Study On Hardness Prediction Model Of Tin - Based Alloy Based On Metallographic Image

Tin-based alloy is used as bearing materials working in high speed and overloaded condition because low friction coefficient, moderate hardness and better tenacity. Its hardness largely depends on the microstructure, including hardness of each phase, the relative amount and distribution of metallurgical organizations. Establish a quantitative correlation model between the hardness of the alloy and its microstructure can thoroughly analyze the regular pattern between the change of each phase parameters and hardness of the alloy, provide reference and basis for studying a new alloy and formulating aging technique of alloy production.Different microstructure will be eroded by the etchant in varying degrees, thus showing different colors or brightness in the microscope. Automatic image analyzers gray the metallographic picture and select the appropriate threshold values to distinguish phases accordingly this principle. However, for multi-phase alloys, it can’t fully identify its metallurgical composition by gray-scale difference alone, such as tin-based alloy, there are two distinct phases in the white area of the metallurgical photo. Usually, different generation order, different morphology, so it can identify different phases which has small difference of gray by morphological differences. Currently, more of these multi-phase alloys microstructure are manually identified by using image processing software, tint each phases and count the relative amount of each phase. It’s slow, inefficient and heavy workload. Therefore, studying automatic metallographic identification method based on image processing has high practical value and significance.In quantitative metallographic analysis, the 2d tissue parameters of interface are converted to 3d space parameters according to the theory of solid geometry and stereology. This paper has given a metallurgical image recognition method based on the characteristics of tin-based alloy microstructure, established its hardness predictive model by the extracted characteristic parameters of alloys. Firstly, analyzed the factors and mechanisms influence hardness by experimental data. Secondly, identified the tin-based (α solid solution) by using Otsu threshold segmentation. Then, extracted the white area organizations in the metallographic photo by using morphological edge detection, scan line seed filling algorithm. An improved intersection counting method is given for the problem of the polygon contain. Calculated its compactness and aspect ratio of the minimum area circumscribed rectangle, set up a SVM classifier based on these two characteristic parameters to distinguish other two microstructure in the white area. Finally, extracted relative amount, distribution of each phase in the phase diagram, a total of six parameters, linear and nonlinear prediction models of hardness are constructed based on the parameters by partial least squares regression analysis.Experimental results show that the improved intersection counting method can better solve the image misrecognition problem caused by area containing. It can effectively identify the metallurgical composition and distribution of tin-based alloys through gray difference and nonlinear SVM classifier based on graphics features. The nonlinear PLS model can effectively predict the hardness of tin-based alloy, the accuracy of test samples is up to 98%.