Smart Robotics-Driven Predictive Maintenance in Smart Factories

Thesis Abstract

The rapid digitization of manufacturing environments has elevated the complexity of maintenance strategies in modern plants, where unplanned downtime, safety concerns, and energy inefficiencies constrain productivity. This study addresses the challenge of achieving reliable, cost-effective maintenance in smart factories by integrating autonomous robotics, edge computing, and data-driven analytics to deliver predictive maintenance that minimizes unscheduled outages while maximizing equipment availability. The aim is to develop and validate an end-to-end framework that leverages robotic sensing, real-time data fusion, and machine learning to predict asset health and optimize maintenance actions in a dynamic production setting. Specific objectives include (i) designing a multi-sensor data acquisition architecture combining vision, vibration, thermal, and current sensors mounted on collaborative and autonomous robots; (ii) developing a predictive maintenance model that fuses time-series sensor data with robot-instrumented condition monitoring to forecast remaining useful life (RUL) and failure probability for critical assets; (iii) evaluating the framework’s impact on plant-level metrics such as Overall Equipment Effectiveness (OEE), maintenance cost, and energy consumption; (iv) assessing operational feasibility, safety, and human–robot collaboration implications in the production environment; and (v) formulating guidelines for scalable deployment across different manufacturing lines. The methodology adopts a mixed-methods, multi-site research design conducted in three manufacturing cells within a high-mix, low-volume electronics assembly facility and a parallel set of operations in a mid-volume automotive components plant. The study targets a population of 180 autonomous/partially autonomous robots and 4200 peripheral assets (motors, conveyors, pumps, CNC modules) across the two sites. A stratified random sample of 60 robots and 120 critical assets is selected, with data collected over a 12-month monitoring period. Data collection instruments include (i) robotic-integrated sensor suites (vibration, temperature, acoustic, current, and visual data) with onboard preprocessing and edge analytics, (ii) a centralized data lake aggregating ERP, MES, and maintenance management system (MMS) records, (iii) standardized maintenance logs and downtime records, and (iv) semi-structured interviews and structured surveys with maintenance engineers and production supervisors to capture process-level insights on robot-assisted maintenance workflows. Validity and reliability are ensured through instrument calibration, cross-validation of sensor readings, and triangulation between sensor data, MMS records, and interview findings. Analytical techniques encompass time-series forecasting (LSTM and Prophet models) to estimate remaining useful life, survival analysis to model failure probabilities, and Bayesian updating to refine predictions with new data. A multivariate regression framework tests the relationship between predicted RUL, maintenance interventions, and OEE outcomes. Model performance is assessed using mean absolute error (MAE), root mean square error (RMSE), area under the ROC curve (AUC) for failure prediction, and cost-benefit analysis comparing predictive maintenance against baseline corrective and preventive strategies. A qualitative component employs thematic analysis of interview transcripts to identify barriers and facilitators to human–robot collaboration, safety, and change management. Anticipated findings indicate that the integrated robotics-driven predictive maintenance framework reduces mean time between failures by 28–35%, lowers maintenance costs by 15–22%, and improves OEE by 6–10% relative to traditional maintenance approaches. The study is expected to reveal that edge-enabled data fusion from robotic sensors enhances early fault detection, while centralized analytics improve decision timeliness and maintenance prioritization. The contribution to knowledge lies in (i) presenting an empirically validated, scalable architecture for intelligent maintenance in smart factories that synergistically combines robotics, edge computing, and data science; (ii) advancing methodological rigor for industrial predictive maintenance by integrating time-series forecasting, survival analysis, and Bayesian updating within a unified framework; and (iii) providing practical governance and human–robot interaction guidelines to support safe and efficient deployment in complex manufacturing environments. The study concludes with recommendations for standardizing robot-assisted maintenance protocols, enhancing data governance, and pursuing phasedrollout strategies across manufacturing networks to sustain gains in reliability, productivity, and energy efficiency.