Detecting early warning signs of failure in mechanical systems is key to keeping operations safe and running smoothly. We introduce In our paper a practical method that combines two different techniques to detect anomalous vibration patterns, one based on data features using Isolation Forest and another that looks at how much a signal differs when rebuilt using Autoencoders. We worked with 10000 real vibration signals on which we applied cleaning steps, scaling and unsupervised learning to find hidden signs of possible issues. Methods to unveil common patterns that could indicate mechanical problems. Problems we tested different threshold levels 5%, 7%, 10%, and 12%. We also explored the results using PCA plots, visual tools to highlight anomalies and extra checks like overlap analysis and heatmaps. Using this two-part system we ensured a more reliable detection of these failures. Overall, it’s a clear and flexible solution for finding faults in smart maintenance and industrial monitoring.