Assessing assembly instructions quality using operator behavior

Highly customized products and rapid delivery are essential in modern assembly systems, while efficient and error-free assembly processes are crucial. Supporting assembly operators with assembly instructions is paramount, but creating and maintaining qualitative instructions for high-variety, low-volume production systems is complicated and labor-intensive. Consequently, quality problems in assembly instructions are frequent and usually result in severe adverse outcomes. Recognizing and addressing these problems is challenging, as available assembly information on the shop floor is often not monitored, and instruction authors are overwhelmed by the workload. This research proposes a novel methodology for automatically and objectively evaluating the quality of assembly instructions based on the Behavioral Observation Theory. A behavioral coding scheme for manual assembly activities was developed, and the effects of low-quality instructions were explored through an empirical study. In a series of assembly experiments, subjects performed tasks guided by intentionally flawed instructions. The most impactful behavior types were identified and used as input features for a classifier algorithm, which computes a probability score for quality issues in an instruction version. This approach enabled us to identify low-quality instructions via operator behavior with an accuracy of 64% and estimate the type of quality problem with 54% accuracy. While there is room for improvement in accuracy, the most significant quality problems were successfully identified. By highlighting these issues, the method engineer can address them effectively, substantially enhancing the set of assembly instructions. We pinpointed areas for further research and improvements, which will enhance the monitoring of the quality of assembly instructions.