Quantification of paste homogeneity by vision-based identification method: Case study for an industrial mixer

Paste mixing is crucial in Cemented Paste Backfilling (CPB) to ensure both the strength and flowability of the backfill. Improper mixing can reduce strength and cause pipeline blockages. Properly mixing paste and cement is vital for enhancing backfill strength and paste flowability. However, existing methods lack real-time monitoring and intuitive usability. We propose a data-driven, non-contact system to evaluate paste homogeneity visually. A collaborative end-cloud device captures real-time images of the mixer tail, formalizing homogeneity evaluation as a semantic image segmentation task. Our method detects non-paste and non-homogeneous areas, defining a non-homogeneity factor as their proportion to the paste area. Using Gaussian process regression, the system predicts the factor’s probability and defines a homogeneity metric. Experiments at an industrial paste filling station show the system’s efficiency, accuracy, and alignment with subjective assessments. This method enables real-time monitoring of paste homogeneity, providing engineers with quantitative data to optimize the mixing process.