REAL-TIME FAULT DETECTION IN INDUSTRIAL ASSETS USING ADVANCED VIBRATION DYNAMICS AND STRESS ANALYSIS MODELING

Abstract

Real-time fault detection plays a pivotal role in condition-based maintenance (CBM) by enabling early identification of abnormal equipment behavior before failures escalate into safety, reliability, or productivity losses. This study evaluates the technical performance and organizational acceptance of an integrated real-time monitoring framework that combines vibration dynamics and stress analysis modeling for rotating and load-bearing industrial assets. Using a quantitative, cross-sectional, case-study–based design, the research draws on two complementary data sources: high-frequency vibration and stress measurements linked to documented fault events, and perceptual data from maintenance engineers, reliability specialists, supervisors, and operators collected through a structured five-point Likert survey (N = 120). Composite vibration and stress health indices were constructed from standardized signal features and used to model variations in fault severity, detection accuracy, detection time, and false-alarm rates. Findings show that both vibration and stress indicators were strong and significant predictors of fault behavior (r = 0.61 and r = 0.55 with fault severity, respectively). Multiple regression analysis demonstrated that vibration (β = 0.43, p < .001) and stress (β = 0.35, p < .001) indicators jointly explained 58% of the variance in fault detection performance, while their interaction (β = 0.18, p = .012) provided additional diagnostic value, indicating that multi-modal sensing detects faults more accurately and more quickly than either modality alone. The integrated system achieved an average detection accuracy of 89.5% and an average detection time of 4.2 hours. User perceptions were similarly positive: perceived reliability (M = 3.98), usability (M = 3.86), and usefulness (M = 4.05) exceeded the neutral midpoint, with strong internal consistency (α = 0.86–0.91). In perceptual modeling, perceived reliability (β = 0.38, p < .001) and usability (β = 0.29, p < .001) emerged as the strongest predictors of user acceptance (R² = 0.69), alongside contributions from perceived usefulness and the system’s objective detection performance. The study provides robust empirical evidence that integrating vibration and stress analysis enhances real-time diagnostic capability and strengthens user trust and acceptance. These findings contribute to prognostics and health management (PHM) theory by demonstrating how multi-sensor health indicators and user-centered perceptions jointly shape the effectiveness and adoption of real-time monitoring systems in industrial environments.