Ontology-guided knowledge retrieval in an automobile assembly environment

In the body shop of an automobile assembly plant, having access to correct and timely information is very important in solving problems encountered in the assembly process. The Variation Reduction Adviser (VRA) system used within General Motors (GM) is a database containing problems encountered in this process and their possible solutions. The VRA acts as an electronic logbook that shares information across shifts within a plant as well as across multiple plants. The VRA also serves as a problem-solving tool by which solutions to problems encountered may be retrieved and reused. To function effectively as a problem-solving tool, it is important that relevant information is quickly retrieved from the VRA database. Traditionally, keyword-based retrieval strategies have been used. In these approaches, the user types in a list of words or phrases and those records in the database that contain those words or phrases exactly as typed are retrieved. The problem with this approach is that records containing words or phrases that are semantically related to the ones typed in but not exactly the same are not retrieved. For instance, if the user types "left-hand side," the traditional keyword search will not find records that contain the abbreviation "LHS." This paper describes a search mechanism based on a thesaurus (a simple version of an ontology) that overcomes this problem. It describes the standard criteria to measure the effectiveness of a search, defines a new criterion, and shows that in terms of these criteria, the ontology-guided approach gives better search results than the exact match mechanism. The results are shown in the context of real searches during the use of the VRA in a GM assembly plant.