The mold injection control method comprises the steps that three-dimensional point clouds of a current injection molding part and an adjacent previous injection molding part are obtained, the global difference degree between the current injection molding part and the adjacent previous injection molding part is calculated through overall registration, if the global difference degree exceeds a threshold value, a difference
mask is generated and covers a real-time image of a cavity, a difference image is formed, and the difference image is subjected to injection molding. And performing
hybrid clustering on the difference image and a known defect type
database to obtain candidate defect types and membership degrees, calculating
posterior probability contribution degrees of the sensing nodes to the candidate defect types by taking the membership degrees as input and combining a Bayesian
reasoning algorithm, if the contribution degree of any sensing node exceeds a threshold value, outputting a defect
root cause report, and if the contribution degree of any sensing node exceeds the threshold value, outputting a defect
root cause report. And a
machine table or a mold execution mechanism is driven to carry out correction. According to the invention, the problems of low detection efficiency, large subjective deviation and lagging
root cause judgment due to the fact that automobile part defect detection and
root cause analysis depend on manual
visual detection or single-dimensional automatic detection are solved.