Empirical relationships between algorithmic SDA-M-based memory assessments and human errors in manual assembly tasks

Abstract

The majority of manufacturing tasks are still performed by human workers, and this will probably continue to be the case in many industry 4.0 settings that aim at highly customized products and small lot sizes. Technical systems could assist on-the-job training and execution of these manual assembly processes, using augmented reality and other means, by properly treating and supporting workers’ cognitive resources. Recent algorithmic advancements automatized the assessment of task-related mental representation structures based on SDA-M, which enables technical systems to anticipate mistakes and provide corresponding user-specific assistance. Two studies have empirically investigated the relations between algorithmic assessments of individual memory structures and the occurrences of human errors in different assembly tasks. Hereby theoretical assumptions of the automatized SDA-M assessment approaches were deliberately violated in realistic ways to evaluate the practical applicability of these approaches. Substantial but imperfect correspondences were found between task-related mental representation structures and actual performances with sensitivity and specificity values ranging from 0.63 to 0.72, accompanied by prediction accuracies that were highly significant above chance level.