Performance prediction of axial pump based on wavelet neural network

Axial pumps have been widely used in hydraulic and agricultural engineering. During the process of manufacturing and transformation of axial pumps, it is very important to predict the essential performance and obtain essential performance curve. However, due to the short flow channel and the rapid change of flow pattern, the internal flow becomes extremely complex. Compared to centrifugal pump, it is very difficult to obtain the essential performance curve. So, the processes of design and manufacturing are cumbersome and the economy is very poor. Commonly, manufacturers use the processes which are the circles of designing, prototype, test and improvement to obtain the essential performance curve, while it is tedious, cumbersome and laborious. If the essential performance and essential performance curve can be predicted, the associated costs will be reduced significantly, and the cycle of the design, manufacture and renovation will be shorten obviously. To obtain essential performance curve, the traditional methods are using model test or similar conversion, and their drawbacks are expensive cost, long-cycle test and over reliance on assumption accuracy and satisfaction degree. Neural network is a new idea and method to solve the performance prediction of axial pump. Today, wavelet neural network has been widely used in various engineering fields, such as water conservancy, energy and electron. Wavelet neural network uses continuous wavelet function instead of back propagation(BP) neural network activation function. It inherits the advantages of wavelet transform and BP neural network, and has many characteristics, such as self-learning, self-adaptive, nice time-frequency, and strong modeling capabilities. Using wavelet neural network method, the prediction model was established, which was suitable for axial pump to predict essential performance and obtain essential performance curve. The author selected the Morlet wavelets as wavelet function, increased the momentum item and adopted the gradient descent learning algorithm. The adaptability, convergence and accuracy of the model were examined by 2 models′ training. Compared with the BP neural network, this model showed the advantages that the convergence speed was accelerated and the accuracy was improved greatly. Through the contrast and analysis between predicted data and tested data in actual engineering renovation, the prediction accuracy, stability and practicability of the model were further proved. The results showed that the model had higher precision and stability, the transformation cycle was also shortened effectively, and the renovation cost was reduced greatly. According to the above, the essential performance curve can be obtained and the essential performance can be predicted economically and reliably. So, the research based on the wavelet neural network has a higher practical engineering value, and it can provide technical support and reference during the process of design, manufacture and renovation of axial pump.