APPLICATION OF MACHINE LEARNING MODELS FOR PREDICTIVE MAINTENANCE OF BIOTECHNICAL SYSTEMS

Ensuring the reliability and efficiency of agricultural machinery is critical for modern farming operations. Traditional maintenance strategies, including corrective and preventive approaches, often lead to unexpected downtime or excessive servicing costs. This study explores the application of machine learning-based predictive maintenance for agricultural equipment, focusing on the hydraulic system of a Massey Ferguson 7700 S tractor. Real-time sensor data was collected, with hydraulic pressure selected as the primary diagnostic metric for detecting early signs of mechanical degradation. A predictive maintenance framework was developed using seven machine learning models: Isolation Forest, One-Class SVM, KMeans, DBSCAN, Autoencoders, Convolutional Neural Networks (CNNs), and XGBoost. These models were individually applied to identify pressure anomalies indicative of potential failures. To enhance detection accuracy, a "Council of the Wise" ensemble approach was introduced, where an anomaly was validated only if at least four of the seven models agreed on its presence. This consensus-based method reduced false positives and improved fault identification reliability. Results demonstrated that integrating multiple models effectively distinguished significant anomalies from noise, capturing both transient mechanical instabilities and gradual wear-related failures. The findings highlight the potential of machine learning-driven predictive maintenance in optimizing maintenance schedules, reducing unplanned downtime, and extending equipment lifespan. This study establishes a scalable, data-driven maintenance approach that enhances the operational resilience of agricultural machinery, ensuring greater efficiency and sustainability in farming operations.