Analysis and Expending Data Clouds to Real-Time IIoT: Dimensionality Reduction, Anomaly Detection, and Predictive Maintenance

Large-scale real-time sensor data collection throughout industrial operations is made possible by IIoT, yet predictive maintenance procedures become more difficult. The study looks into how predictive analytics and machine learning work together to optimize maintenance decisions in IIoT frameworks. With an accuracy of 0.95 and 0.80, respectively, Random Forest's predictive maintenance analysis outperformed GBM. Both algorithms performed better with Random Forest; its recall was 0.90 compared to 0.85 for GBM, and its precision got up to 0.85, whilst GBM's only managed 0.75. Early failure identification was made possible by the anomaly detection system, which was made up of One-Class SVM and Isolation Forest. It was able to identify anomalous behavior in temperature values ranging from 70°C to 84°C and pressure variations between 118 psi and 124 psi. In order to maximize performance speed, the principle component analysis (PCA) method retained 25% of the original data variation within the first two principal components, reducing the dimensions of the data. As storage capacity rose gradually from 100GB to 600GB, cloud-based data intake showed rising storage demand, indicating system scalability. By integrating cloud storage and machine learning methods with data preparation techniques The findings demonstrate that IIoT systems have improved predictive maintenance capabilities, which reduce downtime and increase operational efficiency while saving money.