Visual Detection of Fouling in Wood-Fired Boilers Using a Lightweight Convolutional Neural Network for Energy Efficiency in Small-Scale Industries

Abstract

The tempeh industry, as part of the Micro, Small, and Medium Enterprises (MSME) sector, plays a vital role but often faces challenges in energy efficiency, particularly in wood-fired boilers. A significant decline in boiler efficiency is caused by the accumulation of soot and ash (fouling), which hinders heat transfer. Boiler cleaning practices that are still manual and based on fixed schedules lead to fuel wastage and inconsistencies in the production process. This study proposes an innovative and cost-effective fouling monitoring system using a Convolutional Neural Network (CNN) with a MobileNetV2 architecture. The system utilizes an endoscope camera installed in the Flue Gas Chamber to capture images of the internal condition of the boiler. This area is strategically selected because it functionally represents the cumulative condition of the fire tubes while providing more practical and holistic visual access. The trained CNN model analyzes these images to objectively classify the level of fouling, enabling a shift from reactive maintenance strategies to predictive, condition-based maintenance. Implementation results in an MSME tempeh production facility show significant impacts: a 10% reduction in wood fuel consumption and up to a 50% decrease in manual cleaning frequency. Economically, this translates to an annual operational cost saving of IDR 6,300,000 and a return on investment (ROI) period of less than one year. This study demonstrates that the application of AI-based computer vision is an effective, feasible, and impactful solution to enhance the competitiveness and sustainability of MSMEs.