Predictive maintenance is becoming increasingly important in modern manufacturing enterprises with highly automated processes. In this study, we propose a cyber-physical system (CPS) for predictive maintenance of bakery plant equipment based on a distributed sensor infrastructure. The system integrates a data acquisition and preprocessing module with an agent-based analysis and failure prediction module. Sensor data on temperature, vibration, pressure, and related parameters are grouped by devices and processed to recover missing values and smooth anomalies. For failure prediction, a hybrid approach combining condition-based maintenance (CBM) models and expert rules is applied. The system architecture is designed as a multi-agent system (MAS), which ensures modularity, scalability, adaptability, and interpretability of predictions. Experimental evaluation demonstrated stable operation, robustness to missing or corrupted data, and reliable prediction results. As a consequence, the proposed system improves equipment reliability, reduces downtime, and optimizes maintenance operations in continuous production cycles.