Abstract: Mechatronic systems, including robotic arms and CNC machines, can experience operational failures, increased downtime, and financial losses due to software flaws. Through the analysis of sensor data, system logs, and operational metrics, this study suggests a hybrid machine learning (ML) framework for anomaly identification and fault prediction in mechatronic systems. Three main modules make up the framework: (1) a feature extraction module that uses time-series and statistical analysis to extract important indicators; (2) an anomaly detection module that uses Autoencoders and Isolation Forest to find anomalous patterns; and (3) a fault prediction module that uses a Random Forest and Multi-Layer Perceptron (MLP) ensemble for accurate fault classification. Real-world industrial datasets and benchmark datasets, including the NASA Bearing Dataset and PHM Data Challenge datasets, are used to assess the suggested approach. According to experimental results, the Fault Prediction Module achieves 96.3% accuracy with an AUC of 0.98, while the Anomaly Detection Module achieves 94.5% accuracy. The framework improves predictive maintenance techniques, lowers false alarms, and effectively detects anomalies caused by software. This study demonstrates how ML-driven fault detection can enhance industrial mechatronic systems' dependability, minimize downtime, and optimize maintenance schedules. For additional improvement, future research will concentrate on edge computing integration, real-time deployment, and adaptive learning models.

Key words: software fault prediction, anomaly detection, industrialization, mechatronic systems, predictive maintenance, autoencoders