The health condition of hydraulic turbines is one of the most critical factors for the operation safety and financial benefits of a hydro power plant. After the massive entrance of intermittent renewable energies, hydropower units have to regulate their output much more frequently for the balancing of the power grid. Under these conditions, the components of the machine have to withstand harsher excitation forces, which are more likely to produce damage and eventual failure in the turbines. To ensure the reliability of these machines, improved condition monitoring techniques are increasingly demanded.

In this article, the feasibility of upgrading condition monitoring of Pelton turbines using novel vibration indicators and data-driven techniques is discussed. The new indicators are selected after performing a detailed analysis of the dynamic behavior of the turbine using numerical models and field measurements. After that, factor analysis is carried out in order to assess which are the most informative indicators and to reduce the dimension of the input data.

For the validation of the proposed method, monitoring data from an actual Pelton turbine that suffered from an important fatigue failure due to a crack propagation on the buckets have been used. The novel condition indicators as well as classical indicators based on the spectrum and harmonics levels have been obtained while the machine was in good operation, during different stages of damage and after repair. All of these have been used to train an artificial neural network model in order to predict the evolution of the crack until failure occurs. The results show that using the improved monitoring methodology enhances the ability to predict the appearance of damage in comparison to typical condition indicators.

水轮机的健康状况是影响水力发电厂运行安全和财务效益的最关键因素之一。在间歇性可再生能源大量涌入之后，水电部门不得不更加频繁地调节其输出功率，以平衡电网。在这些条件下，电机的组件必须承受更强的激励力，这很可能在涡轮机中造成损坏并最终导致故障。为了确保这些机器的可靠性，日益需要改进的状态监视技术。

在本文中，讨论了使用新型振动指示器和数据驱动技术升级Pelton涡轮机状态监测的可行性。在使用数值模型和现场测量对涡轮机的动态行为进行了详细分析之后，便选择了新的指标。此后，进行因子分析以评估哪些是最有用的指标，并减小输入数据的维数。

为了验证所提出的方法，已使用来自实际Pelton涡轮机的监测数据，该数据由于叶片上的裂纹扩展而遭受了严重的疲劳破坏。在机器处于良好运行状态，损坏的不同阶段以及维修后，已经获得了基于频谱和谐波水平的新颖状态指示器以及经典指示器。所有这些都已用于训练人工神经网络模型，以便预测裂纹的发展，直到出现故障为止。结果表明，与典型的状况指标相比，使用改进的监控方法可以增强预测损坏外观的能力。