Stepped fishways are structures that allow the free movement of fish in transversal obstacles in rivers. However, the lack of or incorrect maintenance may deviate them from this objective. To handle this problem, this research work presents a novel low-cost sensor network that combines fishway hydraulics with neural networks programmed in Python (Keras + TensorFlow), generating the first autonomous obstruction/malfunction detection system for stepped fishways. The system is based on a network of custom-made ultrasonic water level nodes that transmit data and alarms remotely and in real-time. Its performance was assessed in a field study case as well as offline, considering the influence of the number of sensing nodes and obstruction dimensions. Results show that the proposed system can detect malfunctions and that allows monitoring of the hydraulic performance of the fishway. Consequently, it optimizes the timing of maintenance on fishways and, thus, has the potential of automatizing and reducing the cost of these operations as well as augmenting the service of these structures. Therefore, this novel tool is a step forward to achieve smart fishway management and to increase their operability.
Keywords: clogging; fishways; hydraulic modeling; low-cost; neural networks; water-level sensors.