Founded in 2008, LLK is a solutions and development company that has brought technological innovation since the beginning of its history, involving the most diverse fields of engineering. Specialized in computer vision, monitoring and protection of equipment and processes, it develops specific solutions in hardware and software in order to meet the demands of the plant, process, management and teams. We have already developed several technologies in partnership with major mining companies such as Vale, Samarco, YamanaGold, Kinross, Gerdau and Usiminas.

Challenge

Vale needs to optimize the sampling process, reducing the number of samples that are sent to the laboratory daily with an automatic measurement process, in the field and in real time, which allows for more efficient decision making. Today, a sample is collected in the field and usually has a response time of 2 hours, on average, by the laboratory. The time it takes to obtain the information does not meet the needs of the plant, which during the analysis may have processed tons of ore outside the desired specifications.

Solution

LLK presents a computer vision solution, based on spectrometry, capable of quantifying the chemical elements that make up the material on a conveyor belt. The tested hypothesis is the mapping of iron in ore samples using hyperspectral sensors with excellent degree of reliability and in real time, obtaining calibration models of the iron content present in the sample. LLK developed during the POC:
Analysis of the spectral signatures of iron ore samples in different particle sizes and grades;
Acquisition of hyperspectral images in real time for analysis; Segmentation algorithm and signature characterization.
Algorithm capable of analyzing the spectral signature of the material and returning the iron content.

Results

Vale provided a total of 47 samples for the development of the solution and analysis of results. Spectral readings were acquired with a spectroradiometer and hyperspectral camera that collects a spectral range of 350-2,500nm. Calibration and model creation processes were also performed at POC, involving pre-processing of measurement data for radiometric and step corrections from the sensors. LLK uses machine learning algorithms to process the measured data in order to obtain the results of iron content in the samples. The model developed by LLK had an error of 3.2% in the analysis of the iron content in ore samples, which is considered a good result considering the amount of samples available and the high learning capacity of the method when applied in the field and with feedback analysis.

The system, when implemented in the field in the final post-POC project, will have very positive impacts that will generate quality, safety and financial returns for the mining company:

Reduction of more than 2 hours in the response time of the chemical composition of iron and silica of the material;

Elimination of human and manual influence in the analysis results;

Possibility of using the laboratory Reduction of production losses caused by contamination of piles with ore outside the desired specification; 100% of data automatically integrated with plant software.