The imbalance of abnormal data in the manufacturing process remains a challenge for applying deep learning methods in the quality control domain. The existing shallow machine learning methods for quality control of mu...
详细信息
The imbalance of abnormal data in the manufacturing process remains a challenge for applying deep learning methods in the quality control domain. The existing shallow machine learning methods for quality control of multivariate nonlinear manufacturing process have some problems, such as low monitoring efficiency and poor real-time performance. This paper introduces a novel deep learning approach for online quality control of multiple time-series processes with imbalanced data. The original process data is initially transformed into grayscale quality images to address imbalanced data and visualize the status of abnormal processes. Next, a convolutional block attention module (CBAM) is integrated into the adversarial classifier generative adversarial network (ACGAN) to tackle the issue of imbalanced data by augmenting the minority abnormal images. By utilizing a balanced dataset, a SA-MCNN recognition model is established, which consists of a multi-scale convolutional neural network (MCNN) with self-attention(SA) module to extract multi-scale features from multiple time series processes. Subsequently, a comprehensive quality control framework is proposed for online monitoring of multiple time series processes with imbalanced data and diagnosing root causes in the manufacturing process. An empirical study is conducted using data from an actual pasting process to validate the augmented images, assess the discrimination ability of the SA-MCNN model, evaluate the performance enhancement of the attention module, and measure the effectiveness of online quality control. The results of the study demonstrate that the proposed method effectively monitors abnormal processes with high accuracy and outperforms other imbalanced time series classification methods.
暂无评论