Elsevier

Journal of Cleaner Production

Volume 196, 20 September 2018, Pages 505-516
Journal of Cleaner Production

Data-driven modelling of the flocculation process on mineral processing tailings treatment

https://doi.org/10.1016/j.jclepro.2018.06.054Get rights and content

Highlights

  • A data-driven model was proposed to predict the flocculation of tailings.

  • This model combined gradient boosting machine and firefly algorithm.

  • Two studies were conducted with and without the effect of chemical characteristics.

  • Predictive performance was validated and relative importance was investigated.

  • Considering chemical characteristics of tailings may improve prediction accuracy.

Abstract

The clarification of tailings slurry using polymer flocculants has been widely used in the mining industry to promote the cleaner production of mineral resources. In this paper, a data-driven prediction model was proposed using gradient boosting machine (GBM) for the non-linear relationship modelling and firefly algorithm (FA) for GBM hyper-parameters tuning. Two studies were performed, among which the main study omitted the influence of chemical characteristics of mineral processing tailings (MPT) while the supplementary study considered. For the main study, 27 types of MPT and 4 types of anionic flocculants were used to prepare the dataset. The flocculation performance was represented by the initial settling rate (ISR) and its influencing variables were selected to be the particle size distribution (PSD) of MPT, the solids content of tailings slurry, the flocculants type, and the flocculants dosage. For the supplementary study, the chemical characteristics of 7 types of MPT were also considered as influencing variables and its influence on the predictive performance of GBM was investigated. The main study shows that the optimum GBM model achieved a correlation coefficient of 0.841 between the predicted and experimental ISR values on the testing set, denoting it was robust in predicting the ISR of the flocculation. Compared with the solids content, the flocculants dosage and the flocculants type, the PSD of MPT was found to be the most significant influencing variable for the flocculation with an importance score of 0.420 out of 1. The supplementary study shows that the predictive performance of GBM could be improved considering chemical compositions of MPT, which were also important influencing variables for the flocculation process.

Introduction

Mineral processing tailings (MPT) are a major source of contamination to the environment, especially in global arid and semiarid regions (Valentín-Vargas et al., 2018). They are inevitable by-products of the hardrock mining that often in a slurry form with a mixture of fine particles and water. According to (Sun et al., 2018; Xie et al., 2009), more than 25 billion tons of MPT have been produced in China, resulting in approximately 12,000 tailings dams. Similar statistics illustrating the poor disposal of MPT can be observed in Australia, Brazil, southern North America and Canada (Cross and Lambers, 2017; Lu et al., 2016). These surface-disposed tailings destroy mining land resources and pose serious environmental concerns, severely limiting the cleaner production of the mining industry. Recycling MPT as cemented paste backfill (CPB) has been proven to be an ideal option for the safe and environmental disposal of MPT (Chen et al., 2017, 2018; Edraki et al., 2014; Kesimal et al., 2005; Liu et al., 2018; Lu et al., 2018; Qi et al., 2018a, 2018b, 2018c; Yilmaz et al., 2009; Yılmaz and Ercikdi, 2016).

CPB is an engineered composite material produced with MPT, hydraulic binders and mixing water (Fig. 1). Before CPB preparation, the tailings slurry from the milling or processing operation needs to be dewatered as it contains dispersed particles that hinder the consolidation and strength gain of CPB (Bussière, 2007). Dewatering process is important and complementary during recycling MPT as CPB, which can provide at least two environmental benefits: (1) recycle water by settling the colloidal particles and (2) recycle dewatered MPT as CPB to underground stopes. Consequently, many solid-liquid separation techniques have been proposed, among which settling the fine tailings using the polymer flocculants has been widely used in mining practice (Lee et al., 2012).

The effectiveness of flocculation process depends on various factors (e.g. the flocculants type and the tailings composition) that involves different destabilization mechanisms (Hogg, 2000; Hogg et al., 1994; O'Shea et al., 2011). Significant progresses have been achieved in the literature through numerous studies related to the flocculation of MPT (Grabsch et al., 2013; Ji et al., 2013; Nasser and James, 2007; Reis et al., 2016; Salam et al., 2016; Taylor et al., 2002; Thompson et al., 2017). For example, Lu et al. (2016) proposed a two-step flocculation process using two oppositely charged polymers for the oil sands tailings treatment. Efficient and cost-effective flocculation relies on the determination of appropriate polymers and their dosage for a specific type of MPT (Grabsch et al., 2013; Salam et al., 2016). Thus, lots of flocculation experiments need to be performed for each type of MPT, which are often cumbersome and costly. Furthermore, the full potential of experimental results is rarely realised as data-driven techniques that can deal with large quantities of data are lacking. Therefore, a prediction model for the flocculation process on MPT treatment is desirable for polymers selection and dosage optimisation.

Nowadays, machine learning (ML) algorithms have been widely used to model the relationship between inputs and outputs in many fields of cleaner production, such as simulating landfill leachate (Bagheri et al., 2017) and investigating solar degradation (Bararpour et al., 2018). To the best of the authors’ knowledge, there is no ML-based prediction model for the flocculation process on MPT treatment in the literature. Among all widely used ML algorithms, gradient boosting machine (GBM) offers different advantages that makes it an appealing technique for the flocculation process modelling: (1) it is an ensemble method that combines a large number of single tree models to improve the predictive performance (P Anderson et al., 2006), (2) it is relatively robust to outliers, (3) it incorporates interaction between influencing variables. Moreover, the GBM algorithm has been proven to have better predictive performance compared with other ML algorithms on several datasets in the literature (Qi et al., 2017; Qi and Tang, 2018; Roe et al., 2005). As the successful application of GBM relies on the careful tuning of its hyper-parameters, firefly algorithm (FA) can be used for an efficient hyper-parameters tuning of the GBM algorithm.

The current study aims to propose a data-driven method, the FA-GBM, for modelling the flocculation process on MPT treatment based on GBM and FA. In the main study, 27 types of MPT and 4 types of polymer flocculants were used for the dataset preparation. The initial settling rate (ISR) was selected as the output variable. The particle size distribution (PSD) of MPT, the solids content of tailings slurry, the flocculants type, and the flocculants dosage were selected as the input variables. The performance of regression models was measured by the correlation coefficient on both the training set and the testing set. In the supplementary study, the chemical characteristics of 7 types of MPT were also determined and considered as the influencing variables for the flocculation process. The rest of this paper is organised as follows. Section 2 presents the materials and methods used for flocculation modelling. Section 3 details the results and discussion and Section 4 describes the conclusions.

Section snippets

Mineral processing tailings

The MPT used in the main study were taken from 27 mines in China. For each sampling site, the MPT were obtained from the processing plants and homogenized on-site. The representative MPT samples were transported to the Analysis and Test Centre of Central South University, China. Experiments were immediately performed once each type of MPT was received in order to preserve its physical and chemical properties.

The particle size distribution (PSD) of MPT was determined using the wet sieving method

Results of wet sieving, XRD and settling tests

Table 2 shows the PSD of MPT from 27 mines in China. As we can see, the PSD of different MPT showed huge variance. For example, 53.37% of the T1 was smaller than 37 μmwhile only 8.29% of the T2 was smaller than 37μm. Similar findings could also be seen from other categories (i.e. the comparison of T7 and T16 in the >150 μmcategory). This is also the objective during the MPT selection as the generalisation capability of the prediction model can be improved when different types of MPT were

Conclusions

A data-driven prediction model, the FA-GBM, was proposed for the flocculation process of MPT using polymer flocculants. The model was composed of GBM and FA, in which GBM was used for non-linear relationship modelling and FA was used for GBM hyper-parameters tuning. Flocculation tests have been conducted on 27 types of MPT and 4 types of anionic flocculants to provide the main dataset and the results from 7 types of MPT were used to prepare the supplementary dataset. The ISR was selected to be

Conflicts of interest

None.

Acknowledgement

This study was financially supported by the 13th Five Years Key Programs for Science and Technology Development of China (No. 2017YFC0602900) and the Fundamental Research Funds for the Central Universities of Central South University (No. 2016zzts092). The first author was supported by China Scholarship Council (No. 201606420046).

References (64)

  • M. Edraki et al.

    Designing mine tailings for better environmental, social and economic outcomes: a review of alternative approaches

    J. Clean. Prod.

    (2014)
  • A. Golbraikh et al.

    Beware of q2!

    J. Mol. Graph. Model.

    (2002)
  • A.F. Grabsch et al.

    The impact of achieving a higher aggregate density on polymer-bridging flocculation

    Int. J. Miner. Process.

    (2013)
  • R. Hogg

    Flocculation and dewatering

    Int. J. Miner. Process.

    (2000)
  • Y. Ji et al.

    Effect of solution salinity on settling of mineral tailings by polymer flocculants

    Colloid. Surface. Physicochem. Eng. Aspect.

    (2013)
  • A. Kesimal et al.

    Effect of properties of tailings and binder on the short-and long-term strength and stability of cemented paste backfill

    Mater. Lett.

    (2005)
  • C. Kim et al.

    Particle-size dependence on metal (loid) distributions in mine wastes: implications for water contamination and human exposure

    Appl. Geochem.

    (2011)
  • C.E. Lee et al.

    Structural recovery behaviour of kaolin, bentonite and K-montmorillonite slurries

    Powder Technol.

    (2012)
  • X. Li et al.

    Numerical simulation and experimental verification of particle coagulation dynamics for a pulsed input

    J. Colloid Interface Sci.

    (2003)
  • L. Liu et al.

    Experimental investigation on the relationship between pore characteristics and unconfined compressive strength of cemented paste backfill

    Construct. Build. Mater.

    (2018)
  • Q. Lu et al.

    A two-step flocculation process on oil sands tailings treatment using oppositely charged polymer flocculants

    Sci. Total Environ.

    (2016)
  • H. Lu et al.

    A new procedure for recycling waste tailings as cemented paste backfill to underground stopes and open pits

    J. Clean. Prod.

    (2018)
  • M. Nasser et al.

    Effect of polyacrylamide polymers on floc size and rheological behaviour of kaolinite suspensions

    Colloid. Surface. Physicochem. Eng. Aspect.

    (2007)
  • J.-P. O'Shea et al.

    Temperature responsive flocculation and solid–liquid separations with charged random copolymers of poly (N-isopropyl acrylamide)

    J. Colloid Interface Sci.

    (2011)
  • C. Qi et al.

    Slope stability prediction using integrated metaheuristic and machine learning approaches: a comparative study

    Comput. Ind. Eng.

    (2018)
  • C. Qi et al.

    An intelligent modelling framework for mechanical properties of cemented paste backfill

    Miner. Eng.

    (2018)
  • C. Qi et al.

    Neural network and particle swarm optimization for predicting the unconfined compressive strength of cemented paste backfill

    Construct. Build. Mater.

    (2018)
  • C. Qi et al.

    A strength prediction model using artificial intelligence for recycling waste tailings as cemented paste backfill

    J. Clean. Prod.

    (2018)
  • C. Qi et al.

    Pressure drop in pipe flow of cemented paste backfill: experimental and modeling study

    Powder Technol.

    (2018)
  • L.G. Reis et al.

    Using acrylamide/propylene oxide copolymers to dewater and densify mature fine tailings

    Miner. Eng.

    (2016)
  • B.P. Roe et al.

    Boosted decision trees as an alternative to artificial neural networks for particle identification. Nuclear Instruments and Methods in Physics Research Section A: accelerators, Spectrometers

    Detect. Assoc. Equip.

    (2005)
  • A.M. Salam et al.

    Determination of the optimum polymer dose for dewatering of oil sands tailings using UV–vis spectrophotometry

    J. Petrol. Sci. Eng.

    (2016)
  • Cited by (55)

    • Recovery of strategically important critical minerals from mine tailings

      2022, Journal of Environmental Chemical Engineering
    • The impacts of high salinity and polymer properties on dewatering and structural characteristics of flocculated high-solids tailings

      2022, Journal of Cleaner Production
      Citation Excerpt :

      Tailings are generally comprised of water and complex gangue remaining after the target mineral is extracted, including various clays that can be difficult to treat (Reid et al., 2009; Sabah and Cengiz, 2004). Among several techniques available for treating fine-particle tailings, bridging flocculation via the addition of water-soluble polymeric flocculants in gravity thickeners is an effective and well-established process which achieves solid-liquid separation by increasing the settling rate of individual particles by forming larger aggregated structures (Arjmand et al., 2019; Fellows and Doherty, 2005; Grabsch et al., 2013; Qi et al., 2018). The subsequent injection of flocculant solutions at high dosages (significantly greater than applied in thickener feedwells) to high solids tailings streams can result in additional aggregation and water release on deposition compared to conventional low-solids thickening alone (Anstey and Guang, 2017; Fawell et al., 2015; Li et al., 2021; Wells et al., 2015).

    • A data-driven optimization model for coagulant dosage decision in industrial wastewater treatment

      2021, Computers and Chemical Engineering
      Citation Excerpt :

      In that, they presented an inductive workflow to combine the statistical design of experiments and predictive modeling, and to demonstrate its effectiveness in the development of anionic polymeric flocculants for the treatment of a real effluent from the potato crisps manufacturing industry. Likewise, by using gradient boosting machine (GBM) and firefly algorithm to model the non-linear relationship, Qi et al. (2018) proposed a data-driven prediction for GBM hyper-parameters tuning where the polymer is its flocculants. In other previous studies, the use of evolutionary algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO) to solve complex optimization problems in the field of water resources engineering has greatly increased, which is proved to be flexible and potential tools for solving several complex water resources problems (Reddy & Kumar, 2020).

    • Utilization of low-alkalinity binders in cemented paste backfill from sulphide-rich mine tailings

      2021, Construction and Building Materials
      Citation Excerpt :

      In the last two decades, cemented paste backfills (CPB) had become a promising alternative for the safe and environmentally responsible disposal of tailings [1–4]. CBP was a mixture of dewatered tailings with a solid percentage between 70% and 85%, a hydraulic binder (often ordinary Portland cement, OPC), and water (fresh and/or mine processed) [5,6]. The fresh CPB mixture should be pumpable fluid (workability index) to enable efficient pumping/delivery from the surface paste-fill plant to the underground mined-out openings [7].

    View all citing articles on Scopus
    View full text