A Complete MCDM Model for NPD Performance Assessment in an LED-Based Lighting Plant Factory

1Department of Industrial Management, Chung Hua University, No. 707, Sec. 2, Wufu Rd., Hsinchu 30012, Taiwan 2Ph.D. Program of Technology Management, Chung Hua University, No. 707, Sec. 2, Wufu Rd., Hsinchu 30012, Taiwan 3Department of Marketing and Logistics Management, Ling Tung University, No. 1, Ling Tung Rd., Taichung 40852, Taiwan 4General Education Center, Hsing Wu University, No. 101, Sec. 1, Fenliao Rd., Linkou District, New Taipei City 24452, Taiwan


Introduction
Global resource demands are constantly on the rise as a result of global population growth and the development and expansion of economies worldwide.People now face the danger of depleting fossil fuel reserves.Since the industrial revolution, industrial and economic prosperity came at the cost of severe environmental problems like global warming, ozone depletion, air and water pollution, and more.The effects of the weather warming include the melting of the ice sheet in the Arctic and climate change.In addition to dwindling resources and global warming, reduced food security resulting from overuse of land and chemicals is a very serious problem.A 2007 Intergovernmental Panel on Climate Change (IPCC) climate report predicted that the global surface temperature is likely to rise a further 1.1 to 6.4 ∘ C between 2007 and 2100.It is also predicted that 15.0% to 40.0% of the Earth's species will become extinct if the average temperature of Earth's atmosphere rises by about 2 ∘ C. If Earth's mean surface temperature increases by 4 ∘ C, around 3 billion people will lose access to safe drinking water.The Kyoto Protocol or Kyoto Treaty hoped to coordinate a response to global warming, including mitigation by emission reduction, in an international convention attended by more than 50 countries in Kyoto, Japan, on December 11, 1997.The treaty was validated in 2002 and the resolution was passed on February 2005, with countries committing to endeavor to reduce emissions of CO 2 , CH4, N2O, HFC, PFC, and SF6 to 5.2% of the amount in 1990.The Copenhagen Accord was created by the COP 15 on December 7 and 8, 2009, to further strengthen the commitment of countries party to the Kyoto Agreement's goals.Afterwards, IPCC's Working Group II (WGII) did the literature survey on the wide range of impacts and risks of climate change and identified impediments and opportunities and warned that failure to make "substantial and sustained" reductions in greenhouse gas (GHG) emissions will increase the likelihood of severe and irreversible impacts [1].In February 2015, they further assessed submissions related to the frequency and scheduling of reports in light of structure and operations [2].However, three actions are to be suggested for the IPCC's decisionmakers: (1) incorporating more practitioners to enhance the awareness and understanding in terms of climate change; (2) allowing a practitioner-led IPCC Special Report representing good-practice responses to climate change; (3) reporting good practices on timely climate response strategies [3].Above all, understanding anthropogenic factors to assist decision-makers in implementing policies and actions on a weather event or climatic process and abiding by the Paris Agreement, enhanced understanding, action, and support, are needed in several areas related to addressing loss and damage, including comprehensive risk assessment and building the resilience of communities, livelihoods, and ecosystems, which can reduce the likelihood of extreme events happening [4].
In recent years, the environmental problems described above have prompted countries to develop green-related products and enterprises and to adopt strategies of renewable energy for new product development.Climate change and food security are among those threats considered to have the most significant impact on people, especially in the Afro-Asian regions, and so any new product able to overcome or alleviate these problems will be distinguished in the mainstream market.Since new product development in the 21st century needs to meet emission reduction and energy saving requirements, light-emitting diodes (LEDs), a semiconductor light source, are a typical choice and are being increasingly used for most forms of artificial lighting.For example, LEDs are used in traffic lights at intersections, outdoor billboards, landscape design, mobile phones, laptops, vehicles, televisions, and vegetative growth controlled by considerable parameters variability under different LED spectrum supplementation [5].In other words, it seems that LEDs have become the new light sources for modern ecoenergy.The Fukushima Daiichi nuclear disaster served as a dire warning that high pollution and nonrenewable energy pose a lasting threat to the environment and must be replaced.Even though Taiwan has been deficient in many kinds of energy resources and highly dependent on imported energy, it has mature optic engineering and agriculture technology.Efforts to integrate these advantages that meet green-related product criteria have thus received significant attention recently.
Enterprises have to seek beneficial investments such as international cooperation, new product development, and equipment maintenance in order to survive in this highly competitive global economy.New product development (NPD) is deemed the most difficult of such investments.In general, the cost of new product development, the number of development team members, and the duration of a project are in direct proportion.In addition, enterprises need to pay for equipment and tools for manufacturing.It is not easy for enterprises to develop an unflawed product, and it actually takes a great deal of time and money to successfully develop a new product.New product development challenges include decision-making, regulation change, manufacturing details, time stress, market economy, producing schedule, customer needs, member diversity, and team work.As a result, performance assessment for new product development is very important in any industry.The proposed LED-based lighting plant factory can be composed of a light source system, an environmental control system, and a nutrient liquid feeding system, as shown in Figure 1.The light source system includes variations for the light quality, amount, and photoperiod.The environment control system controls the air temperature, humidity, and carbon dioxide concentration.The nutrient liquid feeding system controls fertilizer concentration, pH levels, and the quantity of dissolved oxygen.
Team members working on the new product development usually work in a specific project group.Bad decisions by the group will result in significant financial losses for the enterprise.Thus, the financial risks involved in decisionmaking must be reduced as far as possible, and this can be achieved with an assessment system.LEDs are likely to be in the spotlight in the 21st century since illumination products are becoming popular, lighting technologies are improving, and costs are decreasing.LEDs offer many advantages over other forms of lighting, including their flexible emitted light color design and low energy consumption.Of significant value are their potential use in agriculture and versatility and robustness in terms of weather and geography.This study conducts a performance assessment of new product development in the LED-based lighting plant factory whose contributions are described below: The remainder of this study is arranged as follows.Section 2 examines the literature on the subjects and methodologies adopted in this study.Section 3 establishes a hybrid MCDM performance evaluation model for NPD in an LEDbased lighting plant factory.A case study of an LED-based plant lighting factory in Taiwan is carried out to verify the practicality of the proposed model in the next section.The final section sums up this research and offers conclusions.

Literature Review
As noted above, in order to eliminate as much financial risk as possible in new product development, this study aims to develop an integrated MCDM evaluation model for new product development in LED-based lighting plant factories.This section will discuss previous studies in new product development (NPD), plant factory and related lightemitting diode (LED) lighting system, fuzzy Delphi method (FDM), fuzzy decision-making trial and evaluation laboratory (FDEMATEL), fuzzy analytic network process (FANP), and composite priority vector (CPV).
New product development (NPD) is a specific process which has a lifecycle with a high level of diversity from creativity in the concept stage to logistics actions in the commercialization stage.NPD is not a single act of invention but often described interchangeably with innovation or creativity, and NPD has a distinct purpose, characteristics, and applications.In addition, NPD encompasses the entire process with numerous critical success factors bringing a new product or service to the marketplace [6] where this process identifies consumers' wants and needs and transforms them into a commercialized product.In other words, new product development (NPD) is considered the most powerful weapon against competitors and also one of the most important MCDM issues for evaluating the sustainable development of enterprises [7].NPD is one of the core operational management processes in enterprises and requires the capabilities of creation, innovation, enforcement, technology, teamwork, and integration.Each step during the development procedure has a significant impact on NPD, including customers' needs, competitors' intimidation, technical risk, cross-department cooperation, and resource allocation.Some environmental factors of NPD process can be noted in four aspects: firstly, the effect of the market context on business performance generally has been neglected; secondly, most of the NPD projects only focused on "main effects" relationships without concerning the their environments, especially the external environment; thirdly, further research on the interaction between product design and market uncertainty is required at last, numerous unanswered questions with respect to the effects of environmental uncertainty on NPD still exist [8].As new product development (NPD) is also a high-risk and costly process with significant failure rate, NPD success factors normally can be classified into firm internal environment, organizational capability, NPD process, level of new product's competitive advantage, and market environment; and among company-related factors, management commitment to NPD projects and managerial capabilities can be deemed two crucial factors [9].Successful NPD is important in the survival and competitive advantage of enterprises [10]; it not only means sustainable development for a business, but also helps as a buffer in times of recession.NPD involves two highly related criteria, technology feasibility and market profitability, whereas effectiveness and efficiency are the main criteria for NPD performance.Enterprises may employ different methods to assess performance in light of their features, strategies, and designs.These methods include performance measurement balance matrix, performance measurement questionnaires, the Cambridge performance measurement design process, performance criteria systems, or performance measurement integration models.This study integrates FDM, FDEMATEL, and FANP with composite priority vector (CPV) to create a performance assessment model for NPD utilized in a Taiwanese LED lighting plant factory.
Plant factories are so-called artificially controlled environment systems which control the environmental factors such as lighting, temperature, humidity, water, and the concentration of carbon dioxide and can stably produce high-quality crops in an indoor space.In Asia, plant factory systems are able to cultivate high-profit fruits, vegetables, seedlings, and herbs with outstanding merits of high output, being pollution-free, safety, no pesticides, chemicals, and ecofriendliness [11,12].Normally, the configuration of plant factories is a closed plant production system which consists of six principal structural elements: thermally well-insulated and nearly airtight warehouse, multitier system equipped with lighting devices, air conditioners, and fans, a CO 2 delivery unit, a nutrient solution delivery unit, and an environmental control unit including EC (electric conductivity) and pH controllers for the nutrient solution [13].Previously, the conventional light sources such as fluorescent lamp, metal halide lamp, and high pressured sodium (HPS) lamp were generally used to promote plant growth, in which some unnecessary light that mismatches with photosynthesis action spectrum (PAS) is included.To enhance the photosynthetic efficiency of the plants, it is vital to generate artificial lighting sources with coincided spectra with PAS.It is worth mentioning that the light qualities including peak position and intensity are able to be adjusted using wavelength tunable LEDs.
LEDs, typical energy saving, high efficiency, long-lifetime, and environmental-friendly illuminating sources, have come to play a significant role in plant lighting.LED lighting systems have many advantages over lamps currently used in plant growth.LEDs can control the spectral output of a lighting system for specific crops and even be modified over the course in view of a photoperiod or growth cycle.Special lighting modes can also be used to solve plant disease or injury problems [14].Additionally, LED lighting systems can be configured to produce very high light levels, have a very long operating life, and can be turned on and turned off instantly without warm-up time.Moreover, LEDs have the potential for significant cost savings, do not contain mercury, which needs to be safely disposed of, do not have glass envelopes or high surface temperatures that can cause injury, and do not produce damaging ultraviolet wavelengths.As LEDs replace existing lamp technologies in lighting applications, significant cost decreases will be driven by economies of scale.These advantages mean that LEDs meet the standards of environmentally friendly energy.LEDs present a different technology from lamps currently used in plant lighting.LED technology is considered one of the most valuable advances in plant lighting because of the capabilities mentioned above.Many previous studies have shown the development and benefits of LED lighting systems for plant production.In the late 1980s and early 1990s, testing of LEDs for plant growth was conducted with lettuce, potato, spinach, and wheat.LEDs began to be explored for tissue culture systems in Japan [15].LEDs were discovered to be particularly effective in germinating seeds and rooting cuttings in the Netherlands [16] and later in other researches [17,18].These early developments quickly led to the development of LEDbased systems for plant physiology experimentation, since blue LED technology did not offer sufficient levels of blue irradiance.Researchers worked with NASA at the Kennedy Space Center (KSC) on plant-based regenerative life-support systems for future Moon and Mars bases and investigated the effect of LED-based light systems on several crop plants such as wheat, radish, spinach, lettuce, and peppers [19][20][21].
The Delphi method was developed in 1960 to combat some shortcomings of traditional prediction methods such as theoretical or quantitative models or trend extrapolation.However, ambiguity and uncertainty problems persisted in survey questions and responses [22].In order to address these problems, fuzzy set theory was first incorporated with the Delphi method in 1985.The Delphi method was originally proposed by Dalkey and Helmer in 1963 and has been used in a wide range of research applications.The Delphi method systematically shows that group judgments are more valid than individual ones.In general, the standard procedure requires experts in the relevant field to answer questionnaires in two or more rounds.The researcher then provides an anonymous comment on the experts' forecasts from the previous round after each round of the survey.The experts are encouraged to revise their earlier answers according to the responses of other members of their panel.It is believed that the group will converge towards the correct answer, since the range of the answers will decrease during this process.Eventually, the process is repeated several times until a consensus emerges and the mean scores of the final round determine the result.Although the conventional Delphi method offers much scaffolding, the method still includes ambiguity and uncertainty problems in survey questions and responses [22][23][24].The incorporation of fuzzy set theory with the traditional Delphi method is one of the approaches to solving these problems [25][26][27].Murray et al. [28] first applied fuzzy theory to the traditional Delphi method in 1985.Ishikawa et al. [29] utilized the cumulative frequency distribution function and fuzzy integration to convert the expert judgments into fuzzy numbers and employed the "gray zone," the overlap section of the triangular fuzzy numbers, to develop the maximum membership degree and the FDM.FDM has been applied in many different fields of study.Chang et al. [24] established the key successful factors (KSFs) of knowledge management for university students using e-portfolio by using FDM and FAHP.Mohammad [30] developed educational system strategies in university student applications by using a hybrid fuzzy Delphi AHP.The research identifies elements which affect the admission and application of students in an Iranian University.
The decision-making trial and evaluation laboratory (DEMATEL) was proposed by Fontela and Gabus [31] for the application of solving complex problems.DEMATEL aims to yield casual dimensions and intensity of impact by using matrix computation through direct comparison of the interrelation between criteria.These structure and casualty matrices are used to express the relationship between properties in order to find the core issues of an evaluation system [32].These methods thus benefit the decision-maker in terms of execution.DEMATEL is frequently applied to multiple criteria decision-making (MCDM) to understand the core problems and construct evaluation performance models using the mutual impact between factors and the cause-and-effect diagram drawn based on their significance.J.-K.Chen and I.-S.Chen [33] discussed the innovative performance of Taiwan's advanced education institutes in academic research and used DEMATEL, FANP, and TOPSIS to determine the relative weight of each measurement criterion and were thereby able to evaluate how to form the ideal solution and technology that will support the system through development innovation.Büyüközkan and C ¸ifc ¸i [34] noted that enterprises responded to increased public awareness following environmental impacts, while at the same time environmental (green) criteria and strategies became more important.The environmental performance of enterprises is not only related to the internal efforts of enterprises, but also subject to impact from the environmental performance and image of suppliers.Nonetheless, the choice of suppliers is a complex and multiple criteria decision-making problem.Therefore, green supply chain management and the capacity for green supply china management should be employed as a basis to combine MCDM with DEMATEL in order to find the core factors to building a green supplier evaluation model for automobile companies using ANP and TOPSIS.
The Analytic Hierarchy Process (AHP), proposed by Saaty in 1971, assumes that criteria (effecting factors) in the same level are mutually independent.However, because this assumption does not reflect reality, Saaty and Takizawa [35] indicated dependence and independence from linear hierarchies to nonlinear networks using the composite vector of priorities (or composite priority vector, CPV), which illustrated how to generate priorities for decisions in view of the dependence of criteria on criteria, criteria on alternatives, and alternatives on alternatives, based on the hierarchical feedback system framework.They distinguish two types of dependence: either functional, semantic, or qualitative dependence or structural or quantitative dependence.Moreover, they represented two kinds of dependence (outer dependence: dependence between components; inner dependence: interdependence within a component combined with feedback between components) in a nonlinear network diagram.Saaty [36] proposed the analytic network process (ANP) to extend AHP by gathering dependency and feedback with AHP.The purpose of the ANP method is to improve the traditional AHP structure in which the criteria in the same level are not interactive and dependent.In the real world, the factors influencing decision-making are no longer limited by a linear top-down structure.In other words, factors interact more like a network than a linear relationship, as one hierarchy may dominate or be dominated by other hierarchies, namely, the feedback effects.ANP tolerates criteria with feedback in the same cluster (inner dependence) and allows criteria feedback in different clusters (outer dependence).Researchers can identify the relationship among criteria in a cluster and the relationship among the clusters and then infer the priority of the alternatives.
Many previous studies have used FANP to solve complex problems in decision-making models, choosing the best alternative or strategy by fuzzy weights.Mikhailov and Singh [37] published an extended ANP approach that includes personal preference, transfers fuzzy evaluation values to range values by -cut method, and obtains the weights in light of the range value to design a decision-making system.Promentilla et al. [38] applied the ANP technique to analyze polluted yards and assess improvement strategies.The procedure of their methodology is to apply the -cut method, interval arithmetic, and fuzzy optimism index to the definite matrix and then find the priority weights by calculating the eigenvalues.Each value in the pairwise comparison matrix represents the subjective opinions of a decision-maker.Associating ANP with fuzzy theory can demonstrate the fuzzy consensus from the group evaluators' point of view on the importance between any two criteria.Büyüközkan et al. [39] introduced a fuzzy number to the supermatrix using linguistic assessment and a fuzzy algorithm to solve the fuzzy problems arising from the criteria selection and judgment process.Mohanty et al. [40] employed the FANP method to analyze the risks and uncertainty for investments, calculating the weights based on scope analysis and selecting the best R&D project by fuzzy cost analysis.An FANP method provided enriched insights into strategic management in the Turkish airline industry [41].Chen and Chang [42] employed ISM (Interpretative Structural Modeling) to obtain the dimension-dimension and criterion-criterion dependence relationship and used the fuzzy analytic network process (Fuzzy ANP) to determine the top priority weight for assessment improvement in new product development solutions.Song et al. [43] depended on an integrated AHP and ISM method to make an exploration of the vulnerability factors.

Proposed Methodologies and Procedures
This study employs the fuzzy Delphi method (FDM), fuzzy decision-making trial and evaluation laboratory (FDEMA-TEL), fuzzy analytic network process (FANP), and composite priority vector (CPV) to make better decisions in new product investments in LED plant lighting industries.The procedure consists of two stages.In the first stage, major perspectives and all performance criteria affecting new product design will be collected from previous studies and analyzed.Experts from different LED plant lighting industry areas are then selected and invited to take part in the study, according to problems identified in the analysis.The experts are also asked to complete a questionnaire.After determining the indicators for the hierarchical structure by literature review and expert interviews, an FDM is applied to screening those perspectives and criteria for the performance assessment model.In the second stage, FDEMATEL is used to obtain the relationships among the perspectives (outer dependence) and among the performance criteria (inner dependence).A supermatrix is created by using an FANP, and the weights of perspectives and all criteria are computed.A determination of the ranking weights shows the importance of investments in a new product project.The complete MCDM research flowchart is presented in Figure 2.
Step 1 (explore and define research questions).The first thing to be done is to collect data pertaining to the LED plant lighting industry through literature reviews in order to identify the problems encountered during new product development.Other LED applications including product development, manufacturing, marketing, and recycling will also be discussed.
Step 2 (establish an expert committee).Since the field of this study is the LED plant lighting industry, the challenges in designing renewable energy-related new products will be addressed through related data collection, expert interviews, and questionnaires.
Step 3 (analyze the responses of the surveys).A fuzzy Delphi method (FDM) is used to obtain a reasoned consensus and consolidate individual judgments systematically.First, the questionnaire is completed by the selected experts for their suggestions and then reframed by those experts until a consensus is reached.This study uses the FDM proposed by Ishikawa et al. [29] and Lee et al. [44], and the process of the FDM is briefly explained as follows.
Step 4 (collect the possible evaluation criteria). = {  |  = 1, 2, 3, . . ., }, where   is criterion .Once the initial perspectives and criteria are established through literature review, this study obtains critical criteria for new product development.A 0-10 point evaluation scale is designed in the questionnaires and sent to the experts.Each expert is asked to give a value   = {(   ,   value    are then processed, resulting in a triangular fuzzy number. As indicated above, this study creates two triangular fuzzy numbers,   = (   ,    ,    ) and   = (   ,    ,    ), for each criterion   with triangular values of the remaining most conservative value    and the qualified most optimistic value    .If they overlap, the area of the overlap is the "gray zone" [22,45] shown in Figure 3.
According to the gray zone of each criterion   , operate the importance of consensus value   , and examine the expert opinions to see if they arrive at a valid consensus.The consensus value   expresses how important a criterion   is.The higher the   value, the more important the criterion   .Once    and    are obtained from Step 4, the consensus value   can be met as the following formula.
(1) If the two triangles do not overlap (   ≤    ), the criterion   achieves consensus among the experts.The consensus importance value   will be the average of    and    .
(2) If the two triangles overlap (   >    ), there is a gray zone   in the criterion   and   =    −    .The value ranging between the geometric mean of the most optimistic value and the geometric mean of the most conservative value can be shown as   =    −    .The relationship between   =    −    and   =    −    , meanwhile, ought to be identified as follows: (a) If   <   , it seems that there is no consensus among the experts, but the extreme opinions attributed by the experts are considered not far from other opinions given by the rest.For this reason, let the consensus on the importance   of the criterion   be the fuzzy relationship in order to find the fuzzy sets, and then obtain the maximum membership degree [44].  (  ) represented below is the membership function of the fuzzy triangular numbers   and   : (b) If   >   , it is certain that there is no consensus among the experts, and the extreme opinions attributed by the experts are considered far from other opinions given by the rest.Thus, the criterion   which is not converged will provide the experts with the ranging value   for the next questionnaire.All criteria are expected to reach a convergence until the consensus value   is generated.
As   above indicates, a threshold  is set to estimate whether a criterion   is qualified or not.That is, compare a consensus value   with ; a criterion   will be selected if   > ; otherwise, it will be eliminated [46].In general, a threshold  is suggested from 6.0 to 7.0, according to literature review; however, this is subject to change in order to make a better choice [46,47].
To sum up, the FDM procedure described above can be organized into three main phases as follows.

Phase 2. Identify the gray zone (yes or no).
Phase 3. Set the threshold.
Step 5. Identify the relationships among the perspectives and among the criteria by utilizing a fuzzy decision-making trial and evaluation laboratory (FDEMATEL) method, and build a network relationship map.
Step 6. Derive the weights of the perspectives and criteria in new product design by using a fuzzy analytic network process (FANP).This study employs a 9-point scale of relative importance, proposed by Saaty, to design the FANP questionnaire.Experts are then asked to represent decisionmakers in LED plant lighting industries for the purposes of the survey.The FANP is processed as follows.
(1) Form the network structure in which the goal, the perspectives and criteria are well defined, and the exterior relationship among the perspectives and the interior relationship among criteria are determined.
(2) Form pairwise comparison matrices with the 1 to 9 scores received from the expert responses in the questionnaires.Analyze consistency.The priority of the elements can be compared by the computation of eigenvalues and eigenvectors: where  is the eigenvector, the weight vector, of matrix , and  max is the largest eigenvalue of matrix .
(3) Check the consistency of pairwise matrices with consistency index (CI) and consistency ratio (CR).The consistency property of the matrix is then checked to ensure the consistency of judgments in the pairwise comparison.
As suggested by Saaty [48], the upper threshold CR values are 0.05 for a 3 × 3 matrix, 0.08 for a 4 × 4 matrix, and 0.10 for larger matrices.If the consistency test is not passed, the original values in the pairwise comparison matrix must be revised by the decision-maker.
(4) Construct fuzzy positive matrices.The pairwise comparison scores are transformed into linguistic variables, which are represented by positive triangular fuzzy numbers.According to Buckley [49], the fuzzy positive reciprocal matrix can be defined as where R is a positive reciprocal matrix;  is the number of decision-makers; r is relative importance between elements  and ; In order to minimize the fuzziness of the weight, two constants,    and    , are chosen as follows: The upper and lower bounds of the weight are defined as The upper and lower bound weight matrices are By combining  *   ,    , and  *   , the fuzzy weight matrix for decision-maker  can be obtained and is defined as W  = ( *   ,    ,  *   ),  = 1, 2, . . ., . (6) Integrate the opinions of decision-makers.Geometric average is applied to combine the fuzzy weights of decisionmaker opinions: where W is the combined fuzzy weight of decision element  of  decision-makers, W  is the fuzzy weight of decision element  of decision-maker , and  is the number of decision-makers.
(7) Defuzzify the synthetic triangular fuzzy numbers into crisp numbers by centroid method.
Form pairwise comparison matrices using the defuzzification values, and combine each submatrix with priority vectors to be an initial supermatrix.As they may not fit the column stochastic rule, normalize each column matrix to make a weighted supermatrix.Calculate a limited supermatrix by taking the weighted supermatrix to 2 + 1 powers so that the supermatrix converges into a stable supermatrix.
Step 7. The weight values acquired from FANP with CPV are ranked, and the best alternative is selected.Moreover, the significant elements (factors) that LED-based lighting plant factories try to use in developing a new product will be indicated and will provide decision-makers with important guidance.
Step 8. Conclusions and recommendations will be summarized.

Case Study
This study aims to construct an investment decision-making model using the fuzzy Delphi method (FDM), Fuzzy DEMA-TEL (FDEMATEL), and the fuzzy analytic network process (FANP) for new product development (NPD).An empirical case study from the LED-based lighting plant industry is presented as follows: (1) integrating crucial objectives and criteria of new product development obtained by literature review and expert interviews; (2) exploiting the FDM to screen the elements of objectives and criteria and identifying the cause-and-effect relationships among them; (3) determining the weights of priority vectors of objectives and criteria using FANP, which can facilitate the decision-making quality and enhance the NPD investment benefits in the LED-based lighting plant factory.
As this study is primarily concerned with decisionmaking in new product development in the LED-based lighting plant factory, it invites industrial and academic experts experienced in electromechanics, agriculture, and venture capital.Relevant literature surveys and repeated discussions between experts and scholars allow the study to identify crucial elements of potential objectives and criteria of new product development in an LED-based lighting plant factory, and the proposed evaluation model can be further utilized in the future development of new products.
In the first phase, the expert questionnaires for the fuzzy Delphi method (FDM) are implemented to select the critically affected elements of objectives and criteria in new product development.The expert questionnaire for the fuzzy decision-making trial and evaluation laboratory (FDEMATEL) is next employed to determine the degree and direction of impact from each performance index and to draw the network structure of each index via causality relationships in the second phase.In addition, a fuzzy analytic network process (FANP) is applied to analyze the objective and criteria weights and to determine the priorities of critical impact factors in the third phase.Through the above process, the critical priority factors are identified in this research, which will affect investment in new product development in an LED-based lighting plant factory.
In addition to the literature survey and expert interviews, this study arranges five objectives and twenty-two criteria as FDM factor-screened samples.This is sufficient to obtain the cumulative frequency distribution function and the fuzzy integration to turn the experts' judgments into fuzzy numbers using 10 to 15 participants.Using more than 10 participants in the FDM process effectively reduces the error among them [50].A total of 17 experts from different professional and academic fields take part in this study, including the photovoltaic industry, agricultural industry, machinery industry, and academia.The various FDM participants' backgrounds and qualifications are listed in Table 1.
In recent years, a plant factory-related research has received significant attention as an emerging industry.This study focuses on an LED-based lighting plant factory in view of the current market situation and limited human resources, with the aim of meeting potential consumers' NPD demands.Based on the research methodology of the previous section, the study allows experts and scholars to proceed to the related FDM questionnaires and identify the objectives and criteria using brainstorming discussions, which are described in Tables 2 and 3.
By integrating the results of comparing extreme values with their threshold values, the trimmed new product development decision-making model for an LED-based lighting plant factory can be generated by using the Fuzzy Delphi method (FDM) and represents 4 objectives and 11 criteria, as shown in Table 4.Moreover, the decision-making hierarchy schematic model for an LED-based lighting plant factory is depicted in Figure 4.
The impact interactions of the above 11 key indicators (i.e., A to K) can be identified through the Fuzzy DEMA-TEL survey, which invited six domestic industrial experts and scholars in the plant factory field to take part in the questionnaires.They were asked to identify, based on their expertise, the direct effects exerted by each element on other elements, using a scale ranging from 0 to 4, where "0," "1," "2," "3," and "4," respectively, mean "no influence," "low influence," "medium influence," "high influence," and "very high influence."The larger the value of impact indicators, the greater the influence of the factor on another factor!After completing the above questionnaires, each expert evaluates the direct impact effects and places them in direct-relation matrices, and the corresponding triangle fuzzy number can be seen in Table 5.
The direct-relation matrix is developed by summing all vectors and taking the sum of maximum vectors as the benchmark of form.The initial direct-relation matrix  is shown in (9).The normalized direct-relation matrix () can be obtained by calculating (10) and (11).Substitute the direct-relation matrix  into (12) to calculate the total-relation matrix (), while the direct-relation matrix can also be addressed as the total-relation matrix.
Particularly, the unit matrix  is defined in (13), and all criteria direct-relation matrices are combined to become the totalrelation matrix: At the outset of constructing the experts' questionnaires, the elements of objectives such as management team, financial funding, product and technology, and marketing are used to obtain the relevance and level of impact through directrelation analyses, as shown in Table 6.The sum of   on each column and the sum of   on each row, respectively, are then calculated, as shown in (14), where   denotes the level of impact directly and affects other criteria;   denotes the level; and criterion  is directly affected by other criteria.  +   denotes relevance, signifying the intensity of the relationship between criteria.  −   denotes level of impact, also known as the intensity with which criteria have effect or are affected.
Using two-dimensional coordinates as the base of a causal diagram and taking   +   as the -axis and   −   as the -axis, the relevance and level of impact of each criterion are marked on the coordinates.The causality between criteria is drawn according to the threshold, in addition to analysis.In particular, if   −   is a positive value, it suggests that  affects criterion ; conversely, if   −   is a negative value, it suggests that  affects criterion .In order to obtain the relevance and level of impact among objectives, the matrix values are substituted into (14) to yield   and   , which, in addition to a summary, are found in Table 16.In particular,   +   represents the relevance value between objectives, whereas   −   denotes the level of impact or impact value.
The relevance value and impact value are set up in twodimensional coordinates, where relevance value (  +   ) is filled in the -axis, and impact value (  −   ) is filled in the -axis.This study configures a threshold to highlight the causality.In particular, the threshold value is 1.073, which is the maximum value of diagonal elements within the total-relation matrix ().Through the configuration of the threshold, a number smaller than the threshold will be deleted, while a number greater than or equal to the threshold will be drawn on the coordinates to form the directrelation causal diagram for objectives.The causal diagram for objectives is shown in Figure 5.
The relevance and level of impact of management team, financial funding, product and technology, and marketing are shown in Tables 7-10, and causal diagrams for management team, financial funding, product and technology, and marketing are shown in Figures 6-9.
In addition to the above, a fuzzy analytic network process (FANP) is used to calculate the relative weights of relationships between objectives and criteria, so that enterprises or   R&D departments can select the most important elements in the new product development decision-making scheme.This study constructed dependency relationships between hierarchical relationships of the key elements of objectives and criteria in new product development of an LED-based lighting plant factory using Fuzzy DEMATEL.Moreover, the study composes pairwise relative weights of the FANP architecture, while the questionnaire responses of 10 experts are assessed with the fuzzy semantic variables integrated by  geometric mean, and the holistic process of the FANP is illustrated as follows.Step 1 (establish the network structure).An analytics network process structure consisting of the business goal and objectives and criteria well defined in the first step of this case study is established.This structure includes the exterior relationship among criteria and the interior relationship among criteria under each objective, as shown in Figure 10.
Step 2 (form pairwise comparison matrices).Based on the (1-9) scores from the expert responses to the FANP questionnaires, the pairwise comparison matrices for the objective and criteria are formed.Meanwhile, the entries in each column are normalized and the eigenvectors are obtained.The fuzzy and defuzzy pairwise comparison matrices from 10 experts' answers to the four objectives are shown in Tables 11  and 12.
Step 3 (examine the consistency).Once the eigenvectors are obtained, the largest eigenvalue for each matrix can be computed.Then, the consistency index (CI), random index (RI), and the consistency ratio (CR) are calculated.
According to Saaty's suggestion, the consistency is satisfied if the CR is smaller than 0.1.Thus, the expert responses are   analyzed using the above process.The results show that all CR values are less than 0.1.This means that the questionnaires are validated.As an example, a consistency examination including CI, RI, and CR from the same experts' answers to the four objectives is shown in Table 13.
Step 4 (construct the supermatrix).Each matrix is defuzzified to obtain the weights of the matrix, and then the weights are brought into the supermatrix, and an iterative computing model is used to generate the limited matrix from the supermatrix.Table 14 is an unweighted supermatrix, and a weighted supermatrix is shown in Table 15; after iterative computing, the limited supermatrix is generated, as shown in Table 16.

Conclusion
This study first described objectives and performance criteria affecting new product design, collected and analyzed from previous studies and expert interviews and questionnaires.Experts were selected for their expertise and experience in various fields relating to LED plant lighting.Moreover, a reasoned consensus and consolidated individual judgments were systematically obtained by analyzing the responses of the surveys of a fuzzy Delphi method (FDM), and a fuzzy decision-making trial and evaluation laboratory (FDEMATEL) was managed to determine the relationships among the objectives and performance criteria.Secondly,  environmental green standards can be met using the objectives and criteria obtained, which indicate the relationships among new product development objectives and criteria in the LED plant lighting industry.As a result, the weights of potential risk factors and top priority factors with FANP and the composite priority vector (CPV) can be generated to help enterprises establish an efficient decision-making assessment system, able to determine the most suitable alternatives for new product development in the LED-based lighting plant factories.

Figure 1 :
Figure 1: The schematic structure of an LED-based lighting plant factory.
(i) Discovering the exact objectives and criteria which meet environmental requirements (ii) Indicating the relationships between objectives and criteria in view of new product development in the LED-based lighting plant factory (iii) Evaluating the weights of potential critical factors and top priority factors to help enterprises' selection of trade-offs in new product development (iv) Constructing a novel performance assessment system for new product development in the LED-based lighting plant factory.

Figure 3 :
Figure 3: Gray zone in two triangular fuzzy numbers.
Calculate fuzzy weights.Based on the Lambda-Max method proposed by Csutora and Buckley in 2001, calculate the fuzzy weights of decision elements.The procedures are as follows.Apply -cut.Let  = 1 to obtain the positive matrix of decision-maker , R  = [r  ]   , R  = [r  ]   , and let  = 0 to obtain the lower and upper bound positive matrices of decision-maker , and R  = [r  ]   .Based on the weight calculation procedure proposed in ANP, calculate weight matrices    = [  ]   ,    = [  ]   , and    = [  ]   ,  = 1, 2, . . ., .

Figure 4 :
Figure 4: The decision-making hierarchy schematic diagram of an NPD project.

Figure 6 :
Figure 6: Causal diagram for management team.

Figure 10 :
Figure 10: An analytics network process structure of the proposed MCDM model.

Figure 11 :
Figure 11: The structure of composite priority vector.

Figure 12 :
Figure 12: Comparisons of weights of objectives and criteria between FANP and CPV.
and    are the lowest value and the highest value for the criterion   scored by the expert .These two values also, respectively, indicate the quantitative scores of )}, where   should be an integer; the criteria for each expert's most conservative value and most optimistic value.The lowest level    and the highest level    of all expert judgments for each criterion   are then statistically calculated, and any extreme values greater than twice the standard deviation will be removed from the criteria [22].The triangular values (the minimum    , the geometric mean    , and the maximum    ) of the remaining most conservative value    and the triangular values (the minimum    , the geometric mean    , and the maximum    ) of the qualified most optimistic

Table 2 :
The definition of objectives in the FDM process.

Table 3 :
The definition of criteria in the FDM process.

Table 4 :
The trimmed NPD decision-making model.

Table 5 :
Triangle fuzzy number of identified direct effect and scale range.

Table 6 :
Relevance and level of impact among objectives after integration.

Table 7 :
The relevance and level of impact (6 experts) for management team.

Table 8 :
The relevance and level of impact (6 experts) for financial fund.

Table 9 :
The relevance and level of impact (6 experts) for product and technology.

Table 10 :
The relevance and level of impact (6 experts) for marketing.

Table 11 :
The fuzzy pairwise comparison matrix (10 experts) for objectives.

Table 13 :
The consistency assessment of fuzzy pairwise comparison matrix.

Table 17 :
Comparison of FANP and composite priority vector.