Abstract:

Many small and medium-sized manufacturers still record downtime, scrap and output on paper forms, and the indicators used to manage production, such as overall equipment effectiveness (OEE), are only as reliable as the transcription of these forms. This paper describes an information system that turns scanned handwritten production forms into structured records and OEE indicators, and it examines where data-quality risk remains in such a workflow. The system reads the fields of a known form, assigns each field a calibrated confidence, accepts confident values automatically and sends the rest to a person for review; every stored value keeps its quality state, the software version that produced it and the reviewer’s action. Form fields are mapped to the time elements of ISO 22400-2, and the indicator calculation reproduces 30 independently hand-computed scenarios. On 39 forms written by 13 volunteers and processed by the deployed service, about half of the 585 fields were accepted without review, one of which was wrong, and a form was processed in a median of 11 seconds on a server without a graphics processor. Of the 63 wrong values, 62 reached the review queue, but 24 of these were approved or edited incorrectly by the single reviewer. The residual error therefore lay mainly in the reviewed fields. The paper derives implications for the design of review steps, audit sampling and the presentation of indicators whose inputs have different levels of verification.