SENSOR-DRIVEN DEEP LEARNING FRAMEWORK FOR REALTIME PREDICTIVE MAINTENANCE AND FAILURE PROGNOSIS IN SMART INDUSTRIAL ROBOTIC SYSTEMS

In industrial robotics, studies show that over 30% of unplanned downtime is due to equipment failure, with robot-related faults accounting for nearly 20% of total maintenance costs annually. Additionally, predictive maintenance powered by AI can reduce repair costs by up to 25% and increase uptime by 10–20%. These figures underline the growing need for intelligent fault diagnosis systems to improve reliability and efficiency in robotic environments. Traditional manual fault diagnosis techniques are time-consuming, require skilled operators, and often fail to detect early-stage failures in dynamic industrial settings. Moreover, they lack realtime adaptability and consistency, especially in high-speed, sensor-rich robotic operations. These shortcomings result in costly production halts and delayed maintenance responses. To address these limitations, a robust AI-driven fault diagnosis system is proposed that uses sensor-driven datasets including force, torque, voltage, and current parameters from robotic arms. The collected data undergoes a comprehensive preprocessing stage involving outlier removal, normalization, and partitioning into training, validation, and testing sets. Two machine learning models are explored: an existing K-Nearest Neighbors (KNN) classifier and a proposed Dense Neural Network (DNN), which is trained to classify system states into normal or failure conditions. The DNN, designed with multiple hidden layers, effectively learns complex relationships in sensor data, enabling high-accuracy classification. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. Finally, the trained system is used for real-time fault prediction and pattern analysis, enabling preventive maintenance strategies and improving the operational safety, reliability, and productivity of industrial robots.