The sophistication of industrial systems and the rapid introduction of Industry 4.0 technologies make it increasingly important to develop more sophisticated predictive maintenance (PdM) and troubleshooting processes. In this paper, we propose an AI-based predictive maintenance and automated troubleshooting framework to analyse logs in Industrial IoT systems. The framework leverages AI to extract features, parse logs, and apply prediction and detection models like LSTM for failure prediction and Isolation Forest for anomaly detection. This may involve generating time-stamped log data depending on different types of faults (e.g., component wear, sensor failures) in simulated industrial environments to train the models. F1 Score of 0.91 for predictive maintenance is better than traditional method, Top-3 Accuracy of 88% for troubleshooting recommendations is also competitive. Well, to summarize as from above result it shows how AI powered log analysis can enhance maintenance efficiency, cut back downtime and guide in fault resolution. But there are significant important limitations, and more research is required to handle the challenges of real-world data complexities like scaling up or down as well as adapting the model to different contexts in real time. The scope of our future work will include the integration of online learning, reinforcement learning, and explainable AI to enhance both the versatility and interpretability of these models.