OHT AI Fault Diagnosis Case: Minimizing Line Interruption Risks

Even with the same model, each unit had different characteristics,
making it impossible to cover all equipment with a single model

In a semiconductor manufacturing line, the OHT is core equipment that transfers wafers between processes, and a failure of just one unit can halt the entire production line. The problem was that even with the same model, the torque characteristics generated during operation varied from unit to unit, making it difficult to universally apply a single diagnostic model to all equipment. Even if a breakdown occurred, the exact cause—whether it was a wheel or a gear—could not be specified, leaving the person in charge in a situation where they had to sequentially inspect the entire equipment even after an anomaly was detected.

Application of a deep learning diagnostic model that self-calibrates
deviations between units and interprets the cause of failures

We developed and applied a deep learning diagnostic model that automatically calibrates operational deviations for each unit while classifying fault types. The model extracts fault-related features from torque signals and is designed so that field engineers can directly check which signals were used as the basis for the judgment. Through this, even if a new unit is added, the same model can be applied immediately without separate training.

Stable diagnosis and improved operational efficiency of OHTs
across the entire manufacturing line

Covering all equipment with a single model Regardless of differences in operational characteristics for each unit, the same diagnostic model can now be applied to all OHTs across the manufacturing line. The burden of maintaining individual models for each unit has disappeared. Direct interpretation of fault causes on-site Unlike the existing method that stopped at simply detecting anomalies, it is now possible to immediately check on-site down to the root cause level, determining whether it is a wheel defect or a gear defect. Focused inspection of the exact area is possible without unnecessary complete inspections. Improved equipment availability Early diagnosis and accurate cause identification have accelerated response speeds, and reduced the risk of unexpected line interruptions, thereby increasing the overall availability of the OHTs.