Studies Of Ultrasonic Casting Power Based On The Dynamic Matching Of Neural Network

Aluminum alloy ultrasound casting is a new technology which is pollution-free and high efficient as well as has great potential. While the ultrasonic power is a crucial part in the ultrasonic processing system, its performance will affect the quality of aluminum alloy processing directly. Ultrasonic transducer is a capacitive load and its internal dynamic parameters will be changed by the impact of casting conditions or tool head wear, resulting in the resonant frequency of the transducer drifting and resonant system detuning. Therefore, we must make reasonable dynamic impedance matching for the ultrasonic transducer in order to make the power supply have the highest efficiency. So this paper mainly focuses on studying the dynamic matching network of ultrasonic power and its control algorithm.According to the performance requirements of the power required for aluminum alloy ultrasound casting and from the perspective of easy implement and stable work, an uncontrolled rectifier of single-phase diode, MOSFET half-bridge inverter and the main circuit topology of level-adjustment function by high-frequency transformer was employed.Through analyzing the resonant frequency characteristics of the piezoelectric transducer, the series resonant frequency is chosen to be the system frequency. After some analysis and comparison of the matching programs, an alterable inductance of dynamic matching of LC network that the matching capacitance had a fixed value and the matching inductor could adjust the air gap was chosen. Moreover, the circuit and control method for implementing this program was presented.By using BP neural network to predict the amount of the matching inductor and employing the variable step method to search the maximum current, the real-time tracking of resonant frequency of ultrasound transducer with the output frequency of ultrasound power can be achieved. By combination of the two intelligent algorithms, the dynamic matching of ultrasonic casting power together can also be realized. Furthermore, some simulation about the prediction process of neural network and its power system with MATLAB were done and it is found that the BP network can predict accurately the amount of matching inductance, demonstrating the dynamic matching algorithm here is effective and can meet the expectations. So the research of the paper has an important significance of study on the development of ultrasonic casting power.