ACTIVITY BASED COSTING FOR MARITIME ENTERPRISES 1, a 2, b 3, c 4, c Sualp Omer Urkmez David Sockon Reza Ziarai Erdem Bilgili a, b De Monfor Universiy, UK, c TUDEV, Insiue of Mariime Sudies, Turkey 1 sualp@furrans.com.r 4 erdembilgili@udev.com 2 sockon@dmu.ac.uk 3 rziarai@udevedu.com Absrac. This paper is concerned wih he developmen of an Aciviy Based Cosing (ABC) sysem for use by SMEs in he shipping indusry in Turkey. I repors on he applicaion on an Arificial Neural Nework in ABC/ABM wih he inenion of deermining coss more accuraely. Currenly many companies are using cosing sofware and some projec managemen or scheduling sofware o define aciviies and iming and in he majoriy of he case companies leave he cosing o a parallel exercise where each aciviy is cosed using hisorical or empirical daa. The analysis of cosing sysems in he ship building companies has shown ha he hisorical daa has no been effecively used for fuure ship building projecs cosing. A review of neural nework applicaions indicaed ha such neworks could provide a means of accumulaing hisorical daa and also a decision making ool in cosing aciviies and deermining ime aken for each of hese aciviies. To his end, a neural nework which had been used successfully elsewhere was adaped in he developmen of an ABC sysem. This paper repors how daa used in consrucing five ships was applied o configure a neural nework wih he inenion of esablishing, on he one hand, a relaionship beween he cos of he aciviies in building a ship and he ship s ideniy parameers (lengh, widh, onnage, ec.) and on he oher hand, o esablish a relaionship beween he hours of he aciviies for a ship applying he same ideniy parameers. Keywords: Aciviy Based Cosing, ship building and neural neworks. 1. Inroducion The research presened in his paper has shown ha he exising cosing sysems used in shipping indusry are no dissimilar o hose used in design and manufacure of oher engineering producs and services. However, he cosing sysems employed in he secor, invesigaed o-dae, do no provide ools ha can be used o eiher idenify or o accuraely deermine and/or allocae overhead coss (Cokins, 2001). This is of paricular ineres in he shipbuilding indusry where overheads are ofen esimaed using pas pracices and accoun for a large % of he overall coss. The second weakness and probably more serious is ha in convenional cosing sysems indirec coss are no disribued o each and every aciviy. Inroducion of Aciviy Based Cosing (ABC) echniques in recen years has shown ABC can provide an appropriae means of idenifying and deermining overhead coss, and allocaing hese o each and every aciviy in he design and manufacuring processes as well as associaed pre design and managemen aciviies (Ziarai, 1989). As ye, no cases have been
noed o apply ABC in he shipbuilding secor. There are also no published repors or papers on he applicaion of ABC echniques in he shipbuilding companies. Increase in he applicaion of ABC led o he realisaion ha some of he limiaions of radiional cos accouning sysems could be overcome by ABC cosing pracices. There are many papers reporing on how ABC could form he basis for an overall cosing pracice (Beajon and Singhal, 1990; Turney, 1992). ABC has primarily wo problems viz., i does no have general crieria o selec relevan cos drivers, and secondly ABC provides a linear sysem and hence unsuied in siuaions when cos behaviour shows non-linear characerisics (Ziarai, 1989). The inenion in his repor is o esablish general crieria for cos drivers and apply non-linear mechanisms such as neural neworks o cases where he cosing is non-linear. The analysis of cosing sysems in he shipbuilding indusry has shown ha he hisorical daa has no been effecively used for cosing he fuure shipbuilding projecs. This is because generally here are almos 400 aciviies in he manufacure of a ship. A review of Arificial Neural Nework (ANN) applicaions indicaed ha such neworks could provide a mechanism for accumulaing hisorical daa and be used o aid decision making process (Ziarai and Sockon, 2003). However, of equal imporance, if no more, is he ANN capabiliy of allowing cos relaionships for he inended ABC o be deermined. ANN herefore overcomes boh limiaion of ABC and provides a means of using pas daa in an effecive manner. Inpus Neural Nework Oupus Aciviy Cosing Daa Daa Time Taken Daa MODEL ARITHMATIC/ MATHEMATICAL AND/OR NUMERICAL Cos Relaionships Figure 1. The Basic Sysem for Esimaing Aciviy Cosing/Time Taken (Source: Ziarai, 2003) Cos Relaionships for he Inended ABC The firs ask was eiher o idenify an exising ANN or o develop one for knowledge acquisiion and presenaion. The lieraure search led o an exising neural nework (Ziarai, 2003, Bilgili, 2004) which had been used successfully elsewhere. This nework was adaped in he developmen of he ABC sysem. x 1 W k1 i n p u P a e r n x m-1 x m W km-1 W km SUM M ATION Bias ne TRANSFER FUNCTION y=f(ne) o n e x l a y e r b k S pnaic weighs Figure 2 General Block Diagram of he ANN (Source: Ziarai 2003)
The above ANN is designed o carry ou wo basic asks. The firs ask esablishes a relaionship beween he cos of he aciviies for a ship and ship s ideniy parameers such as lengh, widh ec. The second ask esablishes a relaionship beween he aciviy periods for consrucing a ship and again ideniy parameers as before. The research overall is concerned wih he developmen of an Aciviy Based Cosing (ABC) sysem for use by Small and Medium size Enerprises (SMEs) in he shipping indusry in Turkey. I represens a review of Arificial Inelligence (AI) echniques and pracices and is applicaion in ABC paricularly in deermining coss more accuraely. The ulimae aim of his work is o design and develop a more reliable cosing model. 2. Research Work The iniial work involved a review of cosing echniques, curren pracices and processes applied in he mariime SMEs in Turkey. Several companies were included in a preliminary phase which led o he developmen of a quesionnaire. The quesionnaire was used in followup inerviews wih managers wihin he secor. ANN can only produce accepable oupu (knowledge presenaion) if he inpus (knowledge eliciaion) are admissible. Afer all an ANN is a sysem and like all sysems i requires admissible inpus and sable sae o produce accepable oupus (Ziarai, 2003). The review of he oucomes of he survey clearly indicaes ha he companies no longer use laborious manual calculaions repored by Ziarai (1989). The emergence of inexpensive compuerised accouning sysems has provided opporuniies for many companies o updae heir previous spreadshee based pracices. The business of he companies involved in his phase of he invesigaion is in ship managemen and shipbuilding. I was ineresing o noe ha none of he companies were aware ha indirec overheads (managemen coss) may have been considered when allocaing coss o specific projecs and/or aciviies. The companies involved in ship managemen in he shipping indusry in Turkey were found o use similar cosing sofware and a given mehodology (Urkmez, 2007) which is in line wih he inernaional pracice. This inernaional pracice does no allocae indirec coss o projecs/aciviies. The companies involved wih shipbuilding were found (Urkmez, 2007) o have heir own mehod of cosing and make use of a variey of sofware produc wih varying degree of complexiy. However, no company of hose conaced/invesigaed currenly uses ABC/M or has a sysemaic mehod of disribuing indirec coss o each aciviy wihin a given projec. Typical cosing sysems used by companies, for differen cos headings are repored in Urkmez, (2007). Alhough hese companies have similar convenional cos sysems, here are wo essenial differences. In one approach more aenion is paid o deailing direc coss and in he oher more effor is exered in deailing indirec coss. I is easier o adap ABC for laer han he former. Analysis of he daa colleced so far shows ha he mehod ofen used in cosing ship building is relaively simple; he shipyard simply liss direc equipmen coss by use of spreadshee based sofware e.g. MS Excel. The labour coss are acceped as he average cos for building such a ship based on per uni kg in shipyards in Turkey. To follow he work flow and
esimae he delivery ime of he ship, aciviies are lised and an approximae period is aribued o each aciviy, leading o an esimaion of cos. Some Companies also make use of commercial cosing projec managemen and scheduling sofware o define aciviies and iming and leave he cosing o a parallel exercise where each aciviy is cosed using hisorical or empirical daa. One company used MS Projec sofware raher innovaively where all aciviies were ploed agains ime schedule and hen coss were esimaed for each aciviy. Generally, here are almos 400 aciviies in he manufacure of a ship. Alhough he aciviies are cosed, i is no an ABC sysem, as indirec coss have no been disribued o each aciviy. The analysis of cosing sysems of hese ship building companies has shown ha he hisorical daa has no been effecively used for fuure ship building projec cosing. A review of neural nework applicaions indicaed ha such neworks could provide a means of accumulaing hisorical daa and also a decision making ool. To his end, a neural nework which had been used successfully elsewhere (Ziarai, 2003) was adaped o develop an ABC sysem. 3. Daa collecion and analysis In he company seleced for measuring and disribuing he cos of manpower and he operaional coss of machines as accuraely as possible, i was necessary o design a form o gaher he necessary daa. A review of he arrangemens in various deparmens necessiaed wo forms o be developed. One form was for some deparmens which calculaed he cos for he whole deparmen raher han calculaing he cos of each member of saff individually. This form was given o he appropriae heads of deparmens who compleed he form on a deparmenal basis. The second form was given o he deparmens where he cos of each saff member could be calculaed individually. Once he coss were deermined hese coss were hen disribued o aciviies/projecs. In addiion, he daa obained using he wo forms were complemened by specific repors by deparmens, i.e. he daa from he CNC deparmen was cross-referenced wih he daa included in seel cuing repors. The accouns deparmen of his shipyard is preparing monhly maerial coss for each ship building projec, sub-conracor paymens, salaries of shipyard personnel and oher coss; his daa was colleced and sored using he MS-Access daabase projec. I is planned o calculae he exac coss of each aciviy in any of he ship building projecs. 4. Neural Nework Configuraion and Training This arificial neural nework underakes wo basic asks, i esablishes he relaionship beween he: i) cos of he aciviies for a ship and he ideniy parameers of he ship, e.g. lengh, widh, onnage, ec., and ii) hours of he aciviies for a ship and he ideniy parameers of he ship. Archiecures of boh neural nework configuraions are idenified excep for he weighs of he processing elemens wihin he neural neworks. The ANN has he following properies:
a) Consiss of muliple neurons. b) There are hree layers ermed inpu layers, hidden layers and oupu layers. The number of neurons a he inpu layer mus be equal o number of he inpu parameers so ha each inpu is represened by a given neuron. In he work presened here here are 11 inpu parameers. Each inpu parameer represens a specific propery such as lengh of he ship, widh of he ship, ec. Since here are 11 inpu parameers, he number of he nodes for he inpu layer of he nework is also 11 in order ha he above rule is saisfied. c) There are no resricions or analyical formula for he number of nodes in a hidden layer. I has been se o 20 nodes. d) For he oupu layer, he number of neurons mus be equal o he number of oupu parameers since each node represens an oupu parameer. There are 395 aciviies for ships being esed here. Therefore, he number of he neuron in he neural nework has been se o 395. e) Neural neworks process he normalised values of he inpu parameers and produce he normalised values of he oupu parameers. f) The neural nework mus be rained enabling reliable relaionships beween inpu and oupu parameers o be esablished. Alhough here are several raining mehods, he mos common mehod, i.e. back-propagaion, is used for hese neural neworks. The main aim in he back-propagaion learning algorihm is o achieve minimum Sum Square Error. 5. Discussions and Conclusions Figure 3 demonsrae ha neural nework reached a seady sae afer some 9000 epochs, reaching a olerance error of less ha 0.001 hence providing a sable sysem for he inended experimens. The Figure 4 shows he regression line represening he relaionship beween original raining daa and neural neworks oupu. As can be seen for ship 1 he wo ses of daa converge and have srong and posiive correlaion. The second figure shows relaive and average % error for all he 395 aciviies in consrucing he firs ship. Similar experimens were carried ou wih he oher four ships. The resuls are promising, clearly showing ha neural nework can reliably be used o predic he cos of each aciviy in building a vessel and also forecas he ime aken for each of hese aciviies. Similar experimens were carried ou for he four ships. Full aciviy chars idenifying cos of aciviies and ime aken for each of hese aciviies for all seven ships are available bu due o he shear size of hese chars, hese are no presened in his paper bu given in Urkmez (2007). The resuls obained were used in he consrucion of he wo new vessels. The research work is coninuing and is expeced he work would lead o improved daa gahering sysem enhancing he qualiy of he daa. In parallel he neural nework developed as par of his research programme is being incorporaed ino a knowledge-based-sysem wih a view o improve he qualiy and reliabiliy of forecased aciviy coss and aciviy ime as well as abiliy o predic lead-imes and projec managemen of ships buil in he fuure.
10 0 Performance is 0.00099854, Goal is 0.001 10-1 Training-Blue Goal-Black 10-2 10-3 10-4 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 9009 Epochs Figure 3. Changes in error during raining of neural neworks. M=1.008 C=-1.717 M: Gradien, C:Poin of Inersecion Inpu value: Ideniy Parameers, (e.g. lengh, widh,ec) Average error: 3.03 % Figure 4. Regression Line for raining daa Vs. Acual oupu values 6. References Beajon, G. J. and V. R. Singhal, 1990, Undersanding he Aciviy Coss in an Aciviy Based Cos Sysem., Journal Of Cos Managemen, For The Manufacuring, Indusry, vo1.4, No. 1, pp.51-72 Turney, P. B. B. (1992), Aciviy Based Managemen,.Managemen Accouning, pp. 20-25. Ziarai, R. e al 1989, keynoe speech, Cosing Pracices in SMEs, ManTech Conference, Euroecne 89, Souhampon Insiue. Ziarai M., Sockon, D., Ucan, O. N. and E. Bilgili (2002a) Applicaion of Neural Neworks in Logisic Sysems, in Proceedings of he Inernaional Conference on Fuzzy Sysems and Sof Compuaional Inelligence in Managemen and Indusrial Engineering, Isanbul Technical Universiy, Isanbul, Turkey Ziarai M, Improving he Forecasing Process in he Auomoive Afer-Marke Supply Chain, PhD Thesis, De Monfor Universiy, UK, 2003. Gary Cokins, 2001ABC Managemen an Execuive Guide 2001 Urkmez, S 2007, Aciviy Based cosing for Mariime Enerprises, MPhil/PhD Transfer Repor, De Monfor Universiy.