AI-Based Fault Prediction for Boiler Feed Pump in Al-sabiya Steam Power Plant in Kuwait Using Logistic Regression and TinyML

Abstract

The reliability of boiler feed pumps (BFPs) is critical to the continuous operation of steam power plants, where unplanned downtime can lead to significant economic and operational losses. This study proposes an artificial intelligence (AI)-driven fault prediction model utilizing logistic regression (LR) within a supervised learning framework. The model targets the classification of BFP operational states into four categories: Normal, Abnormal, Early Maintenance, and Annual Maintenance. The primary aim is to implement an end-to-end predictive maintenance solution using TinyML technology, thereby enabling low-latency, edge-based fault detection on resource-constrained hardware. A dataset comprising five critical features—temperature, pressure, flow, running hours, and alerts—was collected and preprocessed. The model was trained using TensorFlow in a cloud environment and subsequently optimized through quantization into TensorFlow Lite (TFLite) format for deployment on an ESP32 microcontroller. Comparative evaluation revealed that while the cloud-based TensorFlow model achieved a classification accuracy of 99%, the TFLite model on ESP32 preserved a respectable 95% accuracy with significantly reduced inference latency and memory footprint. This paper also includes a comparative literature analysis across anomaly detection, healthcare diagnostics, and smart agriculture, establishing the broader applicability and competitiveness of the proposed approach. Through architectural illustrations, performance benchmarks, and deployment case studies, the research demonstrates that integrating TinyML with predictive maintenance for BFPs can deliver real-time decision-making capabilities while minimizing computational overhead. These findings suggest that such lightweight, edge-deployable AI systems hold strong potential for industrial automation, particularly in developing countries seeking scalable, cost-effective digital transformation strategies.