Adopting Machine Learning Techniques for Predictive Maintenance in the Industrial Internet of Things (Iot)

Abstract

Maintenance performed before problems occur is essential. Through specialized strategies, the machine learning (ML) approach allows systems to predict and reduce various machine failures. Predictive maintenance (PdM) has become an essential tool to improve the efficiency and reliability of industrial machines to improve maintenance management. By using ML algorithms to proactively detect and correct potential equipment failures, PdM minimizes unplanned downtime, reduces maintenance costs and improves operational efficiency. The Internet of Things (IoT) refers to the network of everyday objects connected via the Internet, global data and real-time responses. Invented by Kevin Ashton in 1999, the IoT has since revolutionized industries with examples including self-driving cars, smart TVs and RFID supply chains. As this technology evolves, it continues to have a significant impact on many industries. Recently, research on deep learning (DL) techniques, including convolutional neural networks (CNN), closed recurrent units (GRU), and long-term memory (LSTM), have focused on anomaly detection. This thesis proposes a new DL-based hybrid strategy to perform critical security assessments before cyber-attacks on IoT-embedded infrastructure. The models integrate techniques such as XGBoost, Autoencoders, CNN, GRU and LSTM to build, train and validate efficiently.