Predictive maintenance is an advanced approach to maintenance that utilizes data-driven methods to predict potential failures in machinery, enabling timely intervention. This paper presents a novel system for predictive maintenance of servo motors by leveraging Apache Kafka for real-time data streaming and a Convolutional Neural Network (CNN) implemented in PyTorch for anomaly detection in motor sounds. The system captures acoustic signals using a high-fidelity microphone mounted on the servo motor body, providing an effective way to monitor the health of the motor through sound analysis. These audio signals are processed and streamed in real-time to a Kafka producer, which ensures continuous data flow and robust data handling. The Kafka consumer then extracts Mel Frequency Cepstral Coefficients (MFCCs) from the audio data, which are fed into the CNN model to detect anomalies indicative of potential motor failures. The system integrates a Kafka consumer with a Modbus server to provide real-time predictions to industrial monitoring systems. This method enhances maintenance efficiency and reduces downtime, thereby increasing operational efficiency. The use of acoustic data provides a non-invasive, cost-effective means of monitoring, offering a significant advantage over traditional methods that require physical access to the machinery.