This challenge aims to apply knowledge learnt from artificial intelligence in a real world application. The injection molding machine is used to produce a transparent mold where the goal of this challenge is to build a neural network classifier to monitor the quality of transparent mold. The challenge of this problem lies in its non-stationary characteristic because two machine parameters, namely holding pressure and injection speed, are varied during the process runs.
The dataset file contains part of a real-world manufacturing dataset collected from the injection modling dataset
Task 1 utilizes SGD optimzer under mini-batch update to result in the smallest testing errors
Task 2 analyzes effect of network structures based on the proposed algorithm
Task 3 identifies the effects of learning rates as well as providing a plausible adaptive learning rate for the best set up
Task 4 studies the effect of batch sizes on the proposed algorithm