Innovative Method Combines Spectroscopy and Machine Learning for Rapid Molybdenum Ore Grade Detection

The efficient determination of ore grade is crucial for optimizing the beneficiation process. However, existing methods for detecting molybdenum ore grade have been inadequate in keeping pace with beneficiation efforts. To address this, researchers from Northeastern University in China have introduced an innovative approach that combines visible-infrared spectroscopy and machine learning to achieve rapid and accurate molybdenum ore grade detection.

In their study, published in Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, the team collected spectral data from 128 molybdenum ore samples, leveraging visible-infrared spectroscopy (1). Partial least squares analysis was then employed to extract 13 latent variables from the 973 spectral features. To identify the non-linear relationship between the spectral signal and molybdenum content, the Durbin-Watson test and runs test were utilized to analyze the partial residual plots (1).

Recognizing the need for non-linear modeling due to the complex behavior of the spectral data, the researchers adopted Extreme Learning Machine (ELM) as an alternative to linear modeling methods (1). To optimize the ELM parameters effectively, they devised a novel approach called the Golden Jackal Optimization of adaptive T-distribution (MTSVD-TGJO-ELM). Additionally, to address ill-posed problems associated with ELM, they incorporated a modified truncated singular value decomposition (1).

The MTSVD-TGJO-ELM method is a novel approach developed for rapid detection of molybdenum ore grade; it combines several techniques to enhance accuracy (1). Firstly, the spectral data is processed using partial least square to extract latent variables. Then, the extreme learning machine (ELM) algorithm is employed for modeling the grade of molybdenum ores, considering their non-linear behavior (1). To optimize the ELM parameters, the Golden Jackal Optimization of adaptive T-distribution (TGJO) is utilized (1). Additionally, the method employs the modified truncated singular value decomposition (MTSVD) to decompose the ELM output matrix, addressing ill-posed problems (1). This integrated approach offers improved accuracy and provides a valuable tool for efficient ore grade determination in the mining industry (1).

Comparative analysis demonstrated that the proposed MTSVD-TGJO-ELM method outperformed other classical machine learning algorithms in terms of accuracy (1). By offering the highest level of precision, this novel approach presents a promising solution for the rapid detection of molybdenum ore grade, thereby facilitating the optimization of ore recovery rates in the mining industry (1).

The integration of visible-infrared spectroscopy and machine learning techniques holds significant potential for enhancing the efficiency of ore grade determination. With its ability to swiftly and accurately assess molybdenum ore grade, the proposed method paves the way for improved beneficiation processes, ultimately leading to higher ore recovery rates. This research marks a significant advancement in the field and showcases the power of innovative approaches in resource extraction and utilization.

Reference

(1) Xie, H.-f.; Mao, Z.-z.; Xiao, D.; Li, Z.-n. Rapid detection of molybdenum ore grade based on visible-infrared spectroscopy and MTSVD-TGJO-ELM. Spectrochimica Acta Part A: Mol. Biomol. Spectrosc. 2023, 298, 122789. DOI: 10.1016/j.saa.2023.122789