| Silicon single crystal is a vital semiconductor material,which is widely used in integrated circuits fields.How to optimize silicon single crystal growth equipment to improve the quality of silicon single crystal is one of the critical issues faced by the industry at present.CZ is the main technical method for producing large-size,low-defect,high-quality semiconductor silicon single crystal,the lifting system is a critical component in the silicon single crystal growth equipment,and it must make the seed crystal ratate and lift smoothly in the growth of silicon single crystal.However,as the use time of silicon single crystal furnace becomes longer,the running state of the lifting system will change under long-term heavy load,which will affect the dislocation-free growth state of the crystal and the quality of the silicon single crystal,so it is necessary to identify and predict the running state of the lifting system in time.Therefore,this paper studies the identification of the operating state and reliability prediction of the lifting system based on its vibration signal,which provides a basis for the growth process of silicon single crystal and realizes the healthy management of silicon single crystal growth equipment.The main research contents are as follows:1.Aiming at the problem that the vibration signal of the lifting system is easily covered by noise,which leads to insufficient feature information extraction,a method is studied to enhance the vibration signal and then extract the features.First,a signal enhancement algorithm MCKD based on MVO is designed to adaptively optimize the L and M to remove the noise components in the vibration signal to the maximum extent.Then,a feature extraction method combining adaptive VMD and multi-scale fuzzy entropy is proposed,which enriches the feature information and can better characterize the time-frequency characteristics of the lift system.Finally,the bearing data set of CWRU and the vibration data of the silicon single crystal lifting system test bench were used to verify the effectiveness of the algorithm.2.Aiming at the problem that there are many normal samples and few abnormal samples of silicon single crystal growth equipment,a state identification method under the condition of sample imbalance is studied.Firstly,the linear kernel function and RBF function are used to construct a multi-kernel function in the SVM,so that the generalization ability and learning ability are taken into account in the constructed model.Secondly,the PSO is introduced to optimize the weight of the multi-kernel functions to improve the identification rate.Finally,the bearing dataset of CWRU and the vibration data of the silicon single crystal lifting system test bench are used to verify the validity of the constructed model,and the identification results are compared with those of unimproved algorithm.3.Aiming at the problem that the historical data of the lifting system is insufficient,a reliability prediction method combining CRPF with BP is proposed.Firstly,an intelligent collaborative resampling strategy is designed to replace the traditional resampling in CRPF,and a prediction model of multi-cooperation CRPF is constructed.Then,a model combining multi-cooperation CRPF with BP is constructed to improve the prediction accuracy,BP is used to correct the prediction error of multi-cooperation CRPF.The contructed model combines the advantages of CRPF and BP,which not only realizes the modeling of few data of lifting system but also has adaptive ability.Finally,the reliability verification is carried out by using the dataset of life prediction and silicon single crystal lifting system test bench,which shows that the proposed method can realize the reliability prediction of the operation state of the lifting system,and provide a decision-making basis for system operation and maintenance management. |
