Supervised Learning-Based Excess Particle Weight Estimation Technique

The remainder in sealed electronic components is one of the main factors hindering the reliability of electronic systems.To solve the problem of remainder,the existence of remainder must be judged,and the production process of components must be evaluated in accordance with the properties of remainder particles,so as to reduce the remainder in the production.At present,there are few analyses on the weight of the remainder.Based on the supervised learning,this thesis studies the weight estimation technology of remainder based on supervised learning.First of all,this thesis proposes a research scheme for estimating the weight of remainder based on the principle of the detection of particle collision noise.According to the weight range of common remainders and the sensitivity of testing equipment,this thesis designs and select two types of sample libraries of sealed relays and sealed electronic components containing remainders.Then,this thesis extracts the weight features of the remainder in the way of the Hilbert Huang transform combined with the traditional method of extracting features of the remainder.After selecting feature and eliminating the features with low correlation and low importance,a feature set describing the weight of the remainder is formed.After verification,the new feature set can more comprehensively describe the information of remainder weight.Based on the classification and regression methods of supervised learning,this thesis respectively conducts the qualitative and the quantitative analysis of the weight of remainder particles on the sealed relay and electronic component data sets.Different qualitative analysis models of remainder weight are established by eight classification algorithms,and the hyper parameters of the models are optimized to obtain the best analysis performance.Different quantitative analysis models for the weight of the remainder are established with five regression algorithms,so as to realize the quantitative estimation of the weight of the remainder.Finally,on the basis of the above research results,a platform for estimating the weight of remainder based on Py Qt5 is established.According to the experimental results,the supervised learning-based redundant weight estimation technology can effectively evaluate the weight of the remainder..