| Heavy launch vehicle is the basic vehicle for the medium and long term development of the next generation of space in China.Fuel tank is the main load-bearing structure of the rocket,so it has a high requirement for weld joint.As a new solid state welding method,friction stir welding(FSW)can obtain higher weld joint strength,and is the preferred welding method of tank.The process of FSW involves complex changes of multiple physical fields.The welding quality is easy to be unstable due to the change of welding process conditions.It is of great significance to improve the stability of weld performance to study the characteristics of welding process state signal and extract the characteristic parameters related to weld performance and apply them to the prediction of weld performance.In this paper,the weld surface texture and real-time temperature signal of 2219 aluminum alloy FSW welding were studied and analyzed,and the weld performance prediction was carried out.The main research contents and innovations of this paper are as follows:1.Based on the single-factor variable method,welding experiments with different process parameters were designed to study the correlation between the roughness of weld surface texture and the stability of weld performance.The theoretical calculation formula of roughness Ra under smooth welding condition is obtained.A method for weld state evaluation based on weld surface texture roughness is proposed.The analysis of Ra test results of weld roughness with different process parameters shows that the deviation between measured Ra value and theoretical Ra value can be used to evaluate the weld state.The maximum deviation between Ra and the theoretical value is only 4.3μm for the stable weld,while the minimum deviation is 6.6μm for the unstable weld.The experimental results show that the surface texture and local contour are regular and uniform when the welds without wear get good weld performance.When the weld performance declines due to weld wear,the regularity of local contour features of weld surface texture gradually changes,so the local contour features of texture can be used to evaluate the wear state of mixing head and the performance state of weld.2.The fractal characteristics of weld images obtained by industrial cameras were studied,and a method to obtain texture roughness information from weld surface images was proposed.Based on fractional Brownian motion model,fractal dimension features are obtained from the power spectrum of gray-scale distribution of texture images.The variation rule of the power spectrum of the weld surface image shows that,for the weld with uniform texture,when the roughness Ra increases due to the process parameter factors,the main frequency Fm of the power spectrum decreases;when the texture is chaotic due to the unstable welding state,the corresponding amplitude of the main frequency Fm of the power spectrum decreases,which will lead to the increase of the extracted fractal dimension.It is verified that the fractal dimension extracted from the weld image can better reflect the variation rule of the weld texture roughness Ra.The light sensitivity and noise robustness of the algorithm are further studied,and the algorithm can achieve better light sensitivity and noise robustness through homogenization and mean filtering.3.A block multi-scale enhanced complete local binary mode coding method is proposed to extract local texture contour features from weld images.Through different process parameters test weld image coding mode of statistical analysis,the results show that the traditional method of coding mode of non-uniform proportion is as high as 15.41%,based on the binary sequence jump in the number of times for the four patterns for further subdivision,improved method of non-uniform pattern types of slope to 1.54%,can obtain more texture information.Through local coding of different scale radii and feature extraction method of texture image region segmentation,the extracted feature vector contains multi-scale and spatial location information,which improves the recognition ability of local features of weld surface texture.In order to improve the noise robustness of the algorithm,the Dixon outlier test strategy is introduced and the outlier is corrected by means of the mean value of neighborhood pixels.The k-nearest neighbor model was used to study the feature recognition performance of the weld texture local contour feature extraction method.The results show that the texture recognition accuracy of the traditional rotation invariant uniform binary mode algorithm is only 81%,and the proposed method can improve the texture recognition accuracy to 96%,and the algorithm has better noise robustness.4.A real-time monitoring and wireless signal transmission system for welding temperature was developed.Temperature signals on the friction interface between tool shoulder and workpiece during welding were obtained by embedding thermocouple sensors on the tools,and the error fluctuation of the testing system was within 5℃.The correlation between welding temperature signal and weld microstructure and mechanical properties is studied.The results show that real-time welding temperature is an important parameter to evaluate the performance of weld.When the welding condition is unstable,the temperature signal will show a large amplitude of low frequency fluctuation.When hole defects occur in welding,there is a higher amplitude fluctuation in welding head rotation frequency.Based on the wavelet packet method,the real-time temperature signal is time-frequency analyzed,and the signal characteristic components of different frequency bands can be obtained,and the decomposition and reconstruction of temperature characteristics and noise signals reflecting the stability and welding state of the welding process can be realized.5.Research on the prediction of weld performance of FSW based on feature fusion of multi-source information was carried out.The fusion of visual features of weld surface texture,real-time temperature characteristics and welding process parameters was used as feature vector,and the weld performance grade was determined by the tensile strength coefficient of weld joints.A weld classification prediction model based on genetic optimization support vector machine was established.After model prediction analysis of different kernel function types,it is determined that gaussian kernel function can obtain the best prediction accuracy.The comparative study of the prediction performance with different information characteristics as input shows that the prediction accuracy is only 73.2% when only using the process parameter information to predict the weld performance grade.The prediction accuracy of welding seam performance classification using multi-source information feature fusion method can be improved to 96.8%.By introducing principal component analysis,the running time of the algorithm is reduced from 1.74 s to 0.78 s.It can effectively eliminate the correlation between multi-source information,improve the real-time performance of the algorithm,and obtain the welding seam performance prediction accuracy of 96.3%. |
PDF Full Text Request