A novel reinforcement learning-based framework for defect identification and analysis in remote sensor networks

In remote sensing networks, defect detection and analysis are essential to preserve the networks' performance. A sensor based defect detection and diagnostic system for the internet of things (IoT) network is the focus of this research. However, this goal presents issues because of the sensors' limited processing power, storage capacity and operational energy. Furthermore, defect detection is crucial, but recognition is sometimes time-sensitive. In this research, we propose a unique framework to tackle this issue using reinforcement learning (RL). Data gathering from the remote sensor network is the initial stage in our approach. To build a large dataset for analysis, we collect sensor data. To gather data, various network-deployed sensors' readings must be gathered. Data preparation includes tokenization, which is performed on the acquired data before analysis. Data tokenization is the process of dividing data into more manageable portions. An essential part of our methodology is feature extraction, where we translate the tokenized information into a format appropriate for defect identification. We use wavelet transform, a robust method for gleaning informative features from time-series data. This procedure improves the quality of the data participation, making it more appropriate for further analysis. We present a novel classification method based on the Enhanced Squirrel Optimized Effective Reinforcement Neural Network (ESSO-ERNN) in our study. To categorize data examples into defect or non-defect categories, this neural network model integrates optimization approaches with the concepts of RL. Fault detection performance is evaluated by determining the F1-measure, detection accuracy (DA), precision and recall rates by using python software. In addition, the metric for classification accuracy demonstrates the efficacy of defect diagnostics.