| Hydrogen crushing of neodymium iron boron is a commonly used milling method inthe permanent magnetic material production process. The production operation data, thealloy state and the hydrogen content in alloy, cannot be detected online in the NdFeBhydrogen crushing process; resulting in NdFeB hydrogen crushing process is difficult toachieve a high level of automatic control. Workers can only rely on experience to judge,extend the hydrogen absorption time to ensure that the alloy completely crushed, resulting inthe production cycle is prolonged, the resources are wasted, production costs increase. Thestate of alloy in the absorbing hydrogen reaction process is unknown. So it is difficult toadjust the optimal control parameters, resulting in the quality of the alloy powder will beaffected. In order to guarantee the product quality and production efficiency, it is needed toimplement online detection of NdFeB hydrogen crushing control system parameters to meetprocess control requirements.This paper makes thorough analysis and research on the NdFeB hydrogen crushingprinciple, process, and important factors affect hydrogen crushing product quality andproduction cycle as well as the difficulty to control. Aiming at the problem that the importantparameters of crushing process can not be online detection, proposed to use the supportvector machine method to realize the prediction of hydrogen content in NdFeB alloy, andaccording to the forecast results to adjust the control parameters, to shorten the productioncycle, improve the quality of the products.Support vector machine is developed based on the theory of statistics, converts lowdimensional nonlinear problem into a linear problem in high dimensional space by kernelfunction calculations, can solve the small sample, is an important soft measurement tools tosolve nonlinear problems. With its strong learning ability and generalization ability, it is widely concerned by scholars in the field of intelligent algorithm. It has been successfullyapplied in many fields. The main contents of this paper include:Firstly, this paper analyses the factors affecting product quality and production cyclethrough researching NdFeB hydrogen crushing principle and process.Secondly, this paper introduces the theory foundation and basic principle of supportvector machine briefly, uses historical data of production process to establish support vectormachine prediction model, simulates and verifies the model by MATLAB software.Thirdly, with the objective of minimum mean square error of support vector machineprediction model based on cross validation, support vector machine model parameters wereoptimized by particle swarm optimization algorithm. Comparing with grid search method,particle swarm optimization algorithm has higher prediction accuracy and bettergeneralization ability.Moreover, hydrogen crushing condition database has been built, exchanges data withMATLAB through ODBC method.It is convenient for MATLAB prediction model toobtain real-time data. The prediction results are stored in the database for analysis and use.Last but not the least, this paper designed the monitor screen of control system,configured variables, and completed the prediction data interaction between the model andthe WINCC through the OPC communication. The prediction model is Applied in thecontrol system, so WinCC can online monitor unpredictable process parameters. It is toconvenient for workers to implement corresponding control. |
