Size Prediction Of Cladding Tracks In Directed Energy Deposition Based On Support Vector Regression

In the field of additive manufacturing,directed energy deposition is efficient and can manufacture large work pieces.The main factors for evaluating the quality of directed energy deposition manufacturing are shape accuracy and internal quality.To realize the controllability of the quality of directed energy deposition manufacturing through intelligent modeling has gradually become one of the research hotspots for many scholars at home and abroad.This research focused on intelligent modeling of directed energy deposition,and studied the algorithm of predicting the size of cladding tracks and the method of optimizing the process parameters.The main research contents are as follows:Firstly,the paper introduced the directed energy deposition experimental platform,including Robots,laser devices,etc.Taking the laser power and other process parameters as inputs,and the width and height of cladding tracks as outputs,the experiments were designed,and data for training models was obtained.By analyzing the experimental data,it was known that the laser power had the greatest influence on the width and height of cladding tracks,and the correlation coefficients were 0.81 and 0.72 respectively.Secondly,it was proposed to build a support vector regression(SVR)model to predict the size of cladding tracks,and optimize the conventional SVR model hyperparameters searching algorithm.The algorithm was improved and the results shown that the improved algorithm had a faster searching speed.Comparing the SVR model with the neural network,the results shown that the predicted width and height of cladding tracks with SVR model had higher accuracy and the average relative errors were 4.58% and 5.33%.On the basis of predicting the width and height of cladding tracks accurately,the method of predicting the aspect ratio of cladding tracks were studied,and an algorithm based on recurrent neural network was proposed.And the algorithm could accurately predict 70% of the test set data,which indicated a feasible research direction,and pointed out that to further improve the predicting accuracy of the aspect ratio of cladding tracks,more factors needed to be considered as inputs,for example the changes of temperature.Finally,the paper built a model combination system by using Python for optimizing the process parameters,so as to obtain the corresponding process parameters according to the target size of cladding tracks.In the key step of adjusting the process parameters,a search method based on breadth first search(BFS)algorithm was proposed.Searching the same and ideal results,the grid search algorithm needed 584 ms while the BFS algorithm only needed 32 ms.The experiments were conducted on the basis of the model combination system.And the results shown that the average relative errors of the actual width and height of cladding tracks and the target size of cladding tracks were 4.12% and 6.15% respectively,which proved the effectiveness of the model combination system.That had laid a foundation for ensuring the accuracy of different positions in complex structural parts manufacturing,realizing automatic online optimization of the process parameters,and further realizing intelligent manufacturing.