Design and Verification of an Interpretable Low-Code Maintenance Decision Support Application for Automotive Assembly

Abstract

Industrial maintenance digitalization is often associated with sensor-rich predictive models, yet brownfield factories may first need reliable decision support from sparse historical records and familiar software. This study designs and verifies an interpretable low-code application for maintenance scenario assessment in automotive assembly. The application combines planned production and cyclic calendar context, selects three similar historical situations, and displays normalized downtime, incident, efficiency, and failure-family indicators with an audit trail. The evaluation used twelve synthetic monthly records from which confidential and equipment-specific information was excluded. The application correctly reproduced all twelve stored monthly cases, generated valid results for thirty-six scenarios, rejected nine categories of invalid input, and responded correctly to forty-eight changes in planned production. An additional test estimated each month using the remaining eleven months. The proposed configuration produced an average difference of 5.52 points on the 0-100 indicator scale. It performed better than methods based only on production similarity or the historical average, although a simple calendar-based method achieved a lower difference. Alternative parameter settings produced comparable results. These findings confirm that the application operates consistently and transparently, but further testing with real industrial data and maintenance users is required before predictive effectiveness can be claimed.