QUANTITATIVE STUDY ON MACHINE LEARNING-BASED INDUSTRIAL ENGINEERING APPROACHES FOR REDUCING SYSTEM DOWNTIME IN U.S. MANUFACTURING PLANTS

Abstract

This study addresses the problem that unplanned equipment downtime continues to erode productivity and maintenance budgets in smart manufacturing, yet many organizations implement machine learning solutions without clear evidence about which capability mix most strongly translates analytics into measurable downtime reduction. The purpose of the study was to quantify how machine learning enabled industrial engineering capabilities influence downtime reduction and to identify the most influential predictors that organizations should prioritize. Using a quantitative, cross sectional, case-based design, data were collected from manufacturing plant cases operating enterprise and shop floor environments that typically integrate CMMS, SCADA, and ERP workflows. The final sample included 214 usable respondent cases from U.S. plants. Key variables included predictive maintenance capability, monitoring and anomaly detection capability, data quality and availability, integration capability across maintenance and production systems, workforce readiness, and management support, with downtime reduction as the dependent outcome measured as a composite Likert 1 to 5 index capturing reductions in downtime frequency and duration and faster response and recovery. The analysis plan applied descriptive statistics, reliability testing, Pearson correlation analysis, and multiple regression to estimate unique effects while controlling for overlap among predictors. Reliability was acceptable to strong, including downtime reduction (Cronbach alpha = 0.88). Descriptive results indicated moderate to high implementation levels (grand mean M = 3.62, SD = 0.71) and moderate perceived downtime reduction (M = 3.59, SD = 0.72). Correlations showed that downtime reduction was positively associated with all predictors (r range = 0.44 to 0.61, p < .01), with the strongest relationships observed for predictive maintenance (r = 0.61) and integration capability (r = 0.58). In the multivariate model, the predictors collectively explained 56% of the variance in downtime reduction (R2 = 0.56; adjusted R2 = 0.55; F(6,207) = 44.19, p < .001). The headline findings were that predictive maintenance (beta = 0.29, p < .001) and integration capability (beta = 0.25, p < .001) were the strongest unique predictors, followed by management support (beta = 0.18, p = .001), monitoring and anomaly detection (beta = 0.14, p = .010), and data quality (beta = 0.11, p = .034), while workforce readiness was not statistically significant after controlling for the other factors (beta = 0.07, p = .157). These results imply that organizations seeking tangible downtime reduction should prioritize robust predictive maintenance routines and end to end workflow integration that converts model outputs into executed maintenance actions, reinforced by leadership governance and sustained investment in data readiness.