Evaluation of different defect-inspection setups for injection molding parts based on the deep learning method

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Quality Control

Background:

• The shift to small-batch production in injection molding increases product variety and necessitates efficient quality control.
• Current quality control often relies on manual inspection, which is inefficient for complex defects.

Purpose of the Study:

• To propose a methodology for evaluating deep learning-based automatic optical inspection (AOI) strategies.
• To compare the effectiveness of different AOI setups for detecting surface defects in injection-molded parts.

Main Methods:

• Assessed three AOI setups: static frontal imaging, belt conveyor inspection, and robotic-assisted inspection.
• Utilized deep learning algorithms for defect detection.
• Developed a systematic workflow for evaluating and optimizing inspection parameters.

Main Results:

• Significant differences in defect detection capabilities were observed across the AOI methods.
• Robotic-assisted inspection demonstrated superior performance, achieving higher accuracy.
• Flexibility in camera angles and positions contributed to the robotic system's effectiveness.

Conclusions:

• The robotic-assisted AOI approach is highly effective for detecting complex surface defects in injection molding.
• The proposed methodology aids in informed decision-making for AOI system design and implementation.
• This research bridges the gap between AI development and industrial application in quality control.