Detecting broken Absorber Tubes in CSP plants using intelligent sampling and dual loss

TL;DR

This work tackles automated detection of broken absorber tubes in CSP plants under severe class imbalance by constructing ATSet from UAV-derived thermal data and sensor readings across seven real plants. It proposes a Deep Residual Network with a dual loss and Dense-Sparse-Dense training, alongside traditional methods such as Random Forest and Histogram Gradient Boosting, augmented by re-sampling. Results show notable gains in minority-class recall (up to ~5–8 percentage points) and stable or improved macro-F1 when using dual-loss with DSD and under-sampling—reaching a macro-F1 of 87 with Balanced HGBC and Random Under-Sampling. The dataset and methods offer a first automated, real-plant-focused solution for CSP fault detection, with potential to standardize evaluation and guide maintenance decisions, while suggesting future work with image fusion and GAN-based data augmentation.

Abstract

Concentrated solar power (CSP) is one of the growing technologies that is leading the process of changing from fossil fuels to renewable energies. The sophistication and size of the systems require an increase in maintenance tasks to ensure reliability, availability, maintainability and safety. Currently, automatic fault detection in CSP plants using Parabolic Trough Collector systems evidences two main drawbacks: 1) the devices in use needs to be manually placed near the receiver tube, 2) the Machine Learning-based solutions are not tested in real plants. We address both gaps by combining the data extracted with the use of an Unmaned Aerial Vehicle, and the data provided by sensors placed within 7 real plants. The resulting dataset is the first one of this type and can help to standardize research activities for the problem of fault detection in this type of plants. Our work proposes supervised machine-learning algorithms for detecting broken envelopes of the absorber tubes in CSP plants. The proposed solution takes the class imbalance problem into account, boosting the accuracy of the algorithms for the minority class without harming the overall performance of the models. For a Deep Residual Network, we solve an imbalance and a balance problem at the same time, which increases by 5% the Recall of the minority class with no harm to the F1-score. Additionally, the Random Under Sampling technique boost the performance of traditional Machine Learning models, being the Histogram Gradient Boost Classifier the algorithm with the highest increase (3%) in the F1-Score. To the best of our knowledge, this paper is the first providing an automated solution to this problem using data from operating plants.

Figures (9)

• Figure 1: Aerial image of a Parabolic Through Collector system captured by a drone with a traditional RGB camera (a), and an infrared camera (b). Images contain receivers with broken and non broken glass. Better viewed in color.
• Figure 2: Decomposition of a CSP plant (using PTC systems) by its fundamental parts. (a) geospatial image of a CSP plant, divided in 8 subfields, labeled from A to H. (b) loop of the HTF, passing through 144 HCEs divided in 4 stages. Blue rectangles define where the temperature of the HTF is measured by the plant. (c) subset of HCEs aerial images captured by the UAV.
• Figure 3: Components from a typical parabolic through receiver (extracted from espinosa2016vacuum).
• Figure 4: Data distribution for the dual problem. On the left, the data concerning to our main problem: broken pipes detection; on the right, the distribution of the data associated to the 7 plants in the surveys.
• Figure 5: Violin plots of four descriptors of the ATSet: (a) for the temperature of the fluid, (b) for the temperature of the glass, (c) for the hydrogen pressure, and (d) for the HCE efficiency. The efficiency plot shows the negative impact of the broken samples in the performance of the plants.