| Summary: | In manufacturing processes, sensors are often implemented to collect data and achieve defect detection. To establish correlation between collected data and quality measurements, machine learning methods are widely used. However, when a new process is developed, it often require sufficient data instances for training to build a model with high accuracy, which costs time and resources.
In this paper, transfer learning methods including TrAdaboost and Joint Domain Adaption (JDA) are used to establish correlation for a new manufacturing process. Data collected from previous processes can be utilized in transfer learning and therefore could save time and resources in data collection.
Also, process optimization will be achieved through the application of optimization algorithms. Traditional optimization method for manufacturing like Design of Experiment (DOE) would require the experiment to run multiple times to determine the optimal parameters. By using optimization algorithms like Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Differential Evolution (DE), we are able to time and resources and achieve in-process monitoring for process optimization.
|
|---|