Why did Seagate rebuild EAVS defect detection with machine learning?
Since the 2010s, manufacturing vision systems have rapidly advanced, with AI integration offering significant benefits. In Seagate's slider factory, EAVS machines were initially seen as limiting throughput, measured by unit per hour (UPH) and first pass yield (FPY) for air bearing surface (ABS) and poles, relative to the predetermined overhead rate (POR). The team identified image processing time as a key area for improvement to meet POR targets.
Cost was another challenge. Enhancing EAVS detection capabilities required a strong return on investment (ROI) without compromising quality. The strict "C=0" (no defects found in the accepted sample size) quality assurance (QA) standard meant a single defect in sample lots triggered full rescreening and recleaning of shipments.
What were the EAVS throughput and rescreening goals?
The slider team set both short-term and long-term goals. Short-term goals included replacing conventional image processing with a machine learning-integrated solution to improve defect detection in the final slider inspection. Also planned was improving outgoing defective parts per million (DPPM) and lot acceptance rates (LAR) to ensure downstream slider quality. An additional aim was to not impact EAVS machine UPH throughput or costs.
Long-term goals focused on QA and traceability improvement for downstream failure analysis. For QA, the team planned to replace physical parts handling by using images captured with the EAVS system. Using EAVS images for in-process quality assurance (IPQA) inspection would not only help reduce the physical handling of parts, but also improve slider cleanliness.
