鞠萍华,陈资,冉琰,涂顺泽
(机械传动国家重点实验室(重庆大学), 重庆 400044)
摘要:
为解决传统质量功能展开(QFD)在实际运用过程中存在关于顾客需求和工程技术之间的关系评估,顾客需求权重的确定和工程技术的优先级排序等方面的固有缺陷,提出一种概率语言环境下考虑专家心理行为的QFD方法. 运用概率语言连乘层次分析法(PL-MAHP)确定顾客需求初始权重,针对顾客需求之间的关联关系,运用模糊认知图(FCM)对顾客需求进行推理分析并获取其最终权重;为有效地处理QFD团队专家评估信息中的模糊性和不确定性,使用概率语言术语集(PLTS)表征顾客需求和工程技术之间的关联强度,并将交互式多属性决策(TODIM)拓展到概率语言环境中,依据各个工程技术的全局占优度来最终确定工程技术的重要度,充分考虑了专家心理行为对工程技术的优先级排序的影响.在电动汽车产品开发实例中运用本文方法结果表明,该方法能合理有效地确定最终重要度. 与现有其他方法对比分析,验证了本文方法的优越性.
关键词: 质量功能展开 概率语言术语集 交互式多属性决策 模糊认知图 工程技术 顾客需求
DOI:10.11918/201907218
分类号:N94
文献标识码:A
基金项目:国家自然科学基金(1,8);国家重大科技专项(2018ZX04032-1,6ZX04004-005)
A novel QFD method considering expert’s psychological behavior character under probabilistic linguistic environment
JU Pinghua,CHEN Zi,RAN Yan,TU Shunze
(State Key Laboratory of Mechanical Transmissions (Chongqing University), Chongqing 400044, China)
Abstract:
To overcome some inherent drawbacks regarding the assessment of relationships between customer requirements and engineering characteristics, the determination of customer requirements weights and the prioritization of engineering characteristics in application of traditional QFD method, a novel QFD method considering expert’s psychological behavior character under probabilistic linguistic environment was proposed. Firstly, the initial weights of customer requirements were determined by using the probabilistic linguistic multiplicative analytic hierarchy process (PL-MAHP). With respect to the interrelationships among customer requirements into account, fuzzy cognitive map (FCM) was used to analyze and determine their final weights. Secondly, the probabilistic linguistic sets (PLTS) was applied to express the uncertainty and hesitancy of subjective assessments in evaluating the interrelations between customer requirements and engineering characteristics. in addition, the TODIM method was extended to probabilistic linguistic environment, the overall dominance of engineering characteristics was utilized to determine the importance ranking of engineering characteristics considering the psychological behaviors of decision makers. Finally, the proposed QFD method was applied in an empirical case concerning the product development of electric vehicle, the results proved that the method can effectively prioritise engineering characteristics in QFD, and comparison study with other relevant methods was also performed to show its merits.
Key words: quality function deployment (QFD) probabilistic linguistic sets (PLTS) interactive multi-attribute decision making (TODIM) fuzzy cognitive map (FCM) engineering characteristics customer requirement
鞠萍华, 陈资, 冉琰, 涂顺泽. 概率语言环境下考虑专家心理行为的QFD方法[J]. 哈尔滨工业大学学报, 2020, 52(7): 193-200. DOI: 10.11918/201907218.
JU Pinghua, CHEN Zi, RAN Yan, TU Shunze. A novel QFD method considering expert's psychological behavior character under probabilistic linguistic environment[J]. Journal of Harbin Institute of Technology, 2020, 52(7): 193-200. DOI: 10.11918/201907218.
基金项目 国家自然科学基金(51835001, 51705048);国家重大科技专项(2018ZX04032-001, 2016ZX04004-005) 作者简介 鞠萍华(1974—),男,副教授,硕士生导师 通信作者 冉琰,ranyan@cqu.edu.cn 文章历史 收稿日期: 2019-07-27
Abstract Full text Figures/Tables PDF
概率语言环境下考虑专家心理行为的QFD方法
鞠萍华, 陈资, 冉琰, 涂顺泽
机械传动国家重点实验室(重庆大学),重庆 400044
收稿日期: 2019-07-27
基金项目: 国家自然科学基金(51835001, 51705048);国家重大科技专项(2018ZX04032-001, 2016ZX04004-005)
作者简介: 鞠萍华(1974—),男,副教授,硕士生导师
通信作者: 冉琰,ranyan@cqu.edu.cn
摘要: 为解决传统质量功能展开(QFD)在实际运用过程中存在关于顾客需求和工程技术之间的关系评估,顾客需求权重的确定和工程技术的优先级排序等方面的固有缺陷,提出一种概率语言环境下考虑专家心理行为的QFD方法.运用概率语言连乘层次分析法(PL-MAHP)确定顾客需求初始权重,针对顾客需求之间的关联关系,运用模糊认知图(FCM)对顾客需求进行推理分析并获取其最终权重;为有效地处理QFD团队专家评估信息中的模糊性和不确定性,使用概率语言术语集(PLTS)表征顾客需求和工程技术之间的关联强度,并将交互式多属性决策(TODIM)拓展到概率语言环境中,依据各个工程技术的全局占优度来最终确定工程技术的重要度,充分考虑了专家心理行为对工程技术的优先级排序的影响.在电动汽车产品开发实例中运用本文方法结果表明,该方法能合理有效地确定最终重要度.与现有其他方法对比分析,验证了本文方法的优越性.
关键词: 质量功能展开 概率语言术语集 交互式多属性决策 模糊认知图 工程技术 顾客需求
A novel QFD method considering expert's psychological behavior character under probabilistic linguistic environment
JU Pinghua, CHEN Zi, RAN Yan, TU Shunze
State Key Laboratory of Mechanical Transmissions (Chongqing University), Chongqing 400044, China
Abstract: To overcome some inherent drawbacks regarding the assessment of relationships between customer requirements and engineering characteristics, the determination of customer requirements weights and the prioritization of engineering characteristics in application of traditional QFD method, a novel QFD method considering expert's psychological behavior character under probabilistic linguistic environment was proposed. Firstly, the initial weights of customer requirements were determined by using the probabilistic linguistic multiplicative analytic hierarchy process (PL-MAHP). With respect to the interrelationships among customer requirements into account, fuzzy cognitive map (FCM) was used to analyze and determine their final weights. Secondly, the probabilistic linguistic sets (PLTS) was applied to express the uncertainty and hesitancy of subjective assessments in evaluating the interrelations between customer requirements and engineering characteristics. in addition, the TODIM method was extended to probabilistic linguistic environment, the overall dominance of engineering characteristics was utilized to determine the importance ranking of engineering characteristics considering the psychological behaviors of decision makers. Finally, the proposed QFD method was applied in an empirical case concerning the product development of electric vehicle, the results proved that the method can effectively prioritise engineering characteristics in QFD, and comparison study with other relevant methods was also performed to show its merits.
Keywords: quality function deployment (QFD) probabilistic linguistic sets (PLTS) interactive multi-attribute decision making (TODIM) fuzzy cognitive map (FCM) engineering characteristics customer requirement
质量功能展开(Quality function deployment,QFD)是一种用于将产品顾客需求(Customer Requirements,CRs)转为工程技术(Engineering Characteristics,ECs)的产品规划方法[1],尽管传统QFD方法因其简单易操作的特点而被广泛应用于汽车、电子和服务等各个领域[2-4],但在实际应用中仍存在许多问题.本文主要关注的问题有:1)使用清晰的数值量化顾客需求和工程技术之间的关联程度,无法描述专家评估信息的模糊性和不确定性;2)确定顾客需求权重时忽略了顾客需求之间存在的关联关系;3)使用加权平均算法推导工程技术的重要度,忽略专家有限理性的心理行为,不适用于对工程技术重要度的精确排序.
针对问题1),三角模糊数[5]、犹豫模糊集[2]、区间直觉模糊数[6]、直觉模糊集[7]和毕达哥拉斯模糊集[8]等模糊集理论被广泛应用于表征顾客需求和工程技术关联强度.但在实际应用过程中,由于人类认知的固有模糊性和不确定性,QFD团队专家更倾向于使用语言术语(Linguistic Term Set, LTS)进行判断评估,作为犹豫模糊语言集(Hesitant Fuzzy Linguistic Term Set, HFLTS)[9]的拓展.文献[10]提出的概率语言术语集(Probabilistic Linguistic Term Sets, PLTS)不仅包含决策专家对多个语言术语的犹豫信息,而且还可以通过对不同语言术语增加概率信息来反映不同程度的偏好,有效地避免了偏好信息的丢失,提高了语言信息表达的灵活性,更加适用于描述专家评估信息的模糊性和不确定性[11].
针对问题2),当前研究主要使用决策实验与评估法(decision making trial and evaluation laboratory,DEMATEL)和网络分析法(analytic network process,ANP)刻画顾客需求间的关联关系,文献[12]提出集成DEMATEL和复杂比例评估法(complex proportional assessment, COPRAS)的QFD模型,研究了顾客需求间自相关关系对工程技术重要度确定的影响.文献[2]运用犹豫模糊DEMATEL分析顾客需求间的相互关系并确定其权重.文献[13]对顾客需求间存在的关联关系问题,提出综合ANP和QFD供应商选择模型,尽管DEMATEL和ANP是分析因素关联行为的有效工具,但存在专家多次主观判断易产生不一致性,不能描述顾客需求间存在正、负因果影响关系和顾客需求动态变化特性等问题.
针对问题3),QFD方法中工程技术重要度排序问题可以被视为一种多属性决策问题(multiple criteria decision making, MCDM),因此,COPRAS[12]、偏好结构排序法[14]和灰色关联法[15]等MCDM方法已被用于确定工程技术重要度.文献[16]提出的TODIM方法是一种基于前景理论价值函数为基础的有效行为决策方法,相比于其他MCDM方法,TODIM充分考虑了专家有限理性的心理行为对工程技术重要度排序的影响,使得工程技术重要度排序结果更加合理可靠.
本文提出了一种概率语言环境下考虑专家心理行为的QFD方法,该方法综合考虑专家主观经验知识和顾客需求间自相关两方面因素,结合概率语言连乘层次分析法(probabilistic linguistic multiplicative analytic hierarchy process, PL-MAHP)和模糊认知图(fuzzy cognitive map,FCM)确定顾客需求权重,利用PLTS评估顾客需求和工程技术之间的关联程度,并基于PL-TODIM确定工程技术重要度.最后将本文所提的QFD方法应用于电动汽车产品开发实例中,验证了该方法的有效性和适用性.
1 概率语言术语集定义1[10]?设LTS为S={sα|α=-τ, ..., -1, 0, 1, ..., τ},为了描述专家评估时的犹豫和不确定性,定义一个概率语言术语集PLTS为
$\begin{array}{*{20}{l}}{L(p) = \{ {L^{(l)}}({p^{(l)}})|{L^{(l)}}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \in {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} S, {p^{(l)}} \ge 0, l = 1, 2, \cdots , }\\{{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \# L(p), \sum\nolimits_{l = 1}^{\# L(p)} {{p^{(l)}} \le 1\} } .}\end{array}$
式中:L(l)(p(l))为概率信息为p(l)的语言术语L(l), #L(p)为所有L(p)中包含的语言术语的数目.
1.1 PLTS的标准化定义2[10]?若PLTS中∑l=1#L(p)p(l)<1,则概率信息标准化的PLTS?(p)定义为
$\dot L(p) = \{ {L^{(l)}}({\dot p^{(l)}})|l = 1, 2, \ldots , \# L(p)\} .$
式中:?(l)=p(l)/∑l=1#L(p)p(l),令a(l)是语言术语L(l)的下标,则PLTS内所有元素按照α(l)p(l)的值升序排列;若PLTS内含有两个或多个具有相等α(l)p(l)值的元素,则按α(l)的值升序排列.
定义3 [10]? L1(P)={l1(l)(p1(l))|l=1, 2, ..., #L1(p)}和L2(P)={l2(l)(p2(l))|l=1, 2, ..., #L2(p)}是两个不同的PLTS,若#L1(p)>#L2(p),则将#L1(p)-#L2(p)个语言术语添加到L2(p)中,其中添加的语言术语是L2(p)中最小的语言术语,且其概率为0,使得L1(p)和L2(p)中包含的语言术语数相等.
基于定义2和定义3,可以获得标准化的PLTS,L(p)={L(l)(P(l))|l=1, 2, ..., #L(p)},其中P(l)=p(l)/∑l=1#L(p)p(l).
1.2 PLTS的集聚定义4?[17]?设M={Mq|q=1, 2, …, Q}是由Q位专家组成的专家团队,其权重向量为[λ1, λ2, …, λQ]T,且满足∑q=1Qλq=1.专家Mq提供的PLTS为Lq(p)={Lαq(pαq)|Lαq∈S},其中pαq为语言术语Lαq的概率信息,则综合PLTS为
$\begin{array}{*{20}{c}}{L(p) = \{ L_\alpha ^{(l)}({p^{(l)}})|L_\alpha ^{(L)}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \in {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} S, {p^{(l)}} = \sum\nolimits_{q = 1}^Q {v_\alpha ^q{\lambda _q}} , }\\{l = 1, 2, \ldots , \# L(p)\} .}\end{array}$ (1)
式中:vαq为Lq(p)中语言术语Lαq的权重
${v^q} = \left\{ {\begin{array}{*{20}{l}}{p_\alpha ^q, }&{{\rm{ if}}{\kern 1pt} {\kern 1pt} {\kern 1pt} L_\alpha ^{(l)}{\kern 1pt} {\kern 1pt} {\kern 1pt} \in {\kern 1pt} {\kern 1pt} {\kern 1pt} {L^q}(p);}\\{0, }&{{\rm{ if}}{\kern 1pt} {\kern 1pt} {\kern 1pt} L_\alpha ^{(l)}{\kern 1pt} {\kern 1pt} {\kern 1pt} \notin {\kern 1pt} {\kern 1pt} {\kern 1pt} {L^q}(p).}\end{array}} \right.$ (2)
例1?假设3位专家提供的PLTS分别为L1(p)={s1(0.5), s2(0.5)},L2(p)={s2(0.8), s3(0.2)},L3(p)={s2(1)},专家权重向量为[0.3, 0.2, 0.5]T, 则依据定义4聚集得到的综合PLTS为L(p)={s1(0.15), s2(0.81), s3(0.04)}.
1.3 PLTS间的比较定义5[17] ?设基于S={sα|α=-τ, ..., -1, 0, 1, ..., τ}的PLTS为L(p)={L(l)(p(l))|l=1, 2, ..., #L(p)}, α(l)是语言术语L(l)的下标,则其期望值E(L(p))和方差值σ(L(p))分别定义为
$\begin{array}{*{20}{c}}{E(L(p)) = \sum\nolimits_{l = 1}^{\# L(p)} {(\frac{{{\alpha ^{(l)}} + \tau }}{{2\tau }}{p^{(l)}})} /\sum\nolimits_{l = 1}^{\# L(p)} {{p^{(l)}}} , }\\{\sigma (L(p)) = {{(\sum\nolimits_{l = 1}^{\# L(p)} {((} \frac{{{\alpha ^{(l)}} + \tau }}{{2\tau }}E(L(p)))}^2}{p^{(l)}})/\sum\nolimits_{l = 1}^{\# L(p)} {{p^{(l)}}} {)^{1/2}}.}\end{array}$
利用期望值E(L(p))和方差值σ(L(p))构造PLTS的大小比较规则:若E(L1(p))>E(L2(p)),则L1(p)>L2(p);若E(L1(p))=E(L2(p)),则进行方差值的比较,若σ(L1(p))>σ(L2(p)), 则L1(p)<L2(p);如果σ(L1(p))=σ(L2(p)),则L1(p)=L2(p).
1.4 PLTS间的距离定义6[18]?设LTS为S={sα|α=-τ, ..., -1, 0, 1, …, τ}的任意两个不同标准化PLTS分别为
$\begin{array}{*{20}{l}}{{{\bar L}_1}(p) = \{ {{\bar L}_1}^{(l)}(\bar p_1^{(l)})|l = 1, 2, \ldots , \# {{\bar L}_1}(p)\} , }\\{{{\bar L}_2}(p) = \{ {{\bar L}_2}^{(l)}(\bar p_2^{(l)})|l = 1, 2, \ldots , \# {{\bar L}_2}(p)\} , }\end{array}$
则二者之间的距离公式为
$\begin{array}{*{20}{l}}{d({{\bar L}_1}(p), {{\bar L}_2}(p)) = }\\{{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \sqrt {\sum\nolimits_{l = 1}^{\# {{\bar L}_1}(p)} {{{(\frac{{\alpha _1^{(l)} + \tau }}{{2\tau }}\bar p_1^{(l)} - \frac{{\alpha _2^{(l)} + \tau }}{{2\tau }}\bar p_2^{(l)})}^2}} /\# {{\bar L}_1}(p)} .}\end{array}$
2 概率语言环境下考虑专家心理行为的QFD方法针对传统QFD方法中存在的问题,本文提出一种概率语言环境下考虑专家心理行为的QFD方法,该方法主要包括两个阶段:基于PL-MAHP和FCM确定顾客需求权重;基于PL-TODIM的工程技术重要度排序.本文提出的方法为流程图如图 1所示.
Fig. 1
图 1 提出的QFD方法的流程图 Fig. 1 Framework of the proposed QFD mode
2.1 基于PL-MAHP和FCM确定顾客需求权重 2.1.1 基于PL-MAHP确定顾客需求初始权重连乘层次分析法(MAHP)是一种基于属性成对比较矩阵求取属性主观权重的方法[19],相比于AHP,MAHP最大的特点是具有传递特性,不需要对比较矩阵进行一致性检验[20].传统MAHP使用精确数值来表示两个属性之间的相对重要性,导致无法真实准确地反映专家判断信息,考虑到PLTS是表征专家判断的有效工具,本文引用文献[17]提出的PL-MAHP方法确定顾客需求初始权重.
步骤1?确定专家综合概率语言偏好矩阵
设QFD专家团队由Q位专家Mq(q=1, 2, ..., Q)组成,依据专家知识结构和领域经验的不同由AHP法确定专家权重向量为[λ1, λ2, …, λQ]T,且满足∑q=1Qλq=1,对于顾客需求集合{C1, C2, …, Cn},专家Mq给出概率语言偏好矩阵Dq=(Ljkq(P))n×n(j, k=1, 2, ..., n),其中Ljkq(P)= {Ljkq(l)(Pjkq(l))|1, 2, ..., #Ljkq(P)}是基于S={sα}|α=-τ, ..., -1, 0, 1, ..., τ}的PLTS,表示顾客需求Cj对比顾客需求Ck的相对重要性,其满足特性:
$\begin{array}{*{20}{l}}{P_{jk}^{q(l)} = P_{kj}^{q(l)}, L_{jk}^{q(l)} = neg (L_{kj}^{q(l)}), }\\{L_{jj}^{(q)}(P) = \{ {s_0}(1)\} , \# L_{jk}^q(p) = \# L_{kj}^q(p).}\end{array}$
通过式(1)、(2),聚集Dq得到综合概率语言偏好矩阵D =(Ljk(P))n×n.
步骤2?基于PL-MAHP计算顾客需求初始权重
根据MAHP,偏好矩阵D =(Ljk(P))n×n中Ljk(P)与顾客需求Cj和Ck的相对权重关系为
${\tilde w_j} = {\rm{exp}}(\frac{{ {\rm{In}} \sqrt 2 }}{n} \times \sum\nolimits_{t = 1}^n {(\sum\nolimits_{l = 1}^{\# L(p)} {{\alpha ^{(l)}}{p^{(l)}}} /\sum\nolimits_{l = 1}^{\# L(p)} {{p^{(l)}}} ))} , $ (3)
归一化处理得顾客需求Cj初始权重
$w_j^0 = {\tilde w_j}/\sum\nolimits_{j = 1}^n {{{\tilde w}_j}} , {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} j = 1, 2, \ldots , n.$ (4)
2.1.2 基于FCM确定顾客需求最终权重FCM是一种常用的属性间因果关联分析方法,相比于DEMATEL和ANP,FCM能清晰反映属性间存在正、负因果影响关系,并通过迭代推理获得属性间动态变化特性[21].考虑到顾客需求间相互影响关系,本文通过构建模糊认知图表征顾客需求间关联关系,以PL-MAHP得到的顾客需求初始权重作为FCM中顾客需求的初始状态值,并根据FCM推理机制进行推演分析,最终归一化处理顾客需求稳定状态值确定其最终权重.
步骤3?确定顾客需求的初始状态值
设顾客需求初始权重作为FCM中的顾客需求初始状态值,即
步骤4?确定综合顾客需求自相关矩阵
设专家Mq给出顾客需求自相关矩阵为Wq=[ejkq]n×n(j, k=1, 2, ..., n, ejkq∈[-1, 1]),其中ejkq为顾客需求Cj对顾客需求Ck的影响程度,其正负分别表示顾客需求间存在的正、负影响关系,绝对值的大小反映影响程度大小.通过对Wq集聚得到综合顾客需求自相关矩阵W =[ejk]n×n,其中,
${e_{jk}} = (\sum\nolimits_{q = 1}^Q {{\lambda _q}e_{jk}^q} )/Q.$ (5)
步骤5?计算顾客需求的稳定状态值
FCM的推理机制为
$A_{{C_j}}^{t + 1} = f(A_{{C_j}}^t + \sum\nolimits_{k = 1, k \ne j}^n {{e_{jk}}} \times A_{{C_k}}^t).$ (6)
式中:ACjt为顾客需求Cj迭代次数t时的状态值,f为阈值函数,阈值函数种类多样,其中sigmoid函数能够真实反应出各个节点的活动状态趋势及变化程度,并且保证每次的节点迭代状态值输出在[0, 1],所以本文选用sigmoid函数f(x)=1/(1+exp(-αx))作为阈值函数.
经过一定次数的迭代,节点状态值达到稳定状态或循环状态,甚至混沌状态时则迭代停止[22].
步骤6?确定顾客需求的最终权重
通过对顾客需求稳定状态值归一化处理获得顾客需求的最终权重
${w_j} = {A_{{C_j}}}/(\sum\nolimits_{j = 1}^n {{A_{{C_j}}}} ), $ (7)
式中:ACj为顾客需求Cj迭代停止时的稳态状态值.
2.2 基于PL-TODIM确定工程技术重要度排序在建立顾客需求和工程技术的关联关系过程中,专家不同的心理行为将直接影响最终工程技术重要度.以前景理论价值函数为基础提出的TODIM方法最主要优点是能够有效地刻画决策者心理行为.因此,本文将TODIM拓展到概率语言环境下以确定工程技术重要度,其步骤如步骤7~11.
步骤7?确定专家标准综合概率语言关联矩阵
设R q=[Lijq(p)]m×n为专家Mq提供的概率语言关联矩阵,其中Lijq(p)为专家Mq对顾客需求Cj(j=1, 2, ..., n)和工程技术Ei(i=1, 2, ..., m)间的关联程度评价.根据式(1)、(2)聚集各个专家组概率语言关联矩阵R q=[Lijq(p)]m×n得综合概率语言关联矩阵R=[Lij(p)]m×n,基于定义2和3,对R进行标准化处理得标准综合概率语言关联矩阵R=[Lij(p)]m×n.
步骤8?计算顾客需求的相对权重
令Cr为参考顾客需求,顾客需求Cj相对权重为
${w_{j{\rm{r}}}} = {w_{\rm{j}}}/{w_{\rm{r}}}, j = 1, 2, \ldots , n.$ (8)
式中:wr=max{wj|j=1, 2, ..., n}为Cr对应的参照权重.
步骤9?计算工程技术Ei优于Eh的占优度
在顾客需求Cj下,工程技术Ei优于Eh的占优度计算表达式为
$\begin{array}{l}{\phi _j}({E_i}, {E_h}) = \\\left\{ {\begin{array}{*{20}{l}}{\sqrt {{w_{j{\rm{r}}}}d({{\bar L}_{ij}}(p), {{\bar L}_{hj}}(p))/\sum\nolimits_{j = 1}^n {{w_{j{\rm{r}}}}} } , }&{{{\bar L}_{ij}}(p) > {{\bar L}_{hj}}(p);}\\{0, }&{{{\bar L}_{ij}}(p) = {{\bar L}_{hj}}(p);}\\{ - \frac{1}{\theta }\sqrt {\sum\nolimits_{j = 1}^n {{w_{j{\rm{r}}}}} d({{\bar L}_{ij}}(p), {{\bar L}_{hj}}(p))/{w_{j{\rm{r}}}}} , }&{{{\bar L}_{jj}}(p) < {{\bar L}_{hj}}(p).}\end{array}} \right.\end{array}$ (9)
式中:d(Lij(p), Lhj(p))是Lij(p)和Lhj(p)之间距离, j=1, 2, .., n; i, h=1, 2, ..., m.参数θ(θ>0)表示损失规避系数,需要根据专家的风险偏好调整值的大小,若专家是风险爱好型,则损失影响的程度将扩大,取0<θ<1;若专家是风险规避型,则损失的影响将减小,取θ>1;若专家为风险中立型,则为真实损失影响,取θ=1.
步骤10?确定工程技术Ei优于Eh的综合占优度工程技术Ei优于Eh的综合占优度为
$\delta ({E_i}, {E_h}) = \sum\nolimits_{j = 1}^n {{\phi _j}({E_i}, {E_h})} .$ (10)
步骤11?获取工程技术Ei全局占优度ξi
工程技术Ei全局占优度ξi计算公式为
${\xi _i} = \frac{{\sum\nolimits_{h = 1}^m {\delta ({E_i}, {E_h})} - \mathop {{\rm{min}}}\limits_i (\sum\nolimits_{h = 1}^m {\delta ({E_i}, {E_h})} )}}{{\mathop {{\rm{max}}}\limits_i \sum\nolimits_{h = 1}^m {\delta ({E_i}, {E_h})} - \mathop {{\rm{min}}}\limits_i (\sum\nolimits_{h = 1}^m {\delta ({E_i}, {E_h})} )}}.$
根据ξi对工程技术重要度Ei进行排序,ξi值越大,则工程技术Ei重要度越高.
3 实例分析随着全球能源危机和环境污染问题的日趋严重,节能环保的电动汽车的研发已经成为未来汽车产业战略发展方向,某公司为了提升电动汽车产品市场竞争力,选择3位专家作为电动汽车产品开发的QFD团队,其中包括1位产品研发工程师M1、1位销售经理M2和1位制造车间组长M3,依据专家知识结构和领域经验的不同分别确定专家权重λ=(0.5, 0.25, 0.25). QFD团队通过全面市场调研确定电动汽车顾客需求:动力性能好(C1),购买价格低(C2),使用费用低(C3),续航里程长(C4).依据专业知识和实践经验确定电动汽车工程技术指标:最高车速(E1)、最大爬坡度(E2)、加速时间(E3)、续驶里程(E4)、系统成本(E5).
$\mathit{\boldsymbol{D}} = \left[ {\begin{array}{*{20}{l}}{{s_0}(1)}&{{s_{ - 1}}(0.45), {s_0}(0.55)}&{{s_1}(0.55), {s_2}(0.45)}&{{s_{ - 1}}(0.80), {s_0}(0.20)}\\{{s_0}(0.55), {s_1}(0.45)}&{{s_0}(1)}&{{s_1}(0.43), {s_2}(0.57)}&{{s_{ - 1}}(0.15), {s_0}(0.85)}\\{{s_{ - 2}}(0.45), {s_{ - 1}}(0.55)}&{{s_{ - 2}}(0.57), {s_{ - 1}}(0.43)}&{{s_0}(1)}&{{s_{ - 2}}(0.95), {s_{ - 1}}(0.05)}\\{{s_0}(0.20), {s_1}(0.80)}&{{s_0}(0.85), {s_1}(0.15)}&{{s_1}(0.05), {s_2}(0.95)}&{{s_0}(1)}\end{array}} \right].$
3.1 计算过程与结果步骤1? 3位专家使用S1={s-2:不重要,s-1:稍不重要,s0:一样重要,s1:稍重要,s2:重要}对顾客需求间进行偏好比较,得到概率语言偏好矩阵D q,并通过式(1)、(2),聚集D q得到综合概率语言偏好矩阵
步骤2、3?依据式(3)、(4),计算得到顾客需求初始权重为wj0=(0.246, 0.285, 0.157, 0.311),将顾客需求初始权重设为FCM中顾客需求初始状态值
步骤4?通过式(5),集聚顾客需求自相关矩阵为
$\mathit{\boldsymbol{W}} = \left[ {\begin{array}{*{20}{c}}0&{ - 0.575}&{0.250}&{0.625}\\0&0&0&0\\0&{ - 0.250}&0&0\\{0.725}&{ - 0.600}&{ - 0.275}&0\end{array}} \right].$
步骤5、6 ?通过式(6)进行迭代运算,其中考虑到计算方便与各状态收敛的速度,取阈值函数中参数α=1,经过9次迭代后顾客需求达到稳定状态值ACj=(0.796 3, 0.307 7, 0.764 8, 0.782 5).并根据式(7)计算出顾客需求的最终权重为wj=(0.300, 0.116, 0.288, 0.295).
步骤7? 3位专家使用S1={s-2:低,s-1:稍低,s0:中等,s1:稍高,s2:高}评价顾客需求和工程技术之间的关联程度,收集3位专家提供的概率语言关联矩阵R q(q=1, 2, 3)汇总于表 1中.通过式(1)、(2)聚集Rq得到综合概率语言关联矩阵R,并依据定义2和定义3标准化处理R得到标准综合概率语言关联矩阵R,如表 2所示.
表 1
表 1 3位专家提供的概率语言关联矩阵 Tab. 1 Probabilistic linguistic relationship evaluation matrix provided by 3 experts Mq Ei C1 C2 C3 C4
M1 E1 {s0(0.2), s1(0.8)} {s0(0.9), s1(0.1)} {s-2(0.2), s-1(0.8)} {s-1(0.2), s0(0.8)}
E2 {s1(1.0)} {s-1(0.2), s0(0.8)} {s-2(0.5), s-1(0.5)} {s-1(0.5), s0(0.5)}
E3 {s0(0.1), s1(0.9)} {s0(0.6), s1(0.4)} {s-1(1.0)} {s-1(1.0)}
E4 {s-1(0.7), s0(0.3)} {s0(0.2), s1(0.8)} {s-1(0.8), s0(0.2)} {s2(1.0)}
E5 {s0(1.0)} {s1(0.3), s2(0.7)} {s0(0.5), s1(0.5)} {s1(0.5), s2(0.5)}
M2 E1 {s1(0.5), s2(0.5)} {s0(0.4), s1(0.6)} {s-2(0.5), s-1(0.5)} {s-1(0.5), s0(0.5)}
E2 s-1(0.2), s0(0.8)} {s0(0.3), s1(0.7)} {s-2(0.8), s-1(0.2)} {s-1(0.7), s0(0.3)}
E3 {s0(0.8), s1(0.2)} {s1(0.8), s2(0.2)} {s-1(0.7), s0(0.3)} {s-1(0.8), s0(0.2)}
E4 {s0(0.7), s1(0.3)} {s1(1.0)} {s-1(1.0)} {s1(0.2), s2(0.8)}
E5 {s0(0.5), s1(0.5)} {s2(1.0)} {s0(0.3), s1(0.7)} {s1(0.8), s2(0.2)}
M3 E1 {s1(1.0)} {s0(0.5), s1(0.5)} {s-2(0.7), s-1(0.3)} {s-2(0.1), s-1(0.9)}
E2 {s0(0.5), s1(0.5)} {s-1(0.2), s0(0.8)} {s-1(1.0)} {s-1(1.0)}
E3 {s0(0.5), s1(0.5)} {s1(1.0)} {s-2(0.3), s-1(0.7)} {s-2(0.5), s-1(0.5)}
E4 {s-1(0.1), s0(0.9)} {s0(0.2), s1(0.8)} {s-2(0.4), s-1(0.6)} {s1(0.1), s2(0.9)}
E5 {s0(0.8), s1(0.2)} {s1(0.5), s2(0.5)} {s0(0.8), s1(0.2)} {s1(0.7), s2(0.3)}
表 1 3位专家提供的概率语言关联矩阵 Tab. 1 Probabilistic linguistic relationship evaluation matrix provided by 3 experts
表 2
表 2 标准综合概率语言关联矩阵 Tab. 2 Normalized group probabilistic linguistic relationship evaluation matrix Ei C1 C2 C3 C4
E1 {s0(0.10), s2(0.12), s1(0.78)} {s0(0), s0(0.65), s1(0.35)} {s-2(0), s-2(0.40), s-1(0.60)} {s-1(0.45), s-2(0.03), s0(0.52)}
E2 {s-1(0.05), s0(0.33), s1(0.62)} {s-1(0.15), s0(0.68), s1(0.17)} {s-2(0), s-2(0.45), s-1(0.55)} {s-1(0), s-1(0.68), s0(0.32}
E3 {s0(0), s0(0.38), s1(0.62)} {s0(0.35), s2(0.05), s1(0.60)} {s-1(0.85), s-2(0.07), s0(0.08)} {s-1(0.83), s-2(0.12), s0(0.05)}
E4 {s-1(0.18), s0(0.75), s1(0.07)} {s0(0), s0(0.15), s1(0.85)} {s-1(0.83), s-2(0.07), s0(0.10)} {s1(0), s1(0.08), s2(0.92)}
E5 {s0(0), s0(0.83), s1(0.17)} {s1(0), s1(0.28), s2(0.72)} {s0(0), s0(0.53), s1(0.47)} {s1(0), s1(0.63), s2(0.37)}
表 2 标准综合概率语言关联矩阵 Tab. 2 Normalized group probabilistic linguistic relationship evaluation matrix
步骤8?选择参照权重wr=0.300,通过式(8)得到相对权重wjr=(1.000, 0.386, 0.960, 0.983).
步骤9?通过式(9),考虑到专家为风险中立型,取损失规避系数θ=1,计算在顾客需求Cj(j=1, 2, 3, 4)下,工程技术Ei优于Eh的占优度j(Ei, Eh)(i, h=1, 2, 3, 4, 5),进而得到每个顾客需求下的占优度矩阵Φj(j=1, 2, 3, 4).
${{\mathit{\boldsymbol{\phi }}} _1} = \left[ {\begin{array}{*{20}{c}}0&{0.147}&{0.155}&{0.318}&{0.305}\\{ - 0.489}&0&{ - 0.232}&{0.284}&{0.270}\\{ - 0.515}&{0.070}&0&{0.281}&{0.265}\\{ - 1.057}&{ - 0.945}&{ - 0.935}&0&{ - 0.426}\\{ - 0.102}&{ - 0.899}&{ - 0.883}&{0.128}&0\end{array}} \right], $
${{\mathit{\boldsymbol{\phi }}} _2} = \left[ {\begin{array}{*{20}{c}}0&{0.096}&{ - 1.368}&{ - 1.497}&{ - 1.541}\\{ - 0.862}&0&{ - 1.497}&{ - 1.686}&{ - 1.741}\\{0.159}&{0.174}&0&{ - 1.132}&{ - 1.341}\\{0.174}&{0.196}&{0.131}&0&{ - 0.886}\\{0.197}&{0.202}&{0.156}&{0.103}&0\end{array}} \right], $
${{\mathit{\boldsymbol{\phi }}} _3} = \left[ {\begin{array}{*{20}{c}}0&{0.046}&{ - 0.694}&{ - 0.677}&{ - 0.817}\\{ - 0.158}&0&{ - 0.686}&{ - 0.670}&{ - 0.827}\\{0.200}&{0.198}&0&{ - 0.617}&{ - 0.964}\\{0.915}&{0.913}&{0.048}&0&{ - 0.952}\\{0.236}&{0.239}&{0.278}&{0.275}&0\end{array}} \right], $
${{\mathit{\boldsymbol{\phi }}} _4} = \left[ {\begin{array}{*{20}{c}}0&{0.196}&{0.209}&{ - 1.149}&{ - 0.984}\\{ - 0.665}&0&{0.206}&{ - 1.221}&{ - 0.645}\\{ - 0.707}&{ - 0.696}&0&{ - 1.345}&{ - 1.102}\\{0.339}&{0.360}&{0.397}&0&{0.342}\\{0.290}&{0.190}&{0.325}&{ - 1.160}&0\end{array}} \right].$
步骤10?通过式(9)计算工程技术Ei优于Eh的综合占优度δ(Ei, Eh),从而得到综合占优度矩阵
$\mathit{\boldsymbol{\delta }} = \left[ {\begin{array}{*{20}{c}}0&{0.485}&{ - 1.698}&{ - 3.006}&{ - 3.038}\\{ - 2.138}&0&{ - 2.209}&{ - 3.293}&{ - 2943}\\{ - 0.863}&{ - 0.255}&0&{ - 2.364}&{ - 3.141}\\{ - 0.349}&{ - 0.196}&{ - 0.359}&0&{ - 1.922}\\{ - 0.310}&{ - 0.268}&{ - 0.124}&{ - 0.654}&0\end{array}} \right]$
步骤11?通过式(10)获取工程技术Ei全局占优势度ξi:ξ1=0.360,ξ2=0, ξ3=0.429, ξ4=0.841, ξ5=1.000.因此,根据全局占优势度ξi对工程技术重要度顺序为:E5>E4>E3>E1>E2.
3.2 损失规避系数的影响损失规避系数θ反映专家对损失的规避心理行为,文献[23]大量实验研究表明损失规避系数θ取值[1.0, 2.5]是更为合理的,为验证系数θ对工程技术重要度排序结果的影响,通过选取θ=1.0, θ=1.5, θ=2.0, θ=2.5对本案例进行分析,并计算各工程技术全局占优势度和重要度排序结果,如表 3所示.
表 3
表 3 针对不同系数θ的工程技术全局占优度和重要度排序 Tab. 3 Overall dominance and importance ranking of ECs with different values of θ θ各工程技术全局占优度ξi重要度排序
E1 E2 E3 E4 E5
θ=1.0 0.360 0 0.429 0.841 1.000 E5>E4>E3>E1>E2
θ=1.5 0.384 0 0.419 0.847 1.000 E5>E4>E3>E1>E2
θ=2.0 0.405 0 0.411 0.837 1.000 E5>E4>E3>E1>E2
θ=2.5 0.421 0 0.404 0.834 1.000 E5>E4>E1>E3 >E2
表 3 针对不同系数θ的工程技术全局占优度和重要度排序 Tab. 3 Overall dominance and importance ranking of ECs with different values of θ
由表 3可知,工程技术E5、E4和E2的重要度排序结果不受损失规避系数θ的影响,一定程度证明该案例运用本文提出的方法得到的排序结果具有一定稳定性.另外,随着损失规避系数θ的递增,工程技术E3和E1的排序发生了互换,意味着专家不同风险态度对工程技术重要度排序结果有重要影响,因此,在现实工程技术重要度排序过程中,需要根据专家对损失的不同规避程度来合理确定适当的θ值.
3.3 比较和分析为验证本文提出的改进QFD方法的合理性和有效性,将其与传统QFD方法、文献[2]提出的改进QFD方法作对比分析,工程技术重要度排序结果如图 2所示.
Fig. 2
图 2 不同QFD方法的工程技术重要度排序比较 Fig. 2 Importance ranking of ECs under different QFD methods
由图 2可知,虽然3种方法获得的工程技术重要度排序结果不完全相同,但都将E2和E5分别确定为重要度最高和最低的工程技术,一定程度上验证了本文方法的有效性.与传统FMEA方法相比,本文提出的方法重要度排序顺序E3和E4发生互换,而与文献[2]方法排序结果对比,则是E3和E1的排序产生变化.导致这些差异可能原因为:
1) 对比传统QFD采用清晰的数字和文献[2]的方法采用的HFLTS表征顾客需求和工程技术之间的关联程度,本文使用的PLTS更能准确真实地反映专家评估信息的模糊性和不确定性,避免信息的丢失;
2) 对比传统QFD忽略了顾客需求之间存在的关联关系,文献[2]的方法采用的犹豫模糊DEMATEL分析顾客需求间的自相关关系,本文综合PL-MAHP和FCM确定顾客需求的权重,充分考虑了专家主观经验知识,更真实地描述了顾客需求间存在的正、负影响关系和动态变化.
3) 传统QFD和文献[2]分别使用加权平均算法和犹豫模糊VIKOR确定工程技术的重要度排序,都是建立在假设专家是完全理性的基础上,忽略了专家心理行为在评估过程中发挥的重要作用,本文采用PL-TODIM方法考虑了专家对损失规避的心理行为,使得到的工程技术重要度排序结果更加合理可靠.
4 结论1) 本文提出一种概率语言环境下考虑专家心理行为的QFD方法.运用PLTS评估顾客需求和工程技术之间的关联程度,不仅适应了专家的语言表达习惯,而且还能解决专家评估信息丢失的问题,并真实地刻画了专家评估信息的模糊性和不确定性.
2) 从专家主观经验知识和顾客需求间客观关联两方面出发,综合PL-MAHP和FCM确定顾客需求的相对重要性,更真实的描述了顾客需求间存在的正、负影响关系和动态变化.
3) 基于PL-TODIM对工程技术的重要度排序,不仅避免了传统QFD中加权平均算法的不合理性,而且考虑了专家对损失规避的心理行为,使得工程技术的重要度排序结果更贴近实际情况.
尽管本文所提的QFD方法为产品规划提供了一种有效实用的工具,但仍有一些问题需要在未来研究中加以解决:在实际情况中,工程技术之间彼此存在关联性,因此在未来研究中可以考虑工程技术之间自相关关系对工程技术的重要度排序影响;可以考虑其他MCDM方法整合到QFD中以获取更准确的工程技术重要度排序;可以引入共识达成过程评估以改善QFD团队专家意见分歧问题.
参考文献
[1] KWONG C K, BAI H. Determining the importance weights for the customer requirements in QFD using a fuzzy AHP with an extent analysis approach[J]. IIE Transactions, 2003, 35(7): 619. DOI:10.1080/07408170304355
[2] WU Songman, LIU Huchen, WANG Lien. Hesitant fuzzy integrated MCDM approach for quality function deployment: a case study in electric vehicle[J]. International Journal of Production Research, 2017, 55(15): 4436. DOI:10.1080/00207543.2016.1259670
[3] WANG C H. Incorporating the concept of systematic innovation into quality function deployment for developing multi-functional smart phones[J]. Computers and Industrial Engineering, 2017, 107: 367. DOI:10.1016/j.cie.2016.07.005
[4] BUTTIGIEG S C, DEY P K, CASSAR M R. Combined quality function deployment and logical framework analysis to improve quality of emergency care in Malta[J]. International Journal of Health Care Quality Assurance, 2016, 29(2): 123. DOI:10.1108/IJHCQA-04-2014-0040
[5] JIA Weiqiang, LIU Zhenyu, LIN Zhiyun, et al. Quantification for the importance degree of engineering characteristics with a multi-level hierarchical structure in QFD[J]. International Journal of Production Research, 2015, 54(6): 1627. DOI:10.1080/00207543.2015.1041574
[6] YU Liying, WANG Linyu, BAO Yuhang. Technical attributes ratings in fuzzy QFD by integrating interval-valued intuitionistic fuzzy sets and Choquet integral[J]. Soft Computing, 2018, 22(6): 2015. DOI:10.1007/s00500-016-2464-8
[7] 杨强, 李延来. 基于直觉模糊集的质量屋顾客需求优先度及灵敏度分析[J]. 计算机集成制造系统, 2018, 24(4): 978.
YANG Qiang, LI Yanlai. Analysis method of customer requirements' priority ratings and sensitivity based on intuitionistic fuzzy sets in house of quality[J]. Computer Integrated Manufacturing Systems, 2018, 24(4): 978.
[8] HAKTANIR E, KAHRAMAN C. A novel interval-valued Pythagorean fuzzy QFD method and its application to solar photovoltaic technology development[J]. Computers & Industrial Engineering, 2019, 132: 361. DOI:10.1016/j.cie.2019.04.022
[9] RODRIGUEZ R, MARTINEZ L, HERRERA F. Hesitant fuzzy linguistic term sets for decision making[J]. IEEE Transactions on Fuzzy Systems, 2012, 20(1): 109. DOI:10.1109/TFUZZ.2011.2170076
[10] PANG Qi, WANG Hai, XU Zeshui. Probabilisticlinguistic term sets in multi-attribute group decision making[J]. Information Sciences, 2016, 369: 128. DOI:10.1016/j.ins.2016.06.021
[11] WU Xingli, LIAO Huchang. An approach to quality function deployment based on probabilistic linguistic term sets and ORESTE method for multi-expert multi-criteria decision making[J]. Information Fusion, 2018, 43: 13. DOI:10.1016/j.inffus.2017.11.008
[12] YAZDANI M., CHATTERJEE P, ZAVADSKAS E, et al. Integrated QFD-MCDM framework for green supplier selection[J]. Journal of Cleaner Production, 2017, 142(3728). DOI:10.1016/j.jclepro.2016.10.095
[13] TAVANA M, YAZDANI M, CAPRIO D D. An application of an integrated ANP-QFD framework for sustainable supplier selection[J]. International Journal of Logistics, 2017, 20(3): 254. DOI:10.1080/13675567.2016.1219702
[14] HOSSEINI M S M, BEHZADIAN M, IGNATIUS J, et al. Fuzzy PROMETHEE GDSS for technical requirements ranking in HOQ[J]. The International Journal of Advanced Manufacturing Technology, 2015, 76(9/10/11/12): 1993. DOI:10.1007/s00170-014-6233-5
[15] SONG Wenyan, MING Xinguo, HAN Yi. Prioritising technical attributes in QFD under vague environment: a rough-grey relational analysis approach[J]. International Journal of Production Research, 2014, 52(18): 5528. DOI:10.1080/00207543.2014.917213
[16] GOMES L F A M, LIMA M P. TODIM: Basics and application to multicriteria ranking of projects with environmental impacts[J]. Foundations of Computing and Decision Sciences, 1992, 16(4): 113.
[17] WU Xingli, LIAO Huchang, XU Zeshui, et al. Probabilistic linguistic MULTIMOORA: a multi-criteria decision making method based on the probabilistic linguistic expectation function and the improved Borda rule[J]. IEEE Transactions on Fuzzy Systems, 2018, 26(6): 3688. DOI:10.1109/TFUZZ.2018.2843330
[18] MAO Xiaobing, WU Min, DONG Jiuying, et al. A new method for probabilistic linguistic multi-attribute group decision making: application to the selection of financial technologies[J]. Applied Soft Computing, 2019, 77: 155. DOI:10.1016/j.asoc.2019.01.009
[19] LOOTSMA F A. Scale sensitivity in the multiplicative AHP and SMART[J]. Journal of Multi-Criteria Decision Analysis, 1993, 2(2): 87. DOI:10.1002/mcda.4020020205
[20] TRIANTAPHYLLOU E. Two new cases of rank reversals when the AHP and some of its additive variants are used that do not occur with the multiplicative AHP[J]. Journal of Multi-Criteria Decision Analysis, 2001, 10(1): 11. DOI:10.1002/mcda.284
[21] BAYKASOGLU A, GOLCUK I. Development of a novel multiple-attribute decision making model via fuzzy cognitive maps and hierarchical fuzzy TOPSIS[J]. Information Sciences, 2015, 301: 75. DOI:10.1016/j.ins.2014.12.048
[22] HAJEK P, FROELICH W. Integrating TOPSIS with interval-valued intuitionistic fuzzy cognitive maps for effective group decision making[J]. Information Sciences, 2019, 485: 394. DOI:10.1016/j.ins.2019.02.035
[23] KAHNEMAN D, TVERSKY A. Prospect theory: an analysis of decision under risk[J]. Econometrica, 1979, 47(2): 263.