Slm printing defect detection and repair method and system based on deep learning network

A selective-laser-melting (SLM) printing defect detection and repair method based on a deep learning network, the method comprising:

• S1, acquiring a molten pool image in an SLM printing process, preprocessing the molten pool image, and making a defect dataset according to a defect type;
• S2, training a first neural network model through the defect dataset to obtain a defect recognition model for recognizing a defect and determining a defect type;
• S3, in a printing process of a part to be detected, collecting the molten pool image in real time, and performing online defect recognition through the defect recognition model;
• specifically,
• if a current layer has no defects, performing S4;
• if a defect is recognized in the current layer, selecting whether to repair the defect according to a defect type, wherein the defect types include a short crack, a pore, a deformation, and a long crack, the short crack and the pore belong to the common defect category and the deformation and long crack belong to the serious crack category;
• if a defect is identified in the current layer, selecting whether to perform defect repairing based on the defect type, wherein the step specifically includes:
• acquiring a type, output by the defect recognition model, of each defect and a corresponding defect confidence level in real time;
• if the defect type contains a serious defect, directly ending printing;
• if all defect types are common defects, computing a final defect confidence level of the common defects:
• FCL
• =
• w
• ×
• N
• +
• (
• 1
• -
• w
• )
• ×
• M
• wherein FCL is a final defect confidence level, w is a weight, w∈[0, 1], N is an average of defect confidence levels corresponding to pore defects, and M is an average of defect confidence levels corresponding to short crack defects;
• determining whether the final defect confidence level is greater than a pre-set confidence threshold;
• if so, selecting defect repair;
• otherwise, performing S4 to carry out powder spreading and printing on a next layer;
• and
• if defect repair is selected, after the current layer is printed, suspending powder spreading once, predicting laser remelting parameters by adopting a pre-trained second neural network model, and performing laser remelting repairing on the defect by adopting the laser remelting parameters until the current layer has no defects;
• and
• S4, continuing to perform powder spreading and printing on a next layer, returning to S3, repeating the processes of online defect recognition and laser remelting repairing until the part to be detected is printed.