GeChuang Tianshu AI Replaces 75% of Manual Inspection at TCL CSOT (Suzhou) - 95% Accuracy, 1,000+ Industrial Scenarios

Title

GeChuang Tianshu AI Replaces 75% of Manual Inspection at TCL CSOT (Suzhou) - 95% Accuracy, 1,000+ Industrial Scenarios

Content

GeChuang Dongzhi, the industrial AI subsidiary of TCL, deployed its self-developed Tianshu AI ADC visual quality inspection system at TCL CSOT (Suzhou), replacing more than 75% of traditional manual inspection and reducing the traditional AOI equipment misjudgment rate of 60-70% to 95% accuracy, driving industrial quality inspection into a "super-intelligent era" across 1,000+ industrial scenarios. TCL CSOT (Suzhou) was awarded the "2025 Excellence-Level Intelligent Manufacturing Demonstration Factory" certification by China's MIIT for its "AI-driven new display panel intelligent factory." The 540,000-square-meter production facility features highly automated robotic arms and overhead transport rails, with the Xingzhi large model for product innovation and AI digital employees serving as the quality engineer's "eyes" and "brain." In the critical micrometer-level defect inspection workflow, AI digital employees operate through a closed-loop "ADC+ADR" dual-system: the ADC automatic defect classification system uses AI vision technology to identify and classify over 4,000 types of micrometer-level defects at 95% accuracy; the ADR automatic defect repair system drives high-precision tools to fully automatically repair repairable defects. The Tianshu AI platform encapsulates complex deep learning operations into ready-to-use productivity tools, delivering "millisecond-level reclassification" with significant efficiency improvements. From "human-wave reclassification" to "AI second-classification," traditional AOI equipment still has misjudgment rates above 60%, while the Tianshu AI system breaks through human eye limits with identification accuracy above 95%. Technical architecture spans three platforms: a training platform with built-in tens-of-millions-level defect feature library supporting low-code model iteration, an inference platform with lightweight edge computing boxes achieving millisecond-level single-image processing (5-10x faster than human eyes), and resource scheduling with dynamic GPU allocation supporting 100+ algorithm models running concurrently at TCL CSOT. Operational scale covers 1,000+ industrial scenario deployments, with the 540,000-square-meter TCL CSOT (Suzhou) factory as a flagship site. The system helped one semiconductor client reduce labor input by 90% and achieve defect detection rate above 99%, with the inference platform covering visual inspection imaging needs across 30+ industries. Looking ahead, GeChuang plans to deepen collaboration with the TCL Industrial Research Institute on complex defect pattern discovery and cross-domain knowledge transfer, scale to additional industrial scenarios, and expand presence in semiconductor, semiconductor display, and SMT industry segments.