Research on running state recognition method of hydro-turbine based on FOA-PNN

Chaofeng Lan,Shuijing Li,Huan Chen,Wu Zhang,Hui Li

DOI: https://doi.org/10.1016/j.measurement.2020.108498

IF: 5.6

2021-02-01

Measurement

Abstract:<p>To effectively monitor the operating state of hydro-turbine, a diagnosis strategy based on the operating conditions and pressure pulsation of the turbine is proposed. The improved Hilbert-Huang Transform (HHT) method is used to study the characteristics of pressure pulsation under different operating conditions. The physical parameters of pressure pulsation are extracted through the mutual information theory. Procedures include optimizing the smoothing factor σ of the Probabilistic neural network (PNN) network through the Fruit fly optimization algorithm (FOA), constructing the FOA-PNN network model, classifying the unit operating status. The result shows that when σ = 0.23, the prediction accuracy of the FOA-PNN network is 100%, and the training time is 0.336372 s. It is proven that the FOA-PNN can predict the running state of the turbine in a short time and monitor the running malfunction in real time.</p>

engineering, multidisciplinary,instruments & instrumentation

What problem does this paper attempt to address?

The paper aims to address the issues of monitoring the operating state and diagnosing faults in hydraulic turbines. Specifically, the study proposes a diagnostic strategy based on pressure pulsation signals and operating conditions to effectively monitor the operating state of hydraulic turbines. The main issues addressed include:

1. **Nonlinear and Unstable Signal Processing**: Due to the special and complex nature of hydraulic turbine systems, their vibration signals exhibit high nonlinearity and instability, making it difficult for traditional fault diagnosis methods to accurately identify these signals.

2. **Minor Differences Under Different Operating Conditions**: Although the same fault type may occur under different operating conditions, the characteristic differences are minimal, making it difficult for traditional methods to distinguish them.

3. **Reliability of Real-Time Fault Diagnosis**: Existing methods mainly rely on experience to analyze and process the collected pressure pulsation signals, which cannot ensure the reliability of real-time fault diagnosis.

To overcome the above issues, the study proposes the following improvements:
- Use an improved Hilbert-Huang Transform (HHT) method to analyze the pressure pulsation characteristics under different operating conditions;
- Extract the main fault features of the pressure pulsation signals through mutual information theory and construct vibration feature vectors;
- Utilize a Probabilistic Neural Network (PNN) for training to predict the operating state of the hydraulic turbine;
- Apply the Fruit Fly Optimization Algorithm (FOA) to optimize the smoothing factor in the PNN network, improving classification accuracy.

Through the above methods, the study demonstrates that the FOA-PNN network can accurately predict the operating state of the hydraulic turbine in a short time and monitor operating faults in real-time.