Evaluaton of Qualty Management Performance n Offce of Presdent usng Modfed Publc Sector Management Qualty Award (PMQA) Model Thoedtda Thpparate 1 untcha Kongkaew, Sunantha. Ompan 1. Lecturer, Faculty of Management Scences, Prnce of Songkla Unversty Hatya, Songkhla, Thaland Emal: thoedtda.t@gmal.com Abstract Ths paper presents the applcaton of adaptve neuro-fuzzy nference systems (AFIS), usng Sugeno AFIS to measure the qualty management performance of Offce of Presdent (OOP). Attrbutes assocated wth Publc Sector Management Qualty Award (PMQA) crtera are consdered as the nput and output varables for the man-model and sub-models, respectvely. The dependent varable s the qualty management performance. Staffs n the Offce of Presdent provded data for ths study. The result can be used to mprove ts qualty as well as to acheve hgh performance organzaton. Keywords: qualty, adaptve neuro-fuzzy nference systems, Sugeno AFIS Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 1
1. Introducton Most educatonal projects are of unque and dynamc nature. The educatonal project has been contnually crtczed for not achevng the level of mprovement n performance and productvty shown by other types of projects. The Offce of Presdent has been under consderable pressure to mprove the effcency of the educatonal process. Pressure s also ncreasng from clents who demand better outputs and outcomes n shorter duraton. Ths has created a unversty need for reform to challenge the change for qualty management n the Offce of Present. ew challenges requre new approaches. A vson of change for qualty management wthn the perspectve of revalung the Offce of Presdent s very necessary to develop a culture of contnuous mprovement. The global compettveness, organzatonal culture and change, usage of IT, performance measures and benchmarkng for contnuous mprovement, best practces for educatonal management, and sustanable development are the key ssues affectng educatonal projects. Durng the 1990s, the man emphass of Thaland s manufacturng ndustry was on mplementng ISO 9000 standards. Although dfferent TQM approaches and tools, such as just-n-tme, total productve mantenance, have been appled snce the 1980s [1], a number of foregn-owned companes wthn the electroncs sector, and few Tha-owned groups, have successfully mplemented TQM [2]. Consderng TQM postonng, Thaland ranks n the mddle of the developng countres of Southeast Asa. Its status n TQM s hgher than Indonesa or Phlppnes, and lower than Malaysa or Tawan [2]. For the mplementaton of TQM n Thaland, lttle emprcal research has been rarely conducted n Tha manufacturng companes. The current stuaton of TQM mplementaton n Tha manufacturng companes s vague. As a result of the fact that the emprcal studes n the area of TQM mplementaton are lmted, t s dffcult for Tha manufacturng companes to obtan suffcent nformaton to support ther TQM mplementaton process. Thus, Tha manufacturng companes are experencng numerous dffcultes, and even falures n mplementng TQM. Many ndustres should take up the concept of total qualty management (TQM) n order to mprove ther performance, especally for the constructon ndustry. As Tyler and Frost [3] have ponted out, qualty assurance (QA) has been taken serously only recently n the UK constructon ndustry, and even then only n the large contractng companes. For example, Bhattar [4] showed that none of the medum-szed constructon companes had ntroduced QA systems. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 2
One of the key emphases of a qualty award s, for a company to acheve sustanable fnancal success. In the nstance of MBQA, the award wnnng frms reported a 44% hgher stock-prce return, 48% hgher growth n operatng ncome, and a 37% hgher growth n sales than the control group of frms [5]. Organzatons are usng varous crtera to help them durng mplementaton efforts to evaluate themselves aganst crtera to determne how well ther mprovement efforts are progressng. Sets of crtera that the majorty of organzatons uses nclude Demng prze categores, Juran s ten ponts, Crosby s fourteen ponts, and the MBQA crtera [6]. The present study wll adopt a case from Thaland for examples. Accordngly, Publc Sector Management Qualty Award (PMQA) crtera wll be adopted as the qualty management performance evaluaton crtera. The PMQA has a herarchcal structure. It has seven strategc crtera. Each strategc crteron has ts assocated sub-crtera. Currently, several studes have been done on development of the forecastng models whch have focused on engneerng performance, qualty performance, nnovaton performance and sustanablty performance. However, the approprate ndcators for the Offce of Presdent are rarely concerned. Therefore, the ntegraton of qualty management objectves and other objectves are not determned. Wang [7] elaborated an nnovaton performance forecastng model wng an adaptve neural network (A), t stll exsted some problems, such as choosng the nfluence ndcators, solvng the lngustc character of sources, explanng the tranng procedure of outcome, descrbng how to smulate the rules for predcton, and fndng robust forecastng technques. To overcome these drawbacks, Chen, Wang et, al. (2010) [8] proposed an adaptve neuro-fuzzy nference systems (AFIS) to measure the nnovaton performance through knowledge management objectve and nnovaton objectve. However, many scholars have suggested that both total qualty management (TQM) and organzatonal learnng can ndvdually and effectvely promote nnovaton. The study on the measurement of mpact of TQM on the qualty management performance of the Offce of Presdent s not much concerned. The research focus s to nvestgate the factors affectng qualty management performance n the Offce of Presdent. The man objectve s to (1) determne the ntegraton of PMQA crtera, (2) evaluate mpacts of PMQA crtera on qualty management performance (3) propose solutons for hghway project manager to mprove the project s qualty as well as to acheve hgh performance organzaton. A newly developed system s to automate and enhance the process of determnng or measurng a reasonable qualty performance for projects from the offcers' perspectve. Projects developed by the Offce of Presdent have been used as data sources for ths study. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 3
2. Research Framework As mentoned earler, the objectves of ths study are; frst, to determne the ntegraton of PMQA crtera. The PMQA has seven strategc crtera. Each strategc crteron has ts assocated sub-crtera. Ths study used the PMQA crtera as a gudelne for developng a model. Fgure 1 presents 31 sub-crtera and the ntegraton among sub-crtera or sub-attrbutes assocated wth each crteron or attrbute n a proposed model. In addton, the ntegraton among attrbutes was consdered. The second objectve s to measure mpact of PMQA crtera on the qualty management performance of the Offce of Presdent. In order to realze ths objectve, a research framework s developed. The framework s a smple lnear model of the relatonshp between the ndependent and dependent varables. Attrbutes assocated wth PMQA crtera are consdered as the ndependent and dependent varables for the man-model and sub-models, respectvely. For the man-models, the ndependent varables conssts of sx blocks ncludng 1) leadershp (LD), 2) strategy plannng (SP), 3) customer and stakeholder (CS), 4) nformaton technology (IT), 5) human resource (HR), and 6) process management (PM). The dependent varable s the qualty management performance consdered from the combnaton of sx varables ncludng 1) prestge measurement, 2) unque compettve ablty ganng performance, 3) customer satsfacton, 4 nformaton management performance, 5) human resource development performance, and 6) process management performance. In total, there are 31 sub-crtera as shown n Fg.1. For the sub-models, the ndependent varables assocated wth each of these sx blocks are determned separately. For example, the leadershp model has fve ndependent varables ncludng 1) operatonal value and deals restructure, 2) organzatonal qualty mprovng msson, 3) top managers leadng style, 4) qualty culture constructon and, 5) ncreasng of socal contrbuton. Dependent varable s Prestge measurement. Sub-attrbutes assocated wth each block are shown n Fg 2. 3. euro-fuzzy Model The neuro-fuzzy system attempts to model the uncertanty n the factor assessments, accountng for ther qualtatve nature. A combnaton of classc stochastc smulatons and fuzzy logc operatons on the A nputs as a supplement to artfcal neural network s employed. Artfcal eural etworks (A) has the capablty of self-learnng, whle fuzzy logc nference system (FLIS) s capable of dealng wth fuzzy language nformaton and smulatng judgment and decson makng of the human bran. It s currently the research focus to combne A wth FLIS to produce fuzzy network system. AFIS s an example of such a readly avalable system, whch uses A to accomplsh fuzzfcaton, fuzzy nference and defuzzfcaton of a fuzzy system. AFIS utlzes A s learnng mechansms to draw rules from nput and output data pars. The system possesses not Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 4
only the functon of adaptve learnng but also the functon of fuzzy nformaton descrbng and processng, and judgment and decson makng. AFIS s dfferent from A n that A uses the connecton weghts to descrbe a system whle AFIS uses fuzzy language rules from fuzzy nference to descrbe a system. The AFIS approach adopts Gaussan functons (or other membershp functons) for fuzzy sets, lnear functons for the rule outputs, and Sugeno s nference mechansm [9]. The parameters of the network are the mean and standard devaton of the membershp functons (antecedent parameters) and the coeffcents of the output lnear functons as well (consequent parameters). The AFIS learnng algorthm s then used to obtan these parameters. Ths learnng algorthm s a hybrd algorthm consstng of the gradent descent and the least-squares estmate. Usng ths hybrd algorthm, the rule parameters are recursvely updated untl an acceptable level of error s reached. Each teraton ncludes two passes, forward and backward. In the forward pass, the antecedent parameters are fxed and the consequent parameters are obtaned usng the lnear least-squares estmaton. In the backward pass, the consequent parameters are fxed and the error sgnals propagate backward as well as the antecedent parameters are updated by the gradent descent method. Based on the orgnal AFIS study [10]; the learnng mechansms should not be appled to determne membershp functons n the Sugeno AFIS, snce they convey lngustc and subjectve descrptons of possbly ll-defned concepts. Hence, the choce of membershp functon should depend on the specfc types of data. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 5
Organzaton of operaton strategy plannng (C6) Operaton structure adjustment (C7) The qualty mprovement of strategy (C8) Operatonal values and deals restructure (C1) Organzatonal qualty mprovng msson (C2) Strategy plannng (SP) Human resource plannng (C18) Human resource development (C19) Human resources utlzaton (C20) Employee relatonshp management (C21) Knowledge management (C22) Human resource (HR) Prestge measurement (C33) Unque compettve ablty ganng performance (C32) Top managers leadng style (C3) Qualty culture constructon (C4) Leadershp performance (LD) Customer satsfacton (C26) Informaton management performance (C30) The ncreasng of socal contrbuton (C5) Human resource development performance (C29) Market operaton strategy development (C9) Manufacturng process management (C23) Process management performance (C31) Busness relaton management (C10) Customer relatonshp management (C11) Customer and stakeholder (CS) Supportve actvty plannng (C24) Cross-unt (Department) Management (C25) Process management (PM) Process redefnton of R&D and nnovaton (C12) Input of R&D and nnovaton (C13) Evaluaton of R&D and Internet applcatons nnovaton results (C14) (C16) Unversty technology nformaton recevng channel (C15) Unversty technology nformaton utlzaton (C17) Informaton technology (IT) Fg 1 Crtera and sub-crtera of Publc Sector Management Qualty Award (PMQA) model Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 6
Operatonal values and deals restructure (C1) C1A1 C1A2 C1A3 R1 R2 R3 R4 Organzatonal qualty mprovng msson (C2) C2A1 C2A2 C2A3 R5 R6 R7 R8 Top managers leadng style (C3) C3A1 C3A2 C3A3 R9 R10 R11 R12 Leadershp performance (LD) Qualty culture constructon (C4) C4A1 C4A2 C4A3 R13 R14 R15 R16 The ncreasng of socal contrbuton (C5) C5A1 C5A2 R17 R18 R19 C5A3 R20 Fg 2 etwork of leadershp performance by the AFIS 3.1. AFIS Archtecture The acronym AFIS derves ts name from adaptve neuro-fuzzy nference system. The AFIS has fve layers. The frst layer calculates the degree of membershp of all nputs. The second layer calculators examne the ftness of each rule. The thrd layer calculators determne the normalzed value of the ftness. The fourth layer calculates the output of each rule. The ffth layer produces the output of the fuzzy system. In ths network, both the characterstc parameters and the concluson parameters are ncluded. Durng the tranng process, AFIS dynamcally adjusts these parameters. As a result, the network can accurately descrbe the mappng between the nput and output data [11]. An AFIS archtecture s equvalent to a two-nput frst-order Sugeno fuzzy model wth nne rules, where each nput s assumed to have three assocated membershp functons (MFs) [10]. 3.2. Input/output Indcators The frst step n the modelng process based on an AFIS nvolves the dentfcaton of the nput and output varables. The nput/output ndcators are the nput/output vectors of the AFIS. Generally, the decson makers adopt the nput vector, along wth output vector, to tran the AFIS and subsequently to obtan the weghts. For example f a two nput (x, y) Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 7
and one output (f) fuzzy nference system s consdered, then for a frst-order Sugeno fuzzy model, a typcal rule set wth two fuzzy IF/THE rules can be expressed as Rule 1: If x s A 1 and y s B 1, then m1 x1 y1 z1 Rule 2: If x s A 2 and y s B 2, then m2 x2 y2 z2 where x1, x2, y1, y2, z1 and z 2 are lnear parameters, and A 1, A 2, B 1 and B 2 are nonlnear parameters. The AFIS structure ncludes fve layers as shown n Fg 3. The frst layer calculates the degree of membershp of all nputs. The second layer calculators examne the ftness of each rule. The thrd layer calculators determne the normalzed value of the ftness. The fourth layer calculates the output of each rule. The ffth layer produces the output of the fuzzy system. The nput, output and node functons of each layer, s dscussed n the subsequent sectons. Each of these layers conssts of a number of nodes connected through drect lnks. Each node represents a process unt, and the lnks between nodes specfy the causal relatonshp between the connected nodes. Some parts of the nodes are adaptve whle others are fxed. Adaptve nodes are denoted by usng square box whle the crcular nodes denote fxed nodes. Each adaptve node has a set of parameters that performs a partcular functon (node functon) on the ncomng sgnal. The learnng rules specfy the parameters of adaptve nodes that need to be changed to mnmze a prescrbed error measure. However, the change n values of the parameters results n change n the shape of membershp functons assocated wth the fuzzy nference system. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 8
Fg 3. The AFIS structure Layer 1 s the fuzzy layer, n whch m and n are the nput of nodes A1, B1 and A2, B2 respectvely. A1, A2, B1 and B2 are the lngustc labels used n the fuzzy theory for dvdng the membershp functons. The membershp relatonshp between the output and nput functons of ths layer can thus be expressed as: m O1, A ( = 1, 2) (1) O 1, j B j n (j = 1, 2) (2) where, O 1, and O 1,j denote the output functons and A and B j denote the membershp functons. Also, each node n Layer 2 multples the ncomng sgnals from the varous nputs. In layer 2, each node s a representaton of one rule and the nputs are degrees of membershp functons whch are multpled through a T-norm operator to determne the degree of fulfllment of w of the rule. Hence each node output s the frng strength of the rule. m n O2, w A B ( = 1, 2) (3) where, O 2, denotes the output of layer 2. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 9
Further, all the nodes n thrd layer are fxed nodes and for every th node, the rato of the th rule's frng strength n nference layer to the sum of all the rules' frng strengths s calculated as O 3, w w 2 1w ( = 1, 2) (4) The output, O 3, of ths layer s called the normalzed frng strength. Furthermore, the fourth layer s the de-fuzzy layer whose nodes are adaptve. The output equaton for ths layer becomes w xm yn z, where x, y and z represent the lnear parameters or so called consequent parameters of the node. The de-fuzzy relatonshp between the nput and output of ths layer can thus be defned as: O 4, w f w x m y n z where, O 4, denotes the output of layer 4. ( = 1, 2) (5) The fnal output n the layer 5 computes the summaton of all ncomng sgnals. Thus there s just one node n ths layer wth a smple summng functon, defnes as: w f 5 w f w ( = 1, 2) (6) O, where, O 5, denotes the output of layer 5. For the man-model, nput varables nclude 1) leadershp (LD), 2) strategy plannng (SP), 3) customer and stakeholder (CS), 4) nformaton technology (IT), 5) human resource (HR), and 6) process management (PM). The output varable s the qualty management performance. There are sx sub-models assocated wth the nput varables of the man-model. Table 1 shows the nput/output varables of the sub-models. Ths study consders factors havng the mpact on qualty management performance, thus attrbutes assocated wth the qualty management performance become the nput varables. 3.3. Tranng AFIS For provng the applcablty of the model and llustraton, the proposed model was appled n thrty of the research projects n Thaland. The frst step to apply the model was to construct the decson team. The decson team ncluded offcers n Offce of Presdent. For tranng the AFIS, some experences about the system behavor are necessary. For ths am, a questonnare was desgned ncludng dfferent combnatons of crtera. The decson team Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 10
was asked to gve a score to them f possble at all, based on ther knowledge about the system. Then, they rated the educatonal projects wth respect to each crteron. The nput parameters are represented on the unt unverse [0,10] wth trangular or trapezodal membershp functons descrbng the lngustc varables such as the operaton performance, for example: "low", "medum" or "hgh" (Fg. 4). The system was bult n the Matlab Fuzzy Toolbox and Smulnk envronment. A Matlab programme was generated and compled. The pre-processed nput/output matrx whch contaned all the necessary representatve features, was used to tran the fuzzy nference system. Fg 5 show membershp functons descrbng the lngustc varables such as the Operatonal values and deals restructure (C1), for example: "low", "medum" or "hgh" after tranng the network. Fg 6 shows the structure of the AFIS; a Sugeno fuzzy nference system was used n ths nvestgaton. Based on the collected data, 50 data sets were used to tran the AFIS and the rest (41) for checkng and valdaton of the model. For the frst sub-model (Leadershp) for 5 attrbutes generated by the proposed approach, the decson makers derved 6 completed questons, as much as possble for decson team to answer. Subtractve clusterng was used for rule generaton, where the range of nfluence, squash factor, acceptance rato, and rejecton rato were set at 0.5, 1.25, 0.5 and 0.15, respectvely durng the process of subtractve clusterng. The traned FIS ncludes 14 rules (clusters). Because by usng subtractve clusterng, nput space was categorzed nto 14 clusters. Each nput has 14 Gaussan curve bult-n membershp functons. Fg 6 shows the surface of AFIS after tranng. The tranng error of the AFIS was 0.8074 after 30 epochs as shown n Fg 7. The traned fuzzy nference system ncludes 14 rules (clusters) as present n Fg 8. Table 1 Input/output varables of sub-models Sub-model Input varable Output varable Operatonal values and deals restructure (C1) Organzatonal qualty mprovng msson (C2) Leadershp Top managers leadng style (C3) Prestge measurement (LD) qualty culture constructon (C4) The ncreasng of socal contrbuton (C5) Strategy Organzaton of operaton strategy plannng (C6) Unque compettve ablty plannng (SP) Operaton structure adjustment (C7) The qualty mprovement of strategy (C8) ganng performance Customer and Market operaton strategy development (C9) Customer satsfacton stakeholder Busness relaton management (C10) Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 11
(CS) Informaton technology (IT) Human resource (HR) Process management (PM) Result based management (RM) Customer relatonshp management (C11) Process redefnton of R&D and nnovaton (C12) Input of R&D and nnovaton (C13) Evaluaton of R&D and nnovaton results (C14) Constructon technology nformaton recevng channel (C15) Internet applcatons (C16) Constructon technology nformaton utlzaton (C17) Human resource plannng (C18) Human resource development (C19) Human resources utlzaton (C20) Employee relatonshp management (C21) Knowledge management (C22) Manufacturng process management (C23) Supportve actvty plannng (C24) Cross-unt (Department) Management (C25) Customer satsfacton (C26) Human resource development performance (C27) Informaton management performance (C28) Process management performance (C29) Unque compettve ablty ganng performance (C30) Prestge measurement (C31) Informaton management performance Human resource development performance Process management performance Qualty management performance Fg 4. Membershp functons of the leadershp sub-model (before Tranng) Fg 5. Membershp functons of the operaton performance (after Tranng) Fg 9 depcts a three dmensonal plot that represents the organzatonal qualty mprovng msson (n2) and ncreasng of socal contrbuton (n5) to prestge measurement (out1). The assumpton s that the collected data are representatve of the features of the data that the Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 12
traned FIS s ntended to model. However, the accuracy of the model s affected by the nherently nsuffcent and tranng data whch cannot cope wth every feature of the data that should be presented to the traned model. Fg 6. etwork of leadershp performance by the AFIS Fg 7. tranng error of the AFIS Fg. 8. Traned man AFIS surface of supply chan performance Fg. 9. Traned man AFIS surface: qualty management performance 4. Concludng Remarks A major objectve of ths paper was to propose an AFIS model to predct the qualty management performance for educatonal projects n the Offce of Presdent. The fndngs confrm factors assocated wth the PMQA crtera are the sx nput dmensons, whle qualty management performance s the output varable for the proposed model. To face the challenge n the dynamc envronment, offcers should understand the qualty actvtes so as to sustan ther compettveness. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 13
Ths paper examnes a method named Adaptve uero-fuzzy Inference System (AFIS) as a tool to capture expert judgements on how the qualty management performance can be assessed. Educatonal projects are used to demonstrate the models developed regardng the PMQA crtera. For example, fve ndependent varables assocated wth the leadershp performance are modeled usng the proposed model. The value of leadershp performance can then be calculated. From ths study t can be recommended that the AFIS can be used as tool to assess the qualty management performance. However, each project s unque. The forecasted values are depended on a combnaton of expertse and the actual stuaton of a partcular educatonal project beng consdered. The developed model can provde the value used as a gudelne for assessng the qualty management performance. 5. References [1] Department of Trade and Industry, Management Best Practce Managng nto the 90s, Department of Trade and Industry, 1990. [2] Insttuton of Structural Engneers, Gude to Good Management Practce for Engneerng Desgn Offces, Insttuton of Structural Engneers, March 1991. [3] Tyler, A.H. and Frost, D.T. (1993) Implementaton of a constructon ndustry qualty assurance system, Internatonal Journal of Qualty & Relablty Management, Vol.10, o.4, pp.9-18. [4] Bhattar, D.P. (1990) Problems of qualty control n constructon. MSc thess, Loughborough Unversty of Technology, 1990. [5] Krasachol, L., Wlly, P.C.T. and Tannock, J.D.T. (1998) The progress of qualty management n Thaland, The TQM Magazne, Vol.10, o.1, pp.40-4. [6] Tabucanon, M.T. (1993) Thaland s manufacturng sector: ssues on development, technology, and management, Journal of Manufacturng Systems, Vol.12, o.3, pp.199-203. [7] Wang, T.Y., and Chen, S.C. (2006) Forecastng nnovaton performance va neural networks A case of Tawanese manufacturng ndustry, Technovaton, Vol.26, pp. 635 643. [8] Chen, S.C., Wang,T.Y., and Lu, S.L. (2010) Applcaton of neuro-fuzzy network to forecast nnovaton performance The example of Tawanese manufacturng ndustry, Expert Systems wth Applcatons, Vol.37, pp.1086 1095. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 14
[9] Takag,T., and Sugeno, M. (1985) Fuzzy dentfcaton of systems and ts applcaton to modelng and control, IEEE Transactons on Systems Man and Cybernetcs, Vol.15, pp.116 132. [10] Jang, J.S.R. (1993) AFIS: adaptve- network based fuzzy nference system, IEEE Transactons on Systems Man and Cybernetcs,Vol.23, pp.665-685. [11] Zhang, Z., Dng, D., Rao, L., and B,Z. (2006) An AFIS based approach for predctng the ultmate bearng capacty of sngle ples. Foundaton Analyss and Desgn : Innovatve Methods (GSP 153) ASCE. Aprl 21 st, 2012 Faculty of Lberal Arts, Prnce of Songkla Unversty 15