ARDIS: knowledge‐based architecture for visual system configuration in dynamic surface inspection

Abstract

This work presents an approach to dynamic surface inspection in laminated materials based on the configuration of a visual system to obtain good quality control of the manufacturing surface. It aims to overcome some of the limitations of the single-use visual inspection systems by integrating and differentiating knowledge. The configuration task for surface inspection is solved as a Configuration-Design task according with the CommonKADS methodology. This task is analysed at the knowledge level and is decomposed into simple subtasks to reach the inference level. The generic domain knowledge involved in the configuration and revision tasks is differentiated in six types: initial, global, environment, real-time, image quality and computer vision techniques. The main goal of the proposed architecture is to reuse this partitioned knowledge for the configuration and revision of different visual surface inspection systems. The architecture has been implemented in a platform that distinguishes several operation modes: visualization of specific surface inspection results, system configuration and knowledge injection. Results obtained in the configuration and revision tasks for two surface inspection systems, stainless-steel and wood, are presented.

40 years of cognitive architectures: core cognitive abilities and practical applications

“…The procedural memory contains knowledge about how to get things done in the task domain. In symbolic production systems, procedural knowledge is represented by a set of if-then rules preprogrammed or learned for a particular domain (3T [160], 4CAPS [562], ACT-R [321], ARDIS [353], EPIC [279], SAL [225], Soar [330], APEX [173]). Other variations include sensory-motor schemas (ADAPT [49]), task schemas (ATLANTIS [179]) and behavioral scripts (FORR [138]).…”

“…The remaining action selection mechanisms include finite-state machines (FSM), which are frequently used to represent sequences of motor actions (ATLANTIS [179], CARACaS [250], Subsumption [77]) and even to encode the entire behavior of the system (ARDIS [353], STAR [293]). Probabilistic action selection is also common (iCub [569], CSE [223], CogPrime [190], Sigma [556], Darwinian Neurodynamics [156], ERE [272], Novamente [193]).…”

“…The long-term knowledge in planners is usually referred to as a knowledge base for facts and problem-solving rules, which correspond to semantic and procedural long-term memory (e.g. Disciple [524], MACsi [257], PRS [185], ARDIS [353], ATLANTIS [179], IMPRINT [70]). Some architectures also save previously implemented tasks and solved problems, imitating episodic memory (REM [382], PRODIGY [563]).…”