FDAS: A KNOWLEDGE-BASED FRAMEWORK FOR ANALYSIS OF DEFECTS IN WOVEN TEXTILE STRUCTURES
Knowledge-based systems are computer-based tools for representing and utilizing domain-specific knowledge to solve problems. By linking the power of computers to the richness of human expertise, these systems enhance the value of expert knowledge by making it readily and widely available. These systems are also permanent, consistent, easy to access, transfer, document and less expensive when compared to human experts. A less obvious but important advantage of knowledge-based systems is the acceleration of the development, clarification and expansion of human knowledge itself.
The objective of the research is to build a knowledge-based system for the analysis of defects in woven fabrics (Fabric Defects Analysis System FDAS). FDAS should be capable of identifying a fabric defect from a description of its visual attributes and provide the user with a list of probable causes and remedies. The initial domain for FDAS is denim fabrics. Expansion of the domain to other fabrics will also be explored. In addition to being a useful tool for the control of defects in fabric manufacturing and a good training aid for personnel responsible for fabric defects analysis, it will serve as part of an overall system for the analysis of defects in garment manufacturing.
A knowledge-based system for the chosen task is justified for the following reasons: (i) it requires the knowledge of more than one expert, (ii) human expertise is scarce and required at multiple locations and (iii) the task solution has a high payoff. A textile plant with 200 weaving machines, producing 32,000 lb of sheeting fabrics per machine-year, loses approximately $36,000 per year for every 1% of fabrics sold as seconds. A survey of trouser manufacturers has shown that price realization on defective goods is only about 45% of first quality goods. Maintaining high levels of quality is absolutely essential for the U.S. textile/apparel industry to remain globally competitive.
An exhaustive list of fabric defects occurring in denims and in most yarn and fabric dyed goods has been compiled from literature and contacts with industrial experts. A new scheme has been proposed for classification of these defects. The proposed scheme marks a departure from traditional methods of classifying fabric defects as warp and filling defects or as spinning and weaving defects. The traditional methods presuppose knowledge of the exact identity of the defect and its origin. However, this is precisely the information sought by a user from a defects analysis system. The classification scheme in FDAS depends only on the visual attributes of fabric defects; the identity of the defect, its probable cause and suggested remedy are contained in its output. Knowledge about defects is represented in FDAS using a hierarchy of classes with multiple inheritance and forward changing rules. Inference is controlled by hierarchical classification of the defects with early pruning and dynamic resetting of the priority of classes of defects. The architecture of the system meets the requirements for modularity and reusability such as clear distinction of domain and control knowledge. Menus for interaction with FDAS have been designed to minimize cognitive load on the user.