Participatory Design of Warning Symbols Using Distributed Interactive Evolutionary Computation
Type of Degreedissertation
Industrial and Systems Engineering
MetadataShow full item record
Safety warnings play an important role in communicating risk via product labels and environmental signs. With the diversification of cultures and languages in the United States, and with the increasing globalization of most industries, emphasis on the communication of this risk through symbols and other non-written forms has increased. Both ANSI and ISO have developed voluntary standards for the production and evaluation of warning symbols, but many symbols currently in use have been found deficient with respect to the comprehension and effectiveness guidelines found in these standards. In other cases, commonly used symbols have not undergone effectiveness evaluation at all. Thus, there remains a need to produce warning symbols shown to be effective in communicating risk to a multicultural, multilingual, global society. Though the ANSI and ISO standards fail to specify a technique for developing symbol designs, three techniques were identified from the literature. Of these three, the focus group method was claimed by its developers to be the most effective in producing high quality symbol designs because it involves realistic users of the symbols in more aspects of the design process than either of the other techniques. The focus group method requires human participants to sort and filter many designs into a single proposed symbol. This type of search task is well suited to machine computation, and this research will model the focus group method of human design generation and consolidation as a distributed interactive genetic algorithm which will evaluate and generate designs using simple simultaneous feedback from a group of human users. The literature revealed a similar interactive evolutionary computation algorithm used to design safety symbols in a prior study, although that algorithm used a single participant and still required human designers to evaluate many symbols by hand to determine the best design. The proposed distributed interactive genetic algorithm will remove the designer’s input at this stage of the design process by allowing the users and the algorithm to determine a final design for the group without designer interference. First, a survey was administered to 145 university students and safety professionals to determine an ordered list of safety messages (or referents) sorted by their perceived difficulty to convert into symbols. From this list, two referents were chosen for the study, one easy (“Hot Exhaust”) and one difficult (“Do Not Touch with Wet Hands”). Seventy American university students, 35 born in the U.S. and 35 born in India, were recruited to sketch symbol designs for each of the two referents. These designs were evaluated by a panel of safety professionals to identify the graphical attributes contained in each drawing, and the presences or absence of each identified attribute in a given symbol created a binary attribute matrix for each referent. These matrices were summed and clustered using a K-means clustering algorithm to determine the centroid values of each cluster of symbol drawings. Thirty-five attributes were identified by the panel among the “Hot Exhaust” drawings, and the clustering revealed that only three of them were present among the centroid values of each of the five identified clusters. Likewise, 28 attributes were identified for the “Do Not Touch with Wet Hands” drawings, but only five were present in the centroid values of the four clusters identified for this referent. From these centroidal attributes, a version of the distributed interactive genetic algorithm was created for each referent. Forty-six participants, divided into four groups of 10-12 by country of origin, designed symbols using the algorithm, and the symbol most representative of each group was compared by 401 participants from around the globe to symbols generated using a traditional method and to symbols in use currently. The results indicated that for the easier referent, “Hot Exhaust”, the algorithm produced symbols that performed as well or better than symbols produced by other means, including the symbol currently in use. However, for the more difficult referent, “Do Not Touch with Wet Hands”, other symbols performed better than those produced by the algorithm. Additionally, the algorithm generally converged in 20 generations or less, which falls within the recommended limitations of such algorithms within the literature. However, the algorithm converged faster for U.S. and multinational groups than for groups of participants from other single nations. In summary, the distributed interactive genetic algorithm technique showed promise as a design tool for developing symbols that perform as well or better than current design methods. Furthermore, the algorithm’s performance may vary depending on the difficulty level of the referent tested as well as on the composition of the participant groups used in the design process. Further research is needed to confirm and characterize these relationships.