Investigations on Sensory Data Analytics Using Machine Learning and Applications

Abstract

Sensory data analytics is indispensable in numerous sectors, offering essential newlineinsights that drive decision-making processes, optimize operations, and facilitate monitoring newlinetasks. Despite significant progress in algorithms and techniques, persistent challenges newlinelimit the full potential of sensory data analytics systems. This research adopts a newlinemultifaceted approach to address these challenges. newlineA significant aspect addressed in machinery fault detection using multiple vibration newlinesensors necessitates hierarchical classification. An issue with current approaches is newlinethe decrease in overall detection probability with an increase in the number of classes newlineand layers. To tackle this, a novel hierarchical classification method is proposed, integrating newlineAnt Colony Optimization (ACO) to optimize classifier block order and utilizing newlinethe Adaptive Boosting algorithm (Adaboost) for enhanced detection, particularly in newlinemulti-class scenarios. Monte Carlo simulation and experimentation with engine vibration newlinedata demonstrate improved performance within hierarchical structures. newlineAdvancements in LiDAR technology, combined with deep learning techniques, newlinepresent novel opportunities in object detection, human activity recognition, and mobile newlinerobot navigation. However, achieving pinpoint accuracy and privacy in object location newlineremains challenging. This study explores privacy-preserving object tracking using newlineLiDAR technology, aiming to maintain privacy and confidentiality in enclosed environments. newlineGaussian process regression (GPR) surpasses deep learning techniques for object newlinedetection, with Particle Swarm Optimization (PSO) enhancing regression performance newlinethrough hyperparameter optimization. newlineVibration detection, an essential parameter in scientific research and engineering newlineapplications, is captured using piezoelectric sensors. Despite progress, there is room for newlineimprovement in sensor selection. This research focuses on sensor placement optimization to reduce expenses and maintenance. Regression analysis identifies representative newlinesensors