Research On The Identification Method Of Inclusion Type In Casting Based On Random Forest Algorithm

Cylindrical castings have been widely utilized in many areas,such as defense and military,transportation,metallurgy and chemistry,etc.However,inclusions or pores are easily formed in the production process,which affect the mechanical properties and corrosion resistance of castings.Therefore,the non-destructive testing of castings is one of the important means to ensure the quality of products manufactured.In this paper,on the basis of the traditional ultrasonic testing methods for cylindrical aluminum castings,the acoustic field characteristics,simulation technology,echo characteristics and inclusion type recognition algorithms for inclusion-containing castings are analyzed.Guided by the acoustic propagation principle of the ultrasonic inside the cylinder casting,a two-dimensional ultrasonic simulation model containing inclusion castings is established with the help of COMSOL finite element simulation software.Based on the verification of the correctness of the simulation model,the study obtains inside the casting four kinds of inclusions（Al₂O₃,TiN,Mg,Fe）,four inclusion length values（4mm,6mm,8mm,10mm）and four width values（0.5mm,1mm,1.5mm,2mm）,whose echo signals provide data support for subsequent echo analysis and inclusion type recognition.Aiming at the different inclusion defects in aluminum castings,the ultrasonic echo signal feature extraction and inclusion type recognition algorithms are deeply studied.Firstly,the time-domain and frequency-domain characteristics of the flaw echo and the bottom echo are analyzed when the probe is directly facing the inclusion.It is found that both the time-domain amplitude and the frequency-domain peak can reflect the inclusion characteristics.Secondly,in order to obtain the corresponding relationship between the time-domain amplitude,frequency peak value and inclusion type of the echo,excitation from different angles and stimulation calculations are conducted on half circumference that is180 degrees.In addition,the EMD method is used to decompose the defect echo and the bottom echo,and the energy distribution of the echo is selected as the input feature of the inclusion type recognition according to the physical meaning of the feature.The corresponding relation between time domain amplitude and frequency domain peaks and inclusions can be used as the input feature vector for inclusion type identification.A decision tree algorithm is used to identify the type of inclusions in the casting.In order to further improve the generalization performance of the model,this paper proposes a strong recognition model of random forest algorithm composed of multiple decision trees to identify the types of inclusion,and compares the identification performance of random forest algorithm and three decision tree algorithms（ID3,C4.5,CART）to the inclusion types.It is found that the recognition effect of random forest algorithm is the best.