Machine Learning in Industrial Quality Control of Glass Bottle Prints

TL;DR

Two ML-based approaches for quality control of bottle prints were developed and evaluated, which can also be used in this challenging scenario as numerous factors can negatively affect the printing process.

Abstract

In industrial manufacturing of glass bottles, quality control of bottle prints is necessary as numerous factors can negatively affect the printing process. Even minor defects in the bottle prints must be detected despite reflections in the glass or manufacturing-related deviations. In cooperation with our medium-sized industrial partner, two ML-based approaches for quality control of these bottle prints were developed and evaluated, which can also be used in this challenging scenario. Our first approach utilized different filters to supress reflections (e.g. Sobel or Canny) and image quality metrics for image comparison (e.g. MSE or SSIM) as features for different supervised classification models (e.g. SVM or k-Neighbors), which resulted in an accuracy of 84%. The images were aligned based on the ORB algorithm, which allowed us to estimate the rotations of the prints, which may serve as an indicator for anomalies in the manufacturing process. In our second approach, we fine-tuned different pre-trained CNN models (e.g. ResNet or VGG) for binary classification, which resulted in an accuracy of 87%. Utilizing Grad-Cam on our fine-tuned ResNet-34, we were able to localize and visualize frequently defective bottle print regions. This method allowed us to provide insights that could be used to optimize the actual manufacturing process. This paper also describes our general approach and the challenges we encountered in practice with data collection during ongoing production, unsupervised preselection, and labeling.

Figures (7)

• Figure 1: Cropped images of glass bottle prints: (left) acceptable print, (middle) unacceptable smeared print, and (right) unacceptable rotated print.
• Figure 2: Photography setup: (left) system overview and (right) a captured image. The camera and lighting remain stationary during the capture, and the bottle is rotated.
• Figure 3: Cropped images: (left) image without filter, (middle) image with Sobel, and (right) image with Canny. Reflections and the luminous background are reduced by applying these filters.
• Figure 4: Visualization of our two approaches: (AP1) ORB, Filters, IQMs, & Classifiers and (AP2) Transfer Learning with CNN Models. Some methods from our approaches can be used beyond binary classification (Insights).
• Figure 5: ROC Curves of our most accurate models: SVM (AP1) and VGG-11 (AP2).