Elsevier

Minerals Engineering

Volume 170, 15 August 2021, 107023
Minerals Engineering

Froth image feature engineering-based prediction method for concentrate ash content of coal flotation

https://doi.org/10.1016/j.mineng.2021.107023Get rights and content

Highlights

  • A systematic feature engineering of froth image was designed and developed.

  • The morphology and color space features have a higher correlation with ash content.

  • The fusion of multi-dimensional features of froth image can be used to predict ash content well.

  • This approach showed a good potential for coal flotation monitoring.

Abstract

Machine vision and machine learning have been researched widely in froth flotation and the technology continues to benefit from advances in computer technology. The feature engineering leads to a predicted resistance in the machine learning pipelines, especially in coal flotation. This study sought to examine the performance of feature engineering of coal flotation froth image on ash content prediction with an industrial dataset. In order to evaluate the practical use in industry, the morphoscopic (3 features), statistical (gray levels histogram (5 features), gray level co-occurrence matrix (24 features), statistical modeling (48 features)) and color spaces (18 features) are used to prepare feature engineering. Correlation matrix are used to investigate the relationship between features and ash content. The support vector regression is used to predict ash content. The evaluation of the model performance shows that the principal component analysis can effectively improve the accuracy. When the feature dimension is reduced to 14 by the principal component analysis, the optimal RMSE is 0.6331, the R2 value is 0.78. The feature engineering of coal flotation froth image in this paper can make a good prediction of the coal flotation concentrate ash content. Furthermore, the results can be used as the theoretical basis for the intelligent construction of flotation.

Introduction

Coal pulp flotation is an effective separation technology based on the differences in surface hydrophobicity between coal and gangue particles (Niu et al., 2018, Laskowski, 2001). In this process, the coal particles adhere to the bubbles and float to the flotation cells surface, which leads to the information of flotation conditions being reflected on the froth layer (Ozdemir et al., 2009, Hubbard. Arthur. , 2004). Unlike other minerals, the quality of coal flotation concentrate is expressed by ash content (Demir et al., 2008). The conventional method of ash content determination is based on the time-consuming processing of sampling, filtrating, drying, sample preparing and burning to ash in a muffle furnace. If flotation variables are adjusted according to this delayed result, the ash content of the flotation concentrate could not meet the production requirements. Hence, the operator still controls the flotation variables based on the froth visual features observed at the flotation cells surface. But the accuracy of this subjective control is still poor. In addition, there are some quick detection methods, such as γ-ray transmission (Yazdi and Esmaeilnia, 2003, Fookes et al., 1983), neutron analysis (Sibiya et al., 2014). But due to the high cost, harm to the operator and the environment, there are few applications in the flotation field.

Machine vision as an efficient, environmental technology has been rapidly applied in object recognition (Serban et al., 2020), object detection (Zhang et al., 2020) and medical image analysis (Zhang et al., 2020). With the assistance of machine vision technology, many systems and pipelines have been already or expectantly applied in the industrial field in recent years (Aldrich et al., 2010, Morar et al., 2012, Nakhaei et al., 2019). Of course, the combination of machine vision and mineral flotation has been paid more attentions (Aldrich et al., 2018). Copper flotation froth image recognition has been the most widely studied. The copper concentrations in flotation were predicted based on the flotation variables containing bubble size (Saghatoleslam et al., 2004, Hosseini et al., 2015, Navia et al., 2016, Jahedsaravani et al., 2017, Cheng et al., 2020), bubble rate (Morar et al., 2006), burst rate (Morar et al., 2006), color (RGB) (Jahedsaravani et al., 2014), stability (Jahedsaravani et al., 2017, Morar et al., 2006, Barbian et al., 2007), solids loading (Morar et al., 2012, Morar et al., 2006) and velocity (Jahedsaravani et al., 2016, Geladi, 2010). The machine vision in flotation is also used for other minerals like bauxite (Aldrich et al., 2010, Gui et al., 2013, Cao et al., 2013), sulfide (Bonifazi et al., 2001), zinc (Kaartinen et al., 2006), gold-antimony (Wang et al., 2018, Zhou et al., 2020) and iron (Wang et al., 2014).

In coal production, the ability to predict flotation outcomes is of great importance. At present, control of flotation includes regulation of the density of the flotation feed, the reagent consumption, the air consumption, and the pulp level in the flotation tank (Petukhov V N et al., 2019). The coal products must satisfy the requirements for various applications, the ash content is the most important variable of customer requirements. In addition, accurate control of ash content can improve the recovery of clean coal, prevent material waste (Tan et al., 2016). The ash content of flotation concentrate must be strictly controlled to maximize financial profitability. Generally, the ash content in the flotation concentrate was predicted based on the analysis of the bubble number (Hargrave et al., 1996), mean bubble diameter (Citir et al., 2004, Amankwah and Aldrich, 2014), average gray (Amankwah and Aldrich. 2014), variance, energy (Zhang et al., 2014), smoothness and entropy (Massinaei et al., 2019) using BP neural network. Tan et al. (2016) investigated the relationship between the froth features (average gray level, homogeneity, bubble burst) and the concentrate ash content in the laboratory. More recently, Massinaei et al. (2019) successfully analyzed relationships between the froth features and the operating conditions as well as the ash content in an industrial flotation column. However, these features are relatively single and in a lower dimension, and are not sufficient to systematically investigate the correlation between froth image features and the concentrate ash content.

Feature engineering is the process that integrates feature generation and feature selection aiming to transform the feature dimension to obtain the best prediction effect (Yang et al., 2013). For the flotation froth images, each feature space may comprise redundant, irrelevant and relevant features, reduce feature dimensions was important for extracting useful information in feature engineering (Zhang et al., 2016, Kambhatla and Leen, 1997). Therefore, systematically extracting the multi-dimension features of the froth image and selecting the appropriate number of feature main components is the necessary condition for accurate prediction of coal flotation concentrate ash content.

In this paper, the relationship between the froth image features and the ash content was explored systematically based on the dataset from industrial field. The multi-dimensional features of the froth image were mainly extracted from the feature engineering of morphoscopic, statistical and color spaces. Correlation matrix are used to investigate the relationship between features and ash content. Then the principal component analysis (PCA) algorithm is used for the reduction of feature dimension. Based on that, support vector regression (SVR) was used to train the model of the feature principal components data and ash content. This work can be used to control coal flotation variables and as a basis for further study.

Section snippets

Materials

In general, the hardware of machine vision consists of industrial camera, lens, light source, hood, computer, etc. (Nakhaei et al., 2019) In this study, the five megapixels color industrial camera (BALSER ace-2440-20gc) with charge-coupled device (CCD) sensors, prime lens (COMPUTER, 35 mm), diffuse bowl-shaped light-emitting diodes (LED), hood and computer (Windows 10) were used to capture froth images. The camera was fixed vertically by the bracket 50 cm above the froth layer surface in the

Morphoscopic features generation

The morphoscopic features were important variables describing the flotation performance. After the froth image was segmented by the watershed algorithm, the morphological features of the bubbles were calculated in pixels. The correlation results between morphoscopic features and ash content are shown in Fig. 5. The area (mean bubble area) and diameter (mean circumscribed circle diameter) increase with the ash content as a whole, and the correlation between area and ash content is strongest

Conclusions

The focus of this study was to investigate the feature engineering of the froth image for the accurate prediction of coal flotation concentrate ash content. From the perspective of image processing, the features of morphoscopic (3 features), statistical (gray levels histogram (5 features), gray level co-occurrence matrix (24 features), statistical modeling (48 features)) and color spaces (18 features) were studied systematically.

The correlation matrix shows that morphology has the strongest

CRediT authorship contribution statement

Zhiping Wen: Conceptualization, Methodology, Writing - original draft. Changchun Zhou: Supervision. Jinhe Pan: . Tiancheng Nie: . Ruibo Jia: . Fan Yang: .

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

Authors acknowledge the support of China Natural Science Foundation (No. 51974309) and China Scholarship Council (201806420065).

References (55)

  • A. Jahedsaravani et al.

    Development of a machine vision system for real-time monitoring and control of batch flotation process

    Int. J. Miner. Process.

    (2017)
  • A. Jahedsaravani et al.

    Prediction of the metallurgical performances of a batch flotation system by image analysis and neural networks

    Miner. Eng.

    (2014)
  • D. Navia et al.

    Real-time optimization for a laboratory-scale flotation column

    Comput. Chem. Eng.

    (2016)
  • C. Niu et al.

    Analysis of coal wettability by inverse gas chromatography and its guidance for coal flotation

    Fuel.

    (2018)
  • O. Ozdemir et al.

    Surface chemistry aspects of coal flotation in bore water

    Int. J. Miner. Process.

    (2009)
  • C. Palm

    Color texture classification by integrative co-occurrence matrices

    Pattern Recognit.

    (2004)
  • J.M. Prats-Montalbán et al.

    Multivariate image analysis: A review with applications

    Chem. Eng. J.

    (2011)
  • J. Tan et al.

    The concentrate ash content analysis of coal flotation based on froth images

    Miner. Eng.

    (2016)
  • Q. Wang et al.

    Feature selection in froth flotation for production condition recognition

    IFAC Papers on Line.

    (2018)
  • M. Yazdi et al.

    Dual-energy gamma-ray technique for quantitative measurement of coal ash in the Shahroud mine

    Iran. Int. J. Coal Geol.

    (2003)
  • J. Zhang et al.

    Recognition of flotation working conditions through froth image statistical modeling for performance monitoring

    Miner. Eng.

    (2016)
  • C. Aldrich et al.

    Multivariate image analysis of realgar–orpiment flotation froths

    Miner. Process Extr. Metall. Rev.

    (2018)
  • A. Amankwah et al.

    Automatic estimation of bubble size distributions in flotation froths by use of a mean shift algorithm and watershed transforms

    IEEE Trans. Geosci. Remote Sensing.

    (2014)
  • J. Arrospide et al.

    Log-Gabor filters for image-based vehicle verification

    IET Image Process.

    (2013)
  • L. Chen et al.

    Texture Analysis Using Gabor Filter Based on Transcranial Sonography Image

    Bildverarbeitung Für Die Medizin: Algorithmen-systeme-anwendungen. DBLP

    (2011)
  • Cheng. P., Garcia-Herreros. P., Lalpuria. M., Grossmann. I. 2020. Optimal scheduling of copper concentrate operations...
  • E Jorjani., H Asadollahi Poorali., A Sam., S. Chehreh Chelgani. Prediction of coal response to froth flotation based on...
  • Cited by (26)

    • Evaluation of coal screening performance by spatial image reconstruction method

      2023, Fuel
      Citation Excerpt :

      In recent years, the rapid development of technologies such as artificial intelligence, visual sensing, and image processing has pushed the coal processing industry toward the direction of intelligent automation [9–14]. With the maturing of the digital image processing technology, the use of machine-vision technology in the energy sector has become widespread [15–17]. In recent decades, many studies have employed image analysis techniques to predict the particle size composition of materials, and attained fruitful results [18–25].

    View all citing articles on Scopus
    View full text