| According to the preset geometric model to calculate the theoretical weight of the crystal,the growth weight of laser crystal growth based on the Czochralski method is controlled to achieve the purpose of controlling the diameter indirectly,which the process has the characteristics of time-varying,nonlinear,pure hysteresis and large inertia.With the growth of the laser crystal,the liquid level of the crystal inside the crucible is decrease and the ratio of solid solution in the furmace is change,which cause the temperature fluctuation in the furnace.The melting segregation of solid solution interface leads to the continuous decrease of crystal crystallization temperature.The increase in the length of the crystal which makes the worse of the heat dissipation performance also cause temperature drift in the furmace during the late growth period,resulting in the interruption of crystal growth.In view of the above factors affecting crystal growth,the key control techniques of crystal growth were studied from the perspective of crystal growth control algorithm.First of all,the crystal furmace and crystal were modeled based on crystal growth input/output data.Secondly,crystal growth controller was designed which modeling with three crystal diameter control schemes,such as fuzzy PID control algorithm,BP neural network tuning PID control algorithm and BP neural network tuning PID control algorithm optimized by Adagrad optimization algorithm.Finally,the crystal growth control system simulation model was built based on MATLAB/SIMULNINK,by analyzing the interference factors affecting crystal growth during crystal growth,the equivalent model of the interference factor was added to the system simulation model.Through the simulation operation of SIMULINK,the step response curves and interference response curves of the three control schemes were obtained,and compared to verify the effectiveness of the algorithm.Simulation experiments show that the BP neural network tuning PID control algorithm which improved by Adagrad optimization algorithm can achieve better crystal growth control effect.Compared with fuzzy PID algorithm,it has better dynamic response and the adjustment time is reduced by 45 s.Compared with the traditional BP neural network PID algorithm,the tuning process is more stable,the system overshoot is controlled at 0.13%.It can better overcome the interference during the crystal pulling growth process,keep the growth temperature stable,inhibit the temperature drift and achieve equal diameter growth of the crystal.The actual test of the weighing system shows good precision and linearity,and the weight error is less than ±0.1 g which meets the crystal growth requirements;Data analysis of the crystal growth process shows that the crystal diameter error is less than±1mm and the crystal growth is stable. |
