SBIR/STTR Award attributes
As smart manufacturing has evolved more machines are equipped with smart sensors and meshed with Internet-of-Things. As manufacturing process becomes more complex, more difficulty comes along to clear the data and formulate the right problems to model. Fast reliable industrial inspection is key challenge in manufacturing scenarios. Surface inspection is usually inspected employing machine vision and image processing techniques to detect surface defect for enhanced product quality. However, flexible configuration in modern manufacturing system can shift production from one product to another quickly. Therefore, manually designed features in traditional machine learning technique may lead to insufficient or unsatisfactory inspection performance in complex surface scenarios or dynamic changing process. Addressing these challenges Analatom proposes using deep learning to learn high-level generic features that apply to wide range of textures or difficult-to-detect defects cases. Transfer-learning can extract the knowledge from one source task and then applies the learned knowledge to different but related task. It employs pre-trained deep learning model from relevant task for model initialization and fine-tuning enabling knowledge reuse and updating as transferred deep learning. This allows building inspection systems for novel manufacturing inspection tasks and transitioning to inspect new asset types rapidly and with little skilled oversight from user.
