During actual production line commissioning, we found that AOI inspection must be strictly synchronized with the tact time of the Solder Paste printer and placement machine; otherwise, image smearing or missed detection will occur. Especially when the line speed exceeds 90cm/s, if the light source trigger delay exceeds 15μs, the recognition accuracy of pad edges will drop sharply from 99.2% to 93.7%. The JJY AOI equipment, through a hardware-level IO hard trigger mechanism, stably controls the response error within ±2μs, ensuring that the imaging position of each PCB on the conveyor belt is highly consistent.

Inspection parameter settings cannot be copied directly from standard templates. For example, if the contrast threshold for 0201 component solder joints is set to 45, slight solder balls are easily misjudged as acceptable, but adjusting it to 52 will mark some genuine good products with an offset within 0.08mm as NG. We finally conducted repeated verification on 12 production lines of different models, and confirmed that dynamically binding this parameter to the product of the stencil aperture area ratio and the slope of the reflow oven temperature zone curve, followed by ±3 floating correction, can balance sensitivity and fault tolerance.

AOI inspection results must achieve two-way data penetration with the MES system. It is not simply exporting CSV reports, but writing the pixel deviation value, confidence score, and associated SPI solder paste volume data of each NG coordinate into the MES work order ID field. In this way, when process personnel investigate abnormalities, they can directly retrieve the historical data chain of the same solder joint across the three stages of printing, placement, and reflow, avoiding logical discontinuities such as "AOI reports an offset, but SPI shows the solder paste is centered".

On-site, we often encounter OCR character recognition failures caused by reflection interference on the green solder mask surface. Instead of adopting the symptomatic solution of adding a polarizing filter, we let the JJY AOI perform a low-power pre-flash before exposure to collect the substrate reflection feature map, and then dynamically adjust the main light source angle and RGB channel gain ratio. Actual tests show that the character recognition rates for FR-4, high-frequency PTFE, and metal substrates reach 99.6%, 98.3%, and 97.1% respectively, without the need to replace lenses or recalibrate.

Maintenance frequency directly affects inspection stability. We require that 3 sets of baseline images must be run with a standard glass calibration board before startup each shift, and the 2nd set must forcibly include a ±0.15° rotation disturbance to test the angle compensation module. If the pad center coordinate offset exceeds 0.02mm after two consecutive rotations, the image processing firmware must be restarted instead of just cleaning the lens. This action seems cumbersome, but it can intercept more than 83% of false alarm rate escalation problems in advance.

When the AOI continuously outputs more than 5 pieces of the same NG type, the system will not automatically stop the line, but will pop up a red box flashing prompt on the HMI interface, and simultaneously push an alarm package containing a coordinate heat map to the equipment engineer's mobile phone. Because we found that most batch defects originate from stencil tension attenuation or placement machine nozzle vacuum fluctuations. These physical quantity changes are reflected as gradual feature offsets on AOI images, and must be cross-verified with equipment underlying sensor data to accurately locate the root cause.