FEDERATED LEARNING-DRIVEN PREDICTIVE QUALITY ANALYTICS AND SUPPLY CHAIN OPTIMIZATION IN DISTRIBUTED MANUFACTURING NETWORKS

Abstract

This study addresses a practical problem in distributed manufacturing networks: how to improve predictive quality and supply performance when raw data cannot be centralized due to privacy, sovereignty, and competitive constraints. The purpose is to evaluate whether federated learning–driven predictive quality analytics yields measurable gains in plant quality and supply outcomes. This study foregrounds federated learning–driven predictive quality analytics as a mechanism to fuse tacit, sensitive, or proprietary signals from geographically dispersed partners while reducing breach surfaces and preserving competitive boundaries. We employ a quantitative, cross-sectional, case-based design using a multi-firm survey and embedded vignettes. The sample comprises 204 usable responses from cloud or enterprise implementations across multi-site manufacturers with a 60.9 percent response rate. Key variables include FL adoption maturity, heterogeneity management, communication efficiency, data governance strength, network integration, PQA performance, and aligned outcome indices for quality and supply. The analysis plan specifies descriptives, bivariate correlations, and hierarchical regression with heteroskedasticity-robust errors, followed by moderation tests for governance and integration and mediation via bootstrapped indirect effects. Headline findings show that FL maturity, heterogeneity handling, and communication efficiency are positively associated with PQA (β=0.34, 0.21, 0.17; ΔR²=0.31). PQA in turn predicts better quality outcomes (β=0.49; ΔR²=0.24) and supply outcomes (β=0.44; ΔR²=0.22). Governance strengthens capability to PQA translation (interaction ΔR²≈0.03), and integration strengthens PQA to outcomes translation (interaction ΔR²≈0.04). Mediation tests confirm significant indirect effects from capabilities to outcomes through PQA. Case vignettes triangulate these effects, for example first-pass yield improved by 3.2 points, on-time delivery by 2.5 points, and expedites decreased by 18 percent within six months. Implications are managerial and methodological: treat governance as the control plane, institutionalize integration routines, and use federated, privacy-preserving analytics to convert earlier defect detection into reliable service and leaner buffers without centralizing data.