Predictive Maintenance of Flexible Pipe Connectors in Parabolic Trough Collectors Using Artificial Neural Networks

Abstract

In parabolic trough solar power plants, flexible pipe connectors, specifically Rotation and Expansion Performing Assemblies (REPAs), are critical yet failure-prone components due to their exposure to thermal cycling, mechanical stress, and harsh environmental conditions. This study focuses on the Rotary Flex Hose Assembly (RFHA), one common REPA type. Unexpected RFHA malfunctions, such as leaks in swivel joints, can lead to costly downtime and safety hazards. Despite regular preventive maintenance, current strategies often fail to anticipate such failures early enough to prevent disruption. This study investigates whether vibration analysis, combined with machine learning, can provide early warnings of RFHA degradation. A convolutional neural network (CNN) was trained to detect pre-failure conditions using spectrogram comparisons derived from high-frequency vibration data collected during an accelerated life-cycle test campaign. Structural Similarity Index (SSIM) values were used to quantify differences between twin RFHA vibration patterns over time. The model successfully predicted failures up to 1,000 cycles in advance, equivalent to more than two years of lead time in field operation, achieving an overall accuracy of 0.93 and a precision of 1.0, indicating that all predicted failure events were correct and no false positive alarms were generated. These results demonstrate that SSIM-based vibration comparison and CNN classification can provide a robust foundation for predictive maintenance of RFHAs, offering a scalable solution for enhancing the reliability of CSP plant operation.