The Ausmlian Journal of Constru $20M Location Sydney; Wollongong and lllawara region; Newcastle and Hunter valley region; Mid North Coast; North Coast; North West NSW; South West NSW Complexity High, medium, low Procurement type Construct only; Design & construct; Construction Management; Management Contracting Subcontractor's related Suitable experience relevant to current type of project High, medium, low Track record of competitive pricing Always competitive; average;not competitive Track record of performance during construction Outstanding; average; poor Financial stability High, medium. low Availability of suitable subcontractors High,medium, low Current dispute with main contractor Yes; No Table 1: Project srmrlanty parameters The Australian Journal of Construction Economics and Building{Vol6, No 21 33 In addition to consulting databases of subcontractors used on previous similar tenders, estimators were influenced by how competitive their tenders were. Estimators reviewed selected sub-sets of subcontractors to determine their suitability for the current tender by comparing them to the nominal 'best' subcontractor of the same trade category (e.g. mechanical, electrical,hydraulic, structuralsteel, etc.). A subcontractor was considered the best Appropriateness of each subcontrcictor for a particular project if it could be established that they could potentially submit a highly competitive sub-bid. If the tender is won, the subcontractor needs to be able to complete the subcontract in a technically sound and proficient, financially secure and occupationally healthy and safe manner. The process in which the estimators selected subcontractors for their tenders is described in Figure 1. Identify characteristics o f tender projeels and criteria for selecting suitable subcontrcictors Similarity of current situation & degree of success of past tenders Retrieve solution(s) of the most sirrular projeels Select the most suitabl e subcontractors Figure 1: Process of Selecting Subcontractors for Tendering FRAMEWORK FOR THE CBR SUB- CONTRACTOR SELECTION MODEL CBR can be beneficially used to support the tender subcontract selection process described above. A conceptual framework that drives the development of a case-based subcontractor selection advisory system is shown in Figure 2. The framework consists of three key modules, i.e. input, selection, and output. The Input module provides construction estimators with a means of submitting data. Estimators need to enter similarity parameters for construction tenders and rate their importance. Depending on these similarity parameters and their weightings, similar ses are retrieved by the Subcontractor Select1on module. These similar cases are accumulated over time as the system is used on successive occasions. It therefore follows that the more the system is used the more accurate it becomes. The retrieval process is performed using a nearest neighbour retrieval mechanism (Kolodner, 1993). A list of subcontractors used in similar past tenders and the degree of success of those tenders is provided to estimators for consideration. When data comparable to the current case has been selected, and estimators have found the solution and outcome to be 34 The Australian Journal of Construction Economics and Building (Vol6, No 2] acceptable, they can apply this 'case' to the new - - - tender project. If suitable case data are not available, estimators will need to adapt data to suit the distinctive characteristics of the new tender. Further details on this process are provided in the next section. The selected subcontractors are then reported to estimators through the output module. All data (including the project similarity parameters and their importance weightings) are presented for checking. Details of the new case and the adopted or adapted solution(s) are stored in the CBR database for future reference and retrieval. INPUT Users enter the c haracteristics of the tender and importance weightings of each case attribute Jdmti.fY smi.br hist:m"c OQSIJ - SillCONTRACTOR SILECDON Similar cases of subcontractor select10n retrieved .-.... Solution (including list and details of subcontractors used in previous tender) and outcome (ranking of tender) presented lnbM•IrMI • Solution adapted by reviewing and adjusting the list of subcontractors in accordance with the main contractor's requirements of the potential subcontractors OU1PUT List of suitable c -.::: subcontractors for .... Case-'- new tender Sitnwl m Jw Jimu,uu Figure 2: Conceptual framework of the case-based system for selecting subcontractors 36 The Australian Journal of Construction Economics and Building (Vol6, No 21 ARCHITECTURE OF THE CBR SUB- CONTRACTOR SELECTION MODEL To establish the suitability of CBR approaches in this domain, the conceptual framework described above was developed into a CBR prototype using ART"EnterpriseTM version 10. CASE REPRESENTATION A robust CBR system is largely dependent on a clear representation of constituent cases and an appropriate structure for describing their contents (Aamodt and Plaza, 1994). CASSS comprises three main constituents: problem, solution and outcome (Table 2). Case Attributes Values Characteristics Problem Part Project Category Administrative and civic; commercial; Categorical data with no implied logicalrelationship educational; hospital; industrial; recreationa;l residential; civil engineering; others Construction Type New construction; refurbishment; combination of both Categorical data with no implied logical relationship Size < $0.3M; $0.3M-$0.5M; $0.5M-$3M; $3M- $20M;> $20M Quantitatively measurable Location Sydney; Wollongong and lllawara region; Newcastle and Hunter valley region; Mid North Coast; North Coast; North West NSW; South West NSW Categorical data with no implied logical relationship Complexity High, medium, low Categorical data with implied logical relationship Procurement Type Construct only; Design and construct; Construction Management; Management Contracting Categorical data with no implied logical relationship Solution Part Project Name Name of Tender Text Date Submitted Date Date List of Subcontractors Used Details of each subcontractor for each category including contract details, areas of operation, experience in certain type of works, track record of competitive pricing, track record of performance during construction, financial stability and record of current disputes with main contractor Various types Solution Part Ranking of Tender Submitted 1, 2, 3, 4, etc. Quantitatively measurable Table 2: Characteristics of case attributes The problem part is represented by a collection of tender similarity parameters. The solution part contains a list of subcontractors used in a past tender whereas the outcome provides feedback detailing the degree of success of the tender. The degree of success is measured by the client's ranking of the contractor's tender submi sion prov1des a snapshot of case representation 1n CASSS. As the case attributes for CASSS contain both numerical and linguistic values various case representation schema were ' adopted to ensure case details were meaningfully encapsulated for future retrieval, comparison and reuse. Some of the schemas used are presented below. Quantifiable data: To reduce computational effort and time, ranges were defined for quantifiable data. For instance, in CASSS, "project size" is divided into five ranges: (i) "less than $0.3M"; (ii) "$0.3M- $0.5M"; (iii) :$0.5 - $3M"; (iv) "$3M to $20M"; and (v) over $20M". The values of new and historic cases are considered equal if both are within the same range. Categorical data with no implied logical relationship: Linguistic data is best captured through a precise and consistent categorical representation scheme, as this reduces the likelihood of misunderstanding and typing errors. Data of this type are codified as linguistic categories that best describe their cha cteristics may Include such values as "commercial" "industrial", "residential" and so on. In ' addition, data with Boolean values (i.e. yes or no) belongs in this category too. As no logical relationship exists between the values, they can be regarded as discrete points where an exact match is required. Project Category !Hospital _:j Construction Type Project Size INew Construction _:j D lsJ to $20 millions _:j D Location jNorth West NSW _:j Complexity jMedium _:j D Procurement Method jTraditional Construct 0 :::J D r.,..:n.;_n -li't'fil· :.t.' XYZ Hospital Refurbsi hment Date Submittted: 27/07/2004 ABC Police Station Tender Ranked: 2 out of 4 ABC MedicalCentre Project Category:Hospital XYZ Court Hous CostrY!;ion ProJect S1ze:$3 to $20 millions I._ _ J1 Location: North West NSW Complexity:Medium New _:earch j _:;ubbies IProcurement Method:Traditional Figure 3: Case Representation in CASSS Categorical data with an implied logical relationship: It is not uncommon to describe a concept using linguistic terms. For instance, "high","medium" and "low" are 38 The Australian Journal of Constru""ta.ts of EOCpexien::e Med.imn;has been doin; a lot of retailtefiubis rlt w<»lcs;betw.en 3and 5 years of experience Low;hu onlydore afe.v ofretai.ltefi:ubishnerlt w<»lcs No expexien:e;hu never dane anyetail tefiubiSlurentworlc:s Figure 4: Taxonomy structure for reflecting the relationships of categorical data In ART*Enterprise™ case attributes are represented as non-hierarchal. The major advantage of this organisation is that entire case libraries may be searched during the case matching and retrieval process. As a result, the accuracy of case retrievals is a function of how reliable the matching mechanisms are, whilst adding new cases to the case library is relatively cheap and easy compared to CBR systems which use hierarchical structures (Kolodner 1993). MATCHING AND RETRIEVAL Since flat organisational structures do not justify the use of inductive approaches, CASSS uses a nearest neighbor retrieval mechanism. Similar cases are retrieved from the case library on the basis of the global similarity value (total case score) which ranges from 0 to 100; with 100 representing an exact matching and 0 a total mismatch. The global similarity value is determined by the following formula: Global similarity value= L f(T;, S1) w1 x 100 fori = 1 ton where: T = target case S = stored case n = number of attributes in each case i = an individual attribute from 1 to n w = importance weightings of attribute i f = local similarity between attribute i in cases Tand S The local similarity value (i.e.attribute score), on the other hand, ranges from 0 to 1. For attributes composed of categorical data with no implied logical relationship, the local similarity value is either 1 (when the two values are similar) or 0. However, if there is an implied logical relationship between the data values or in the case of quantifiable data, the local similarity value is calculated in accordance with the positions where the data values of the two cases appear in the taxonomy tree. Thus the proximity of shared common index nodes indicates higher similarity values. Once similarity scores have been generated for all cases, they are ranked and the five cases with the highest similarity scores are presented for further consideration. ADAPTATION A combination of different adaptation strategies was adopted for CASSS. For instance, if users are satisfied that a retrieved case closely resembles the current case (i.e. the tender being worked on), they can employ a null adaptation strategy by simply adopting the matching solution to the new case without any modification. However, when the intrinsic characteristics of the two cases differ, modifications to the historic solutions might be desirable. Critic- based adaptation (Brown and Lewis, 1993) and parameterised adaptation (Schank et al, 1994) strategies are provided to help decision-makers arrive at more appropriate solutions. In the solution part of each retrieved case, a list of subcontractors used in past tenders is presented to users. To facilitate detailed assessment of the suitability of the proposed subcontractors, the details of each subcontractor including area of operation, experience,performance, financial stability, etc. are provided (as shown in Figure 5). lt.techanical Services , l •1'lll'i r21 u\; Mechanical Subcontractor 007 2 Mechanical Subcontractor 021 3 Mechanci al Subcontractor 010 4 r Go Baclc to View Another Case r Choose the Subby for New Tender r. Use Details to Search Similar Subbies OK Name !ielli Rrcea - Contact jAIMifei!X ] Phone Doc)QOQ( ) Fax Address Operating in - 'North West NSW' - Suitable Experience in - 'Hospital' Competitive Pricing C....,.aye Performance OUTSTANDING Financial Stablity Current Disputes with Main Contractor Figure 5: Subcontractor Details Screen in CASSS If users are dissatisfied with the subcontractors proposed for a certain trade, they can search for ou-.er companies from the database of subcontractors using case- based reasoning. CASSS displays an adaptation screen to guide users through this process as shown in Figure 6. Users are required to enter characteristics of the required subcontractors and their importance weightings. Another CBR engine (which is incorporated into the adaptation mechanism of CASSS) allows users to search through the system's database of subcontractors for alternative subcontractors to the ones already proposed. Once users are satisfied with the list of subcontractors to be used in the new tender (case), a report of these subcontractors is available for the next stages of the tender process. The Australian Journal of ConstnKtion Economics and Building [Vol6, NoI 40 The Australian Journal of Construction Economics and Building [Vol6, No 2] - Operating in - 'North West NSW' !Yes :o::J IL9 _, Suitable Experience in - 'Hospital' H,..., IGH iJ'I ._a ..J Competitive Tender Pricing !Always Competitive 3 .1.1._0 . Performance during Construction joUTSTANDING 3 L-la_ , Financial Stability jHIGH o:J Current Disputes with Main Contractor jNo o:J I._s_ , .ls__ , MechanicalSubcontractor 007 Mechani cal Subcontractor 010 Mechanical Subcontractor 021 Mechanical Subcontractor 015 Mechanical Subcontractor 027 Back to Previous Screen r.- d-su._b Y. !" !!. I d OK I Name Phone Competitive Pricing Performance Financial Stability Current Disputes with Main Contractor Figure 6: Critic-based adaptation using user's knowledge and CBR SYSTEM MAINTENANCE As the quality of the advice given by CASSS relies heavily on the quality of the information of past tenders, system maintenance (i.e. recording and updating of subcontractor information) is an important issue. As a feature of case-based reasoning, CASSS has the ability to record information of every new tender case and automatically update its database (i.e. the case base of tendered projects). Furthermore, the system case base is also designed to link with the main database of subcontractors that is normally kept in a construction company. The system case base will automatically be updated if there is any change in the subcontractor information in the subcontractor database. SYSTEM PRACTICALITY An actual tender for the construction of a new hospital (with a value of approximately A$10m) in New South Wales, Australia was used to demonstrate the practicality of CASSS. Some details of this tender are summarised in Table 3. . Tender Details Details Project Category Hospital Construction Type New Construction Project Size Approximately $8,000,000 Location of Site NSW North West Level of complexity of project A simple structure including slab on ground, light weight steel frame, metal roof and external cladding. Services include electrical, security, data, nurse call, ducted air conditioning, water, sewer, stormwater, etc. Procurement method Traditional Construct Only Table 3: Details of test tender case A total of 40 historic construction tenders were collected to train the CASSS model. A set of tender similarity parameters and their importance weightings were identified in accordance with the characteristics of the construction project, the external environment and the main contractor's expectation of potential subcontractors. CASSS then recommended a list of subcontractors for the tender1 To determine whether the solution generated by the model was comparable to that produced by domain experts, four independent, experienced construction estimators with extensive local knowledge of subcontractors in the area were invited to assess the subcontractors chosen by CASSS. As can be seen from the table above, there is generally an unequivocal agreement between the domain experts and CASSS. It is also noteworthy that the list of subcontractors selected by CASSS was quite similar to that prepared by the actual estimator of this particuiar tender, and that the estimator's tender was successful. CONCLUSION This paper has presented a novel way for selecting subcontractors for construction tender projects using CBR. Since CBR is an experience-based approach, the lessons learned in previous cases can be made available to estimators to provide them with an early indication of the likely future outcomes of a tender. Based on the information collected from experts, a conceptual framework for a case- based system for selecting subcontractors at tender time was devised. The framework was subsequently developed into a computer prototype using a CBR shell - ART*Enterprise. The prototype, using trial data, has demonstrated that CBR can provide appropriate recommendations for the tender of a hospital. However, in order for CASSS to be fully functional, further verification and validation of the system are needed. 1 The issue of confidentiality prevented the authors from publishing details of these subcontractors -- ---- --------------- ----The_Austr_al_ian_Jo_u_""''""'"""""'"""""''"''"""'[Vol• 42 The Australian Joumal of Construction Economicsand Building [Vol6, No 2] Subcontractor Category CASSS Recommend ations Expert 1 Expert 2 Expert3 Expert4 Excavation 3 subbies Agreed Agreed Agreed Agreed Concrete 3 subbies Agreed Agreed Agreed Agreed Structural Steel 3 subbies Agreed Added another subby Agreed Agreed Bricklayer 2 subbies Agreed Agreed Agreed Agreed Metal Roofing & Cladding 3 subbies Agreed Added another subby Agreed Agreed Aluminum Windows & Doors 4 subbies Agreed Agreed Agreed Agreed Doors & Frames 3 subbies Agreed Agreed Agreed Agreed Agreed Gyprocker 3 subbies Agreed Agreed Agreed Agreed Carpet & Vinyl 3 subbies Agreed Agreed Agreed Agreed Tiling & Waterproofing 2 subbies Agreed Agreed Agreed Agreed Painting 3 subbies Agreed Agreed Added another subby Agreed Metalworks 2 subbies Agreed Agreed Agreed Agreed Electrical Services 3 subbies Agreed Agreed Agreed Added another subby Mechanical Services 3 subbies Agreed Agreed Agreed Replace one subby Hydraulic Services 3 subbies Agreed Agreed Agreed Agreed Medical Gas Services 3 subbies Agreed Agreed Agreed Agreed Fire Services 2 subbies Agreed Agreed Agreed Agreed Table 4: Results of system reliability test REFERENCES Barletta, R. (1991). An introduction to case- based reasoning, AI Expert, 43-49. Aamodt, A. and Plaza, E. (1994). Case based reasoning:foundational issues, methodological variations and system approaches, Artificial Intelligence Communication, No. 7, 39-59. Brown, S. J. and Lewis, L. M. (1993). A case-based reasoning solution to the problem of redundant resolutions of nonconformances in large-scale manufacturing, Innovative Applications of Artfficiallntelligence 3, eds. R. Smith & C. Scott, AAAI Press, 121-133. Kolodner, J. L. (1993). Case Based Reasoning, San Mateo, Morgan Kaufmann Publishers, CA. Schank, R. C., Kass, A., and Riesbeck, C.K. (1994). Inside Case-Based Reasoning, Erlbaum Associates, Hillsdale, N.J. Shash, A. A. (1998). Bidding Practices of Subcontractors in Colorado, Journal of Construction Engineering and Management , May/June 1998, 219-225. Tam, E. (2003).STAT-USA Market Research Reports, U.S. & Foreign Commercial Service and U.S. Department of State, 2003. Tang, H. {2001).Construct for excellence, report of the construction industry review committee, January 2001. The Printing Department, HKSAR. The Australian Joomal of Constru