Terminal Defect Detection Based On Machine Learning And Feature Extraction

Terminals are now widely applied to various fields,in pass electrical connection effect,a qualified terminal product can bring a lot of convenience,to the production and use to avoid a lot of trouble terminal product defect detection is to detect the defects with the purpose of the terminal blocks,terminals are applied to avoid defects in production and life there are many reasons that cause defective terminal products,such as: less wire and terminal deep pressure,terminal now terminals such as shallow play detecting method is used for manual visual inspection and instrument testing,the inspection method of low cost and high efficiency,less than the requirements of modern production based on machine learning and feature extraction,this paper designs a set of perfect terminal crimping defect detection device to realize the detection of terminal product defects.The main research contents of this paper are as follows.First of all,according to the actual demand for defect detection terminal complete terminal pressure welding defect detection device and the system hardware modular structured design and layout,the main contents include: sensor selection and configuration of the hardware layout pressure acquisition module programming upper main program design and design of human machine interface.Secondly,this paper designed a curve fitting feature extraction algorithms,the algorithm using Fourier function of sensor response curve fitting,the fitting is completed,fitting curve of the model parameters as feature extracting data existing in the original signal can be the actual sensor noise signal,the larger effects on the extraction of the characteristics of the data of the original signal in order to eliminate the sensor noise,before the response curve fitting of sensor,using the method of wavelet de-noising of sensor signal preprocessing,eliminate the high frequency noise signal of sensor signals keep characteristic signal part of the original signal.Again,in order to judge according to the characters of terminal data terminal category,this paper,by using the machine learning can learn from the given data,the characteristics of machine learning classification model is established to study characteristics of terminal data and classification in order to be chosen from among many machine learning classification algorithm is most suitable for the classification of defects detection system algorithm,this paper designed a Naive Bayes,Decision Tree,Support Vector Machine(SVM)and Random Forest algorithm contrast experiment of classification algorithm,four to contrast experiment results accuracy and specificity performance index for evaluating classification algorithm Experimental results show that the Random Forest algorithm is better than other algorithms in the performance of the two indexes and is most suitable for the detection system.Finally,using different specifications of the terminal products to verify the correctness of the test system,the results of validation experiments show that the equipment for defect detection rate reached 100%,terminal products and terminal product testing qualified rate of false positives under 2%,can meet our design expectations,also can meet the requirements of factory production,can be put into mass production.