SYSTEM FOR PREDICTING PRODUCT FAILURE IN PROCESS AND A METHOD FOR GENERATING LEARNING MODEL FOR FAILURE PREDICTION

摘要

Disclosed in-process product failure prediction system and learning model generation method. A system for predicting product defects during a process according to an embodiment comprises: a data collection module that produces and collects measurement information, which is process-specific metadata including voltage, current, and accumulated gas amount; Defects using pattern learning through deep learning algorithms including MLP (Multilayer Perceptron), RNN (Recurrent Neural Network), DNN (Deep Neural Network), and ANN (Artificial Neural Network) by accumulating and storing the collected measurement information for each process. A learning engine that generates a predictive learning model; A probability calculation module for identifying detailed quality information including cracks, pores, slag mixing, and fusion status of a product, and calculating a defect probability for each process of the product, an error range of the defect probability, and an error rate by using a failure prediction learning model; And if the calculated defect probability exceeds the reference value, a process control command including discontinuation of subsequent process entry and manager notification is generated, and products with a defect probability equal to or greater than the reference value are set as a provisional defect determination model, and a provisional defect determination model A process control module for grasping detailed information including the model name and process process of; Includes.