Laboratory

😀 Current Lab Members in MA201

Name: Jia-Hong Chou

Ename: Eric

PhD program - 2nd year student

Name: Samuel Trinata Winnyarto

Ename: Samuel

PhD program - 1st year student

Name: Fantanun Getie Tsegaw

Ename: Fantanun

PhD program - 1st year student

Name: Alfian Nur Hidayat

Ename: Alfian

MS program - 2nd year student

Name: Alebachew Mengistu Worku

Ename: Alex

PhD program - 1st year student

Name: Willam Gomez

Ename: Willam

Postdoctoral researcher

📝 Research Area

Remaining Useful Life Prediction

Remaining Useful Life (RUL) prediction is a critical task in predictive maintenance, involving the estimation of the time or usage remaining before a system or component fails or can no longer perform its intended function. By leveraging historical sensor data, operational conditions, and degradation trends, RUL prediction enables organizations to proactively schedule maintenance, minimize unexpected downtime, and optimize the equipment's lifecycle. This predictive capability is especially valuable in high-stakes industries such as aerospace, manufacturing, and energy, where unplanned failures can lead to significant safety risks and financial losses. Advanced techniques, including machine learning and deep learning, are increasingly employed to improve the accuracy and reliability of RUL predictions.

Domain Adaptation & Transfer Learning

Domain adaptation is a specialized area of machine learning that focuses on improving a model's performance when applied to a target domain that differs from the source domain on which it was initially trained. This technique is particularly important when labeled data in the target domain is scarce or unavailable, while ample labeled data exists in the source domain. By aligning the feature distributions between the source and target domains, often through methods such as adversarial training, feature transformation, or discrepancy minimization, domain adaptation allows models to generalize more effectively across different data distributions. It plays a crucial role in real-world applications where data collected under varying conditions or environments leads to domain shifts, such as in speech recognition, medical diagnosis, and industrial quality control.

Feature Fusion

Feature fusion is a machine learning technique that combines features from different sources or modalities to create a more informative and robust representation for a specific task. By integrating complementary information, such as visual, textual, or sensor data, it helps models better capture complex patterns. Common strategies include early, intermediate, and late fusion, which are particularly useful in applications such as multimodal learning, image classification, and fault diagnosis.

Defect Recognition

Wafer map pattern defect recognition is a crucial process in semiconductor manufacturing that involves identifying and classifying abnormal patterns on wafer maps to detect potential process issues or equipment failures. These patterns, which may include edge clusters, scratches, or center failures, often indicate systematic problems rather than random defects. By leveraging image processing techniques and machine learning algorithms, especially deep learning, this recognition process enables automated, accurate, and real-time analysis of wafer maps. This helps improve yield, reduce downtime, and support root cause analysis, making it an essential component of modern yield management and fault diagnosis systems in semiconductor fabs.

3D-based AI model for Dynamic Defect Detection

A 3D-based AI model for dynamic defect detection leverages spatial and temporal information to identify defects that occur over time, such as flickering or intermittent anomalies in displays or manufacturing processes. Unlike traditional 2D approaches that analyze static frames, 3D models, often built using architectures like 3D convolutional neural networks (3D CNNs), capture motion and sequential changes by processing sequences of image frames as volumetric data. This enables the model to detect subtle, transient defects that are challenging to identify in single-frame analysis. By incorporating the temporal dimension, 3D-based AI models significantly enhance detection accuracy and reliability in real-time applications such as quality inspection, video-based monitoring, and dynamic fault diagnosis..

Page updated

Google Sites

Report abuse