Knowledge acquisition of routine design problems in industrial settings
Marco M. Olthof, Juan M. Jauregui-Becker, Hans Tragter, Fred J.A.M. van Houten University of Twente, Enschede, the Netherlands
[email protected] [email protected] [email protected] [email protected] Abstract: This paper presents a knowledge acquisition method for the identification and formalization of routine design problems at industrial settings. The method aims at supporting the industrial applicability of Computational Synthesis Systems (CSS), which is a technology that allows automating the generation of solutions to design problems. Keywords: Knowledge acquisition, routine design, industrial settings, computational synthesis
1. Introduction Faster response to markets has set high pressure in the development of new and innovative products. Because of the high pace in product development, pressure on the design phases is increasing. As consequence, designers have to design complex products in shorter times, which often results in firsttime-right approaches that tend to sacrifice better design solutions for time efficiency. This has motivated the development of the field of Computational Synthesis (CS), which studies algorithmic procedures to automate the generation of design solutions. The idea is that by combining “low-level” building blocks, “high level” functionalities can be achieved. One of the major challenges facing CS is the translation of industrial design problems into Computational Synthesis Systems (CSS). A CSS is a program that automates one specific design problem, as for example, the design of a compression spring. According to Cagan (2005), this is probably motivated by the fact that initializing a CS process has not received much attention in literature, as most computational synthesis methods are developed to solve one particular design problem. Furthermore, most industrial design problems are not formalized into explicit models, where variables and equation describe the problem. Instead, they are implicitly clear to engineers and designers dealing with them. This paper is part of the research project Smart Synthesis Tools (SST) developed at the University of Twente in cooperation with Delft University of Technology, both in the Netherlands. The project researches the development process of CSS. The goal is to deliver a generic development methodology for dedicated CSS, such that industrial design processes can be better supported (Schotborgh, 2007). SST research focuses on three main topics: (1) methods to develop models of design problems, (2) methods to develop CSS for a given (and modelled) design problem, and (3) methods to identify and formalize design problems in industrial settings. In Jauregui-Becker (2008) and Schotborgh (2008) methods for modelling and developing CSS are described. This paper deals with the development of a knowledge acquisition method to aid the identification and formalization of routine design problems in industrial settings. The purpose of this method is to aid the translation of industrial design problems into CSS. The research consisted in investigating and testing different knowledge acquisition techniques. A quantitative and qualitative analysis of the results of each of the tested techniques was used to define a method for knowledge acquisition, regarded as the S3 method. The S3 method consists of three elicitation phases: a Scanning phase, a Structuring phase and a Specifying phase; where each phase integrates a number of elicitation techniques. The scanning phase allows obtaining an overview of the design problems in the organization. Here, it is also assessed whether the problem is routine or not. Thereafter, a specific design problem can be chosen for its automation. The structuring phase results in a general model of the problem, where the important elements and relations are identified. The specifying phase gathers detailed knowledge on the parameters and specific calculations involved in the problem, and formalize it in appropriated models. The method was tested at the Injection Molding department of PHILIPS Advanced
Technology. Results of applying the method resulted in a structured tree containing all routine design problems involved in the design of injection molding molds. One of the identified problems was chosen to test the method for the formalization phase, namely, runner and gate system design. After applying the method, a problem model including the variables, constraints and relations was formalized. In order to assess the quality and consistency of the elicited knowledge, a CSS for gate and runner design was developed following the guidelines described by Schotborgh (2008). The design solutions generated by the CSS corresponded to those developed by designers, which showed that the problem model extracted with the S3 method was consistent. These results are used to conclude that the S3 method is suitable for identifying and formalizing routine design problems. By scanning, structuring and specifying the design problem, a knowledge engineer should be able to elicit the desired knowledge from experts. 2. Knowledge acquisition method development Knowledge Acquisition (KA) is the process of obtaining data, information and knowledge to get a consistent view of a process where an expert performs a problem solving task. Developing a method for knowledge acquisition accounts for the characterization of three parts, as described by Cordingly (1989): problem definition, knowledge elicitation and knowledge representation. The following sections describe these three parts for the method developed here. 2.1. Problem Definition The first step in developing a method for KA consists of defining the type of knowledge the method targets. In this paper, the focus is set on artifactual routine design problems, which proceed within a known space of functions, expected behaviours and structure variables (Gero, 1990). According to this, the type of knowledge to identify with the method can be classified into the following groups: 1. Identifying artifactual design problems. 2. Identifying which problem is routine. 3. For a routine problem, formalizing its information contents so that it can be properly formulated. The following subsections will clarify the required knowledge for each of these problem definitions. 2.1.1. Artifactual design problems. As presented by Gero (1990), and shown in Figure 1, design artefacts are described by three different types of entities: Vocabulary of elements: represent the physical elements of a system. For the gear device design shown in Figure 1 these are two gears and two shafts. Descriptions of elements: Determine the attributes of the artefact, like the diameter (D) and angular velocity (w) in Figure 1. Configuration of elements: Determine the disposition of the elements in the structure, as for example the connectedness between elements represented by the relations in Figure 1. By listing these types of information, artifactual design problems can be identified.
Figure 1: Gear device: elements, descriptions and configurations.
2.1.2. Routine design characteristics According to the completeness of information described in Greeno (1978), a design problem has to account the following aspects in order to be routine: A known initial state: Design parameters as well as input variables are well defined. A clear goal state: The quality of the solution of the problem can be assessed by a previously defined objective function.
Constrained set of logical state: The constraints of the design solution are known on forehand. Constraint variables: Value ranges of the design variables are known on forehand.
2.1.3. Routine design formulation According to Jauregui-Becker (2009), a routine design problem is structured when its parts and interrelations are formalized as follows: Embodiment elements and scenario elements to describe the initial state of the design, Objective function, performance indicators and analysis techniques to express and assess the goal of the design artifact, Topology relations and physical coherence constraints to indicate the set of logical states that have to hold for the design artifact to exist, Confinement constraints to restrict the values that embodiment, scenario and performance descriptions are allowed to reach. The following definitions are used: Element: Is a class description of a design artifact component. Descriptions: Characterize an element class by representing its attributes in the form of variables. Embodiment: Is the subset of descriptions of an element upon which instances are created to generate design solutions. Scenario: Is the subset of environment variables, attributed to elements in the natural world and considered in measuring a design artifact’s ability to accomplish its function. Topology relations: Define the configuration among embodiment elements and scenario elements. Physical coherence constraints: Assure no physical impossibilities are committed by the artifact being designed. Performances: Are those descriptions used to express and assess the artifacts behavior, and are calculated using analysis relations. Analysis relations: Use known theories, for instance the laws of physics or economics, to model the interaction of the design artifact with its environment and predict its behavior. Objective function: Weighs and adds the performances to result in the overall performance of the design. Confinement constraints: Determine the range in which a description can be instantiated. 2.2. Knowledge elicitation Knowledge elicitation consists of obtaining, analysing and interpreting knowledge from any relevant knowledge source – e.g. literature, examples, and human expert- in a methodological and systematic way. Many techniques and methods exist for eliciting knowledge, in particular techniques for gathering knowledge from experts (Burge, 1998; Cooke, 1994; Diaper, 1989). Choosing an adequate technique is mainly based on the type of knowledge to be retrieved and the type of models used for representing it. This can be seen in Burge (1998), where a categorization of knowledge elicitation techniques is presented according to the type of knowledge that is elicited. The different categories of types of knowledge can be seen in Table 1.
Types
Description Techniques used to determine the steps followed to complete a task and the Procedures (P) order in which they are taken. Problem solving strategies Techniques that elicit a problem solving strategy. These methods attempt to (PSS) determine how the expert makes their decisions. Techniques are used to extract the goals and sub goals (decomposition) that Goals, sub-goals (G) are used when performing a task. These methods are listed separately from procedures since ordering is not necessarily provided. Classification (C) Techniques used to classify entities within a domain. Relationships (R) Techniques that obtain relationships between domain entities. Techniques that are used to evaluate systems, usually prototype systems, or Evaluation (E) other types of knowledge elicitation session results. Table 1: Categories of types of knowledge (Burge, 1998)
With the given problem definition in three stages and the desired types of knowledge for each part the knowledge elicitation process has been divided into three phases: scanning, structuring and specifying. Scanning is done with the aim of getting an overview of the design problems present in the organization by assessing the problem definition of Section 2.1.1. Structuring allows determining which problems are routine and which are not and giving structure to the problem. The definition of Section 2.1.2 is used for this ends. Specifying is done with the aim of further detailing the problem and, in this way, translates it into the formal model presented in Section 2.1.3. To select adequate techniques for retrieving these types of knowledge, a selection criterion has been defined based on: the type of knowledge elicited, the combination between techniques to form one method, the suitability to elicit technical knowledge and the ability to be adjusted to this subject. As result, the following types of techniques are chosen: Scanning phase: Problem Solving Strategies are preferred, as they are good in obtaining a general view of the problem its solving strategies. Structuring phase: the Classification type of techniques are prefered as they allow structuring knowledge easily. Specification phase: Classification, Procedures and Problem Solving Strategies techniques are selected given that the goal is to find the knowledge on a detailed level on all parameters, constrains and relations. As the three different phases are to be integrated into one method, the knowledge that results from one phase is used as input for the next one. Therefore, the techniques to select have to be compatible among them. This accounts for having knowledge type compatibility as well as representation compatibility. Furthermore, the techniques should be capable of handling technical information and models. A combination of direct and indirect techniques is desirable. Defined by Burge (1998), direct techniques directly request the design steps and their sequence, while indirect techniques are to refine this knowledge by obtaining steps and sequences that may be implicit. Direct techniques are used to elicit explicit knowledge, whereas indirect techniques are more effective at eliciting implicit knowledge (Bradley, 2006). Furthermore, Olsen and Biolsi (1990) suggest that a direct method can be used to determine objects in a domain and indirect method used to find the relationship between them. Moreover, research shows that using direct and indirect techniques together obtains more alternative sequences for design steps than using only direct (Burge, 1998). By using this criterion, a number of techniques have been selected for further experimentation. These are presented in Table 2, where D implies a direct technique and I an indirect technique. The other abbreviations stand for the different types of knowledge, as in Table 1, pointing out what the result of the technique will be. Category
Interviews
Verbal Reports
Conceptual Elicitation Techniques
Technique Unstructured interviews (D) (P,PSS) Structured interviews (D) (P,PSS) Twenty questions (I) (PSS) Forward Scenario Simulations (D) (P,PSS) Talk aloud (D) (P,PSS) Laddering (I) (C,R)
Repertory Grid Analysis (I) (C)
Description Unstructured interviews are free-form interviews in which neither the content nor the sequencing of the interview topics is predetermined. Structured interviews follow a predetermined format. It is a technique based on the traditional parlour game with the same name. The elicitor selects a situation, diagnosis, fault or state and the expert has to try to ascertain the specific concept that the elicitor has in mind by asking yes/no questions The elicitor or the expert supplies an example situation or case and the expert describes, usually step by step, what would be done in the situation or when handling the case Requires the expert to say aloud everything that he or she would say to him or herself as if conveying an internal conversation Is similar to the structured interview technique, goal decomposition (Goal decomposition involves having the expert work backwards form a single goal to the evidence that leads to that goal), except that the concepts are not restricted to goals and sub goals. Is based on Kelly's Personal Construct Theory. Define elements, use triadic approach to define constructs, scale all elements at the constructs, this way the expert builds a grid.
Category Conceptual Elicitation Techniques Data Collection Method Structural analysis
Technique Step Listing (D) (C) Chapter Listing (D) (C)
Description The expert is asked to list the steps involved in performing a task The expert lists chapter headings and subheadings for a hypothetical book on the domain
Document Analyses (D,I) (P,C,R)
Involves gathering information from existing documentation.
Direct elicitation of structure (D) (C,R,E) Semantic Nets (D) (C,R)
Involve the construction of a graphical representation of knowledge derived from a structured interview The expert makes a Semantic Network while analysing a product
Structuring and gathering knowledge about the design problem by filling the PFBPSS scheme Additional PCPACK5 is an automated tool that supports retention, sharing, Techniques management and re-use of knowledge. PCPACK5 is an integrated suite PCPACK5 (All) of 10 knowledge tools designed to support the acquisition and use of knowledge. (Milton, 2003) Table 2: Selected Knowledge Elicitation techniques (described by Burge (1998), Cooke (1994) & Corbridge (1994)) PFBPSS (D) (C,R)
2.3. Knowledge representation The final phase of the knowledge acquisition process is the representation of the elicited knowledge. In this research, a Semantic Networks based representation has been chosen. Semantic Networks are attributed graphs, where the nodes represent concepts, arcs represent relations between concepts, and labels are used to represent properties of both nodes and arcs. The following guidelines are used to represent the problem structure: Nodes: represent elements. Have a type, which depends on whether the element belongs to the embodiment or the scenario. Arcs: represent the relation among the elements. Relations also have a type, which depends on whether the relation is topologic, coherence, analysis or objective function. Node labels: uses descriptions and its confinement constraints to elaborate on the class definition of the element. Arc labels: specifies the model of the relation by relating element descriptions and independent descriptions if required. A weight factor is an example of an independent description. In Figure 2, a semantic network is presented to represent the design of the gear device of Figure 1.
Figure 2: Gear device design problem representation (Jauregui-Becker, 2008)
3. Experimental assessment An experimental assessment is done with the aim of validating the consistency and coherency of the elicited knowledge. Validating the knowledge per se involves providing some justification for claiming that the description or implementation of the knowledge is appropriate and not just internally consistent (Firley, 1991).
In this research, knowledge is validated by performing an objective and a subjective assessment. The objective assessment measures the time it takes to elicit and process knowledge. Also the amount of elicited knowledge is counted by considering the number of gathered steps, subjects, and rules. The subjective assessment is carried out by the knowledge engineer and the expert. In this research, the subjective measures indicated by Diaper (1989) where used: Effort: the amount of effort the expert and the knowledge engineer have to put in. Familiarity: the familiarity with the technique used in the session, if it is clear what to do and if it is a well known technique. Connectivity: if the technique elicits the knowledge in a manner that connects with the way the expert is used to perform his/her work. Distortion: the level of distortion of the knowledge elicited. Amount of knowledge on one subject, or the type of knowledge are points to take into account. Satisfaction: the level of satisfaction with the results of the session, seen from the point of view of both the expert and the knowledge engineer. Productivity: if the expert and knowledge engineer feel the session was productive. In order to select the set of techniques best suited for the goal of this research, a number of knowledge acquisition experimentation session where performed. The experiments were performed at PHILIPS Advanced Technology, as this organization is participating in the research program Smart Synthesis Tools. Five experts participated in the experimental sessions. Every session was performed in an isolated room and was tape recorded. Additional notes were made by the knowledge engineer. Because of confidentially reasons, the contents of the elicited knowledge are not discussed in this paper. 3.1. Results of the objective assessment Table 3 shows results from objective assessment. The table indicates the elicited time and amount of knowledge gathered per expert for four of the tested techniques. The letters A-E stand for the different experts. After comparing the required time and the amount of knowledge elicited for each technique, it was concluded that Forward Scenario Simulation, Twenty Questions and Laddering demonstrated to be the best performing. Other tested techniques, as Repertory Grid Analysis shown in Table 3, had a poor performance. Technique: Forward Scenario Simulation, average Eliciting Time: 15-20 minutes Expert Steps and subjects mentioned
A
B
C
D
28
21
28
25
Technique: Twenty Questions, average Eliciting Time: 20-25 minutes Expert
A
D
E
Main subjects
7
8
Division of questions per subject
3-1-3-2-4-3-4
6 5-3-3-2-3-3 (1 loose questions)
Average amount of questions per subject
Average: 2.9
Average: 3.2
2-2-2-3-3-4-2-1 (1 loose questions) Average: 2.4
Technique: Laddering, average Eliciting Time: 45 minutes Expert
A
B
C
D
Main subjects
22
14
17
28
Sub subjects
28
21
37
42
Levels
8
7
10
6
Technique: Repertory Grid Analysis, average Eliciting Time: 60+ minutes Expert
C
D
E
Constructs
7
10
10
Clusters
6 5 6 Table 3: Objective assessment of 4 different techniques
3.2. Results of the subjective assessment Table 4 shows the subjective assessment scores for each of the tested techniques. As it can be seen on the Table, the following techniques performed the best: Forward Scenario Simulation, Laddering, Step Listing and Document Analysis.
PCPack5
Structural Analysis Semantic Nets
Structural Analysis Direct Elicitation of Structure
Data collection Method Document Analysis
Chapter Listing
Step listing
Repertory Grid Analysis
Laddering
Talk aloud
Forward Scenario Simulation
20 Questions
++ + ++ ++ + -+ ++ + +++ ++ + ++ ++ + ++ + ++ ++ +++ + ++ ++ + +++ +++ ++ + ++ + + + ++ +++ + + + ++ + ++ + +++ ++ ++ + + +++ + +++ + +++ ++ Table 4: Subjective assessment of the tested techniques. Scale: [++, +, +-, -, --]
PFBPSS
Effort Familiarity Connectivity Distortion Satisfaction Productivity
Structured Interview
Unstructured Interview
Effort: [++ almost no effort, -- a lot of effort], Familiarity: [++ familiar, -- not familiar], Connectivity: [++ connected, -- not connected], Distortion: [++ no distortion, -- distortion], Satisfaction: [++ satisfied, -not satisfied], Productivity: [++ productive, -- not productive].
++ ++ ++ ++
++ +++ + +
3.4. The S3 method The results obtained in previous sections are used now to define a method for routine design problem identification and formalization. The method is conformed of a Scanning phase, a Structuring phase, and a Specifying phase, and is regarded as the S3 method. Scanning phase The goal is to get an overview of the design problems based on a map of the major subjects and their descriptions. The idea is to chart the size, scope and parts of the problem, without focusing on the details. It was chosen to do so by first using Forward Scenario Simulation and secondly using Twenty Questions. Forward Scenario Simulation allows a quickly scanning of all major subjects. Subsequently, Twenty Questions is used to determine the descriptions. By combining one direct and one indirect technique, a complete scan of the different design problems is obtained. Structuring phase The goal is to get a structured overview of the complete problem with all the subjects, major and minor, with a classified structure. Laddering has been chosen as the knowledge eliciting technique. The purpose of this technique is to create a structured tree, with all the subjects and different parts, so that a structure model of the identified problems is obtained. The subjects from the former phase are used as the starting point of building the tree. Also the sub subjects and other parameters can be put in the tree. The following questions proved to be successful in determining if the problem was or not routine: Are all parameters known?; Is known on forehand how to solve and constrain the parameters?; Does the design problem changes because of the implementation of new technology? Specifying phase The goal of this phase is to obtain a formal model of one problem. Also procedural knowledge about how to solve the parameters, how to determine the different components, and how they are constrained is elicited. When there are more components also topological relations are elicited. The knowledge engineer guides the type of knowledge he/she will elicit based on Lai (2007): Declarative knowledge: Get the knowledge about the parameters and components Procedural knowledge: Get the knowledge about the analysis, tacit knowledge on how to solve a variable, and all about the relations between parameters and components.
First the Laddering technique is used to obtain declarative knowledge that formalizes the problem as described in Section 2.1.3. Then, the Step Listing technique is used to go through each subject step by step and identify procedural knowledge on how to solve the problem. When the analysis techniques are composed of complex equations, the Document Analysis is used. This because this technique guides the knowledge engineer through all relevant documentation where this type of knowledge is often contained. 4. Case Study Injection molding design has been chosen as case study to test and validate the method. Injection moulding is a manufacturing process for producing parts from plastic material. Material is heated and injected into a mold cavity where it cools and hardens to the configuration of the mold cavity. After the product is designed, molds are designed and made by mold specialists in metal. Injection molding is used for manufacturing a wide variety of plastic products. As a large amount of written documentation (e.g. specialized books and internet) exists on this topic, the completeness and consistency of the elicited knowledge can easily be validated. The method is tested at the injection moulding department at PHILIPS Advanced Technology. Five experts in the field participated in the experiments. 4.1 Identification of a design process Figure 3 shows a graphical representation of the structured tree that is made of mold design with the laddering technique. In Figure 3 all the different design parts are mentioned, as well as the subjects within the different design parts. The parts are executed from top to bottom, if parts are next to each other they are executed simultaneous. The different parts are examined and identified in routine and non routine design. In Figure 3 the blue parts are routine design. As can be seen, most of the design steps are routine design. Therefore they are in principle suitable for building Smart Synthesis Tools.
Figure 3: Tree structure of Injection Mold Design obtained at the identification phase
4.2 Formalization of a routine design problem Design of a runner and gate system was selected as case study for testing the specifying phase of the method. Following the definitions from 2.1.3 the results are given in Table 5. A Semantic network representing the problem is shown in Figure 4. Due to confidentiality all specific knowledge has been removed from Table 5. However, the table and the figure show that the method was successful in determining specific knowledge about the problem composition.
Figure 4: Semantic network of runner and gate design problem obtained at the structuring phase
Variable type
Element Part
Scenario
Straight Runner Material Process Straight Runner (runner system)
Parameter Flow Length Wall Thickness Part Volume Length Material parameters Ejection Temperature (Demolding Temperature) Velocity of the flow front Diameter
Diameter 1 Diameter 2 Runner Angle Length Diameter 1 Embodiment Diameter 2 Gate Angle Length Injection time Process Melt Temperature of plastic Mold Temperature Packing time Velocity of the flow front Max Shear rate Max Shear stress Injection Pressure Performance Process Gate freeze time Cool time straight runner Cool time runner Cool time part (plate) Fourier check Table 5: Results of Knowledge Acquisition Sessions obtained at the specifying phase
4.3 Evaluation of S3 method For the evaluation of the S3 method the efficiency and consistency are described. To see how efficient and effective the method and knowledge acquisition is in comparison with gathering knowledge from books, the knowledge acquired with the sessions is compared with the knowledge found in books. For efficiency comparison the knowledge about the three different subjects of the case study are compared. For every subject five areas are assessed: general knowledge, specific knowledge, constraints, design rules and analytic rules. Table 6 shows if knowledge is found through knowledge acquisition from experts or from books. Three different indications are possible: found, some found and not found, meaning that knowledge or rules are found on that specific point from experts or from books. Knowledge Runners General Specific Constraints Design rules Analytic rules Gates General Specific Constraints Design rules Analytic rules Process General Specific Constraints Design rules Analytic rules
Expert
Books
~
~ ~
~
~
~ Table 6: Efficiency check [= found, ~=some found, =not found]
4.4 Computational Synthesis of Runner Gate design This design problem was implemented into a Computational Synthesis Tool so that the procedural knowledge on how to solve the problem could be tested. Figure 5 shows a screenshot of the Result screen for one input set corresponding to a specific existing runner gate design problem. Every blue point represents one design solution. The specification of the value of each variable for the selected solution (point in red) is shown on the right table in the Figure. Designs generated automatically by the program were comparing to those made by designers. The program, based on the elicited knowledge, was capable of reproducing the designs created by expert designers and is now used by them.
Figure 5: Screenshot of Result screen of Runner Gate CSS.
5. Conclusion Results allow concluding that the S3 method is suited for identifying and formalizing routine design problems. By scanning and structuring the design problems, routine design problems can be identified. With further specification of the design problems they can be formalized to fit into the frameworks of CSS. Several tests and the case study show that by using the S3 method, a knowledge engineer should be able to elicit the desired knowledge. However, additional testing of the S3 method is required to establish the general usability and consistency of the method for identifying and formalizing routine design problem in industrial settings. 6. Acknowledgements The authors gratefully acknowledge the support of the Dutch Innovation Oriented Research Program ‘Integrated Product Creation and Realization (IOP-IPCR)’ of the Dutch Ministry of Economic Affairs. The authors especially want to thank PHILIPS Advanced Technology and in particular Fokke J. van der Veen for his effort and energy, contribution to this research project. 7. References Bradley J.H., Paul R. & Seeman E. (2006). Analyzing the structure of expert knowledge. Information and Management 43. 77-91 Burge J. (1998). Knowledge Elicitation for Design Task Sequencing Knowledge. Worchester Polytechnic Institute. Cagan J., Campbell M.I., Finger S. & Tomiyama T (2005), A Framework for Computational Design Synthesis: Model and Applications. Journal of Computing and Information Science in Engineering 5, 171-181 Cooke Nancy J. (1994). Varieties of knowledge elicitation techniques. International Journal Human – Computer Studies 41. 801-849 Corbridge C., Rugg C., Major N.P., Shadbolt N.R. & Burton A.M. (1994). Laddering: Technique tool use in knowledge acquisition. Knowledge Acquisition 6. 315-341 Cordingly, E.S. (1989), Knowledge elicitation techniques for knowledge-based systems, In: D. Diaper, (Ed.), Knowledge Elicitation: Principles, Techniques and Applications, Ellis Horwood. Diaper D. (1989), Knowledge elicitation: Principles, Techniques and Applications, Chicester, England: Ellis Horwood Ltd. Firlej M. & Hellens D. (1991), Knowledge Elicitation, A practical Handbook, UK: Prentice Hall International Ltd. Gero, J. S., 1990, Design prototypes: a knowledge representation schema for design, AI Magazine, American Association for Artificial Intelligence, 11/4:26-36. Greeno, J. (1978), Natures of Problem-Solving Ability, Handbook of learning and cognitive processes, Hillsdale, Lawrence Erlbaum, 5: 239-270. Hart A. (1992), Knowledge Acquisition for expert systems, 2nd ed., New York: McGraw Hill Inc. Jauregui-Becker J.M., Tragter H. & van Houten F.J.A.M. (2008). Structuring and Modeling Routine Design Problems for Computational Synthesis Development. CIRP Design Conference 2008 Jauregui-Becker, J.M., Tragter, H., van Houten, F.J.A.M. Structure and models of artifactual routine design problems for computational synthesis. CIRP Journal of Manufacturing Science and Technology, 2009, 1(3), Design Synthesis, 120-125 Lai, Lien F. (2007). A knowledge engineering approach to knowledge management. Information Sciences 177. 4072–4094 Milton, N. (2003, November 20). Epistemics: PCPACK5 and Knowledge Acquisition, retrieved at 1st of April 2009, form http://www.epistemics.co.uk Okafor E.C. & Osuagwu C.C. (2006). The Underlying Issues in Knowledge Elicitation. Interdisciplinary Journal of Information, Knowledge, and Management Volume 1. 95-108 Olson, J., Biolsi, K. (1990), Techniques for Representing Expert Knowledge, In Ericsson, A., Smith, J. (Eds.) Towards a General Theory of Expertise, Cambridge University Press Schotborgh W.O., Tragter H., Kokkeler F.G.M., van Houten F.J.A.M., Tomiyama T. Towards a Generic Model of Smart Synthesis Tools. In: CIRP Design Seminar 2007, CIRP 2007, 2007, Berlin. Schotborgh, W.O.; Kokkeler, F.G.M.; Tragter, H.; Bomhoff, M.J.; Houten, F.J.A.M. van, “A Generic Synthesis Algorithm for Well-Defined Designs”, CIRP Design Conference on Design Synthesis 2008.