Categorization Errors for Data Entry in Maintenance Work-Orders

Abstract

In manufacturing, there is a significant push toward the digitization
of processes and decision making, by increasing the
level of automation and networking via cyber-physical systems,
and machine learning methods that can parse useful
patterns from these complex architectures. As such, this push
toward being “smart” is largely driven by the availability of
data, for analysis, decision guidance, and the training of AI.
The maintenance team, however, one of the core subsystems
in any production line, remains a largely human endeavor.
Consequently, the historical data needed for research and development
of AI-assisted maintenance frameworks are often
full of misspellings, jargon, and abbreviations. While one
might enforce data-entry into pre-specified functional categories
(generally using some form of controlled vocabulary),
cognitive models, along with consistent reports from industry,
indicate that data entry remains a process fraught with significant
errors, especially when a mismatch occurs between designated
schemas and the technician’s needed semantic flexibility.
This paper offers a framework for understanding and
addressing these issues, with a methodological case study in
applying Human Reliability Analysis (HRA) to quantify and
understand human errors associated with entering maintenance
work-order (MWO) data into structured database (DB)
schema. We subsequently suggest potential mitigation strategies
for each to improve the quality of recorded data throughout
the maintenance-management workflow.