AutoML with Focal Loss for Defect Diagnosis and Prognosis in Smart Manufacturing

In smart manufacturing, spotting defects early and predicting future problems are essential for keeping production yields high. Industrial sensor data sets—like the SECOM data—are hard to work with because they often have a strong class imbalance, many missing values, and require complex feature engineering. This overview looks at a domain‑specific AutoML framework called **AutoGBM** that tackles those issues. The approach combines XGBoost with Bayesian Optimization and adds a focal‑loss function to reduce the impact of the imbalanced classes. For data preprocessing, AutoGBM uses a hybrid feature‑learning step. First, it applies a quick filter to drop irrelevant features, then it runs recursive feature elimination to fine‑tune the remaining set. At the

same time, it selects the best interpolation method to fill in missing sensor readings. B.Tech students building smart‑manufacturing projects can use this framework as a step‑by‑step guide for automated predictive‑maintenance pipelines. By employing Bayesian optimization for hyper‑parameter tuning, the model can automatically adjust to changes on the production line. The implementation focus includes: – building a clean data pipeline, – setting up the focal‑loss function inside a gradient‑boosting model, and – measuring diagnostic accuracy. AutoGBM therefore offers a practical, adaptable way to create reliable defect‑diagnosis systems in modern factories.

Message: Hi FE, I need help with “AutoML with Focal Loss for Defect Diagnosis and Prognosis in Smart Manufacturing” in “Mechatronics Engineering”