The Internet of Things (IoT) has grown rapidly, affecting many industries. However, it also raises serious concerns about data privacy and system security. Ensuring accurate fault prediction in IoT environments is essential to maintaining system performance and minimizing unexpected downtime. Unlike the usual centralized fault finding methods, which can expose sensitive data because they collect everything in one place, this study uses a Federated Machine Learning (FML) framework mixed with local differential privacy (LDP). The idea is to distribute the learning process across the Edge devices and use local differential privacy, which helps keep privacy secure and in the same time keep a good accuracy in predicting faults. The results from the experiments show that this model performs well in keeping sensitive data secure without compromising performance.