by using Equation (1).

$$RPN=O\times S\times D,$$

(1)

where O signifies the occurrence, S denotes the severity, and D represents the detection. Through the values of RPN, the failure modes can be prioritized.

When applying traditional FMEA in real-world cases, there exist some important weaknesses. One significant weakness is that it is difficult to evaluate risk factors of failure modes in terms of exact numbers from 1 to 10. Furthermore, traditional FMEA does not consider the relationship of the risk factors. Moreover, traditional FMEA lacks a mechanism for systematically identifying all failure modes. In addition, different combinations of risk factor scores can produce the same RPN value, disregarding the whole different risk levels of the failure modes. Therefore, many studies have been devoted to mitigating these shortcomings of traditional FMEA [35,36,37]. In this study, we integrated fuzzy set theory, HTA, and SHERPA with traditional FMEA to overcome these shortcomings.

2.3. Fuzzy Set Theory

Fuzzy set theory is a powerful technique for dealing with concepts and rules that concern uncertainty, imprecision, and nonlinearity [38]. A significant feature of fuzzy set theory is its ability to represent vague data [39]. UX involves multiple aspects, some of which are characterized by vague and linguistic variables [40]. There are various types of fuzzy numbers used to represent vague data, including triangular, trapezoidal, G-Bell, and sigmoid functions. In this study, triangular fuzzy numbers were adopted to quantify fuzzy linguistic variables concerning the risk of UX because they are intuitive, straightforward to calculate, and easy to comprehend [41]. A triangular fuzzy number can be defined using a triplet (a₁, a₂, a₃), and its membership function can be formulated as follows [42]:

$$\mu \left(x\right)=\left\{\begin{array}{c}0, x<a_1\\ {\displaystyle \frac{x-a_1}{a_2-a_1}}, a_1\le x\le a_2\\ {\displaystyle \frac{x-a_3}{a_2-a_3}}, a_2\le x\le a_3\\ 0, x>a_3\end{array}\right..$$

(2)

To obtain the crisp output for fuzzy calculation, defuzzification is required. Defuzzification can be performed by using various methods, such as center of gravity, mean of maxima, and center of area. Center of gravity is the most popular and precise method due to its clear geometric significance [43]. Therefore, the center of gravity was adopted for defuzzification in this study, and the formula for computing the defuzzification value is as follows:

$$DF={\displaystyle \frac{{\displaystyle \int x\cdot \mu \left(x\right)\mathrm{d}x}}{{\displaystyle \int \mu \left(x\right)\mathrm{d}x}}},$$

(3)

where DF is the defuzzification value and μ(x) is the membership function.

2.4. The Use of Fuzzy Logic in UX

Fuzzy logic is a mathematical technique for inexact reasoning that enables the modeling of human reasoning processes using linguistic terms [44], and is suitable for analyzing subjective and vague information, which is an important feature of user experience. Many studies concerning UX have been conducted using fuzzy logic. Sulikowski et al. [45] used fuzzy logic for purchase intent modeling based on user tracking for e-commerce recommenders. Li and Zhu [46] combined fuzzy logic with the Taguchi method for multi-objective optimization of user experience in mobile application design.

In this study, a fuzzy logic system was adopted to assess the FRPN of a failure mode. The membership function used for each input and output factor in the fuzzy logic system is the G-Bell shape because it enables the creation of continuously differentiable hypersurfaces for a fuzzy model, thereby making the theoretical analysis of a fuzzy logic system more accessible [47]. The G-Bell shape function can be defined as follows:

$$\mu \left(x\right)={\displaystyle \frac{1}{1+{\left|\frac{x-c}{a}\right|}^{2b}}},$$

(4)

where c determines the center of the curve, and a and b dictate the shape of the curve and are usually positive.

3. Proposed Methodology

The flowchart of the proposed methodology is depicted in Figure 1. The methodology consists of four stages, which are detecting failure modes by employing HTA and SHERPA, analyzing failure modes by using fuzzy linguistic variables and SAM, calculating FRPNs by using fuzzy logic, and modifying the failure modes with high priorities. To carry out the methodology, a multidisciplinary team is required.

3.1. Detecting Failure Modes by Using HTA and SHERPA

HTA and SHERPA are combined to systematically detect all failure modes related to UX in interactive system design. HTA is used to decompose the task into clearly identifiable steps that the user performs when interacting with an interactive system, so as to identify any steps that may have potential failure modes. Subsequently, SHERPA is adopted to identify credible failure modes associated with these steps. In SHERPA, failure modes are classified into five categories which include action, retrieval, checking, information communication, and selection, as shown in

Table 1. Failure mode taxonomy in SHERPA.

Failure Mode Category	Failure Mode Code	Failure Mode Description
Action	A1	Operation too long/short
	A2	Operation mistimed
	A3	Operation in wrong direction
	A4	Operation too much/little
	A5	Misalign
	A6	Right operation on wrong object
	A7	Wrong operation on right object
	A8	Operation omitted
	A9	Operation incomplete
	A10	Wrong operation on wrong object
Checking	C1	Check omitted
	C2	Check incomplete
	C3	Right check on wrong object
	C4	Wrong check on right object
	C5	Check mistimed
	C6	Wrong check on wrong object
Information communication	I1	Information not communicated
	I2	Wrong information communicated
	I3	Information communication incomplete
Information retrieval	R1	Information not obtained
	R2	Wrong information obtained
	R3	Information retrieval incomplete
Selection	S1	Selection omitted
	S2	Wrong selection made

[48]. Based on this classification, the failure mode of each task can be identified.

3.2. Analysing Failure Modes by Using Fuzzy Linguistic Variables and SAM

Analyzing failure modes includes two processes, which are evaluating risk parameters of failure modes and aggregating fuzzy opinions for risk parameter assessments.

3.2.1. Evaluating Risk Parameters of Failure Modes

FMEA is used to analyze failure modes. The risk parameters for FMEA include occurrence, severity, and detection. To evaluate these risk parameters, fuzzy linguistic variables are employed. The fuzzy linguistic variables regarding occurrence, severity, and detection are given in

Table 2. Fuzzy linguistic variables regarding occurrence.

Linguistic Variable	Triangular Fuzzy Number	Description
Very low (VL)	(0, 0, 2.5)	Failure modes are unlikely to occur.
Low (L)	(0, 2.5, 5.0)	Relatively few failure modes are likely to occur.
Medium (M)	(2.5, 5.0, 7.5)	Failure modes occur occasionally.
High (H)	(5.0, 7.5, 10.0)	Failure modes can be repeated.
Very high (VH)	(7.5, 10.0, 10.0)	Failure modes are almost inevitable.

,

Table 3. Fuzzy linguistic variables regarding severity.

Linguistic Variable	Triangular Fuzzy Number	Description
Very low (VL)	(0, 0, 2.5)	The effect of failure mode on usability can be ignored.
Low (L)	(0, 2.5, 5.0)	The effect slightly impacts usability.
Medium (M)	(2.5, 5.0, 7.5)	The effect slightly but noticeably impacts usability.
High (H)	(5.0, 7.5, 10.0)	The effect significantly impacts on usability.
Very high (VH)	(7.5, 10.0, 10.0)	The effect seriously impacts usability.

, and

Table 4. Fuzzy linguistic variables regarding detection.

Linguistic Variable	Triangular Fuzzy Number	Description
Very low (VL)	(0, 0, 2.5)	Almost certain to be detected.
Low (L)	(0, 2.5, 5.0)	Easy to detect.
Medium (M)	(2.5, 5.0, 7.5)	Detected occasionally.
High (H)	(5.0, 7.5, 10.0)	Difficult to detect.
Very high (VH)	(7.5, 10.0, 10.0)	Almost impossible to detect.

, respectively [49].

3.2.2. Aggregating Fuzzy Opinions for Risk Parameters Assessments

The team members may have different opinions about the risk parameters assessments, and the ratings for risk parameters derived by subjective judgment are not very reliable [50]. Therefore, the ratings from different team members need to be aggregated. SAM is adopted to aggregate the fuzzy ratings. SAM was developed by Hsu and Chen [51], and the steps for SAM are as follows.

Step 1. Compute the agreement degree.

$$S\left(R_i,R_j\right)={\displaystyle \frac{{\displaystyle {\int }_x\left(\min \left\{{\mu }_{R_i}\left(x\right),{\mu }_{R_j}\left(x\right)\right\}\right)\mathrm{d}x}}{{\displaystyle {\int }_x\left(\max \left\{{\mu }_{R_i}\left(x\right),{\mu }_{R_j}\left(x\right)\right\}\right)\mathrm{d}x}}},$$

(5)

where $R_i$ and $R_j$ are the fuzzy evaluation values of member i and member j, respectively, and $S\left(R_i,R_j\right)$ is the agreement degree between the two members.

Step 2. Establish the agreement matrix AM.

$$AM={\left[S_{ij}\right]}_{n\times n}, i=1,2,\dots ,n, j=1,2,\dots ,n,$$

(6)

where $S_{ij}=S\left(R_i,R_j\right)$.

Step 3. Compute the average agreement degree $AA{D}_i$ of member i with other members.

$$AA{D}_i={\displaystyle \frac{1}{n-1}}{\displaystyle \sum _{\begin{array}{l}j=1\\ i\ne j\end{array}}^n{S}_{ij}}, i=1,2,\dots ,n.$$

(7)

Step 4. Compute the relative agreement degree $RA{D}_i$ of member i.

$$RA{D}_i={\displaystyle \frac{AA{D}_i}{{\displaystyle \sum _{i=1}^{n}AA{D}_i}}}, i=1,2,\dots ,n.$$

(8)

Step 5. Compute the consensus degree coefficient $CD{C}_i$ of member i.

$$CD{C}_i=\beta \times w_i+\left(1-\beta \right)\times RA{D}_i, i=1,2,\dots ,n,$$

(9)

where w_i is the degree of importance for the ith member, and β ($0\le \beta \le 1$) is a relaxation factor. Due to the heterogeneous characteristics of the team members in this study, β was assigned a value of 0.5.

Step 6. Compute the aggregation value R.

$$R={\displaystyle \sum _{i=1}^{n}CD{C}_i\otimes }{R}_i, i=1,2,\dots ,n.$$

(10)

3.3. Calculating FRPNs with Fuzzy Logic

Fuzzy logic is employed to transform the risk parameters into an FRPN. Fuzzy logic allows for the modeling of human reasoning using linguistic terms. It is a perfect tool for mapping the relationship between system inputs and the output we want [52]. This approach has proven useful across a range of areas including automatic control, decision analysis, and expert systems [53]. The structure of a fuzzy logic system mainly consists of four components including a fuzzifier, a knowledge base, an inference engine, and a defuzzifier. The interconnections among these components and the process are illustrated in stage 3 of the proposed methodology flowchart (see Figure 1). Initially, the fuzzifier employs membership functions to transform crisp inputs into fuzzy sets. Following that, the fuzzy inference engine conducts fuzzy reasoning to generate fuzzy values by using the knowledge base. As a final step, the obtained fuzzy values are converted into a crisp output by using the defuzzifier.

A knowledge base, which contains a set of if–then rules, is built to map the relationship between the inputs and the output. A representative linguistic fuzzy rule is as follows [54]:

Rule i: If x₁ is A_i1, x₂ is A_i2, …, and x_r is A_ir, then y_i is C_i,

(11)

where i = 1, 2, …, M (M is the total number of fuzzy rules), x_j (j = 1, 2, …, r) is the input variable, y_i is the output variable, and A_ij and C_i are fuzzy sets modeled by the membership functions ${\mu }_{A_{ij}}\left(x_j\right)$ and ${\mu }_{C_i}\left(y_i\right)$, respectively. The Mamdani fuzzy inference engine is employed for its advantages of intuitive formation, widespread acceptance, and compatibility with human input [55]. Thus, for a collection of M fuzzy rules, the obtained output can be expressed as follows [56]:

$${\mu }_{c_i}\left(y_i\right)=\max \left\{\min \left[{\mu }_{A_{i1}}\left(x_1\right), {\mu }_{A_{i2}}\left(x_2\right),\dots , {\mu }_{A_{ir}}\left(x_r\right)\right]\right\}.$$

(12)

To derive the crisp output for the fuzzy inference result, the center of gravity is adopted for defuzzification.

3.4. Modifying the Failure Modes with High Priorities

Based on the above analysis, many failure modes can be detected. Modifying one failure mode could exacerbate other failure modes or lead to the emergence of new failure modes. Therefore, we need to modify the failure modes with high priorities. Failure modes with a larger FRPN take on a higher level of priority; the top 50% of failure modes are considered for modification [13]. Modifications are made according to the priorities of the failure modes.

4. Case Study

In-vehicle information systems (IVISs) are typical interactive systems, which support a wide variety of secondary vehicle functions, including navigation, media, and climate control. These functions are incorporated into a single system, usually operated through a touchscreen interface [57]. With the advancement of intelligent vehicles, more functions are being integrated into IVISs, potentially affecting their UX. The design of IVISs, which are connected to driving safety, has garnered significant research attention [15,58,59]. For this reason, an IVIS was selected to illustrate our approach. In this case study, we utilized a focus group for data collection. The group included seven team members: three UX experts, two designers, and two IVIS users, each bringing their unique insights and perspectives. The traits of the team members are listed in

Table 5. Traits of the team members in the study.

Team Member	Professional Position (T1)	Job Experience (T2)	Education (T3)	Age (T4)
TM1	Senior expert	20	PhD	43
TM2	Intermediate expert	17	Master	39
TM3	Senior expert	35	Bachelor	59
TM4	Designer	30	Master	52
TM5	Designer	4	Master	25
TM6	Advanced user	27	College diploma	50
TM7	Ordinary user	5	High school	27

.

4.1. Detecting Failure Modes for IVIS

The primary user interface of the original IVIS is depicted in Figure 2. From this IVIS, many tasks can be executed. In this study, we focused on the commonly used tasks that concerned navigation, climate, media, telephone, and radio. The results of the HTA for these tasks are described in a tabular list, as shown in

Table 6. HTA for the original IVIS regarding media, navigation, telephone, radio, and climate.

Code	Task	Rule
0	Performing relevant tasks	Plan 0: do any of 1, 2, 3, 4, or 5 in any order
1	Media	Plan 1: do 1.1, 1.2, 1.3, 1.4, and 1.5 in order
1.1	Play media	Plan 1.1: do 1.1.1, 1.1.2, and 1.1.3 in order
1.1.1	Click on MENU
1.1.2	Click on Media
1.1.3	Click on Play
1.2	Adjust volume	Plan 1.2: do 1.2.1, 1.2.2 in order
1.2.1	Click on blank area
1.2.2	Slide volume bar
1.3	Switch music	Plan 1.3: do 1.3.1, then do 1.3.2, 1.3.3, 1.3.4, and 1.3.5 in any order
1.3.1	Click on Selection	Plan 1.3.1: do 1.3.1.1 and 1.3.1.2 in order
1.3.1.1	Enter playlist
1.3.1.2	Click on the songs
1.3.2	Click on Previous
1.3.3	Click on Next
1.3.4	Click on Repeat
1.3.5	Click on Shuffle
1.4	Adjust process	Plan 1.4: do 1.4.1 and 1.4.2 in any order
1.4.1	Slide to right
1.4.2	Slide to left
1.5	Close media	Plan 1.5: do 1.5.1 and 1.5.2 in order
1.5.1	Click on Stop
1.5.2	Click on MENU
2	Navigation	Plan 2: do 2.1, 2.2, 2.3, and 2.4 in order
2.1	Open navigation	Plan 2.1: do 2.1.1 and 2.1.2 in order
2.1.1	Click on MENU
2.1.2	Click on Navigation
2.2	Choose place	Plan 2.2: do 2.2.1 and 2.2.2 in order
2.2.1	Set route
2.2.2	Set destination	Plan 2.2.2: do 2.2.2.1, 2.2.2.2, and 2.2.2.3 in any order
2.2.2.1	Save current position
2.2.2.2	Historical records
2.2.2.3	Home address
2.3	Start navigating	Plan 2.3: do 2.2, then do 2.3.1, 2.3.2, and 2.3.3 in order
2.3.1	Click on Selection
2.3.2	Select route
2.3.3	Click on Start
2.4	Close navigation
3	Telephone	Plan 3: do 3.1, 3.2, 3.3, and 3.4 in order
3.1	Open phone options	Plan 3.1: do 3.1.1 and 3.1.2 in order
3.1.1	Click on MENU
3.1.2	Click on Call
3.2	Select contacts	Plan 3.2: do 3.2.1, 3.2.2, or 3.2.3
3.2.1	Dial number
3.2.2	Contacts
3.2.3	Recent
3.3	Make calls	Plan 3.3: do 3.3.1 and 3.3.2 in order
3.3.1	Click on dial button
3.3.2	Click on hang-up button
3.4	Return Home
4	Radio	Plan 4: do 4.1, 4.2, 4.3, and 4.4 in order
4.1	Turn on the radio	Plan 4.1: do 4.1.1 and 4.1.2 in order
4.1.1	Click on MEUN
4.1.2	Click on radio options
4.2	Switch radio	Plan 4.2: do 4.2.1, 4.2.2, and 4.2.3 in any order
4.2.1	Click on bands	Plan 4.2.1: do 4.2.1.1 and 4.2.1.2 in any order
4.2.1.1	Select AM
4.2.1.2	Select FM
4.2.2	Click on the radio list
4.2.3	Click on manual
4.3	Add station	Plan 4.3: do 4.2.2, then do 4.3.1 and 4.3.2 in order
4.3.1	Long press station
4.3.2	Click on blank to add box
4.4	Return home
5	Climate	Plan 5: do 5.1, 5.2, and 5.3 in order
5.1	Turn on the air conditioner	Plan 5.1: do 5.1.1, 5.1.2, 5.1.3, and 5.1.4 in order
5.1.1	Click on MENU
5.1.2	Click on vehicle button
5.1.3	Click on blank
5.1.4	Click on air conditioner button
5.2	Set air conditioner	Plan 5.2: do 5.2.1, 5.2.2, and 5.2.3 in any order
5.2.1	Regulate air volume	Plan 5.2.1: do 5.2.1.1 and 5.2.1.2 in any order
5.2.1.1	Increase air volume
5.2.1.2	Decrease air volume
5.2.2	Regulate temperature	Plan 5.2.2: do 5.2.2.1 and 5.2.2.2 in any order
5.2.2.1	Turn down the temperature
5.2.2.2	Turn up the temperature
5.2.3	Regulate blowing mode
5.3	Return home

. The experimenters observed the processes of the users completing the tasks. After that, SHERPA was employed to identify the failure modes related to UX. A total of 25 failure modes were identified, as listed in

Table 7. Failure modes identified using SHERPA.

Failure Mode No.	Code in HTA	Failure Mode Category	Failure Mode Description	Failure Mode Effect	Failure Mode Cause
1	1.1.1	A9	Users confuse the meaning of icons.	Impact on the following operations.	The meaning of icon is unclear.
2	1.1.2	A9	Users are unable to determine the correct method of operation.	Unable to advance to the next step.	The information architecture is complicated.
3	1.2.1	I3	Users are not aware of how to operate.	It is difficult to complete the task.	The operation is complex.
4	1.2.2	A2	Additional time is required to complete the operation.	Impact on usability.	The operation is complex.
5	1.3.2	I1	Users lack confidence in their operations.	Impact on usability.	Shortage of operational clues.
6	1.5.1	A8	Users omit certain steps in the operational process.	Unable to complete the task.	Insufficient relevant clues.
7	2.2.1.1	A9	Users are not aware of how to operate.	It is difficult to complete the task.	Be an unreasonable design.
8	2.3.1	A4	Users are not aware of how to operate.	It is difficult to complete the task.	The operation process is unreasonable.
9	2.2.2.4	A9	Users are unable to locate their saved records.	Impact on usability.	Be an unreasonable design.
10	2.4	I3	Users are unsure if the navigation has been initiated.	Impact on usability.	Insufficient relevant clues.
11	3.1.2	I2	Users are unable to determine the correct method of operation.	Results in incorrect operation.	The information architecture is complicated.
12	3.2.2	R1	Users are unable to locate the contact information.	Unable to complete the task.	Shortage of operational clues.
13	3.2.3	C3	Users select the wrong objects.	Unable to complete the task.	The meaning of icon is unclear.
14	3.4	A8	Users navigate back to the main interface without disconnecting the phone.	Impact on the following operations.	Shortage of operational clues.
15	3.2.1	A9	Users do not know how to navigate back.	Impact on the following operations.	Insufficient relevant clues.
16	3.2.1	A2	Users are not aware of how to operate.	Results in incorrect operation.	The layout of interface is unsuitable.
17	4.2.1	A9	Users have difficulty understanding the meaning of words.	Unable to advance to the next step.	Overlooking users’ characteristics.
18	4.2.2	I3	Users are unclear about their subsequent actions.	Results in incorrect operation.	Shortage of operational clues.
19	4.3.1	A9	Users are unclear about their subsequent actions.	Results in incorrect operation.	The operation is complex.
20	4.3.2	A9	Users have difficulty understanding the meaning of words.	Results in incorrect operation.	The meaning of words is unclear.
21	5.1.3	I1	Users are uncertain about their next step.	Unable to advance to the next step.	There are too many operational steps.
22	5.1.4	A4	Users are unable to figure out how to use it.	Results in incorrect operation.	The operation process is unreasonable.
23	5.2.1	A9	Users confuse the meaning of icons.	Unable to advance to the next step.	The meaning of icon is unclear.
24	5.2.3	S2	Users confuse the meaning of icons.	Results in incorrect operation.	The meaning of icon is unclear.
25	5.2.2.1	A9	Users are unclear about how to adjust the temperature.	Unable to advance to the next step.	Shortage of operational clues.

.

4.2. Analysing Failure Modes for IVIS

4.2.1. Evaluating Risk Parameters of Failure Modes for IVIS

To evaluate the risk parameters of the identified failure modes, triangular fuzzy linguistic variables were adopted. The evaluation results of the seven team members (TM1–TM7) regarding occurrence, severity, and detection are given in

Table 8. Fuzzy evaluations and aggregation of opinions for occurrence.

Failure Mode No.	Fuzzy Evaluations of Team Members							Aggregation of Fuzzy Opinions	Defuzzification Value
	TM1	TM2	TM3	TM4	TM5	TM6	TM7
1	H	VH	M	VH	M	H	H	(5.072, 7.572, 9.382)	7.342
2	M	M	H	M	M	H	H	(3.428, 5.928, 8.428)	5.928
3	VH	VH	H	VH	VH	VH	VH	(7.215, 9.715, 10.000)	8.977
4	M	M	L	M	M	M	M	(2.227, 4.727, 7.227)	4.727
5	M	M	M	M	H	M	M	(2.660, 5.160, 7.660)	5.160
6	M	M	L	M	M	M	M	(2.227, 4.727, 7.227)	4.727
7	M	H	M	M	H	M	M	(2.981, 5.481, 7.981)	5.481
8	VH	VH	H	VH	VH	VH	H	(6.999, 9.499, 10.000)	8.832
9	M	M	M	H	M	H	M	(3.066, 5.566, 8.066)	5.566
10	M	M	H	M	M	M	M	(2.773, 5.273, 7.773)	5.273
11	H	H	H	H	H	M	H	(4.798, 7.298, 9.798)	7.298
12	H	M	M	H	H	H	H	(4.406, 6.906, 9.406)	6.906
13	M	M	L	M	L	M	M	(1.962, 4.462, 6.962)	4.462
14	M	H	M	M	M	M	M	(2.716, 5.216, 7.716)	5.216
15	M	M	H	M	H	M	M	(3.038, 5.538, 8.038)	5.538
16	H	H	M	H	H	H	H	(4.727, 7.227, 9.727)	7.227
17	M	M	M	M	M	M	M	(2.500, 5.000, 7.500)	5.000
18	H	H	H	VH	H	H	H	(5.271, 7.771, 10.000)	7.681
19	H	H	H	H	H	H	H	(5.000, 7.500, 10.000)	7.500
20	H	H	H	VH	H	H	H	(5.271, 7.771, 10.000)	7.681
21	M	M	M	M	M	M	M	(2.500, 5.000, 7.500)	5.000
22	VH	VH	VH	H	H	VH	H	(6.618, 9.118, 10.000)	8.579
23	H	H	H	H	H	M	M	(4.589, 7.089, 9.589)	7.089
24	H	H	H	H	H	M	M	(4.589, 7.089, 9.589)	7.089
25	M	M	L	M	L	M	M	(1.962, 4.462, 6.962)	4.462

,

Table 9. Fuzzy evaluations and aggregation of opinions for severity.

Failure Mode No.	Fuzzy Evaluations of Team Members							Aggregation of Fuzzy Opinions	Defuzzification Value
	TM1	TM2	TM3	TM4	TM5	TM6	TM7
1	H	H	VH	H	H	M	H	(5.085, 7.585, 9.793)	7.488
2	H	H	H	H	H	H	M	(4.896, 7.396, 9.896)	7.396
3	VH	VH	VH	VH	VH	H	VH	(7.285, 9.785, 10.000)	9.023
4	H	M	H	H	M	H	H	(4.519, 7.019, 9.519)	7.019
5	H	M	M	H	H	H	H	(4.406, 6.906, 9.406)	6.906
6	L	L	L	L	L	L	M	(0.104, 2.604, 5.104)	2.604
7	H	H	H	M	H	H	H	(4.741, 7.241, 9.741)	7.241
8	H	H	H	H	VH	H	VH	(5.389, 7.889, 10.000)	7.759
9	M	M	M	H	M	M	H	(2.967, 5.467, 7.967)	5.467
10	M	M	M	M	H	H	M	(2.967, 5.467, 7.967)	5.467
11	H	H	M	M	H	H	H	(4.364, 6.864, 9.364)	6.864
12	H	H	H	H	M	M	H	(4.533, 7.033, 9.533)	7.033
13	M	M	M	M	M	L	M	(2.298, 4.798, 7.298)	4.798
14	M	M	M	M	H	M	M	(2.660, 5.160, 7.660)	5.160
15	H	H	H	M	H	H	H	(4.741, 7.241, 9.741)	7.241
16	H	H	H	H	H	H	VH	(5.117, 7.617, 10.000)	7.578
17	M	M	M	M	M	M	H	(2.604, 5.104, 7.604)	5.104
18	H	VH	H	H	H	VH	H	(5.543, 8.043, 10.000)	7.862
19	H	H	H	H	VH	VH	H	(5.487, 7.987, 10.000)	7.825
20	VH	H	H	VH	VH	VH	H	(6.548, 9.048, 10.000)	8.532
21	H	M	M	M	H	H	H	(3.904, 6.404, 8.904)	6.404
22	VH	VH	VH	VH	VH	H	VH	(7.285, 9.785, 10.000)	9.023
23	H	H	H	M	H	H	H	(4.741, 7.241, 9.741)	7.241
24	H	H	H	VH	H	H	H	(5.271, 7.771, 10.000)	7.681
25	H	VH	H	H	H	M	H	(5.029, 7.529, 9.793)	7.451

, and

Table 10. Fuzzy evaluations and aggregation of opinions for detection.

Failure Mode No.	Fuzzy Evaluations of Team Members							Aggregation of Fuzzy Opinions	Defuzzification Value
	TM1	TM2	TM3	TM4	TM5	TM6	TM7
1	M	L	M	L	L	L	M	(0.998, 3.498, 5.998)	3.498
2	L	L	L	L	M	L	L	(0.160, 2.660, 5.160)	2.660
3	M	M	M	M	M	M	L	(2.396, 4.896, 7.396)	4.896
4	M	M	M	M	M	M	M	(2.500, 5.000, 7.500)	5.000
5	M	M	L	M	M	M	M	(2.227, 4.727, 7.227)	4.727
6	L	L	L	L	L	L	L	(0.000, 2.500, 5.000)	2.500
7	M	M	M	M	L	L	M	(2.033, 4.533, 7.033)	4.533
8	L	L	L	L	L	L	L	(0.000, 2.500, 5.000)	2.500
9	M	L	M	M	M	M	M	(2.284, 4.784, 7.284)	4.784
10	M	L	L	M	M	M	M	(1.906, 4.406, 6.906)	4.406
11	M	M	M	L	L	L	L	(1.110, 3.610, 6.110)	3.610
12	M	M	L	L	L	L	L	(0.594, 3.094, 5.594)	3.094
13	L	L	M	L	L	L	L	(0.273, 2.773, 5.273)	2.773
14	M	M	M	M	L	M	L	(2.131, 4.631, 7.131)	4.631
15	M	M	L	M	M	L	L	(1.572, 4.072, 6.572)	4.072
16	M	L	M	M	M	L	L	(1.629, 4.129, 6.629)	4.129
17	M	M	M	M	L	M	L	(2.131, 4.631, 7.131)	4.631
18	L	L	M	L	L	L	L	(0.273, 2.773, 5.273)	2.773
19	L	L	L	L	L	L	L	(0.000, 2.500, 5.000)	2.500
20	L	L	L	L	L	L	L	(0.000, 2.500, 5.000)	2.500
21	M	M	M	M	L	L	L	(1.685, 4.185, 6.685)	4.185
22	H	H	H	VH	H	H	H	(5.271, 7.771, 10.000)	7.681
23	M	L	L	M	M	L	M	(1.460, 3.960, 6.460)	3.960
24	H	H	M	H	H	M	M	(4.072, 6.572, 9.072)	6.572
25	M	M	M	M	M	L	M	(2.298, 4.798, 7.298)	4.798

, respectively.

4.2.2. Aggregating Fuzzy Opinions for Risk Parameters Assessments for IVIS

To aggregate the fuzzy opinions of team members for risk parameters assessments, SAM was used. The weighting scores according to the traits of team members are given in

Table 11. Weighting scores according to the traits of team members.

Traits	Classify	Score
Professional position (T1)	Senior expert	5
	Intermediate expert	4
	Designer	3
	Advanced user	2
	Ordinary user	1
Job experience (T2)	≥30 years	5
	20–29	4
	10–19	3
	6–9	2
	≤5	1
Education (T3)	PhD	5
	Master	4
	Bachelor	3
	College diploma	2
	High school	1
Age (T4)	≥50	4
	40–49	3
	30–39	2
	≤29	1

[60]. The weights assigned to each team member based on their traits (see Table 5) are presented in

Table 12. Weights of team members based on their traits.

Team Member	Weighting Traits				Total Score	Weight
	T1	T2	T3	T4
TM1	5	4	5	3	17	0.191
TM2	4	3	4	2	13	0.146
TM3	5	5	3	4	17	0.191
TM4	3	5	4	4	16	0.180
TM5	3	1	4	1	9	0.101
TM6	2	4	2	4	12	0.135
TM7	1	2	1	1	5	0.056

.

For the risk parameter of occurrence, according to Equations (5) and (6), as regards failure mode No. 1, the agreement matrix can be obtained as follows:

$$AM=\left[\begin{array}{ccccccc}1.000& 0.200& 0.143& 0.200& 0.143& 1.000& 1.000\\ 0.200& 1.000& 0& 1.000& 0& 0.200& 0.200\\ 0.143& 0& 1.000& 0& 1.000& 0.143& 0.143\\ 0.200& 1.000& 0& 1.000& 0& 0.200& 0.200\\ 0.143& 0& 1.000& 0& 1.000& 0.143& 0.143\\ 1.000& 0.200& 0.143& 0.200& 0.143& 1.000& 1.000\\ 1.000& 0.200& 0.143& 0.200& 0.143& 1.000& 1.000\end{array}\right]$$

By using Equation (7), the average agreement degree (AAD) can be calculated, and then the relative agreement degree (RAD) can be derived according to Equation (8). On the basis of RAD and the weights of the team members, the consensus degree coefficient (CDC) can be calculated according to Equation (9). The computed AAD, RAD, and CDC for occurrence are given in

Table 13. The computed AAD, RAD, and CDC for occurrence.

Team Member	Average Agreement Degree (AAD)	Relative Agreement Degree (RAD)	Consensus Degree Coefficient (CDC)
TM1	0.448	0.190	0.191
TM2	0.267	0.113	0.130
TM3	0.238	0.101	0.146
TM4	0.267	0.113	0.147
TM5	0.238	0.101	0.101
TM6	0.448	0.190	0.163
TM7	0.448	0.190	0.123

. By using Equation (10), the aggregation result was obtained as follows: $R=\left(5.072, 7.572, 9.382\right)$. Consequently, by using Equation (4), the defuzzification value for the aggregated fuzzy opinions can be obtained, whose result was 7.342.

The aggregated fuzzy opinions and the corresponding defuzzification values for all the failure modes concerning risk parameters of occurrence can be derived, as shown in the last two columns of Table 8. In the same way, the aggregated fuzzy opinions and the corresponding defuzzification values for the risk parameters of severity and detection can also be derived, as shown in the last two columns of Table 9 and Table 10.

4.3. Calculating FRPNs for IVIS

4.3.1. Constructing the Fuzzy Logic System

To calculate the FRPNs, a fuzzy logic system was built. The input variables were the defuzzification values for the risk parameters, and the output variable was FRPNs. According to the suggestion of Chang and Dillon [61], the G-Bell shape function was used to construct the fuzzy logic system, and the membership functions of the input and output variables are illustrated in Figure 3.

Through a focus group, the team members established the relationship between the input variables (occurrence, severity, and detection) and the output variable (FRPN). This relationship is illustrated in Figure 4. A total of 125 fuzzy rules (5 × 5 × 5 = 125) were determined; one of the fuzzy rules is as follows: if the occurrence is very low, the severity is medium, and the detection is very high, then the FRPN is medium. Based on these fuzzy rules, the fuzzy logic system was constructed. The schematic of the constructed fuzzy logic system is illustrated in Figure 5. The 3D surface viewers of the constructed fuzzy logic system are presented in Figure 6, which shows the influences of input variables on the FRPN. It can be observed that these 3D surface viewers conform to the characteristic of monotonicity, and thus the constructed fuzzy logic system can yield valid and comparable results.

4.3.2. Calculating the FRPN

Through the constructed fuzzy logic system, the FRPN was obtained. Figure 7 illustrates the rule viewer of the constructed system, which can receive the defuzzification values of the risk parameters for the three inputs and generate the FRPN for the output. For example, for failure mode No. 1, it can be expressed as follows: if the defuzzification value of occurrence was 7.342, the defuzzification value of severity was 7.488, and the defuzzification value of detection was 3.498, then the FRPN was 6.395. The FRPNs of all the failure models produced by using the constructed fuzzy logic system are presented in column 2 of

Table 14. Prioritization of failure modes.

Failure Mode No.	FRPN	Prioritization
1	6.395	8
2	5.347	19
3	8.078	1
4	6.286	14
5	6.267	15
6	3.602	25
7	6.330	12
8	6.432	6
9	5.153	20
10	5.113	21
11	6.374	10
12	6.289	13
13	4.869	24
14	5.046	22
15	6.093	17
16	6.384	9
17	5.013	23
18	6.467	4
19	6.411	7
20	6.823	3
21	5.889	18
22	8.028	2
23	6.332	11
24	6.450	5
25	6.233	16

, and the prioritizations as per FRPNs are presented in the last column of Table 14.

4.4. Modifying the Failure Modes with High Priorities for IVIS

Corrective actions were developed for the top 50% of failure modes that were ranked from 1 to 12 (as listed in Table 14), with failure mode No. 3 (Users are not aware of how to operate) given the highest priority. This was followed by failure modes No. 22 (Users are unable to figure out how to use it) and No. 20 (Users have difficulty understanding the meaning of words). The last error requiring corrective action was failure mode No. 7 (Users are not aware of how to operate). The original IVIS had numerous functions that were not effectively designed, resulting in a poor UX. To improve its UX, several corrective actions were carried out. These actions were proposed based on the design principles for dealing with error, the laws of UX, and the principle of user interface design [28,62,63]. The detailed actions included simplifying operations to reduce users’ burden and facilitate task completion (for failure modes No. 3 and No. 19), enhancing the system model to better align with users’ mental model (for failure modes No. 8 and No. 22), and providing perceivable signifiers for users to understand possible actions and their execution (for failure modes No. 1, No. 23, and No. 24). Other actions involved using the words familiar to users (for failure modes No. 20), setting clear clues in the appropriate context for each task (for failure modes No. 18), adjusting the interface layout to facilitate the execution of frequently used tasks (for failure mode No. 16), and improving the information architecture for ease of use (for failure mode No. 11). The primary user interface for the revised IVIS is shown in Figure 8.

5. Discussion

Traditional FMEA used exact numbers to evaluate the risk factors and employed the RPN, which was computed through Equation (1), to prioritize failure modes. By contrast, the proposed approach used fuzzy linguistic terms to evaluate the risk factors, and the results were then aggregated through SAM. On the basis of the aggregated values, a fuzzy logic system was built to derive the FRPN that was used to prioritize failure modes. A comprehensive experiment was carried out where the values for occurrence, severity, and detection, along with the corresponding RPNs, were systematically obtained by the team members, following the guidelines stipulated in traditional FMEA. To compare the results, we placed the FRPNs from our approach alongside the RPNs from traditional FMEA, as shown in

Table 15. Comparison of the results derived from two approaches.

Failure Mode No.	RPN Derived from the Traditional FMEA Method				FRPN Derived from the Proposed Approach
	Occurrence	Severity	Detection	RPN
1	8	7	4	224	6.395
16	7	8	4	224	6.384
8	9	8	3	216	6.432
20	8	9	3	216	6.823
11	7	6	5	210	6.374
12	7	5	6	210	6.289
23	6	7	5	210	6.332
7	6	8	4	192	6.330
18	8	8	3	192	6.467
19	8	6	4	192	6.411
15	6	6	4	144	6.093
21	4	6	6	144	5.889
9	6	5	4	120	5.153
10	5	6	4	120	5.113
14	4	6	5	120	5.046
17	6	4	5	120	5.013

. The results show that our approach is more effective in generating unique values. It differentiated between failure modes that were indistinguishable using traditional FMEA. For instance, failure modes No. 1 and No. 16 had identical RPNs (224) with traditional FMEA. However, with our new approach, they had different FRPNs (6.395 and 6.384). In addition, if we only used severity to prioritize failure modes, many failure modes would be indistinguishable. For instance, failure modes No. 10, No. 11, No. 14, No. 15, No. 19, and No. 21 all had an identical severity level of 6. Obviously, the FRPNs from our approach provided a clear ranking of the failure modes, each with different priorities. This clear ranking can offer valuable insights that improve the UX of interactive system design.

The comparison between the proposed approach and two other typical approaches is presented in

Table 16. Comparison of prioritization across different approaches.

Failure Mode No.	Prioritization Based on the Proposed Approach	Prioritization Based on the Combination of FMEA with Fuzzy TOPSIS	Prioritization Based on the Combination of FMEA with Fuzzy Logic
1	8	9	9
2	19	16	19
3	1	2	1
4	14	18	11
5	15	15	11
6	25	25	25
7	12	12	11
8	6	4	5
9	20	19	20
10	21	21	20
11	10	11	10
12	13	13	11
13	24	24	24
14	22	22	22
15	17	14	16
16	9	7	8
17	23	23	22
18	4	6	4
19	7	8	7
20	3	5	3
21	18	20	16
22	2	1	2
23	11	10	15
24	5	3	6
25	16	17	18

. In the approach that combined FMEA with fuzzy TOPSIS [64], the weights of the risk factors—occurrence, severity, and detection—were taken into account when calculating the FRPN. This results in a greater variety of unique FRPN values, thereby facilitating the prioritization of failure modes. However, in this combined approach, the FRPN was derived through fuzzy algebra operations, and the relationship between the risk factors and the FRPN was not fully addressed; therefore, there may be some bias in the results of this combined approach. For example, in the case of failure mode No. 3, the prioritizations obtained from the other two approaches are both ranked 1, while the prioritization from the approach that combined FMEA with fuzzy TOPSIS is ranked 2. In the approach that combined FMEA with fuzzy logic [13], the input values for the fuzzy logic system were derived from the defuzzification values, which were calculated based on the average of the fuzzy evaluation results. Consequently, identical FRPNs may occur, making it challenging to determine the prioritization of failure modes. For example, the prioritizations for failure modes No. 4, No. 5, No. 7, and No. 12 are identical. Clearly, the proposed approach, which integrates FMEA with SAM and fuzzy logic, overcomes the limitations of the two other typical approaches. Based on these findings, we believe our new approach offers several advantages, including more accurate results, ease of understanding, and flexibility in practical applications.

In order to compare the UX of the original IVIS with the newly revised version, an A/B test was conducted, and the system usability scale (SUS) was employed. SUS was developed by Brooke [65] and has been extensively used in measuring the UX of interactive systems. It comprises 10 statements to which users are asked to express their level of agreement. The statements are equally divided between positive and negative wording. In this scale, a score of 0 represents the worst possible outcome, while 100 represents the best. We enlisted 30 participants for this UX test, ensuring a balance in gender and an age range between 30 and 60. Each participant had at least 5 years of IVIS experience, ensuring a solid understanding and familiarity with such systems. The participants were asked to execute the assigned tasks and then assess both versions of the IVIS using SUS. The revised IVIS received a SUS score of 90.12. This was a significantly higher score compared to that of the original IVIS, which only garnered a score of 60.15. The statistical significance of this difference was confirmed (p < 0.05). These results suggest that the recommended measures that were incorporated into the revised IVIS design were indeed effective, and the revised design has proven to be more usable. Therefore, these findings serve to validate the proposed approach, demonstrating its effectiveness in enhancing the UX of the IVIS.

This study primarily focused on identifying and ranking failure modes. Despite the fact that the causes of these failure modes were delineated with clarity, we did not delve into a comprehensive investigation of the underlying mechanisms of these causes. To facilitate a deeper understanding of these mechanisms, one potential approach could involve the application of fault tree analysis, which is often utilized in the exploration of potential causes for failure modes [66,67]. Future research could concentrate on integrating fault tree analysis with the approach proposed in this paper to research the root causes of failure models concerning UX [68]. Note that in this case study, failure mode identification was carried out in a stationary vehicle, not a moving one. This had a minimal impact on the accuracy of error identification results [69]. Future studies could expand this research to include vehicles in motion on highways.

Our approach is slightly complex and comprises several computational steps, but it is properly carried out. More importantly, it is composed of four simplification stages, namely, detecting failure modes, analyzing failure modes, calculating FRPNs by using fuzzy logic, and modifying the failure modes with high priorities. The fuzzy calculation process can be facilitated through programming. In this study, the programming was implemented using the fuzzy logic toolbox for MATLAB. It involved aggregating fuzzy opinions for risk parameter assessments and calculating FRPNs with fuzzy logic. The proposed approach will have broad applications due to these four simplification stages. While this study employed an IVIS as an example, the approach outlined in this research is universal and can be applied broadly to other interactive systems, including mobile apps, software interfaces, and even specific hardware controls, with the aim of improving UX. Possible future works include identifying automated repair methods for the failure modes identified and ranked through the proposed method, identifying how the FRPN values from the proposed method can help with such repair, and adding the fuzziness of human reasoning while interacting with the system.

6. Conclusions

In this paper, we propose a fuzzy FMEA-based hybrid approach that integrates FMEA with SAM and fuzzy logic to improve the UX of interactive system design. Through the proposed approach, the priorities of failure modes concerning UX can be distinguished clearly; furthermore, it allows the subjective and even subconscious aspects of the risk factors of failure modes to be effectively managed. The proposed approach mainly focuses on the UX problem concerning designing for errors. This approach can be regarded as a universal approach for improving the UX of interactive system design and can be applied to a broader field of UX, including interactive product design, app design, and web design. It should be mentioned that in an interactive system, users, technologies, activities, contexts, and design can all lead to failure modes, and this study primarily focused on design. Future research should aim to integrate all components to analyze the risk of failure modes concerning the UX of interactive systems.