Cost-of-Quality Reduction via Data Attribute Recommendation (DAR): Master-Data Stewardship at Scale in Manufacturing

Abstract

The manufacturing businesses are still experiencing a high rate of digital change, which amplifies the need to have precise, consistent, and full master data to aid in production, supply-chain integration, quality control, and overall enterprise analytics. Nevertheless, the quality of master-data has continued to be a edging problem; resulting in unnecessary rework, scrap, warranty requests, non-compliance with regulations, and wastes within the system. All these inefficiencies add up to what is known in industrial circles as Cost of Poor Quality (CoPQ) a direct and indirect economic cost paid by manufacturers. To alleviate this load, the traditional interventions used by organisations have been manual data-stewardship processes, data quality validation using rules, and workforce interventions. However, these methods do not scale to high volume, high-dimensional data that is characteristic of modern business setups represented by mass customisation and multi-variant product configurations. In this paper, I present and empirically test a scalable, machine-learned solution called Data Attribute Recommendation (DAR) an automatic system that suggests values which are the best to put in missing, ambiguous or inconsistent master-data attributes. DAR has a goal to scale stewardship processes and at the same time minimize CoPQ. DAR uses supervised and semi-supervised learning architectures, similarity based clustering, probabilistic inferences and ontology mediated data standards to provide automated attribute completion, anomaly detection and confidence-assessed predictions. Also, DAR is referred to a closed-loop master-data governance architecture, which encourages lifelong learning and cuts back on the human validation overhead. The suggested solution is tested under an industrial setting that entails bill-of-materials (BOM), process variables, material requirements, and supplier metadata of a Tier-1 automobile parts supplier. DAR reached the average accuracy of attribute-recommendation of 92.7 percent that saved 62 percent of workload in manual stewardship and enhanced the completeness of data-quality by 71 percent to 96 percent overall. Simultaneously, the implementation also led to a reduction in CoPQ of an estimate of 14% through a lesser number of specification errors, better procurement precision, and lowered rework in assembling. The research will provide a detailed approach to integrating the principles of data-governance to manufacturing along with AI-inspired attribute-recommendation models. The paper discusses the architecture of DAR, data-pipeline design, modelling strategies, performance metrics and integration blueprint. Also, the paper presents the implications of DAR to future Industry-4.0 quality-engineering principles, such as predictive quality management, autonomous decisioning, and next-generation digital-thread realizations.