Machine learning based fault detection for sustainable wireless sensor networks

The Internet of Things (IoT) is a pivotal component within the paradigm of Industry 4.0. The accuracy of data collected through sensors in IoT systems is of great importance for the reliability and continuity of the system. Incorrect data originating from sensors has the capacity to exert a detrimental effect on system performance and give rise to financial losses. In this context, the rapid detection and correction of errors is vital for the stability of IoT systems. The present thesis is divided into two sections. The first is concerned with the classification of erroneous data in order to identify fault types. The second is concerned with the prediction of data in a cloud environment, with a view to ensuring system continuity following the identification of anomalies. The initial section, the focal point of the study, involved the creation of a comprehensive dataset for the classification of erroneous data. The dataset under consideration encompasses sensor data obtained from an IoT system configured within a controlled environment, in addition to a range of error types derived from mathematical functions. The dataset obtained was then subjected to classification on the embedded system, utilising a range of machine learning algorithms and deep learning models with hyperparameter optimisation. This process was found to be highly accurate in its detection of fault classes and distinction of anomaly conditions. In the subsequent stage, a data prediction model was implemented in the cloud environment to address the detected erroneous data, thus facilitating uninterrupted system operation. The present thesis demonstrated high performance in detecting and eliminating sensor-related errors in IoT systems, thereby providing an effective solution to ensure the dynamic continuity of the system.