Engineering Headway Vol. 38

Abstract: Predictive maintenance is a strategic necessity in the mineral processing industry, particularly for seawater pipelines that are highly susceptible to internal coating degradation. Failure to accurately predict the reduction in coating thickness can result in leaks, significant financial losses, and operational disruptions. While machine learning algorithms hold significant potential to enhance prediction accuracy and improve infrastructure reliability, their application in piping systems especially for coating thickness remains limited. This study develops a data-driven maintenance framework using Artificial Neural Networks (ANN) to predict internal coating thickness degradation based on a decade of historical inspection records from 2014 to 2024. To address the challenges of data-driven modeling in industrial contexts, this research incorporates advanced feature engineering, including temporal decomposition of inspection dates and spatial encoding of manhole positions. Furthermore, to mitigate the risks of overfitting inherent in limited industrial datasets, the model integrates Dropout layers and L2 regularization. The findings demonstrate that a regularized ANN architecture can effectively capture non-linear degradation patterns. The proposed model achieves a Coefficient of Determination, R² of 0.94 and a Mean Absolute Error, MAE of 25.91 µm. These results are expected to reinforce predictive maintenance strategies, optimize shutdown scheduling from 2026 to 2027, and promote the sustainability of industrial operations through more efficient resource allocation and real-time monitoring potential.