Early Fault Detection in Electro-Pneumatic Actuators using Mathematical Modelling and Machine Learning: A Bottling Company Case Study

This research examines the vital problem of detecting anomalies at an early stage within industrial systems by studying an electro-pneumatic double-acting cylinder actuator used in a bottling facility production line. The occurrence of malfunctions in valves leads to operational inefficiencies, and both planned and unplanned downtime, and additional maintenance expenses. The study contributes a new dual method that unites mathematical modelling and machine learning to overcome the limitations of conventional anomaly detection methods. The predictive model created for the actuator assessed its typical operation by evaluating pressure fluctuations, timing behaviour and displacement performance. Establishing baseline parameters through this process allowed the creation of synthetic datasets for normal operational standards. Real-time measurement points were validated through a baseline reference and machine learning models based on support vector machines received training data from labelled sets. The application of feature selection methods helped find essential variables to boost performance metrics in models. The research created valuable insights by reaching 90% success in operational state identification between normal and anomalous conditions across various test scenarios, which leads to an adaptable predictive maintenance system. The bottling company applied the case application, which led to 25% less machine downtime alongside better maintenance schedules, together with improved reliability during production. The research outcomes match the objectives of Agenda 2063 set by the African Union by supporting industrial development alongside innovation and sustainable economic expansion as well as meeting SDG targets such as Goal 9.4 and Goal 12.6 for sustainable industrial practices. This study provides essential information for industrial optimization policies through operational efficiency measures that demonstrate global significance for predictive maintenance systems. The scientific methods alongside their research results deliver important knowledge regarding industrial ecosystems in Africa and across the world by tackling regional and worldwide sustainable productivity issues.