The invention discloses a
fiber composite material defect type and distribution prediction method based on comparative learning. The method comprises the following steps: firstly, deploying a
data acquisition system, acquiring surface temperature distribution of a
composite material by using a thermal imaging camera, monitoring key manufacturing parameters such as
laser power, scanning speed and
layer thickness through a process parameter sensor, and constructing a multi-
modal data set; secondly, designing a double-flow
encoder architecture, extracting defect texture features in thermal imaging through a
convolutional neural network and a full connection layer, performing standardized embedding on process parameters, and introducing a contrast
loss function to realize multi-
modal feature alignment; and then constructing a representation fusion module, inputting the image and parameter features into a defect classifier and a
spatial distribution regression device, and obtaining a prediction network through training. And developing a defect prediction and
evaluation system, deploying a model and evaluating defect influence in combination with a performance
database. And finally, a closed-loop feedback mechanism is established, and process parameters of additive manufacturing equipment are adjusted according to a prediction result to reduce the defect generation probability. According to the method, the defect type and distribution of the
fiber composite material are effectively predicted, the process is optimized, and the product quality is improved.