Service Failure Risk Assessment and Service Improvement of Self-Service Electric Vehicle

Abstract

Electric vehicle sharing is necessary for achieving carbon neutrality. The self-service electric vehicle mode offers unique advantages in terms of freedom of movement and privacy protection. Meanwhile, this mode requires a high-quality service guarantee because of the separation of management and use. The purpose of this study is to propose a framework for the risk control and service optimization of self-service electric vehicles, which includes service life cycle analysis, risk assessment by using a newly integrated fuzzy failure mode and effect analysis, and a consumer satisfaction survey based on the Kano model. Sixteen services were extracted through the service life cycle analysis and online review study, and their corresponding service failures were then ranked through risk assessment. The risk assessment showed that the reliability of vehicle-related services has the greatest impact on safety, followed by financial-related and driving-safety-related services. A Kano model-based survey showed that all kinds of service failures brought significant customer non-satisfaction, while different service improvements brought differentiated satisfaction. To deeply improve service satisfaction, a Risk-Satisfaction analysis was conducted, indicating that services with high risk and high satisfaction deserve further investment.

1. Introduction

Environmental protection is increasingly urgent, as 12% of greenhouse gases and 25% of urban ambient PM2.5 are produced by transportation links [1], so promoting green transportation has become an important means of sustainable development [2]. Shared travel comes in many forms, and it is characterized by low carbon, economic, and great environmental protection value [3,4]. Car sharing has unique advantages in medium and long-distance shared travel, which first appeared in Switzerland in 1948, and its rapid development began around the beginning of the 21st century [5]. Car sharing can replace a certain number of private vehicles and meet people’s travel needs. Research has shown that one shared car can replace 15 private ones [6]. However, car sharing cannot absolutely achieve the purpose of protecting the environment because shared cars tend to be operated for long periods daily, and the development of the car sharing industry may promote the production scale of internal combustion engine (ICE) vehicles, which exert great environmental impacts [7]. The result is a paradox, since the use of ICE vehicles adversely affects human health and living quality, while current climate problems require the reduction of greenhouse gas emissions [8]. As the carbon emissions from electric vehicles (EVs) are lower than that of ICE ones over the lifespan in all scenarios [9], the dilemma can be addressed by EV sharing. Promoting EV sharing can reduce the ownership of vehicles and the pollution from ICE vehicles simultaneously, while exerting the effect of environmental protection [10,11]. In conclusion, EV sharing can meet people’s travel requirements, reduce vehicle congestion, and benefit the control of global warming, which is important to sustainable development [12,13].

Many EV sharing modes are available to consumers. In terms of vehicle operators, they can be divided into ridesharing EVs and self-service EVs (SSEVs). Meanwhile, SSEVs are even more popular with consumers [14]. This is related to the reality that when sharing limited space with strangers during travel, passengers may feel uncomfortable with the awkward atmosphere and proximity [15,16]; moreover, a negative correlation exists between consumers’ loss of autonomy, lack of privacy, and ride-sharing [17]. In contrast, in self-service mode, consumers do not have such psychological barriers, and they could have more freedom while driving by themselves. Therefore, SSEV has an irreplaceable value in shared travel because of its privacy protection and flexibility characteristics.

However, since the management and use of SSEVs are naturally separated, the security issues of using SSEVs are closely related to the quality of their services. First, SSEVs are in a state of continuous use, and their operating intensities are greater than ordinary EVs. EV technologies are not as mature as ICE ones [18,19]; with the absence of drivers’ continuous tracking, some faults and safety risks may not be found in time, thereby leading to losses of consumer interest. Second, SSEVs are like public goods, so consumers may not properly use them or may dirty them. Improper usage may damage the functions of EVs, and sanitation problems can lead to the spread of diseases. Third, consumers who have different driving experiences may drive EVs independently with absolute autonomy, while they may lack the sufficient experience or not possess a complete picture of the EV’s conditions, which exposes them to a variety of risks and causes a higher accident rate [20]. Fourth, the legal and property relations of the usage of SSEVs are complicated, so consumers are in a relatively weak status. This asymmetry is easily prone to inaccuracies and injustices in terms of liability determination, violation handling, dispute settlement, and other matters. In summary, the reliabilities of SSEVs are affected by multiple sources, including frequent use, lack of continuous tracking, uncertainties in sharing, complex relations, and so on. The safety performance of SSEVs requires car sharing companies (CSCs) to provide reliable services to counter uncertainties in sharing to guarantee the interests of consumers. Typically, the service reliability of CSCs directly affects the travel convenience, property safety, and even physical safety of consumers, etc. However, due to cost control measures or managerial negligence, CSCs may provide unqualified services, so the interests of consumers will not be guaranteed.

A survey has shown that consumers are greatly concerned about the reliability of shared EVs, and they will assign low scores due to risks, failures, and other problems; hence, the attention on reliability should not be ignored [21]. In the service process of using shared EVs, if consumers have negative experiences or safety concerns, their enthusiasm will drop, which will sacrifice the opportunity to reduce carbon emissions [22,23]. Other serious consequences will occur with the widespread of negative words and media reports. Specific CSCs, and even the whole EV sharing industry, may be greatly impacted, meaning such a scenario needs to be avoided. Therefore, it is necessary to systematically assess SSEV services to indicate the risk types and risk levels of service failures. Potential risks may exist in the entire service life cycle (SLC) of SSEVs from the registration stage to theuse stage to the final account cancellation stage. Thus, the current study carries out risk assessment from the entire SLC to figure out the important services and their corresponding failures, so as to improve the reliability of the services as a whole.

In pursuit of the reasonable allocation of resources, effective improvements of safety performance, and consumer satisfaction, a novel framework for service risk assessment and safety improvement, which is based on fuzzy failure mode and effect analysis (FMEA) and the Kano model, is proposed. Thus, service improvement strategies can be formulated based on consumer requirements and risk assessment results. CSCs can identify the importance of service security, improve the quality of services, and develop characteristic service strategies through this study; market regulators can supervise CSCs referring to the results; and consumers can choose relatively reliable SSEV service providers accordingly. All of these contribute to the promotion of the development of the SSEV industry and, then, the goal of carbon neutrality.

The remainder of this article is organized as follows: Section 2 presents the literature review, including related car sharing service research, the improvement of fuzzy FMEA, and the introduction of the Kano model. Section 3, firstly, describes the fuzzy mathematical knowledge and symbols used in the research and, then, provides a step-by-step introduction to the methods used in the study, including these: how to determine authority weights, how to verify the evaluation values by consensus inspection, how to calculate the weights of risk factors, and how to rank FMs. Section 4 details the services and service failures through the decomposition of the SLC and study of online reviews. In Section 5, risk assessment criteria are established, risk values are assessed by experts, consensus tests are applied for inspection, and then the FMs are sorted using the methods proposed in Section 3. Moreover, sensitivity and comparative analyses are conducted herein. Section 6 carries out a risk–satisfaction analysis based on the result of risk assessment and an investigation of Kano requirements. Section 7 discusses the conclusion, presents a brief summary of the study, and proposes some research prospects.

2. Literature Review

Car sharing facilitates people’s travel and saves resources, and many scholars have carried out various studies to improve its service quality. Lee et al. studied the service design of shared cars earlier [24], Arena et al. designed different modes of EV sharing services from the perspective of consumers [25], and He et al. designed free-floating shared electric vehicle services from the perspective of CSCs [26]. Compared with private cars, car sharing has greater uncertainties and risks, so sharing safety issues attracted the attention of scholars. Some studies focus on service process safety of specific car sharing modes [27,28], some on driver and passenger safety [20,29,30], and some on technical safety [31,32].

As mentioned above and in the studies shown in

Table 1. Related research of service and safety of car sharing.

Author	Research Topic	Research Content and Finding
Choi [20]	Relationship analysis of car sharing and car accidents	It analyzed the relationship between car sharing and accidents in urban and proved a positively correlation. Based on the findings, security services and measures were recommended as necessary.
Lee [24]	System design of electric car sharing with mobile technology	It designed an electric car sharing system from user perspective and presented necessary infrastructures of the system.
Arena [25]	Service design of EV sharing from consumer perspective	It studied service design for different types of EV sharing in hope of promoting the development of EV sharing.
He [26]	Service design of free-floating EV sharing from CSC perspective	It studied the service design problem for free-floating EV sharing systems and built a model including repositioning, recharging and fleet size determination to guarantee the service level.
Symeonidis [27]	Security and privacy assessment of P2P key-less car sharing system	It assessed security and privacy issues in P2P key-less car sharing system and provided a guide for secure system design.
Hanusik [28]	Identification and risk assessment in ICE car sharing	It identified and evaluated fourteen risks in car sharing from the operator’s perspective, and put forward management suggestions.
Chaudhry [29]	Passenger safety in ride-sharing mode	It analyzed the risks in ride-sharing mode and put forward suggestions to resist risks through security service optimization.
Lee [30]	Effect of car sharing on the crashes of teenage drivers	It discussed the safety and security issues of car sharing platforms and verified the correlation between the number of teenage drivers in sharing and the number of car crashes of teenage drivers.
Park [31]	Design of secure authentication method with bio-information	It designed a security service solution that using fingerprint information to unlock vehicles while preventing information leaks.
Jing [32]	Safety analysis of using car sharing software while driving	It studied the impact of driver age and experiences in software usage on driving safety, and proved that car sharing software has a significant negative impact on driving distraction and usability.

, unqualified car sharing services may cause injuries and losses of passengers, leading to an increase in accident rate, which are tightly related to service qualities and illustrate that safety issues with car sharing cannot be ignored. Therefore, qualified services are required to reduce possible risks and losses. However, through extensive reading, current studies on the safety of electric vehicle sharing services mainly focused on details [20,30,31,32], and existing systematic studies on other sharing modes only considered service design or security severally [24,25,26,27,28,29] but do not fully consider the inevitable relationship between service reliability and consumer satisfaction. Therefore, no scholars have conducted systematic studies on service optimization of SSEVs from a comprehensive perspective. Moreover, previous comprehensive studies on the risk assessment and the design or optimization of car sharing services mainly relied on the judgment of researchers, which did not combine with the real thoughts and expectations of consumers [25,26,28,31]. Thus, to improve the objectivity and authenticity of the research, this study will make up for this deficiency by using real forum data and a consumer survey.

The purposes of this study are, firstly, to indicate the important services in the SLC, secondly, to assess the risk levels of corresponding service failures, and, finally, to provide reference for further service optimization. Considering that services are consumer oriented, it is one-sided to determine which services should be improved based solely on the level of risks. Consumer attitudes towards service failures and service improvements also need to be considered in order to effectively improve the overall service quality. Therefore, in determining the improvement priorities of different services, this study not only evaluates the risk of service failure, but also investigates the degree of change in satisfaction with service improvement. Risk assessment is carried out through a newly integrated fuzzy FMEA approach, and consumer satisfaction is investigated based on the Kano model. Combined with the two-dimensional analysis of risk and satisfaction, this study can provide specific and feasible insights for service improvement.

2.1. Fuzzy FMEA

FMEA was first proposed by NASA in the 1960s [33], which is an effective safety and reliability analysis tool that is widely used in the exploitation of means of transport [34], automotive accessories [35], product redesign [36], and so forth. The FMEA method calculates risk levels through the risk priority number (RPN), which is the product of three risk factors: occurrence, severity, and detection. Occurrence refers to the probability of occurrence of a failure mode (FM), severity refers to the severity of the consequences of corresponding FM, and detection refers to its detecting difficulty. The greater the RPN value is, the higher the risk level [37]. When used in practical applications, this classical FMEA has certain limitations. In some cases, the values of risk factors are difficult to be determined precisely, the relative importance of different risk factors is not always being considered, and multiplying the values of different risk factors may obtain same RPNs [38]. Moreover, as the RPN value is a product of three numbers, a small change in any risk factor may cause a large change in the result [39]; high severity FMs may be ignored when the rest two values are low [40]. All of these problems cause sorting failures.

Therefore, to handle the fuzzy issues and avoid sorting failures, many scholars have proposed fuzzy FMEA methods on the basis of fuzzy information and multi-attribute decision making method [41,42]. For examples, Efe improved a fuzzy FMEA model combined with the VIKOR (Vlse Kriterijumska Optimizacija I Kompromisino Resenje, in German) method under intuitionistic fuzzy environment and, then, utilized it in the analysis of operational safety risks of shipbuilding [43]. Lo et al. proposed an integrated FMEA approach, which constructed risk factor influence graphs on the basis of the DEMATEL (decision-making trial and evaluation laboratory) method combined with the ranking principles of TOPSIS (technique for order preference by similarity to an ideal solution) and, then, ranked FMs using four integrated multiple-attribute ordering methods [44]. Luqman et al. made FMEA improvements in Pythagorean environments using directed graph, matrix techniques, and language information in power plant risk studies [45]. Under a fuzzy-based environment, Liu et al. used fuzzy information, integrated the empowerment method and the combined method of TODIM (Interactive multi-criteria decision making) and PROMETHEE (Preference ranking organization method for enrichment evaluation) to flexibly reflect experts’ opinions, so as to improve the implementation accuracy of FMEA [46].

As discussed above, scholars integrated different methods to make up for the shortcomings of classical FMEA. However, most of these studies improved fuzzy FMEA from a rational point of view, while a minority take into account people’s irrationality. In the case of EV sharing, consumers are important participants, so sufficient attention should be paid to their risk aversion behaviors in the calculation. The value mechanism of consumers is concave for gains and convex for losses, while losses are steeper than gains [47]. Thus, the current research proposes an integrated fuzzy FMEA combined with prospect theory (PT). In addition to the matching of ranking method, this study also needs to construct a more comprehensive evaluation process. Wu et al. made a systematic review of FMEA methods and pointed out that the following problems still widely exist in the current application of fuzzy FMEA: the FM selection procedure is not rigorous enough; the formulation of risk assessment standard is not scientific and not widely used; and, moreover, there is still room for innovation in the objective weight assignment method of risk factors and sorting method [48]. On the basis of the above discussion, this study presents a comprehensive and rigorous process during the analysis process.

This work determines expert authority weight through mutual evaluation; identifies service failures of SSEVs through online review analysis and SLC analysis; constructs risk assessment criteria of the FMs combining with graded language variables; tests expert assessments by consensus theory; develops the TOPSIS entropy method to objectively determine weights of different risk factors; and uses the PROMETHEE-II evaluation method based on the prospect theory (PT), namely PT-PROMETHEE-II, to sort the risk levels of service failures. The service extraction is more realistic because of the combination of online comments and SLC decomposition; the risk ratings stand up better because of rigorous expert assessment; and the rational weight determination of risk factors and the full consideration of consumer psychology make the ranking results more realistic.

2.2. Kano Model

The Kano model was proposed by Kano Noriaki [49], which originated from the motivation-hygiene theory proposed by Herzberg in 1959 [50]. The Kano model is an evaluation method that makes the invisible attributes of service quality explicit (Figure 1). This model defines six different customer requirements, including must-be requirement (M), one-dimensional requirement (O), attractive requirement (A), indifferent requirement (I), reverse requirement (R) and questionable requirement (Q) [51].

As shown in Figure 1, different requirement types have different subjective experiences when the same amount of objective performance changes. The relative important ones of them are the ‘M’, ‘A’, and ‘O’ requirements [52]. The characteristic of ‘M’ is that lower than the consumers’ expectation brings more non-satisfaction, while higher does not bring significant satisfaction. The characteristic of ‘A’ is opposite to ‘M’: lower than the expectation does not lead to significant non-satisfaction, but higher leads to significant satisfaction. The characteristic of ‘O’ is that lower than the expectation leads to significant non-satisfaction, while being higher leads to significant satisfaction. Among the other three requirements, ‘R’ is opposite to ‘O’, ‘I’ does not cause any change in satisfaction or non-satisfaction, and ‘Q’ is often ignored because it does not conform to cognition.

The Kano model divides the differences of consumer demands in detail, surpasses the linear relationship between quality and satisfaction, and presents the changes of consumer demand dynamically. Thus, due to its excellent performance in depicting consumers’ real ideas, it has been widely used in product design, enterprise management, service optimization, and other fields [53,54]. As service quality is widely concerned by consumers, applying the Kano model to the service reliability optimization of SSEVs definitely contribute to a more comprehensive and efficient service improvement. Through the Kano model survey, consumers’ requirements for service security will be clarified. The importance of service security will be revealed, and strategic recommendations for service improvement will be made accordingly.

To sum up, the contributions of this study mainly include three aspects. First, this study makes up for the gap in SSEV service research, which plays a role in promoting SSEV service security and has long-term significance for sustainable development. Second, this study conducts a survey of consumer requirements while evaluating service failure risks, which provides a more comprehensive guide for the improvement of service quality. Third, this study integrates a new fuzzy FMEA approach with a rigorous program, appropriate weight calculation method, and sorting method. The rigorous procedures and real data-based analyses ensure the reliability of evaluation, the weight calculation method ensures objectivity, and the ranking method reflects the reality of risk avoidance.

3. Methodology

To facilitate the risk assessment, this study sets mathematical symbols for the calculation. Let $E_t\left(t=1,2,\dots ,T\right)$ indicate the expert set in the assessment, where $T$ is the number of experts involved in the assessment. Let $R_i\left(i=1,2,\dots ,m\right)$ indicate different risk factors, where $m$ represents the number of risk factors. Let $F{M}_j\left(j=1,2,\dots ,n\right)$ represent the FMs, where $n$ indicates the number of FMs. The evaluation matrix given by expert $E_t$ is defined as $E_t={\left(E_{ij}^t\right)}_{m\times n}$, in which $E_{ij}^t$ is the evaluation value given by expert $E_t$ for risk factor $R_i$ of service failure $F{M}_j$. Given the authoritative differences of different experts, expert $E_t$ is given weight values ${\gamma }_t$ through mutual evaluation. The weight values satisfy ${\gamma }_t\ge 0$ and ${\displaystyle {\sum }_{t=1}^T{\gamma }_t}=1,\left(t=1,2,\dots ,T\right)$. Then, set $w_i$ represents the weights of the risk factors, where $w_i\ge 0$ and ${\displaystyle {\sum }_{i=1}^m{w}_i}=1,\left(i=1,2,\dots ,m\right)$.

The analysis framework contains three main parts. Part one: forming the expert team through mutual evaluation, analyzing the SLC of SSEV by them, and, then, identifying important services and their failure modes based on the principles of grounded theory and service quality function deployment (QFD) by reading online reviews. Part two: a risk assessment of service failures, which consists of three main steps: (1) evaluating risk levels to construct the risk assessment matrix; (2) verifying the evaluation matrix with consensus theory, adjusting the evaluation values until consensus requirements are met, and then integrating the evaluation matrix with authority weight vector; and (3) calculating objective weights by using TOPSIS entropy method, ranking the FMs by using PT-PROMETHEE method, and, then, conducting the sensitivity and comparison analyses to ensure the robustness of the results. Part three: conducting a consumer requirement questionnaire survey and proposing service optimization suggestions based on the results of risk assessment and Kano survey. The overall process of the framework is shown in Figure 2.

3.1. Linguistic Variables

Fuzzy sets were designed to describe complex and uncertainty problems well, which can characterize the ambiguity of human cognition. This study employs linguistic variables to evaluate the experts’ authority and describe the risky extent of different service failures.

A preset odd linguistic variable set is defined as $Lv=\{l_p|p=0,1,2,\dots ,(\frac{L}{2})-1,\frac{L}{2},\frac{L}{2}+1,\dots ,L\}$. The value of $L$ is usually even. $l_p$ represents the $p+1$th language phrase in $Lv$. The above set has the following features:

• Ordering: when i > k, L_i > L_k.
• Reversible: when k = L − i, $neg(L_k)=L_i$, where neg is an inverse operation.
• Extreme value: when $L_i>L_k$, there exists $\max \{L_i,L_k\}=L_i$ and $\min \{L_i,L_k\}=L_k$.

To facilitate the calculation, this study transforms the language information into triangle fuzzy numbers (TFNs); the function is given as follows [55]:

$$\tilde{\mathrm{l}}=(\max \left\{(p-1)/L,0\right\},p/L,\min \left\{(p+1)/L,1\right\})$$

(1)

Given nine hierarchical linguistic variables, i.e., $L=8$, the conversion relationship between linguistic variables and triangular fuzzy numbers according to Formula (1) is shown in

Table 2. Conversion relationship between linguistic variables and TFNs.

Linguistic Variable	Triangle Fuzzy Number
EL (Extremely Low)	(0.0000, 0.0000, 0.1250)
VL (Very Low)	(0.0000, 0.1250, 0.2500)
L (Low)	(0.1250, 0.2500, 0.3750)
SL (Slightly Low)	(0.2500, 0.3750, 0.5000)
F (Fair)	(0.3750, 0.5000, 0.6250)
SH (Slightly High)	(0.5000, 0.6250, 0.7500)
H (High)	(0.6250, 0.7500, 0.8750)
VH (Very High)	(0.7500, 0.8750, 1.0000)
EH (Extremely High)	(0.8750, 1.0000, 1.0000)

.

3.2. The Establishment of the Expert Team

Risk assessment of SSEVs requires experts’ professional knowledge and experiences. So, the experts are required to evaluate the authority values of the others based on known information. To facilitate evaluation, the evaluator’s own evaluation value is uniformly set at F as a reference point (Table 2). In the evaluation, higher language level represents higher authority. As T expert alternatives have the evaluation qualification, $P_t^k,\left(t,k=1,2,\dots ,T\right)$ is set as the authority value of the kth expert given by the tth expert. Then, if the size of the $\frac{1}{T}{\displaystyle {\sum }_{k=1}^T{P}_t^k}$ of an expert is larger than F, the expert will be hired and the value will be used to evaluate weights; otherwise, the candidate will not be invited anymore.

After the invitation, the experts analyze the SLC of SSEVs by reading online reviews, identify necessary services and their failure modes based on grounded theory and QFD, and, then, assess the risk levels of service failures according to assessment criteria.

3.3. Inspection and Integration of FMEA Matrix

After the experts’ evaluation, the decision matrix $E^T={\left(e_{ij}^t\right)}_{m\times n\times T}$ can be formed. To ensure the consistency of the evaluation, a consensus inspection is required. If the evaluation values meet the consensus requirement, the integration calculation will be carried out; if the evaluation values do not reach the consensus level, then the unmet ones need to be re-evaluated until they do [56]. Assuming that the two experts’ assessment values of three risk factors ($i=1,2,\dots ,m$) of the same FM (j) are $e_{ij}^t$ and $e_{ij}^k$, respectively, the similarity matrix of the two experts’ assessment can be defined as:

$$s{m}_{ij}^{kt}=1-d_s\left(e_{ij}^t-e_{ij}^k\right),i=1,2,\dots ,m$$

(2)

where $d_s\left(e_{ij}^t-e_{ij}^k\right)$ is the normalized Euclidean distance of the two evaluation values. Set $\varphi$ as arithmetic average operator, then the aggregation similarity matrix of expert evaluation value can be defined as:

$$c{m}_{ij}=\varphi \left(s{m}_{ij}^{kt}\right)$$

(3)

Let $c{l}_j$ represents the consensus level of FM, then:

$$c{l}_j=\frac{{\displaystyle {\sum }_{i=1}^{m}c{m}_{ij}}}{m}$$

(4)

According to the principle of consistency test, the expert’s assessment should meet the following consensus condition:

$$\min c{l}_j>\epsilon$$

(5)

where $\epsilon$ is the consensus parameter and its value range is $\epsilon \in \left[0.5,1\right]$. This study refers to the study of [57] and takes $\epsilon =0.85$.

After obtaining the required evaluation matrix, the integrated risk factor decision matrix can be obtained by weighted calculation with the weight vector of experts. The weight vector of experts is determined by calculating the Euclidean distance between the authority value and the preset reference point. Setting the distance as ${\gamma }_t^{dis}=d\left(\frac{1}{T}{\displaystyle {\sum }_{k=1}^T{P}_t^k},F\right)$, the weights are calculated as follows:

$${\gamma }_t=\frac{{\gamma }_t^{dis}}{{\displaystyle {\sum }_{t=1}^T{\gamma }_t^{dis}}}$$

(6)

Then, the integrated matrix can be computed as follows:

$$E={\displaystyle {\sum }_{t=1}^T{\gamma }_t}{E}_{ij}^t={\left(e_{ij}\right)}_{m\times n}$$

(7)

Among them, $e_{ij}=\left(e_{ij}^{low},e_{ij}^{mid},e_{ij}^{up}\right)$ is the calculated TFN with conserved language information.

3.4. Calculation of Weights of Risk Factors

The TOPSIS method was proposed by Hwang et al. in 1981. By calculating the relative closeness coefficient (RCC), the TOPSIS method selects the alternative, which is closest to the positive ideal solution [58]. The RCC values eliminate the dimensional differences by calculating the distance differences, realizing normalized processing, and reflecting the discreteness degrees of the distribution of attribute values. According to the principle of the entropy method, the weight size of an attribute is determined by the discrete distribution of the attribute values [59]. Therefore, according to the RCC values, the discretization degree of each risk factor can be calculated using the entropy method, thus, the weights are determined. This proposed method is named as the TOPSIS entropy method, and the calculation steps are as follows.

Step 1: calculate the hierarchical average integral of the TFNs [60].

$${\overline{e}}_{ij}=\frac{e_{ij}^{low}+4e_{ij}^{mid}+e_{ij}^{up}}{6}$$

(8)

Step 2: calculate the values of the positive and negative ideal solutions $f_j^{+}$ and $f_j^{-}$ according to the size of ${\overline{e}}_{ij}$. The maximum fuzzy number of ${\overline{e}}_{ij}$ is a positive ideal solution, and the smallest one is a negative ideal solution.

$$\{\begin{array}{l}{f}_j^{+}=\max \left(e_{ij}^{low},e_{ij}^{mid},e_{ij}^{up}\right)=\left(f_j^{+low},f_j^{+mid},f_j^{+up}\right)\\ f_j^{-}=\min \left(e_{ij}^{low},e_{ij}^{mid},e_{ij}^{up}\right)=\left(f_j^{-low},f_j^{-mid},f_j^{-up}\right)\end{array}$$

(9)

Step 3: calculate the distances between the FM values and their corresponding positive and negative ideal solutions according to the Euclidean distance function.

$$d\left(e_{ij},f_j^{+}\right)=\frac{\sqrt{3}}{3}\sqrt{{\left(f_j^{+low}-e_{ij}^{low}\right)}^2+{\left(f_j^{+mid}-e_{ij}^{mid}\right)}^2+{\left(f_j^{+up}-e_{ij}^{up}\right)}^2}$$

(10)

$$d\left(e_{ij},f_j^{-}\right)=\frac{\sqrt{3}}{3}\sqrt{{\left(e_{ij}^{low}-f_j^{-low}\right)}^2+{\left(e_{ij}^{mid}-f_j^{-mid}\right)}^2+{\left(e_{ij}^{up}-f_j^{-up}\right)}^2}$$

(11)

Step 4: calculate the RCC value of each evaluated value.

$${\rho }_{ij}=\frac{d\left(e_{ij},f_j^{-}\right)}{d\left(e_{ij},f_j^{+}\right)+d\left(e_{ij},f_j^{-}\right)}$$

(12)

Step 5: normalize the RCC values.

$$z_{ij}=\frac{{\rho }_{ij}}{{\displaystyle {\sum }_{i=1}^m{\rho }_{ij}}}$$

(13)

Step 6: calculate the RCC entropy values.

$$e{t}_j=-K{\displaystyle {\sum }_{i=1}^m{z}_{ij}\ln z_{ij}},j=1,2,\dots ,n$$

(14)

where $K=1/\ln n$.

Step 7: calculate the objective weight value of each risk factor.

$$w_j^{\prime }=\frac{1-e{t}_j}{{\displaystyle {\sum }_{j=1}^n\left(1-e{t}_j\right)}},j=1,2,\dots ,n$$

(15)

3.5. Ranking FMs by Using the PT-PROMETHEE-II Method

The PROMETHEE method was first proposed by Brans in 1982 [61]. About two years later, the PROMETHEE-II method was proposed [62,63]. PROMETHEE-II is a ranking method based on a full priority relationship, while PROMETHEE-I is based on a partial priority relationship. Given the nature of risk and consumers’ risk aversion tendency, the PROMETHEE-II method based on PT was proposed [64]. The current study uses the PT-PROMETHEE-II method to assess risks, and the calculation steps are as follows.

Step 1: determine the reference points of the FMs for the PT value calculation. Reference points are used to reflect expert behavioral decision preferences, and each FM value is a mutual decision reference point according to the ranking principle of the PROMETHEE-II method.

Step 2: calculate the distance $d\left(e_{ij},e_{kj}\right)$ between $F{M}_i$ and $F{M}_k$.

$$d\left(e_{ij},e_{kj}\right)=\frac{\sqrt{3}}{3}\sqrt{{\left(e_{ij}^{low}-e_{kj}^{low}\right)}^2+{\left(e_{ij}^{mid}-e_{kj}^{mid}\right)}^2+{\left(e_{ij}^{up}-e_{kj}^{up}\right)}^2}$$

(16)

Step 3: calculate the PT value through the value function of the cumulative prospect theory (CPT) [65].

$$V_j^{+}\left(F{M}_i,F{M}_k\right)=\{\begin{array}{l}{\left(d\left(e_{ij},e_{kj}\right)\right)}^{\alpha },e_{ij}\le e_{kj}\\ 0,otherwise\end{array}$$

(17)

$$V_j^{-}\left(F{M}_i,F{M}_k\right)=\{\begin{array}{l}\lambda {\left(d\left(e_{ij},e_{kj}\right)\right)}^{\beta },e_{ij}>e_{kj}\\ 0,otherwise\end{array}$$

(18)

where there $0<\alpha ,\beta <1$ are the decreasing factors of psychological perception sensitivity. $\alpha$ and $\beta$ indicate that the greater the value is, the less sensitive to gain and loss. $\lambda >1$ indicates that individuals are more sensitive toward losses than gains. Accordingly, $\alpha =\beta =0.88$ and $\lambda =2.25$ [65].

Step 4: determine the level priorities of the FMs with the relative weight of the risk factors.

$$H\left(F{M}_i,F{M}_k\right)={\displaystyle {\sum }_{j=1}^n\left(V_j^{+}\left(F{M}_i,F{M}_k\right)\times w_j\right)}$$

(19)

$$H\left(F{M}_k,F{M}_i\right)={\displaystyle {\sum }_{j=1}^n\left(V_j^{-}\left(F{M}_i,F{M}_k\right)\times w_j\right)}$$

(20)

where $H\left(F{M}_i,F{M}_k\right)$ indicates that the priority of the $F{M}_i$ is below the remaining FMs, and $H\left(F{M}_k,F{M}_i\right)$ indicates that the priority of the $F{M}_i$ is superior to the rest of the FMs.

Step 5: calculate the outflow values and inflow values.

$${\pi }^{+}\left(F{M}_i\right)=\frac{1}{n-1}{\displaystyle \sum _{i\ne k}H\left(F{M}_i,F{M}_k\right)}$$

(21)

$${\pi }^{-}\left(F{M}_i\right)=\frac{1}{n-1}{\displaystyle \sum _{i\ne k}H\left(F{M}_k,F{M}_i\right)}$$

(22)

where ${\pi }^{-}\left(F{M}_i\right)$ represent the outflow values and ${\pi }^{+}\left(F{M}_i\right)$ denote the inflow ones.

Step 6: calculate the net flow value of $F{M}_i$.

$$\pi \left(F{M}_i\right)={\pi }^{-}\left(F{M}_i\right)-{\pi }^{+}\left(F{M}_i\right)$$

(23)

$\pi \left(F{M}_i\right)$ is the comprehensive level priority of $F{M}_i$, and the FMs are arranged in descending order according to the size of the net flow value.

3.6. Calculation of Satisfaction Index

Berger et al. proposed the Better–Worse index analysis method to study the relationship between quality factors and customer satisfaction [66]. The index includes ‘Better’ and ‘Worse’, which represent satisfaction and non-satisfaction of consumers to the same quality attribute, respectively. Their corresponding formulas are given as follows:

$$Better=\left(A+O\right)/\left(A+O+M+I\right)$$

(24)

$$Worse=\left[\left(O+M\right)/\left(A+O+M+I\right)\right]\times \left(-1\right)$$

(25)

Combined with the ‘Satisfaction’–‘Non-Satisfaction’ index and the results of risk assessment, a risk–satisfaction analysis can be carried out. The results of a risk–satisfaction analysis can be used as references for service optimization, striving to be more practical and effective.

4. SLC Analysis of SSEVs

4.1. Construction of Expert Team

Referencing to previous relative studies [57,67], according to the professional level and professional experience of experts, four expert candidates who have more than five years’ relevant working or research experience were invited. The mutual evaluation information is given and shown in

Table 3. Mutual evaluation information of candidate experts.

Expert Mutual Evaluation		Evaluated
		E1	E2	E3	E4
Evaluator	E1	F	SH	VH	H
	E2	H	F	H	H
	E3	H	SH	F	SH
	E4	F	SH	SH	F

. They all meet the assessment requirement.

4.2. Service Life Cycle Analysis

Safety problems cannot be ignored because injuries caused or magnified by service failures will bring serious burdens to customers, the CSCs, and society. Hence, to control the risks, the necessary services and their corresponding service failure modes should be clearly identified first. Such task is organized herein by analyzing the SLC (Figure 3). As shown in Figure 3, the SLC of the SSEVs is divided into three stages: registration stage, use stage, and account cancellation stage. The use stage is divided into three parts according to the using process: starting part, driving part, and stopping part. Different consumers may experience different SLCs. For example, some consumers will cancel after registration or use, while some will not cancel after registration and use, and so forth. However, generally speaking, most consumers will go through the stages of registration, use and cancellation in the SLC. Thus, this study identifies services and their FMs according to the general process.

To identify critical services in the SLC and their FMs, a great deal of time was spent reading online reviews on China’s largest Tieba [68] and auto forum website [69]. These data were written by users based on their own experiences, which span about five years and are relatively comprehensive and truthful. As these online reviews were complex and limited in numbers, using machine learning or artificial intelligence was inefficient and lacked materials. Hence, directly manual reading was adopted. Imitating what grounded theory does in terms of extracting elements [70] and what service QFD does in terms of selecting important services [71], a large number of services and their corresponding FMs were screened out through reading and comprehensive analysis. Experts need to further analyze and summarize them according to the SLC and their professional knowledge, to guarantee the data are being interpreted professionally and comprehensively. Following the principles of importance, independence and integrity, 16 necessary services and their corresponding FMs of the SSEV mode in the SLC were screened out (

Table 4. Necessary services and their corresponding FMs of SSEV mode in the SLC.

Stages		Essential and Reliable Services	Service Failure Modes	Codes
Registration stage		Effective information protection	Information abuse	FM1
		Fair agreement service	Agreement trap	FM2
Use stage	Starting part	Provide reliable quality EVs	Provide defective EVs	FM3
		Professional maintenance services	Careless maintenance	FM4
		Safe and convenient charging service	Unreliable charging service	FM5
		Clear identification of responsibility	Unclear identification of liability	FM6
	Driving part	Professional security identification	Lack of security identification	FM7
		Reasonable and transparent charges	Unreasonable charges	FM8
		Sufficient safety equipment	Inadequate safety equipment	FM9
		Complete and adequate insurance	Inadequate vehicle insurance	FM10
	Stopping part	Convenient and safe parking service	Troubled parking	FM11
		Timely and comprehensive safety alerts	Imperfect security alerts	FM12
		Convenient handling of violations	Complexity in handling violation	FM13
Account cancellation stage		Quick and convenient deposit refund	Troubled in refunding deposit	FM14
		Impartial dispute resolution service	Unfair treatment in dispute	FM15
		Real-time quality customer service	Poor customer service	FM16

). A brief analysis including service differentiation, service features, service FM identification, service demand scenario, and risk consequence analysis is presented later.

4.2.1. Registration Stage

In the registration stage, consumers need to submit personal information. Comprehensive information such as driving license, ID card, and even facial information will be collected, which involves privacy and may be abused by CSCs. Although the CSCs may not abuse the information deliberately, it may be stolen by employees or accessed by criminals via a software virus. Consumers may be harassed or even defrauded because of it. The scope of infringement may be considerably wide, and it is difficult to precisely determine whether these infringements are from shared services.

To guarantee that consumers will not damage the EVs or shirking responsibilities, they usually need to pay a certain amount of security deposit. However, if CSCs do not operate well, or if they deliberately do not refund the deposit, consumers will face losses in the account cancellation stage. At the same time, consumers also need to sign a user protocol, which may have some hidden traps, and it is hard for consumers to be aware of these unfair terms. Under normal circumstance, consumers will face financial or indirect losses from dealing with unfair treatments. However, if CSCs are allowed to stop an EV for maintenance excuses while it is in use, or things such as that, customers will be at risk of faulty operations, travel delays, or accidents.

4.2.2. Use Stage

In the starting part of the use stage, consumers need to carry out a necessary inspection before using an EV, to gain familiarity with its operation. This step is necessary, but it is not all about consumers, so it may be ignored by consumers. CSCs are obligated to provide reliable vehicles, carry out responsible maintenance of vehicles, and guarantee a safe and convenient car charging service. With defective EVs, consumers may face risks related to EV conditions while driving: the breakdown probability will increase, and the safety will not be guaranteed. Consumers will face financial losses and personal injuries with that. EVs have more complex electrical structures and wiring [72], and some potential faults are not easy to identify. With careless maintenance, the sanitary conditions and performances of the EVs cannot be guaranteed. Consumers may have a bad impression when choosing an EV or be infected with diseases because of poor hygiene, and EVs may break down halfway while driving or even have unanticipated accidents due to poor conditions. With an unreliable car charging service, it will take a lot of time and can lead to indirect losses when consumers need to recharge the EVs. Moreover, consumers may face charging accidents including vehicle damages or fires. From the perspective of consumers, if they do not find out existing damages before using an EV, then they will also face a compensation problem, which belongs to the unclear identification of liability.

In the driving part, consumers need to unlock the EVs, drive by themselves, cope with various situations and be billed at the same time. Basically, CSCs should guarantee that the bills are reasonable. If CSCs charge extra fees during the driving process, which is a very dishonest behavior, consumers will stop using the service and spread the word. Meanwhile, CSCs are responsible for preventing dangerous operations from consumers through safety identification. Without security identification service, consumers may have accidents due to rapid acceleration, improper driving, drunk driving, fatigued driving, and other wrong operations. When driving, consumers may face a variety of situations, such as accidents, fires, being trapped in the EV, etc., which may be caused by vehicle failures, passive accidents, or operational errors. Therefore, CSCs need to foresee these situations and equip the EVs with fire extinguishers, tripods, safety hammers, etc. With inadequate safety equipment, consumers may face secondary injury or greater losses. CSCs also need to provide sufficient and complete auto insurance. When taking responsibility in dealing with accidents, if the auto insurance paid by CSCs is incomplete or very low, consumers will face large payouts.

In the stopping part, consumers usually need to park the EV in designated areas or at specific charging stations. In this part, if the positioning fails or lack of parking spaces, consumers probably need to pay extra fees or spend time waiting. They may also choose to park the EV nearby, but this choice may bring troubles to them, such as the loss of dispatch fees. After stopping the EV, if consumers carry some belongings, they also need to take them away and then lock the EV. With imperfect security alerts, consumers may forget their belongings or forget to close the windows, which lead to property losses and liability for compensation. Moreover, without alerts of the battery level, excessive discharge, etc., consumers may not make it to their destination and may be locked in the car because the battery runs out. After parking, if a vehicle is scraped and a new consumer reports the damages in their starting part, the person who is responsible may not be immediately clear. Consumers may face fines in such unclear liability identification cases, while they are actually innocent. After using the SSEV service, consumers may receive violation reminders. CSCs need to assist consumers in dealing with violations, the duration of the process of dealing with the violation typically depends on the arrangement of the CSCs. The complexity of the process may waste too much time, which leads to some unknown indirect losses for consumers.

4.2.3. Account Cancellation Stage

If consumers stop using the EV sharing service of a CSC, they will choose to cancel their account or just ask for a refund of the deposit. When they ask for the deposit, they may encounter losses. For example, some CSCs may not return the deposit for various reasons. After consumers cancel their account, their private information may still remain in the servers of corresponding CSC, which may be sold, stolen, or abused for ignominious purposes and, consequently, leads to consumer losses. Before the cancellation, if some unresolved disputes remain between the consumers and the CSCs, including owing, counterfeiting documents, or borrowing an EV for others, they must deal with these issues first. They may have to pay to settle any troubles, and they may face unfair treatment. Throughout the entire SSEV service process, online or telephone customer service should always be available to answer questions and provide necessary assistance. If customer service is poor, consumers will be helpless and angry, and they will face greater indirect or even direct losses.

5. Risk Assessment of Service Failures

5.1. Construction of Evaluation Criteria and Risk Assessment

After identifying the risky service failures, an important task is to evaluate them. To distinguish differences better, nine levels of language variables are adopted, i.e., L = 8 (Table 2,

Table 5. Assessment criteria of occurrence.

Linguistic Variables	Description of Grading Criteria
EL	Rarely needed and rarely problematic; seldom or never problematic services.
VL	Rarely needed and occasionally problematic; occasionally needed and rarely problematic.
L	Rarely needed and often problematic; occasionally needed and occasionally problematic.
SL	Rarely needed and not provided/always problematic; occasionally needed and often problematic.
F	Occasionally needed and not provided/always problematic; widely needed and rarely problematic.
SH	Widely needed and occasionally problematic; always needed and rarely problematic.
H	Widely needed and often problematic; always needed and occasionally problematic.
VH	Widely needed and always problematic/not provided; always needed and often problematic.
EH	Always needed but not provided or always problematic.

,

Table 6. Assessment criteria of severity.

Linguistic Variables	Description of Grading Criteria
EL	Consumers can usually dispose with ease by themselves without the service.
VL	Consumers can dispose at a small cost or passively suffer a negligible loss.
L	Consumers can dispose at a medium cost or passively suffer a small loss.
SL	Consumers can dispose at a big cost or passively suffer a medium loss.
F	Consumers can dispose at a high cost; or passively suffer a noticeable loss.
SH	Consumers cannot dispose, passively suffer a high loss, or low accident possibility.
H	High accident probability, unpredictable loss, or endless troubles.
VH	Endangers the personal safety of consumers or may be widely transmitted.
EH	Endangers the safety of consumers and others and may cause great social impact.

, and

Table 7. Assessment criteria of detection.

Linguistic Variables	Description of Grading Criteria
EL	Service failures that consumers can intuitively detect and rarely ignore.
VL	Consumers can intuitively detect but occasionally ignore.
L	Consumers can intuitively detect but usually ignore.
SL	Consumers can intuitively detect but do not realize it until too late.
F	Consumers need to be very careful to detect and most ones will ignore it.
SH	Consumers need to be very careful to detect and rarely arouses suspicion.
H	Consumers cannot detect unless they have the expertise or plenty of patience and time.
VH	Consumers cannot detect and need to be certified by professional organizations.
EH	Professional organizations cannot easily identify or costs consumers more than the loss.

).

For occurrence, a high linguistic sequence indicates great probability of service failure (Table 5). In the whole SLC of SSEVs, the required frequencies of basic services and specific services are different. For example, reliable vehicles and good maintenance are always needed, while customer service and violation handling are occasionally or rarely needed. Moreover, the probability of service failures is also different, which is related to the service qualities provided by CSCs. The wide provision of a certain reliable service is closely related to the cost, the number of links that need to be controlled, and the management difficulties, which requires professional judgment by experts. Hence, the failure occurrence can be evaluated by the demand frequency and the service quality.

For severity, a high linguistic sequence represents a series of serious consequences (Table 6). Different service failures have different specific consequences, including trip delays, financial losses, psychological losses, and personal injuries. Some of these consequences only affect drivers and passengers, some affect pedestrians and vehicles around the SSEVs, and some affect potential customers. Thus, the severity can be assessed by analyzing what kind of loss a service failure can bring as well as its scope. Considering that consumers have a certain degree of subjective initiative, the assessment of severity also takes into account the extent to which consumers can handle it themselves.

For detection, the linguistic sequence is positively related to the difficulty of detection (Table 7). Parts of the services are carried by the EVs, while the remaining parts are carried by personnel or invisible processes; some services are always concerns of consumers, while the others are easily ignored. Due to the different service carriers and the different attention of consumers, consumers will have different recognition situations when choosing companies and vehicles, which lead to the difference in the difficulty of detecting the failure of different services.

On the basis of SLC analysis, service identification, and failure analysis, combined with professional knowledge and corresponding grading criteria, experts assessed the risk levels of different service failures. It is prudent to consider that there may be considerable differences between the assessments of different experts. Thus, experts’ assessments also need to be tested by consensus inspection. Through the consensus inspection, the subjective bias of individuals will be controlled within an acceptable range. For further calculation, the language assessment values need to be converted into TFNs according to Function (1), which are shown in Table 2.

5.2. Inspection and Construction of Assessment Matrix

After the conversion, according to Formulas (2)–(5), the consensus level of assessment values can be tested. Through the test, the consensus levels of FM1, FM6, and FM12 are lower than the requirement. Experts need to fine-tune their assessments through further discussion and analysis. For FM1, experts disagree about the severity of the consequences of information abuse. After further discussion, expert E³ raised the evaluation value from EL to L considering the complexity of the consequences, while expert E⁴ lowered the evaluation value from H to F considering the prevention ability of consumers. The consensus coefficient was changed from 0.8039 to 0.8681. For FM6, considering that the unclear determination of responsibility involves the vital interests of consumers, and they are very sensitive to this problem, expert E² adjusted the detection difficulty from L to VL. The consensus coefficient was increased from 0.8477 to 0.8535. For FM12, considering the possibility of multiple consequences without security alerts, expert E⁴ raised their assessment of severity from EL to L. Synchronously, the consensus coefficient was increased from 0.8324 to 0.8681. After consensus inspection, the assessment matrix meeting consensus requirements is obtained and shown in

Table 8. Expert assessment matrix for service failure that meets the consensus requirement.

Risk Factor	Occurrence				Severity				Detection				Consensus Inspection
Expert	E1	E2	E3	E4	E1	E2	E3	E4	E1	E2	E3	E4	Coefficient
FM1	F	SH	SH	F	L	VL	L	F	SL	SL	F	L	0.8681
FM2	L	VL	VL	L	F	SH	SL	F	SL	SL	L	F	0.8889
FM3	SL	F	F	SH	SH	H	VH	EH	H	F	SH	SH	0.8524
FM4	SL	F	F	SH	H	EH	H	SH	SH	H	H	SH	0.8738
FM5	VL	L	L	VL	SH	F	SH	H	F	SL	SL	F	0.9028
FM6	L	F	SL	F	H	F	F	H	EL	VL	L	EL	0.8535
FM7	F	SL	SL	F	VH	H	H	SH	L	SL	SL	SL	0.9097
FM8	SL	F	SL	F	F	SH	F	SL	L	VL	VL	L	0.9028
FM9	SL	VL	L	SL	H	SH	SH	H	SL	F	SL	SH	0.8750
FM10	SL	L	L	SL	SH	SH	F	H	SL	F	F	SH	0.8889
FM11	SH	VH	H	SH	F	SL	L	SL	EL	VL	VL	EL	0.8870
FM12	SL	L	SL	F	SL	SH	L	L	L	SL	L	L	0.8681
FM13	SH	H	SH	H	L	SL	L	L	L	EL	VL	VL	0.9141
FM14	H	SH	SH	F	H	F	SH	SH	L	L	SL	VL	0.8750
FM15	F	SL	L	SL	SH	H	SH	VH	L	SL	VL	VL	0.8611
FM16	F	SL	SH	SH	F	SL	F	SL	SL	VL	L	SL	0.8750

.

To integrate assessment values, the authority weight vector of experts is calculated by Function (6) and resulted in: $\gamma ={\left(0.1606,0.3577,0.2676,0.2141\right)}^T$. Combined with the authority weight vector, the risk factor decision matrix is integrated by using Function (7) and is shown in

Table 9. Integrated risk factor assessment matrix of service failures.

FMs	Occurrence		Severity		Detection
	Integrated Fuzzy Value	Ranks	Integrated Fuzzy Value	Ranks	Integrated Fuzzy Value	Ranks
FM1	(0.4532, 0.5782, 0.7032)	4	(0.1338, 0.2588, 0.3838)	16	(0.2567, 0.3817, 0.5067)	6
FM2	(0.0468, 0.1718, 0.2968)	16	(0.3863, 0.5113, 0.6363)	11	(0.2433, 0.3683, 0.4933)	7
FM3	(0.3817, 0.5067, 0.6317)	6	(0.6919, 0.8169, 0.9151)	1	(0.4754, 0.6004, 0.7254)	2
FM4	(0.3817, 0.5067, 0.6317)	6	(0.6877, 0.8127, 0.8930)	2	(0.5782, 0.7032, 0.8282)	1
FM5	(0.0782, 0.2032, 0.3282)	15	(0.4820, 0.6070, 0.7320)	7	(0.2968, 0.4218, 0.5468)	5
FM6	(0.3014, 0.4264, 0.5514)	9	(0.4687, 0.5937, 0.7187)	9	(0.0335, 0.1116, 0.2366)	14
FM7	(0.2968, 0.4218, 0.5468)	10	(0.6183, 0.7433, 0.8683)	3	(0.2299, 0.3549, 0.4799)	8
FM8	(0.3215, 0.4465, 0.5715)	8	(0.3930, 0.5180, 0.6430)	10	(0.0468, 0.1718, 0.2968)	13
FM9	(0.1271, 0.2521, 0.3771)	14	(0.5468, 0.6718, 0.7968)	5	(0.3482, 0.4732, 0.5982)	4
FM10	(0.1718, 0.2968, 0.4218)	13	(0.4933, 0.6183, 0.7433)	6	(0.3817, 0.5067, 0.6317)	3
FM11	(0.6229, 0.7479, 0.8729)	1	(0.2366, 0.3616, 0.4866)	14	(0.0000, 0.0782, 0.2032)	16
FM12	(0.2320, 0.3570, 0.4820)	12	(0.2792, 0.4042, 0.5292)	13	(0.1697, 0.2947, 0.4197)	9
FM13	(0.5715, 0.6965, 0.8215)	2	(0.1697, 0.2947, 0.4197)	15	(0.0201, 0.1004, 0.2254)	15
FM14	(0.4933, 0.6183, 0.7433)	3	(0.4754, 0.6004, 0.7254)	8	(0.1317, 0.2567, 0.3817)	10
FM15	(0.2366, 0.3616, 0.4866)	11	(0.5982, 0.7232, 0.8482)	4	(0.1095, 0.2345, 0.3595)	12
FM16	(0.3905, 0.5155, 0.6405)	5	(0.3035, 0.4285, 0.5535)	12	(0.1271, 0.2521, 0.3771)	11

. According to the sequence of occurrence, basic services involving complex internal management and financial costs are more prone to problems, while those requiring little investment and infrequently needed ones are relatively low. In order of severity, the service failures that have extensive impacts on driving safety and infringing on consumers’ vital interests rank at the top, while those that consumers can deal with by themselves are not so serious. According to the ranking of detection, service failures involving vehicle safety, and those that are unknown to consumers or can easily be overlooked top the list, while those involving important interests of consumers or requiring consumers’ participation are easy to identify.

5.3. Ranking of FMs through TOPSIS Entropy Method and PT-PROMETHEE-II Method

On the basis of the integrated assessment matrix, the weight vector of different risk factors is calculated by Functions (8)–(15), i.e., $w={\left(0.3099,0.2828,0.4073\right)}^T$.

On the basis of Functions (16)–(18), the PT values can be calculated. Then, with the weight vectors of different risk factors, the outflow and inflow values are calculated using Functions (19)–(22). The results are as follows:

${\pi }^{+}\left(F{M}_i\right)$ = (0.1374, 0.1679, 0.0197, 0.0165, 0.1294, 0.1563, 0.0670, 0.1497, 0.0999, 0.0894, 0.1935, 0.1560, 0.2042, 0.0819, 0.1163, 0.1296)

${\pi }^{-}\left(F{M}_i\right)$ = (0.2523, 0.1484, 0.5887, 0.6785, 0.2249, 0.1276, 0.3004, 0.1228, 0.3075, 0.3159, 0.2651, 0.1096, 0.2221, 0.2796, 0.2052, 0.1596)

Calculating with Function (23), the net flow values are obtained as follows:

$\pi \left(F{M}_i\right)$ = (0.1149, −0.0196, 0.5689, 0.6620, 0.0954, −0.0287, 0.2334, −0.0269, 0.2076, 0.2265, 0.0716, −0.0463, 0.0180, 0.1977, 0.0889, 0.0300)

On the basis of the net flow values, the risk priorities of FMs are ranked in descending order as follows:

FM4 > FM3 > FM7 > FM10 > FM9 > FM14 > FM1 > FM5 > FM15 > FM11 > FM16 > FM13 > FM2 > FM8 > FM6 > FM12.

As shown above, careless maintenance is confirmed as the riskiest service failure, followed by provide defective EVs, lack of security identification, inadequate vehicle insurance, inadequate safety equipment, trouble in refunding the deposit, and others. The FMs such as complexity in handling violation, agreement trap, unreasonable charges, unclear identification of liability, and imperfect security alerts are calculated as having relatively low risk. However, it should be noted that this does not mean that these service failures can be ignored.

For further discussion, the first six are defined as high risk FMs, the middle five as relatively low risk ones, and the last five as low risk ones. It is easy to distinguish that the risk levels of service failures, which when related to driving safety and accident treatments are high, since these risks are accidental, sudden, on-site, and relatively uncontrollable, with unpredictable losses. Service failures that cause minor or indirect property losses are less risky because they are less sudden and easier to limit, with predictable losses as well as being more widely watched and monitored because of the broadly involved consumer interests.

5.4. Sensitivity Analysis

Risk aversion is a natural human psychological behavior. A higher risk aversion coefficient $\lambda$ means that the more conservative consumers are, the less willing they are to take risks. According to the cumulative prospect theory, the value range of the risk avoidance coefficient is $\lambda =\left[1,5\right]$. For set $\lambda =\left\{1,1.5,2,2.5,3,3.5,4,4.5,5\right\}$, the results of the sensitivity analysis are shown in Figure 4.

As the sensitivity analysis results shown, the results change with a different $\lambda$. Combined with the calculation characteristics of the PT-PROMETHEE-II method, the outflow volume is gradually amplified with an increase in $\lambda$. Due to the differences in the inflow and outflow of different failure modes, the variation efficiency of $\pi \left(F{M}_i\right)$ is not consistent, which results in the changes of risk ranking in a convergent way. Combined with the PT, it can be seen that there is a significant difference between rational and irrational consumers. This indicates that risk aversion has an impact on the ranking results, and if consumers are assumed as rational, the risk assessment results will be skewed from reality.

5.5. Comparison Analysis

Herein, the service failures are ranked using the TOPSIS method [73], VIKOR method [74], and CODAS method [75] for comparative purpose, which are proven efficient in fuzzy FMEA and, generally, used in different areas. To ensure the feasibility of comparison, different methods use the same integrated matrix listed in Table 9 and the same weight vector of risk factors as in the previous study. According to the calculation rules of each method, the ranking results of four methods are calculated and shown in Figure 5.

It can be seen from Figure 5 that the results of different methods have good correlation and a few nuances. The consistency proves the effectiveness of the PT-PROMETHEE-II method and the robustness of the result. The nuances in results are related to the differences in ranking principles. The current method performs characterization by considering the irrational behaviors of consumers and conducts a comprehensive comparison. Considering that the current topic is directly connected to consumers’ risk aversion behaviors, the current method is the most appropriate one.

6. Service Improvement Analysis

The usage and management of SSEVs are naturally separated, and CSCs are obliged to guarantee service quality. Providing smooth and reliable services contributes to good risk control and helps CSCs gain consumer trust and good reputation. Based on the Kano model, equal investment in different services can lead to different results. In order to further discuss how to effectively improve the quality of services, this study investigates the characteristics of consumer satisfaction based on the Kano model.

6.1. Kano Investigation

In the Kano model, the quantity of requirement refers to the number aiming at different types of requirements of a certain service (

Table 10. Kano questionnaire evaluation form.

Corresponding Demand Type		Backward Questions
		I like It	It Should Be	It Doesn’t Matter	I Can Stand It	I Don’t Like It
Forward questions	I like it	Q	A	A	A	O
	It should be	R	I	I	I	M
	It doesn’t matter	R	I	I	I	M
	I can stand it	R	I	I	I	M
	I do not like it	R	R	R	R	Q

). In the questionnaire survey, two opposite questions are previously designed for a certain question based on the Kano model, such as ‘what do you think if a service never has problems’ and ‘what do you think if a service always has problems or it is not provided’. Then, the requirement type of a specific service will be identified through consumers’ attitudes toward the two opposite questions (Table 10).

The requirement types of 16 services are differentiated through an online survey of nearly 400 adult respondents who are willing to use SSEVs or have already experienced them in China. Finally, 340 valid questionnaires are obtained by deleting invalid ones. The reliability and validity of the questionnaire are tested by dividing the questions into forward and reverse ones. The Cronbach’s Alpha of the forward and backward questions is 0.849 and 0.910, respectively; both are greater than 0.8, indicating that the reliability of the questionnaire has met the requirements. The validity is divided into content validity and structure validity. The content validity is guaranteed because these services are selected based on careful SLC analysis and the Kano questionnaire is mature. The structure validity also meets the requirements, since the KMO test shows that the KMO values of forward and backward questions are 0.897 and 0.922, respectively, which are both greater than 0.8, while the corresponding Bartlett test values are both 0.000 (i.e., Sig < 0.001). Thus, the questionnaire survey results meet the needs of further analysis.

According to the statistics, the distribution of 16 service requirement types is shown in

Table 11. Different types of requirements for service security based on the Kano model.

Kano	A	M	O	I	Q	R	Type
FM1	5	173	103	25	31	3	M
FM2	6	159	108	22	42	3	M
FM3	3	156	143	14	20	4	M
FM4	11	121	188	16	4	0	O
FM5	11	94	207	16	9	3	O
FM6	7	134	162	16	16	5	O
FM7	68	66	139	41	22	4	O
FM8	6	157	148	19	10	0	M
FM9	25	110	164	24	12	5	O
FM10	28	92	182	29	6	3	O
FM11	30	76	207	19	7	1	O
FM12	44	66	184	28	15	3	O
FM13	26	70	204	32	4	4	O
FM14	12	98	207	11	9	3	O
FM15	7	113	192	23	5	0	O
FM16	15	103	200	14	3	5	O

. The type of requirement for a service is determined by the dominant one in the survey. As shown in Table 11, FM1, FM2, FM3, and FM8 are M-type requirements, while the rests are O-type ones. The results show that consumers attach great importance to service reliability, and their requirements for service quality are either essential or as reliable as possible.

6.2. Risk-Satisfaction Analysis

Combined with the Better–Worse index, the satisfaction and non-satisfaction distributions of different services are calculated by Functions (24) and (25). As shown in Figure 6 (red labels: high risk; blue labels: relatively low risk; green labels: low risk), Worse indexes are mainly distributed in $Worse\in \left[-0.95,-0.65\right]$, and Better indexes are distributed in $Better\in \left[0.34,0.72\right]$. It can be seen from the distribution that risk issues are more likely to cause consumers’ non-satisfaction. This reveals that all kinds of service failures bring great non-satisfaction to consumers, and all service qualities need to meet the most basic requirements. Moreover, as illustrated in Figure 6, different service failures bring various changes in satisfaction. Thus, in pursuit of better satisfaction for consumers, further analyzes are needed.

Services in the first quadrant means that if these services are well done, satisfaction will be rapidly improved. Meanwhile, services in the second quadrant are characterized by the fact that excessive investments will not effectively increase satisfaction. According to the risk levels and satisfaction indexes, services are divided into four categories from the perspective of Risk-Satisfaction analysis: high risk with high satisfaction (HR-HS), high risk with low satisfaction (HR-LS), low risk with high satisfaction (LR-HS), and low risk with low satisfaction (LR-LS) (Figure 7).

First, as mentioned above, according to the distribution of the Better–Worse index, all services should meet basic quality requirements. Second, services in different categories of the Risk-Satisfaction diagram should be distinguished. Since CSCs have limited resources, it is valuable to investigate which services are worth more investments. As higher satisfaction means greater competitiveness, and higher risk means greater potential losses, it is easy to point out that services listed in the HR-HS category should be considered in the first level. Such services include EV maintenance (FM4), security identification (FM7), car insurance (FM10), safety equipment (FM9), and refunding the deposit (FM14). In contrast, services in LR-LS are not recommended to invest in too much, other than the basic guarantee investment. In the HR-LS and LR-HS categories, investment preference depends on the CSCs’ strategic planning. Satisfaction-oriented strategies lead services in LR-HS to surpass HR-LS ones, while safety-oriented strategies do the opposite.

7. Conclusions and Summary

7.1. Conclusions and Discussion

For the pursuit of SSEV promotion and sustainable development, this study proposed a framework to assess risks of service failures and improve service quality of SSEVs. Sixteen important services and their corresponding FMs were identified by SLC analysis. The risk levels of these service failures were determined and ranked by a newly integrated fuzzy FMEA approach. The result showed that service failures related to driving safety, accident handling, and large property losses have a high-risk level. Including providing defective EVs, lack of security identification, inadequate vehicle insurance, inadequate safety equipment, trouble in refunding the deposit, etc. On the basis of the risk assessment, a consumer satisfaction survey was conducted. The analysis of the survey results illustrated that all service failures brought significant non-satisfaction, and there were significant differences in the service improvements that lead to satisfaction. Therefore, all services need to be ensured in basic reliability, and services with high risk and high satisfaction characteristics are worth being invested in more based on the Risk-Satisfaction analysis.

The service recognition and FM recognition were seriously conducted based on forum data and SLC analysis, thus, the comprehensiveness and authenticity of the object identification were both guaranteed. In terms of risk assessment, the process was comprehensively designed. In terms of risk value assignment, it was carefully conducted based on seriously designed assessment criteria and, then, was tested by consensus inspection. Therefore, the result of risk assessment was comprehensive, rigorous, and relatively realistic. Since the risk assessment was conducted based on the characteristics of the service, the result of the risk assessment is suitable for all enterprises who carry out SSEV services. The providers of SSEV services are CSCs, and the carriers of services are tangible EVs, software, and employees, as well as intangible processes. The objects of services are unspecified consumers within a specific range, and the service environments are highly uncertain. For CSCs, it is essential to optimize service quality by providing reliable carriers for unspecific consumers in uncertain environments. In addition to the factors of competition and supervision, it is also the requirement of fulfilling social responsibility. The resources of CSCs are limited, SSEV services are consumer-oriented, and the relationship between service reliability improvement and consumer satisfaction is non-linear. Thus, in order to improve service quality more effectively, consumer-oriented analyzes are required.

Compared with other security studies, the biggest characteristic of this study is that the consumer requirements are taken into account in service optimization. Are the selected services important? The answer, based on the Kano survey, is yes. Consumers are very concerned about the reliability of services, and the failure of any one of the 16 services will cause negative feelings for consumers. The analysis of satisfaction and non-satisfaction proved this point and explained a deeper phenomenon: in terms of how to make consumers more satisfied with service reliability, the improvement effects of different services are different. This is another important contribution of the Kano investigation, which can help CSCs achieve better results with the same investments or the same effects with relatively fewer investments. Since this survey was conducted in the context of China, CSCs who are investing in China can refer to the results of this paper. For CSCs in other countries or regions, a local circumstances-based Kano survey is earnestly recommended.

7.2. Summary and Prospects

To strategically optimize SSEV services, a framework based on risk assessment and the Kano model was proposed, which includes SLC analysis, risk assessment, and satisfaction analysis. SLC analysis contributed to the service identification, risk assessment contributed to the risk ranking of service failures, and the Kano survey contributed to the deep optimization of service satisfaction. High-risk service FMs were selected, and the results have wide applicability. The newly integrated fuzzy FMEA approach has a more comprehensive program and rigorous analysis, plus it objectively calculated the weight and took the risk avoidance factor into consideration when ranking, which has a methodological contribution. The Kano survey showed that consumers are equally sensitive to all service failures, but there are significant differences in satisfaction with service improvements. Therefore, the quality of all services should be guaranteed at the most basic level, and the further improvement of services in the HR-HS category should be considered first. Combined with the results of risk assessment and satisfaction improvement analysis, this work contributes to the risk control and service optimization of SSEV services, so as to ensure the promotion of EV sharing and, then, the achievement of carbon emission reduction and sustainable development.

The consequences of destroying nature are unimaginable, while the results of protecting nature are beneficial to the long-term development of humanity. This research aims to help CSCs optimize service reliability and indirectly contributes to sustainable development, though more relative studies are needed. The research prospects include two aspects: methods and contents. In terms of methods, researchers can carry out innovations based on different types of data and environmental characteristics, optimize the calculation process mathematically, and improve the process from the aspects of objectivity and fuzziness. In terms of contents, studies on service optimization of other means of shared travels and the SSEV service optimization of different countries and regions are recommended.