HanWool Semiconductor的AI系统每分钟检测13,000个MLCC
HanWool Semiconductor于9月1日宣布，其基于人工智能的多层陶瓷电容器（MLCC）视觉检测设备实现了每分钟13,000个的检测速度。该设备用于MLCC生产中的质量检测，其高速检测能力有助于提升MLCC供应链的检测效率。
原文深度解读
本文报道了韩国半导体设备公司HanWool Semiconductor于9月1日宣布其基于人工智能的MLCC视觉检测设备实现每分钟13,000个的检测速度，缺陷检测准确率约98%。该设备用于检测超小型MLCC的外观缺陷，如微裂纹、碎裂、异物和电极异常。文章涉及HanWool Semiconductor、MLCC制造商、外部专业验证机构等主体，关键数字包括检测速度13,000个/分钟、准确率98%（较之前95%提升3个百分点）、MLCC尺寸0.6mm×0.3mm、年底目标15,000个/分钟。该设备与电子元器件供应链相关，因为MLCC是电子元器件的重要类型，其检测效率直接影响MLCC生产质量和供应链效率。
- HanWool Semiconductor于9月1日宣布其AI视觉检测设备实现每分钟13,000个MLCC的检测速度，缺陷检测准确率约98%。
- 该设备为六面视觉检测系统，可检查MLCC整个外观，检测微裂纹、碎裂、异物和电极异常等缺陷。
- 系统将缺陷检测准确率从约95%提升了3个百分点至约98%。
- 系统使用HanWool Semiconductor专有的AI检测平台HawAIe和AI分析引擎AlohaNet，结合传统图像处理技术和AI分析。
- 检测过程中收集的数据用于训练AI，系统根据每个MLCC制造商的产品规格和缺陷类型提高检测准确性，并采用设备端AI技术，在设备内部进行AI计算，无需外部网络。
- 公司计划年底将检测速度提升至每分钟15,000个，并扩大HawAIe的应用范围，包括MLCC集成测量系统、压痕检测设备和超声波无损检测设备。
供应链影响
- 该设备的高速检测能力可能提升MLCC生产线的质量检测效率，从而可能缩短MLCC的交付周期，具体取决于设备在MLCC制造商中的采用率。
- 设备准确率的提升可能减少MLCC缺陷品流入下游供应链的风险，从而可能降低终端电子产品的故障率，但实际影响取决于设备在生产线上的部署情况。
- 公司计划将AI检测技术扩展到其他检测设备，可能影响MLCC生产设备供应链的升级需求，具体取决于MLCC制造商的投资决策。
- 外部验证和销售扩展可能影响MLCC检测设备市场的竞争格局，但市场变化取决于验证结果和客户接受度。
- 设备端AI技术可能降低数据泄露风险，可能增强MLCC制造商对设备采购的意愿，但取决于制造商对数据安全的重视程度。
系统解读，仅作为判断线索，不构成备货、出货、涨价或投资建议。
(照片：HanWool Semiconductor)
HanWool Semiconductor于9月1日表示，其基于人工智能的多层陶瓷电容器（MLCC）视觉检测设备已实现每分钟13,000个的检测速度。缺陷检测准确率约为98%。该公司正在通过外部专业组织验证该设备的性能。
该设备是一种六面视觉检测系统，能够检查MLCC的整个外观。它在高速移动中拍摄尺寸约为0.6毫米×0.3毫米的超小型MLCC，检测微裂纹、碎裂、异物和电极异常等缺陷。该系统将缺陷检测准确率从约95%提高了3个百分点。
由于MLCC体积小且大批量生产，检测速度和检测准确率必须同时实现。随着检测速度的提高，图像捕获和处理的时间减少，需要能够减少漏检和误报的技术。
该系统使用HanWool Semiconductor专有的AI检测平台HawAIe和AI分析引擎AlohaNet。对于位置和形状一致的缺陷，使用传统图像处理技术进行检测，而AI则分析更困难的区域，如微裂纹和不规则表面缺陷。
检测过程中收集的数据用于训练AI。该系统根据每个MLCC制造商的产品规格和缺陷类型提高检测准确性。它还采用了设备端AI技术，在设备内部执行AI计算，使生产数据可以在本地处理，而无需通过外部网络。
HanWool Semiconductor正在通过外部专业组织进行性能测量，以验证每分钟13,000个的检测速率。基于结果，公司计划扩大针对韩国和海外MLCC制造商的设备评估和销售工作。
公司计划在今年年底将检测速度提高到每分钟15,000个，并扩大HawAIe的使用范围。它将逐步将AI检测技术应用于MLCC集成测量系统、压痕检测设备和超声波无损检测设备。
“我们已将检测速度提高到每分钟13,000个，同时提高了基于AI的缺陷检测性能，”HanWool Semiconductor的一位官员表示。“我们计划通过外部专业评估客观验证性能，并建立设备的可靠性。”
(Photo: HanWool Semiconductor)
HanWool Semiconductor said on Sept. 1 that its artificial intelligence-based visual inspection equipment for multilayer ceramic capacitors (MLCCs) has achieved an inspection speed of 13,000 units per minute. Defect-detection accuracy stands at about 98%. The company is having the equipment's performance verified by an external specialist organization.
The equipment is a six-sided visual inspection system capable of examining the entire exterior of an MLCC. It photographs ultra-small MLCCs measuring about 0.6 millimeter by 0.3 millimeter as they move at high speed, detecting defects including micro-cracks, chipping, foreign substances and electrode abnormalities. The system improved defect-detection accuracy by 3 percentage points from about 95%.
Because MLCCs are small and mass-produced, inspection speed and detection accuracy must be achieved simultaneously. As inspection speeds increase, the time available for image capture and processing decreases, requiring technology that can reduce both missed defects and false positives.
The system uses HanWool Semiconductor's proprietary AI inspection platform, HawAIe, and AI analysis engine, AlohaNet. Defects with consistent locations and shapes are inspected using conventional image-processing technology, while AI analyzes more difficult areas such as micro-cracks and irregular surface defects.
Data collected during inspections is used to train the AI. The system improves detection accuracy based on each MLCC manufacturer's product specifications and defect types. It also incorporates on-device AI technology that performs AI computations within the equipment, allowing production data to be processed locally without passing through an external network.
HanWool Semiconductor is conducting performance measurements through an external specialist organization to verify the inspection rate of 13,000 units per minute. Based on the results, the company plans to expand equipment evaluations and sales efforts targeting MLCC manufacturers in Korea and overseas.
The company plans to increase inspection speed to 15,000 units per minute by the end of this year and broaden the use of HawAIe. It will sequentially apply its AI inspection technology to MLCC integrated measurement systems, indentation inspection equipment and ultrasonic non-destructive inspection equipment.
"We have increased inspection speed to 13,000 units per minute while also improving AI-based defect detection performance," a HanWool Semiconductor official said. "We plan to objectively verify the performance through an external specialist evaluation and establish the reliability of the equipment."