MACHINE LEARNING APPROACHES FOR OPTIMIZATION OF LUBRICANT PERFORMANCE AND RELIABILITY IN COMPLEX MECHANICAL AND MANUFACTURING SYSTEMS

Abstract

This quantitative, cross-sectional, case-based study examines the role of machine learning (ML) approaches in enhancing lubricant condition monitoring and, in turn, optimizing lubricant performance and improving equipment reliability across complex mechanical and manufacturing systems. The research is motivated by the persistent underutilization of lubricant-related data in predictive maintenance programs, even as modern industrial assets increasingly incorporate dense sensor arrays capable of generating high-frequency tribological, thermal, and chemical measurements. To address this gap, the study surveyed 214 professionals responsible for maintenance and reliability management in enterprise-scale manufacturing plants operating high-duty rotating and sliding equipment, generating 204 complete, analyzable cases measured on a five-point Likert scale. The conceptual framework was structured around several core latent variables: ML-driven lubrication adoption, lubricant performance, system reliability, data quality, technical readiness, and organizational readiness. Descriptive statistics indicated that while respondents reported only moderate levels of ML adoption (M = 3.21, SD = 0.74), the plants demonstrated comparatively high lubricant performance (M = 3.68, SD = 0.69) and system reliability (M = 3.59, SD = 0.71), suggesting that lubrication improvements and reliability outcomes are already being pursued through traditional means, with ML technologies representing an emergent, rather than fully mature, complement to existing programs. All multi-item measurement scales showed strong internal consistency, with Cronbach’s alpha values ranging from .81 to .89, confirming the reliability of the constructs and supporting their suitability for subsequent correlation and regression analyses. Pearson correlation coefficients revealed that ML adoption was positively and moderately associated with both lubricant performance (r = .52, p < .001) and system reliability (r = .48, p < .001), providing initial empirical support for the theorized relationships. These correlations suggest that plants beginning to integrate ML-enabled lubrication analytics are already experiencing measurable operational benefits, potentially due to improved detection of lubricant degradation, earlier identification of contamination events, or more precise adjustment of lubrication intervals. Multiple regression results further clarified these relationships, demonstrating that the proposed model explained 41.8% of the variance in lubricant performance and 49.2% of the variance in system reliability. ML adoption emerged as a significant predictor across models (β up to .38, p < .001), but data quality and organizational readiness also showed strong predictive influence, highlighting the critical interdependencies between algorithmic tools, underlying data infrastructures, and human/organizational processes.