Forecasting failure rate of IoT devices: A deep learning way to predictive maintenance

Abstract

The Internet of Things (IoT) has become an essential part of everyday life, automating manual systems and providing comfort, security, and privacy. However, IoT systems are susceptible to faults due to hardware, firmware, or software failures, which can interfere with their proper functioning in real-life environments. To ensure fault-free automation of IoT systems, it is important to implement a proactive maintenance strategy that uses deep learning techniques to forecast failures in smart home applications by analyzing each device’s log of events and calculating its failure rate per attempt. This will help service providers and connected devices extend better serviceability of the system while assisting management in making timely decisions. Additionally, the system’s early prediction feature alerts consumers to take necessary actions when devices behave unexpectedly. The performance evaluation of the proposed predictive system using Bi-directional Long Short-Term Memory (Bi-LSTM) and Gated Recurrent Unit (GRU) demonstrates its effectiveness in predicting upcoming failures in smart home applications. Bi-LSTM has a Mean Squared Error (MSE) of 36.99, Root Mean Squared Error (RMSE) of 6.082, and Mean Absolute Error (MAE) of 5.305, while GRU has an MSE of 26.081, RMSE of 5.106, and MAE of 4.282.