DETC FORMALIZING USER ACTIVITY PRODUCT FUNCTION ASSOCIATION BASED DESIGN RULES FOR UNIVERSAL PRODUCTS

Transcription

1 Proceedings of the ASME 2011 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2011 August 28-31, 2011, Washington, DC, USA DETC FORMALIZING USER ACTIVITY PRODUCT FUNCTION ASSOCIATION BASED DESIGN RULES FOR UNIVERSAL PRODUCTS Shraddha Sangelkar Department of Mechanical Engineering Texas A&M University College Station, Texas 77843, USA ABSTRACT 1 products intend to equally serve people with and without a disability. This paper focuses on creating guidelines that are applicable during the early stages of designing universal products. Actionfunction diagrams are used to formally compare existing universal products to their typical counterparts to study the similarities and differences. A data mining technique, particularly association rule learning, generates rules from the universal and typical product comparison data. Generation of function-based association rules for universal design has been performed on a smaller scale using this method; this research seeks to extend and formalize the same method by studying a larger set of universal products. Trends in the generation of rules are analyzed indicating that a finite set of rules should be applicable to an arbitrarily large set of products. Further, the rules are analyzed in detail to evaluate their potential for transferability and reuse from one product to another. Of particular interest is the transferability of the rules across apparently disparate product domains such as garden tools and residential furniture. The conceptual and physical similarity of the rules is discussed in the context of creating universal product families based on a platform of accessible elements. 1. INTRODUCTION AND BACKGROUND design (UD) is a concept intended to promote the development of products and environments to be used effectively by all users without adaptation or stigmatization [1]. Inclusion of persons with a disability in all facets of life is important for ethical and economic reasons. Though there is significant effort to improve universal design [1-4], many challenges remain [5]. As it stands, industry lacks adequate design knowledge for implementation of universal design [6]. 1 Address all correspondence to this author Daniel A. McAdams 1 Department of Mechanical Engineering Texas A&M University College Station, Texas 77843, USA The general objective of universal design is well established. Here, we wish to extend prevailing universal design methods to develop a specific set of guidelines. Our goal is to create universal design guidelines that can be applied in the initial phase of product development when solution concepts are generated and product functions are established. Product function specific rules with an emphasis on the user-product interaction befit to tackle the challenges of designing for disability. The research presented here builds on some recent and closely related research in function-based universal design. Kostovich presents a product analysis framework termed the actionfunction diagram to improve universal design research and practice [7]. Actionfunction diagrams are created in the early stages of design and provide apt representation of the user-product interaction. The study by Kostovich compares a product pair that consists of a universal product and its typical counterpart. Other recent efforts include a formal comparison between universal architectural systems and consumer products at the functional level [8, 9] and applying association rule mining to generate function-based association rules for universal design [10]. In the previous study, fifteen product pairs are studied to explore the applicability of data mining techniques to generate rules for universal design [10]. The study demonstrates the potential of association rule mining for generating rules based on user-product interactions. The interaction between the user and a product as embodied in the actionfunction diagrams gives valuable information for catering to the needs of individuals with a disability. Here, we extend the research method presented by Sangelkar to study a larger dataset consisting of 65 product pairs [10]. Contributions of this research include formalizing the data generation and analysis procedure to obtain a set of universal design rules. Particularly, filtering of the rules given by the Apriori algorithm is formalized to capture the pertinent 1

2 information. The research presented here compares the set of rules obtained from the previous dataset of 15 product pairs to the rules obtained from current dataset of 65 product pairs. This paper aims to determine if we can capture a tractable, or perhaps asymptotically finite, set of universal design rules by studying a larger and varied dataset of product pairs. Further, the rules are analyzed in detail to evaluate their potential for transferability and reuse from one product to another. Of particular interest is the transferability of the rules across apparently disparate product domains. The conceptual and physical similarity of the rules is discussed in the results section. As a compliment to the broad applicability or transferability of the discovered rules, some initial explorations of clustering products are presented. Products are clustered based on the association rules that are applicable to them. These product clusters represent an opportunity to form universal product family. The commonality in the clustered product pairs can serve as a platform of accessible elements. The following section presents background and related work. The next section explains the data generation and data analysis followed by results and discussion. The paper ends with a conclusion and future work section. 2. BACKGROUND AND RELATED WORK Our research approach builds on mining function-based association rules for universal design, exploring the applicability of these rules across a range of different products, and by extension using the rules as a clustering metric for product families. Thus, we provide a brief background on the relevant areas like universal design, association rule mining and product family design Design design of products encompasses a wide array of products and services including information technology and communication products, architectural systems, and public facilities [11]. Here, we focus on universal design of consumer products. Relevant work is performed at the North Carolina State University, the University of Wisconsin-Madison, the University of Cambridge, and the University of Buffalo. The work done at each university is briefly summarized below. A team of researchers organized through the Center for Design at North Carolina State University has compiled seven principles of universal design [1, 12]. These principles have been well received by designers in a range of disciplines. Vanderheiden has developed a set of guidelines for the design of consumer products [13]. These guidelines primarily focus on the accessibility of standard computer hardware and software for people with disabilities. A team of researchers at the University of Cambridge has produced implementable results for universal product design [2, 14-18]. The focus of the Cambridge research group has been in modeling user groups, creating product assessment methods, and extending the needs of universal design to modern product design processes. Housed in the Center for Inclusive Design and Environmental Access at the University of Buffalo is an active group of researchers with focus on universal design [3, 19-21]. Though this group focuses on architectural design and comes from an architectural background, it has performed research on appliances and other applications that extend to product design. In summary, remarkable efforts have been put into improving design for those with a disability. Nevertheless, there is still opportunity for significant contributions to be made in many areas including those focused on early design issues and decisions Association Rule Mining As more products pairs are studied, a formal method to seek patterns in the data is needed. Manual analysis of valuable information becomes difficult as the database of available information grows [22]. Data mining and knowledge discovery in databases (KDD) are methods for extracting models and patterns of interest from large databases [23]. Association rule learning is a data mining techniques that finds associations between variables in large data sets. Here, we employ association rule learning to search for universal design rules from the similarities and differences in the product pairs. Association rule learning mines the data organized as transactions of items. The data contains I = {i 1, i 2,, i n } which is a set of categorical attributes called items and T = {t 1, t 2,, t n } which is a set of transactions. Each transaction in T contains a subset of the items in I. An association rule is defined as an implication of the form X! Y where X, Y " I and X # Y = $. The itemsets X and Y are called the antecedent and consequent of the rule, respectively. Market basket analysis serves as a simple example of association rule learning. The items are goods available at a supermarket and a transaction is what a single shopper puts into their shopping basket for purchase. Mining through a large number of transactions, an association rule might determine that there is a strong association between an itemset of peanut butter and jelly and an itemset of bread: peanut butter and jelly is an antecedent for bread as a consequent. In association rule learning, support (supp) of an itemset X is the percentage of transactions in T that have a specific itemset X, and confidence (conf) is the percentage of transactions in T such that, given they contain X, also contain Y. Formally, conf (X! Y ) = supp (X % Y ) / supp (X) (1) Generating association rules from a database requires an algorithm to search and extract rules typically based on a user supplied minimum level of support and confidence [24]. Several algorithms have been developed and are available for mining association rules from datasets, for example, Apriori, Eclat, and FP-Growth [25, 26]. WEKA workbench is one such collection of the state-of-the-art machine learning algorithms and data preprocessing tools [27]. WEKA allows either 2

3 implementing an existing learning method or writing and testing a custom learning scheme. The Apriori algorithm, as implemented in WEKA, is employed to extract association rules for the research presented here [24] Product Family Design Product family design is a way to achieve cost-effective mass customization by allowing highly differentiated products to be developed from a common platform while targeting products to distinct market segments [28]. Here, mass customization offers an avenue to provide persons with disabilities a wide range of products [29]. Related previous efforts extend methods from mass customization and product family design to create specific methods for universal product family design [29, 30]. Extending the functional representation of universal products, Moon and McAdams develop several methods to create and make decisions about universal product families. The basic framework for the product family is a functionbased modular architecture [31]. Other work includes formal methods for evaluating the accessibility of models [32], the extension of game theoretic approaches to identify the best modules for a product family [29, 30], and the application of formal real value option methods to initially proposed product architectures to enable the creation of a flexile universal product families [33]. 3. RESEARCH APPROACH This section explains the research approach applied to study universal product characteristics. An overview of the research activity is depicted in Figure 2. In the first step, the universal design information from the existing universal products is formalized, which is detailed in section 3.1. products are identified, product pairs are formed, and actionfunction diagrams are created for each product pair. Next, product pairs are compared to study the similarities and differences. In the second step, the data generated in the first step is preprocessed for data mining. The Apriori algorithm mines association rules from the pre-processed data. Generated rules are post-processed to filter the rules applicable for universal design of products. Data Generation!Select universal product pairs!create actionfunction diagrams!compare the product pairs 3.1. Data Generation This section explains how the design information contained in existing examples of universal design is formalized. The following section gives a brief overview about product pairs. Further, the design difference classification and the representation scheme are stated with an example of a can opener product pair. In addition, the concept of actionfunction diagram is clarified as a framework to model the user-product interaction Products Pairs Two products that provide the same overall functionality but differ in their level of accessibility are termed a product pair [7, 34]. The universal product better accommodates the user with a disability by introducing one or more design features that are not observed in the typical version. Figure 1 shows a can opener product pair. The typical can opener is shown on the left and the universal can opener on the right. In a typical can opener, the user needs to grasp the opener with one hand and twist the handle with the other hand. In contrast, an electric can opener utilizes electrical energy to cut open the can. In an electric can opener, the user simply presses the switch - while grasping with one hand - to cut open the can. The electric can opener is more accessible since it allows opening of a can with one hand and eliminates the activity of twisting the handle. FIGURE 1. PRODUCT PAIR OF A CAN OPENER [10] The selection criterion for product pairs is detailed in our prior work [9, 10]. Here, the scope of products studied is limited to those focused on addressing physical impairments. Products pairs in which the universal product addresses user cognitive limitations are not considered. Also, computer and information technology (IT) products are not analyzed. The product pairs studied for this research are listed in the appendix. Data Analysis!Pre-process data!association rule generation!post-process rules generated FIGURE 2. RESEARCH ACTIVITY TO STUDY UNIVERSAL PRODUCT CHARACTERISTICS 3

4 Design Difference Classification Products pairs are compared functionally to identify the design differences that make the products universal. These design differences are classified as parametric, morphological, or functional [8, 34]. A parametric difference between a product pair refers to two products with the same detailed functionality, solution principle, and form but differing in the value for some parameter. A morphological difference refers to two products that retain the same detailed functionality but exhibit a different physical solution principal, form, or geometric topology. A functionally different product pair indicates the addition or deletion of a product function, or the change of some specific product function, to improve its accessibility. If a function is added to the universal design, it is recorded as function addition whereas if the function is deleted from the typical design it is recorded as function deletion. Previous research indicates that classification of changes as parametric, morphological, or functional provides a clear framework for comparing a product pair [9]. The representation scheme to reflect parametric, morphological, or functional difference as a product changes from typical to universal is reproduced in Figure 3. A solid line, a dotted line, and a dashed line represent the material, energy, and signal flows respectively. Dashed boxes in the actionfunction diagram represent user activities with related product functions contained within each activity. The concept of actionfunction diagram is clarified further. ICF User activities Functional basis Material Signal Functional change Morphological change Parametric change FIGURE 3. REPRESENTATION SCHEME AND TO REFLECT PARAMETRIC, MORPHOLOGICAL, OR FUNCTIONAL DIFFERENCE IN ACTIONFUNCTION DIAGRAMS Actionfunction Diagrams The product analysis framework proposed to improve UD research and practice is termed as the actionfunction diagram [7, 8]. In an actionfunction diagram, activity diagrams and functional models are combined into a single graphical representation [35, 36]. This formal user-product interaction representation facilitates comparison between typical and universal products. The actionfunction diagram is explained in detail in our prior work [9, 10]. The work to date shows that, the actionfunction diagram is a useful framework as it provides a clear coupling between the interaction of user activity and customer need, which in this case is a limitation due to disability [9]. Additionally, the framework allows for analysis of user-product interaction early in the design process. The International Classification of Functioning, Disability and Health (ICF), established by the World Health Organization provides a standardized common language for the description of health and health-related states [37]. Product functions are modeled using Functional Basis and user activities are represented with ICF lexicon [8, 37]. Figure 4 reproduces the actionfunction diagrams of a can opener product pair shown in Figure 1. The typical can opener relies on human energy. The universal can opener uses electrical energy for several functions. Thus, the functions like import human energy, transfer human energy, and convert human energy to mechanical energy are deleted in the universal design. The functions added in the universal design are import electrical energy, store electrical energy, supply electrical energy, and actuate electrical energy as well as convert electrical energy to mechanical energy. The magnet in the universal design adds the functionality of exporting the lid away from can. The rotary cutter is morphologically different from the stationary point cutter. Both openers offer morphologically different provisions for grasping with hand. As outlined here for the can opener product pair, actionfunction diagrams of all the 65 product pairs are compared to study the universal design elements. The 3.2. Data Analysis similarities and the differences in the product pairs are further analyzed in detail.as the data analysis follows the data generation step, the input to the data analysis is the actionfunction diagrams of 65 product pairs. This section explains the formal analysis of the information contained in the actionfunction diagram. Data analysis consists of three steps: pre-processing, association rule mining, postprocessing Preprocessing After comparing the actionfunction diagrams, the next step is to tabulate similarities and differences between product pairs. A candidate function set is a product function in a typical product along with the corresponding function in its universal counterpart, which is a possible candidate for a design change that leads to a universal product. A user activity may or may not be associated with a candidate function set. Each row of the table corresponds to a candidate function set. Each candidate function set acts as an input transaction to the Apriori algorithm. The items recorded in a transaction are product functions and user activities in the typical and the universal products in addition to the type of change in the product function. Table 1 records the similarities and differences for the can opener example depicted in Figure 4. 4

5 Actionfunction Diagram: Can Opener Human Import Human Picking up Grasping Turning Hand Import hand Position hand Transfer Human Convert Human to Mech Closed Can Carrying, moving and handling objects Import Solid Manipulating Couple Solid Guide Solid Separate Solid Manipulating Open Can + Can Lid Actionfunction Diagram: Can Opener Mechanical Hand Picking up Import hand Position hand Grasping Actuate Electrical Electrical Import Electrical Store Electrical Supply Electrical Convert EE to Mech Closed Can Carrying, moving and handling objects Import solid Manipulating Couple Solid Guide Solid Separate Solid Export Solid Open Can + Can Lid Mechanical FIGURE 4. THE ACTION-FUNCTION DIAGRAM FOR A TYPICAL CAN OPENER (TOP) AND A UNIVERSAL CAN OPENER (BOTTOM) [10] TABLE 1. PREPROCESSING STEP TO CONVERT INFORMATION FROM ACTIONFUNCTION DIAGRAMS INTO INPUT TRANSACTIONS FOR THE APRIORI ALGORITHM FOR THE CAN OPENER EXAMPLE Product function User activity Product function User activity Change Import Hand Picking up Import Hand Picking up None Position Hand Grasping Position Hand Grasping Morphological Import Human N/A Import Human N/A None Transfer Human Turning N/A N/A Function Deletion Convert Human Turning N/A N/A Function Deletion to Mechanical Import Solid Carrying, moving and Import Solid Carrying, moving and None handling objects handling objects Couple Solid Manipulating Couple Solid Manipulating None Guide Solid N/A Guide Solid N/A None Separate Solid N/A Separate Solid N/A Morphological N/A N/A Actuate Electrical Grasping Function Addition N/A N/A Import Electrical N/A Function Addition N/A N/A Store Electrical N/A Function Addition N/A N/A Supply Electrical N/A Function Addition N/A N/A Convert Electrical to N/A Function Addition Mechanical N/A Manipulating Export Solid N/A Function Addition 5

6 For example, the first row of Table 1 records the user activity of picking up the can opener corresponding to the product function of importing hand into the system. The userproduct interaction of picking up and importing hand remains the same for both designs. The export solid function performed by the magnet to lift the lid, shown on the last row of Table 1, is an addition of functionality. The export solid function has no associated activity in the universal design. However, the typical design requires the user to remove the lid by manipulating. The aim of this exercise is to identify the changes in the product function as the product becomes more accessible. User ability being the prime focus of universal design, it is interesting to know if there is any change in the user activity. For instance, in the electric can opener the activity of turning is eliminated, thus making it universal. In a similar manner, the similarities and differences in all 65 product pairs are tabulated for the study. Product pairs like the universal washer are complex with comprehensive actionfunction diagrams hence more number of input transactions associated with it. In contrast, a simpler product pair like a shovel has fewer input transactions. The average number of transactions per product pair is eight Association rule mining The pre-processed data is input to the association rule learning scheme. The Apriori algorithm requires the user to select minimum threshold values for the support and the confidence. If the minimum value of support and confidence is set to a high threshold, say 5% support and 90% confidence, the rules obtained contain only the most frequent itemsets and all permutations of those itemsets. For instance, the importing functions for the energy, material, and human are most frequent in the dataset. Hence, only the association rules pertaining to importing functions are mined when the support and confidence is set to 5% and 90% respectively. For this study, we choose to set the minimum values of support and confidence to a low threshold value to discover as wide a range of rules and results as possible. Low values of support and confidence allow mining of rules over wide range of activities and functions, giving better insight into universal design. A low value of support is admissible as it implies statistical significance [38]. Though a single transaction occurrence is not generally of interest in the sense that it does not suggest a strong rule, we chose to over explore the potential rule space to find a greater number of strategies for universal design. The minimal support level is set at 0.15 %, such that even itemsets with a single occurrence in the dataset are established as a rule. Consider that antecedent A leads to two different consequent, B and C. Two possible rules are A!B and A!C, and the confidence of each rule is 50%. Due to the variety of product pairs in the dataset, there are instances having the same antecedent but two different consequents. Such rules with 50% confidence gives the product designer a choice of consequent given the antecedent during conceptual stage of design. For example, either a parametric or a morphological change to product function indicate status can make the activity of seeing more inclusive. A designer can choose between the two rules for the optimum solution to address all customer requirements. Hence, the minimum value of confidence is set to 50% for this study Post-processing The Apriori algorithm is set to generate rules with a minimum confidence level of 50% and minimum support level of 0.15 %. The downside of setting low values of confidence and support is that the algorithm mines 13,300 rules from 523 transactions. Hence, the rules require further post-processing to obtain information dense set of rules. In this research, we are interested in the interplay between a user activity, a product function, and any observable design details that make a product more accessible. Notably, the Apriori algorithm generates different permutations of the same itemsets, some of which are shown below. ( Product function in typical = Position Hand, User activity in typical = Grasping)! ( Change = Parametric, User activity in universal = Grasping, Product function in universal = Position Hand) ( Change = Parametric, User activity in typical = Grasping)! ( Product function in typical = Position Hand, Product function in universal = Position Hand) ( Product function in universal = Position Hand, User activity in universal = Grasping)! ( Change = Parametric) ( Change = Parametric)! ( User activity in universal = Grasping) ( Product function in universal = Position Hand)! ( Product function in typical = Position Hand) In the initial stages of design, the information available to a product designer is the product function, the user activity, and the associated user-product interaction. The designer is confronted with the question as what change to one or more product functions would make the user-product interaction better accessible. In other words, given the user activity and product function in typical design (antecedent), what type of change in the product function (consequent) would make it universal? Thus, rules given by the Apriori algorithm are filtered to obtain rules specific for universal design. The rules are filtered such that they contain all the required information with appropriate antecedent and consequent. The filters are explained further. Filter 1: To be a rule with a parametric change, no change, or a function deletion change, the antecedent must contain the product function and the user activity from the typical design. Furthermore, the consequent comprises of the change advocated by the rule. The user activity and the product function remains unchanged for the universal design in the case of a parametric change or no change. For the rules 6

7 stating deletion of a function, both the user activity and the product function are deleted in the universal design. The generic form of the rule is (Product function in typical, User activity in typical)! (Change), where Change = Parametric or No change or Function deletion. For instance, the user-product interaction of grasping and position solid is made universal by parametric improvements to the handle design; this rule is stated as position solid + grasping!parametric. No change is required to product function guide liquid which has no user activity associated with it; this rule is represented as guide liquid + no activity!no change. The function of transferring human energy is deleted in the universal design and so is the turning activity; this rule is represented as transfer human energy + turning!function deletion. Filter 2: To be a rule with a morphological change, antecedent must contain the user activity and the product function of the typical design. The type of change and user activity in the universal design is sought as the consequent of the rule. Inclusion of user activity in the consequent helps to keep track of the changes in user activity, if any. In most of morphological changes, the user activity remains the same but made easier. However, we are particularly interested in the change in the user activity for the morphological changes. The generic form of the rule is (Product function in typical, User activity in typical)! (Change +User activity in universal), where Change = Morphological. Consider the rule, Indicate status + communication using written!morphological + communication using Braille. In this rule, the user activity changes from typical to universal design on incorporating a morphological change to the function indicate status by including Braille script. Filter 3: To be a rule with a function addition change, the antecedent must contain the user activity from the typical product and product function from the universal product. The consequent must be the type of change and the user activity in universal product. The functional addition change does not have a product function in the typical product as they are added in the universal design; hence, the product function form the typical design cannot serve as the antecedent. Rules with functional addition are interpreted as what functions if added to the product makes it universal? The generic form of the rule is (Product function in universal, User activity in typical)! (Change +User activity in universal), where Change = Function addition. The rule, import electrical energy + no activity!function addition, is interpreted as addition of import electrical energy to a product makes it universal. In the same context, any other form of rule, like manipulating! guide solid or parametric!position solid, does not give sufficient information to apply the rule for universal design. Hence, rules providing incomplete information are filtered out. A code is executed in MATLAB to execute these three filters on all the association rules generated by the algorithm. The code outputs a set of function-based association rules for universal design in a format that is easy to comprehend. The post-processing step eliminates the unnecessary statistical information given by Apriori algorithm such as number of cycles performed or the computation time. The code can also be tuned to select only those rules with high values of confidence and support, thus allowing better statistical control on the association rules filtered. 4. RESULTS AND DISCUSSION This section presents the results of the association rule generation for universal product design. The first subsection compares the association rules for universal design from 65 product pairs to our previous work with 15 product pairs and evaluates the methods and results as the product domain studied becomes larger [10]. The second subsection explains clustering of the product pairs based on the rules applicable to them and discusses transferability of the rules Comparison with Previous Results In our previous work, we applied the Apriori algorithm to mine association rules for universal design from 15 product pairs [10]. Here, we compare the association rules obtained from 15 product pairs to the rules obtained from 65 product pairs. Also in our previous work, we applied the data mining software TANAGRA for mining association rules and the post processing of rules was done manually [10, 39]. This paper presents work in which association rules are mined with WEKA software and the post-processing step is automated, the three steps of data analysis as described above are performed on the prior dataset of 15 product pairs to eliminate any inconsistencies. Table 2 compares the results obtained from the previous dataset of 15 product pairs to the results obtained from the current dataset of 65 product pairs. The number of input transactions for the dataset of fifteen products is 135 with average of nine transactions per product pair. Along the same lines, the dataset of sixty-five product pairs consist of 523 transactions altogether, with an average of eight transactions TABLE 2. COMPARISON OF ASSOCIATION RULES OBTAINED FROM THE DATASET OF 15 PRODUCTS PAIRS TO THE DATASET OF 65 PRODUCT PAIRS 15 product pairs 65 product pairs Number of input transactions Total number of association rules Filtered association rules Association rules with support of more than With change one instance No change

8 per product pair. The number of transactions per product pair provides an estimate of the complexity of products pairs studied. The overall complexity of products in both datasets is similar, thus, facilitating the comparison. The total numbers of association rules are all the rules mined by the Apriori algorithm for a confidence of 50% and support that reflects a minimum of one transaction in the dataset. Filtered association rules are those rules obtained after filtering in the postprocessing step. In addition, Table 2 lists the association rules with support of more than one instance both with and without a change. Rules with a change are those recommending either a parametric, a morphological or a functional change to the user-product interaction. Table 3 and Table 4 lists the filtered association rules for the dataset of 15 product pairs and dataset of 65 product pairs respectively. Support of these rules is such that the consequent occurs more than once for a given antecedent. In other words, the rule is observed in at least two product pairs. The confidence of association rules listed is greater than 50%. The antecedent of a rule is the product function and the user activity while the consequent is the change in product function and change in the user activity in the universal design. Rules listed are only those reflecting a change in the product function. Rules resulting in no change are not discussed here for brevity. Table 4 sorts the rules in order of decreasing level of support. Eight rules are common between the two datasets as shown in Table 3 and Table 4. Three rules from Table 3, namely Rule 6, Rule 9 and Rule 11, are not repeated in Table 4. Sixteen association rules are added due to the addition of 50 product pairs to the dataset. Rules that are not mined in the larger dataset do not have the same level of confidence in the larger dataset with 65 product pairs. Rules that are repeated in both product sets do not have the same values of confidence and support, since values of support and confidence depend on the total number of transaction in the dataset. Rules with support of more than one instance imply that the consequent occurs two or more times for the given antecedent. The percentage value of support given in Table 3 and Table 4 depends on the total number of input transactions. Since the total number of input transactions for the dataset of 15 product pairs is different from the dataset of 65 product pairs, the minimum value of support for both datasets is different. The minimum value of support for the rules listed in Table 3 is 2/135 or 1.48%. Similarly, minimum value of support for the association rules listed in Table 4 is 2/523 or 0.38%. Figure 5 and Figure 6 show plots of the number of association rules mined verses the number of products in the dataset. The trend of plots in Figure 5 and Figure 6 enables us to check if the set of association rules for universal design is approaching some finite number as the product set grows arbitrarily large. Figure 5 shows a plot of the total number of association rules generated by the Apriori algorithm against the number of product pairs in the dataset. Figure 6 shows the number of filtered association rules after post-processing against the number of product pairs studied. The algorithm is set to mine rules with support of at least one instance and a minimum value of confidence greater than 50%. Data analysis is repeated on incrementally adding product pairs to the dataset. Here, we have added any 2 random product pairs, without replacement, till all 65 product pairs are included. The rule generation process is repeated 5 times, each for different random order of the product pairs, to calculate the error in the number of the rules generated. In Figure 5 and Figure 6, we have shown the mean and standard deviation of the number of rules generated; the cross represents the mean and the error bar represents the standard deviation. FIGURE 5. TOTAL NUMBER OF ASSOCIATION RULES GENERATED BY THE APRIORI ALGORITHM AGAINST THE NUMBER OF PRODUCT PAIRS IN THE DATASET FIGURE 6. NUMBER OF FILTERED ASSOCIATION RULES AFTER POST-PROCESSING AGAINST THE NUMBER OF PRODUCT PAIRS IN THE DATASET The trends in Figure 5 and Figure 6 shows that the number of association rules generated per product pair reduces as more products are included in the dataset; the slope of the graph decreases gradually and than almost reaches a constant value. After having a considerable quantity and variety of universal products in the dataset, the number of association rule mined do not increase proportionately with addition of every new product pair to the dataset. 8

9 TABLE 3. ASSOCIATION RULES GENERATED WITH MINIMUM SUPPORT OF TWO INSTANCES SUGGESTING CHANGES IN THE PRODUCT FUNCTIONS FOR THE DATASET OF 15 PRODUCT PAIRS ANTECEDENT CONSEQUENT Product Function User Activity Functional Change User Activity Change Confidence Support 1 Import EE No Activity Functional Addition No Activity Convert EE to ME No Activity Functional Addition No Activity Guide Human Standing Functional Addition Same as Guide Human Sitting Functional Addition Sitting Actuate Signal No Activity Functional Addition Pushing with fingers Import Hand No Activity Functional Addition Reaching Transfer HE Manipulating Morphological Same as Convert CE to ME No Activity Functional Addition No Activity Guide Solid Manipulating Morphological Pushing with fingers Separate Solid No Activity Morphological No Activity Guide Solid Pulling Morphological No Activity TABLE 4. ASSOCIATION RULES GENERATED WITH MINIMUM SUPPORT OF TWO INSTANCES SUGGESTING CHANGES IN THE PRODUCT FUNCTIONS FOR THE DATASET OF 65 PRODUCT PAIRS ANTECEDENT CONSEQUENT Product Function User Activity Functional Change User Activity Change Confidence Support 1 Position Hand Grasping Parametric Same as Convert EE to ME No Activity Functional Addition Same as Import EE No Activity Functional Addition Same as Actuate Signal No Activity Functional Addition Pushing with fingers Position Solid Carrying in Hands Parametric Same as Position Hand Reaching Parametric Same as Position Human Maintain Body Position Parametric Same as Import CE No Activity Functional Addition Same as Position Hand Carrying in Hands Parametric Same as Indicate Status Seeing Morphological Same as Separate Solid No Activity Morphological Same as Secure Hand Grasping Functional Addition Same as Store EE No Activity Functional Addition Same as Supply EE No Activity Functional Addition Same as Guide Human Standing Functional Addition Same as Sense Status No Activity Morphological Same as Convert CE to ME No Activity Functional Addition Same as Transfer HE Turning Functional Deletion Same as Guide Solid Pushing with fingers Parametric Same as Position Hand No Activity Functional Addition Reaching Guide Human Sitting Functional Addition Same as Indicate Status Communication Written Morphological Communication Braille Position Hand Manipulating Parametric Same as Transfer HE Manipulating Morphological Same as

10 The first 15 products are not necessarily the dataset of original 15 product pairs. A point called previous results is added in Figure 5 and Figure 6 to depict the data point for the original dataset of 15 product pairs. Though the specific rules resulting from the analysis of the two datasets may be different, the total number of rules is similar indicating a common rate of rule generation for different sets of products. The decreasing rate of association rules mined as depicted in Figure 5 and Figure 6 indicates the potential to capture some tractable set of universal design rules. Thus, the universal design knowledge can be captured and formalized by a set of association rules based on functional representation. As products are added and analyzed, rules that are repeated more often increase in the values of support and confidence Clustering of product pairs and transferability of the rules The general template for the rules generated is that given a product function and user activity, how do a typical and universal product differ. Specifically, we wish to express and understand this difference in a manner that allows a designer to better design, or redesign, a product for accessibility. The rules are to be applied at the concept generation stage of design and are rooted in a functional abstraction of the design problem. In practice, a designer could query for a rule based on the actionfunction diagram of the product being designed. The rule could suggest functional, morphological, parametric, or no change - with associated support and confidence numbers to the designer. The designer is then tasked with applying the suggested change to their design: the designer is transferring design knowledge in the form of rules from prior designs to a new design. To support such design activity, a knowledge base of existing designs could be provided to the designer to understand and implement the rule. In this context, we explore the transferability of rules from one product to the next. We explore the transferability of rules in the context of clusters of products based on sharing common rules. The clustering of products is important for several reasons. Products that contain multiple common rules are likely to be more similar, thus facilitating a tighter analogy between products and simpler rule transfer. Clusters are also important as we wish to explore the potential to create universal product families based on the products that share common rules and in turn common embodiments of those rules. To cluster products, a straightforward rule search is employed. Three iterative loops search through the products. The first loop searches for clusters of products with any four rules in common. The second loop searches for clusters of products with any three rules in common. The third loop searches for clusters of products with two common rules. Clusters of products and sharing of the rules are represented using Venn diagrams. One such cluster is shown in Figure 7. The products contained within each circle represent a cluster. The rules contained within every circle are applicable to all the products within the same circle. Two or more clusters might share a rule. Furthermore, two or more clusters can share a product. Figure 7 consists of clusters of the handheld product pairs that share Rule 1. Rule 1 recommends a parametric improvement in the design of a handle for better grasping while the product positions hand. Figure 8 thru Figure 11 shows products pairs that adopt an ergonomically designed handle for the universal product. Cluster # 2 Cluster # 1 Cluster # 4 Rule 12 Shovel Handheld shower Recliner Lever Box Cutter Chopping Bowl Rule 1 Rule 24 Scissor Pruner Rule 11 Rule 19 Seat Belt Adaptor Syringe Remote Cluster # 3 Rule 1: Position Hand + Grasping =>Parametric Rule 11: Separate Solid + No Activity =>Morphological Rule 12: Secure Hand + Grasping =>Functional Addition Rule 19: Guide Solid + Pushing with Fingers =>Parametric Rule 24: Transfer HE + Manipulating =>Morphological FIGURE 7. CLUSTERS OF UNIVERSAL PRODUCTS SHARING RULE 1 THAT RECOMMENDS A PARAMETRIC CHANGE TO THE FUNCTION POSITION HAND FOR BETTER GRASPING Figure 8 shows product pairs from cluster #1 that share Rule 1 and Rule 24. In addition to ergonomic handle design, these product pairs exhibit morphological change in the way the human energy is transferred for cutting action. scissors incorporate a spring to aid the cutting action. The universal pruner has a four-link mechanism to transfer the human force while cutting. The scissors and pruner provide a close domain design analogy for transferability of rules between office supplies and garden tools. The product pairs used to generate the rules are shown side by side in Figure 8. The transfer of rules from one product to the next moves up and down in the figure. The scissor and pruner example in Figure 8 illustrate the transfer of rules from similar products. Scissors and pruners are functionally and morphologically similar to each other with specific parametric changes implemented to improve their intended applications. The fact that the function-based rule clustering methods 10

11 clusters the scissors and pruner indicate that the method finds close domain products with intuitively transferable rules. Scissor or a medical device. For cluster #3, the physical embodiment of Rule 19 is quite distinct, thus, providing distinct analogies for universal design. Products shown in Figure 10 are from diverse domains. Though they represent a common rule, the physical embodiment of the rule is less similar than in the cluster #2 case illustrated in Figure 8. Pruner Remote FIGURE 8. PRODUCT PAIRS IN CLUSTER #1 SHARING RULE 24 THAT SUGGESTS A MORPHOLOGICAL CHANGE TO FUNCTION OF TRANSFERRING HUMAN ENERGY FOR EASIER MANIPULATING Figure 9 shows product pairs from cluster #2 that share Rule 12 and Rule 1. Rule 12 is the addition of a secure hand function to a product that involves the user activity of grasping. As embodied in a handheld shower nozzle, a shovel, a recliner lever, and a box cutter the result is the addition of a closed loop handle that prevents the product from slipping out of user s hand. As a contrast to the scissor and pruner case, the products shown in Figure 9 are dissimilar. The product domains represent bathroom fixtures, garden equipments, and furniture. Nevertheless, inspection of the design changes in the different products in Figure 9 shows similar implementation of the rule. Thus, designers developing universal design knowledge in one domain can apply that knowledge to a different domain while applying function-based association rules for designing universal products. Hand Held Shower Seat Belt Adaptor Syringe FIGURE 10. PRODUCT PAIRS IN CLUSTER #3 SHARING RULE 19 THAT SUGGESTS A PARAMETRIC CHANGE TO GUIDE SOLID FUNCTION TO AID PUSHING WITH FINGERS Figure 11 shows the product pairs in cluster #4 that share Rule 11 and Rule 1. Rule 11 suggests a morphological change to the function separate solid. A chopping bowl and a box cutter exhibit a morphologically different cutter in universal design. Though products pair from cluster #4 are from a similar domain, all of them are cutting tools, the physical embodiment of rule 11 is distinct Shovel Chopping Bowl Recliner Lever Box cutter FIGURE 9. PRODUCT PAIRS IN CLUSTER #2 SHARING RULE 12 THAT SUGGESTS ADDITION OF A FUNCTIONALITY TO SECURE HAND WHILE THE USER GRASPS Figure 10 shows product pairs from cluster #3 that share Rule 19 and Rule 1. Rule 19 recommends parametrically large tabs or buttons on products that requires a user to push with fingers for operating the device. Rule 19 can be applied for universal design of an electronic device, automobile interiors, FIGURE 11. PRODUCT PAIRS IN CLUSTER #2 SHARING RULE 11 THAT SUGGESTS A MORPHOLOGICAL CHANGE TO THE CUTTER The discussion on transferability of rules continues with more examples demonstrating the physical embodiment of the rules. The Venn diagrams shown in Figure 12, Figure 13, and Figure 14 are similar to Figure 7. A designer can look up for such examples of rules as analogies to generate ideas for designing a universal product. 11

12 Figure 12 shows clusters of products pairs, sharing Rule 5, Rule 6, Rule 7, and Rule 9. Cluster #7 comprises of the washer, the dishwasher, and the oven and they share Rule 5, Rule 7, and Rule 9. Rule 7, which is also shared by cook top, suggests provision of knee space for allowing access to a user on wheel chair. Rule 5 and Rule 9, which is also shared by a refrigerator, recommends some intermediate counter space besides the products for placing hot or cold objects. The cook top and the refrigerator share Rule 6 that specify that controls must be located within reach range of the user. Product pairs stated in cluster #5, cluster #6, and cluster #7 are closely related, as all of them are household appliances. The space around the product is modified to accommodate gross user access to the device. The rules embodied in these products can be conveniently transferred to universal design of other household appliances. Figure 13 shows clusters of product pairs sharing Rule 8, Rule 10, Rule 16, and Rule 17. Cluster #8 contains a kitchen scale, a blood pressure monitor, and a thermometer. Rule 10 and Rule 16 recommend a morphological change to the functions sense status and indicate status such that seeing is easier. Morphological changes recommended by Rule 10 to the function indicate status is obtained by a digital display with large font size. The morphological change suggested by Rule 16 to function sense status is achieved with a sensor using a different solution principle. Rules 8 and Rule 17 suggests addition of functionalities to import chemical energy in form of fuel or batteries and converting chemical energy into mechanical energy. It is worth noting that products pairs from clusters #8 are from distinct product domains, like kitchen equipments and medical devices, but all of them sense, measure and display some physical parameter. In contrast, cluster #9 presents a distant domain analogy for importing external chemical energy. In Figure 14, cluster #10 comprises of a wine opener, a can opener, and an automatic toothbrush and the applicable rules are Rule 2, Rule 3, Rule 13, and Rule 14. These rules suggest addition of functionalities like import, store and supply electrical energy and convert electrical energy to mechanical energy to automate the devices. Cluster #11 consists of a wine opener and a can opener similar to cluster #10 except for the addition of Rule 18. Rule 18 recommends deletion of the turning activity. A toilet auto seat, a recliner, and an armchair share Rule 15 that states functional addition to guide human while standing. Note that, cluster #11 contains product that are closely related and present near analogies. Cluster #12 contains dental care products and kitchen tools that are fairly distinct. Product pairs from very distinct product domains like residential furniture, bathroom fixtures, dental care products, and kitchen tools share Rule 2 and Rule 3 which are external energy related functions. Product pairs from Cluster # 12 and Cluster #13 are quite similar. Washer Cluster # 5 Cluster # 6 Dishwasher Cook Top Rule 7 Rule 6 Oven Rule 5 Rule 9 Refrigerator Oven Washer Dishwasher Cluster # 7 Cook Top Refrigerator Rule 5: Rule 9: Rule 7: Rule 6: Position Solid + Carrying in Hands! Parametric Position Hand + Carrying in Hands! Parametric Position Human + Maintaining Body Position! Parametric Position Hand + Reaching! Parametric FIGURE 12. CLUSTERS OF ARCHITECTURAL PRODUCT PAIRS SHARING RULE 5, RULE 6, RULE 7, AND RULE 9 THAT DEAL WITH A PARAMETRIC CHANGE TO ALLOW GROSS USER ACCESS 12

13 Kitchen Scale Cluster # 8 Thermometer Rule 10 Rule 16 Kitchen Scale Thermometer BP Monitor Rule 8 Blood Pressure Monitor Rule 17 PT Cruiser Cluster # 9 PT Cruiser Rule 10: Indicate Status + Seeing! Morphological Rule 16: Sense Status + Seeing! Morphological Rule 8: Import CE + No Activity! Functional Addition Rule 17: Convert CE to ME + No Activity! Functional Addition FIGURE 13. CLUSTERS OF PRODUCT PAIRS SHARING RULE 8, RULE 10, RULE 16, AND RULE 17 Wine Opener Cluster # 11 Cluster # 10 Rule 13 Rule 14 Wine Opener Can Opener Rule 18 Can Opener Toothbrush Auto Rule 2 Rule 3 Rule 4 Recliner Cluster # 12 Toothbrush Rule 15 Toilet Auto Seat Rule 21 Arm Chair Cluster # 13 Recliner Toilet Auto Seat Arm Chair Rule 2: Convert EE to ME + No Activity! Functional Addition Rule 3: Import EE + No Activity! Functional Addition Rule 4: Actuate Signal + No Activity! Functional Addition + Pushing with fingers Rule 13: Store EE + No Activity! Functional Addition Rule 14: Supply EE + No Activity! Functional Addition Rule 15: Guide Human + Standing! Functional Addition Rule 18: Transfer HE + Turning! Functional Deletion Rule 21: Guide Human + Sitting! Functional Addition FIGURE 14. CLUSTERS OF PRODUCT PAIRS CONSISTING OF VARIETY OF PRODUCTS 13

14 One of our goals is to observe the conceptual similarity of a rule, as it is applicable to products from disparate product domains. The desire is to have sufficient conceptual similarity such that the rules provide insight for universal design across a wide range of products. Such similarity is seen in the product cluster shown in Figure 9. Further, we wish to explore the physical similarity in the context of creating universal product families based on a platform of that common accessible element. Could the actual embodiment of the design result in a common component that can be shared across multiple products serving as a product platform? Garden tools in Figure 15 illustrate such a case. An ergonomic, non-slip type of handle can act as universal product platform that can be shared by other handheld devices. In this case, the product family is built from similar products from a common domain. Products that can be derived from the platform of universal handle are shown in Figure 15. FIGURE 15. OTHER HAND HELD PRODUCTS THAT EXHIBIT RULE 12 AND RULE 1 IN FORM OF AN ERGONOMIC HANDLE DESIGN Figure 16 illustrates transferability of Rule 1 from a spatula to a trowel in form of an accessible add-on module. In this case, the rule is observed in a kitchen tool and applied for universal design of a garden tool. The tool can be initially designed to be universal as shown in Figure 15 or an accessible module can be retrofitted on the product to make it universal as shown in Figure 16. FIGURE 16. TRANSFERABILITY OF RULE 1 BETWEEN A UNIVERSALLY DESIGNED SPATULA (LEFT) AND A TROWEL (RIGHT) IN FORM OF AN ACCESSIBLE ADD-ON MODULE Clustering of product pairs that share the same rule can indicate transferability of rules. The physical embodiment of a rule can be quite distinct from one product to another. Presenting a product designer with examples of the rules can stimulate solution concept generation for universal design. Presenting examples of physical embodiment along with the set of function-based association rules is crucial. 5. CONCLUSIONS AND FUTURE WORK This paper uses a product representation framework called an actionfunction diagram to formally model the user and the product. Actionfunction diagrams aptly represent user-product interaction. Design similarities and differences are formally compared and tabulated for 65 product pairs in this study. The data is pre-processed and mined for function-based association rules. The rules are filtered in the post-processing step to obtain rules that contain all the required information for universal design. The post-processing step is automated such that more product pairs can be conveniently added to the dataset. The set of rules obtained based on the 65 product pairs are compared to the previous results obtained with 15 product pairs [10]. Adding more product pairs to the dataset improves the statistical significance of the rule generated in terms of better values of confidence and support. The rate of rule generation decreases as more products are added to the dataset of product pairs. Thus, the universal design knowledge can be captured and formalized by a set of association rules based on the functional representation of an arbitrarily large set of products. The transferability of rules from one product pair to another is explored. In this context, we are interested in how the function-based rules provide insight to the specific physical embodiment of a product. To perform this exploration, products are clustered based on the rule commonality. Then within these clusters, the products are evaluated for conceptual and physical similarity in the specific embodiment of the rule. In the case of close domain products such as pruners and scissors, the rule implementation is highly analogous. Such a result indicates the complete rule mining methodology, including both representation and mining scheme, is producing expected results. Additionally, the rules are found to be transferable across products in diverse product domains such as plumbing fixtures and furniture. The transferability of the rules across diverse products domains indicates that the universal design knowledge is broadly applicable in product design. Moreover, this result indicates that the method and resulting rules have the ability to support designers for designing new universal products. However while applying these rules, designers should be mindful that the rules given here are design suggestions which may or may not be applicable to a specific design problem. The term rules follows the data mining terminology. Specific study of how the resulting rules can be used to support new product design remains future work. Additional results indicate that the rules, and product clusters based on the rules, may provide opportunities to create product families based on sharing common components that make the product universal. Examples found include a universal handle for garden tools in an existing product family. Also found was the opportunity to create a product family of diverse kitchen and gardening products. More detailed classification of design differences would better serve product platform design. Application of 14

15 association rule mining for design of universal product families needs to be explored further. A rigorous product platform study for the universal product remains as future work. The dataset of products continues to be expanded with more product pairs. A design repository linking the physical embodiment of a rule to the actual rule also remains as future work. REFERENCES [1] Connell, B. R., Jones, M., Mace, R., Mueller, J., Mullick, A., Ostroff, E., Sanford, J., Steinfeld, E., Story, M., and Vanderheiden, G., 1997, The Principles of Design, July 14, 2009, eshtmlformat.html#top [2] Clarkson, P. J., Langdon, P. M., Goodman-Deane, J., and Robinson, P., 2008, Proceedings, 4th Cambridge Workshop on Access and Assistive Technology, Fitzwilliam College, Cambridge. [3] Danford, G. S., 2010, Center for Inclusive Design and Environmental Access, June 15, [4] Vanderheiden, G. C., and Vanderheiden, K. R., 1992, Guidelines for the Design of Consumer Products to Increase Their Accessibility to Persons with Disabilities or Who Are Aging, March 24, nes/toc.htm [5] Bowe, F., 2000, Design in Education, Bergin and Gavey, Westport, CT. [6] Vanderheiden, G., and Tobias, J., 2000, " Design of Consumer Products: Current Industry Practice and Perceptions " Proceedings of the IEA, pp [7] Kostovich, V., McAdams, D. A., and Moon, S. K., 2009, "Representing User Activity and Product Function for Design, Proceedings of the 2009 ASME Design Engineering Technical Conferences & Computers and Information in Engineering, San Diego, CA. [8] Sangelkar, S., and McAdams, D. A., 2010, "Adapting ADA Architectural Design Knowledge to Product Design: Groundwork for a Function Based Approach," Proceedings of the ASME 2010 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, Montreal, QC. [9] Sangelkar, S., and McAdams, D. A., 2011, "Adapting ADA Architectural Design Knowledge to Product Design Using Association Rule Mining: A Function Based Approach," Journal of Engineering Design, In review. [10] Sangelkar, S., Cowen, N., and McAdams, D. A., 2011, "User Activity Product Function Association Based Design Rules for Products " Design Studies, In review. [11] Preiser, W. F. E., and Ostroff, E., 2001, Design Handbook, McGraw-Hill Inc., New York. [12] Story, M. F., 1998, "Assessing Usability: The Principles of Design," Assistive Technology, 10(1), pp [13] Vanderheiden, G., 1997, Handbook of Human Factors and Ergonomics, John Wiley & Sons, Design for People with Functional Limitations Due to Disability, Aging, or Circumstance. [14] Clarkson, J., 2008, Product Experience, Elsevier, San Diego, CA, Human Capability and Product Design. [15] Langdon, P. M., Persad, U., and Clarkson, P. J., 2008, "Operationalizing Analytical Inclusive Design Evaluation," Nottingham, pp [16] Langdon, P. M., Clarkson, P. J., and Robinson, P., 2008, Designing Inclusive Futures, Springer Verlag, London. [17] Waller, S. D., Landon, P. M., Cardoso, C., and Clarkson, P. J., 2008, "Calibrating Capability Loss Simulators to Population Data," Nottingham, UK, pp [18] Clarkson, P. J., Coleman, R., Keates, S., and Lebbon, C., 2003, Inclusive Design: Design for the Whole Population, Springer-Verlag, London. [19] Danford, G. S., 2003, " Design People with Vision, Hearing, and Mobility Impairments Evaluate a Model Building," Generations, 27(1), pp [20] Feathers, D., 2004, "Digital Human Modeling and Measurement Considerations for Wheeled Mobility Device Users," SAE Transactions, 113(1), pp [21] Steinfeld, E., and Mullick, A., 1990, Design: The Case of the Hand. Innovation, The Official Journal of the Industrial Designers Society of America, pp [22] Braha, D., 2001, Data Mining for Design and Manufacturing: Methods and Applications, Kluwer Academic Publishers, Dordrecht, The Netherlands. [23] Fayyad, U., and Stolorz, P., 1997, "Data Mining and KDD: Promise and Challenges," Future Generation Computer Systems. [24] Agrawal, R., and Srikant, R., 1994, "Fast Algorithms for Mining Association Rules in Large Databases " 20th International Conference on Very Large Data Bases, Santiago de Chile, Chile. [25] Han, J., Pei, J., Yin, Y., and Mao, R., 2004, "Mining Frequent Patterns without Candidate Generation," Data Mining and Knowledge Discovery 8(pp [26] Zaki, M. J., 2000, "Scalable Algorithms for Association Mining," IEEE Transactions on Knowledge and Data Engineering, 12(3), pp [27] Witten, I. H., and Frank, E., 2005, Data Mining: Practical Machine Learning Tools and Techniques Morgan Kaufmann San Francisco, CA. [28] Shooter, S. B., Simpson, T. W., Kumara, S. R. T., Stone, R. B., and Terpenny, J. P., 2005, "Toward an 15

16 Information Management Infrastructure for Product Family Planning and Platform Customization," International Journal of Mass Customization, 1(1), pp [29] Moon, S. K., and McAdams, D. A., 2010, "A Platform-Based Strategic Design Approach for Products," The International Journal of Mass Customization, 3(3). [30] Moon, S. K., and McAdams, D. A., 2010, "A Modular and Coalitional Game Based Method for Designing Product Families," The ASME Journal of Mechanical Design, In review. [31] Moon, S. K., and McAdams, D. A., 2009, " Product Platform and Family Design for Uncertain Markets," Proceedings of the International Conference on Engineering Design ICED 09, Stanford, CA. [32] Moon, S. K., and McAdams, D. A., 2009, "A Design Method for Developing a Product Family in a Dynamic Market Environment " Proceedings of the 2009 ASME Design Engineering Technical Conferences & Computers and Information in Engineering, San Diego, CA. [33] Moon, S. K., and McAdams, D. A., 2010, " Product Family Design Valuation in an Uncertain Market Environment," Proceedings of the ASME 2010 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, Montreal, QC.. [34] McAdams, D. A., and Kostovich, V., 2010, "A Framework and Representation for Product Design," The International Journal of Design, In press. [35] Otto, K. N., and Wood, K. L., 2001, Product Design: Techniques in Reverse Engineering and New Product Development, Prentice Hall, Upper Saddle River, NJ. [36] Hirtz, J., Stone, R., McAdams, D., Szykman, S., and Wood, K., 2002, "A Functional Basis for Engineering Design: Reconciling and Evolving Previous Efforts," Research in Engineering Design, 13(2), pp [37] WHO, 2001, International Classification of Functioning, Disability and Health, World Health Organization (WHO), Geneva, Switzerland. [38] Yun, H., Ha, D., Hwang, B., and Ryu, K. H., 2003, "Mining Association Rules on Significant Rare Data Using Relative Support," Systems and Software, 67, pp [39] Rakotomalala, R., 2004, Tanagra Project, April 28, In this section, the product pairs studied in this research are listed. products are shown on the right and its typical version is shown on the left. Appendix A: Product Pairs Studied TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH (CONTINUED) Product TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH 6 Bicycle Product 1 PT Cruiser 7 Cutting Board 2 Box cutter 8 Food storage box 3 Seat Belt Adaptor 9 Toothbrush 4 Ford Focus 10 Trashcan 5 Wash Basin 11 Arm Chair 16

17 TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH (CONTINUED) Product 12 Chopping Bowl 13 Closet 14 Cook Top 15 TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH (CONTINUED) Product 26 Door Knob 27 Lamp 28 Recliner Lever Pizza Cutter 29 Bottle Cap 16 Tooth-brush Dispenser 30 Fountain Drink Lid 17 Power Doors 31 Microwave 18 Eyewear 32 Touch Faucet 33 Blood Pressure Monitor 34 Can Opener 35 Oven 36 Telephone 37 Plug 38 Garlic Press AutoFaucet Jar Opener 1 Nail File Refrigerator 23 Hammer 24 Hoe 25 Iron 17

18 TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH (CONTINUED) Product TABLE A1. PRODUCTS PAIRS STUDIED FOR THIS RESEARCH (CONTINUED) Product 39 Pruner 53 Kitchen sink 40 One-Hole Hole Punch 54 Adjustable height Sink 55 Spatula 56 Syringe 57 Toilet 58 Ice Cube Tray 59 Trowel 60 Bathtub 61 Dishwasher 62 Washer 63 Wrist Watch 41 Recliner 43 Remote Razor Kitchen Scale Scissors Toilet Auto Seat Easy Reach Seat Belt Shovel Hand Held Shower 50 Braille Signs 51 Thermometer 64 Wine Opener 52 Jar Opener Car window controls 18