Research On The Classification Of Screw Locking Features Based On Machine Learning

The parts of the phone are connected by screws,and machine locking screws have become a mainstream trend.In the process of screw locking,the result of screw locking is a crucial step.This paper takes the automatic locking system of mobile phone screws as the object,researches the decision tree algorithm of screw locking feature selection,and researches the weighted kernel K-means algorithm and the Least Squares Support Vector machine(LS-SVM)algorithm of screw locking feature classification.The main work includes the following five aspects:1.Aiming at the screw locking result detection system,a machine learning algorithm is proposed for feature classification.Discussed the hardware control system composition and software framework of the three-axis automatic locking screw machine,as well as the electrical control principle of the machine.According to the hardware composition of the three-axis structure,a three-dimensional structure model is established using CATIA software,and the functions of some hardware and the process of the system are discussed in detail.2.Aiming at the part of data feature selection,an improved inverse distance weighted resampling method based on adjacent data and a normalization algorithm based on threshold culling rule are proposed.ID3 tree building method and pessimistic error pruning are used to simplify the features,and six features are obtained.Experiments show that the improved weighted resampling method and threshold removal normalization algorithm improve the reliability of the data and enhance the accuracy of feature selection by the decision tree algorithm.3.Aiming at the classification of screw locking results,two improved weighted kernel K-means algorithms are proposed.Firstly,a K-means algorithm based on the Kernel Taylor expansion is proposed,which reduces the calculation amount of the kernel function;At the same time,the decision rule of soft maximum activation function is introduced to reduce the number of iterations.Experiments show that the operation speed of the improved algorithm is increased by 34.2%.Secondly,using the kernel function,a feature dimensionality reduction weighted kernel K-means clustering algorithm is proposed to realize the visualization of the distance within the class;at the same time,the mean value of some data is used to replace the initial clustering center to improve the accuracy.Experiments show that the classification accuracy rate is slightly reduced but the classification speed is increased by 34.79%.4.Aiming at the classification of screw locking features,the LS-SVM algorithm developed by Nuclear Taylor is proposed.The kernel function is improved by expanding the kernel function to take the first three items,and at the same time,the grid search algorithm with the step change rule is used to optimize the parameters,and when the optimal parameter combination is selected,not only the highest accuracy rate is considered,but also the principle of less support vectors is considered.The simulation experiment proves that the improved algorithm can maintain the correct rate while increasing the operation speed of the traditional algorithm by 12.7%.5.The simulation experiments are carried out to compare the three feature classification algorithms.In terms of accuracy,the LS-SVM algorithm has a higher accuracy rate,and in terms of operation speed,the improved LS-SVM algorithm has a faster operation speed.Therefore,the improved LS-SVM algorithm was selected to be applied to the classification system of the mobile phone screw locking feature,and the MATLAB human-computer interaction system was established to clearly show the intermediate process and final results of the classification algorithm.