Inspiration -> Lithium-ion batteries (LIB) have revolutionized consumer electronics. They are light-weight and have relatively high power density. This sudden spike in the use of LIBs has led to more mining of Li ores. Moreover, due to interest in electric vehicles, the demand for lithium is expected to further increase in the near future. However like mining of any other minerals, mining of Lithium has environmental, economic and social impacts. Initially the idea wasr to map the end-to-end supply chains for LIBs starting from extraction of Li from the ore to the manufacturing of the finished LIB. Recycling of LIBs is expected to reach about 9% of the total demand of LIBs by 2025. To summarize, the high demand for LIBs in the future calls for increased attention toward mining of Lithium. Usually, there are two methods of extracting Lithium. One method is to use the rock ore (pegmatite) and the second method is to use brine (salt-water).
What it does -> Our model compares the two methods of Li extraction. We have compared two of the largest Li mines in the world. The mine in Australia near Manjimup is based on the rock ore based extraction technique which is known to be economically less viable. Moreover it is also considered to be more resource intensive and hence has a larger footprint. The second mine uses brine as the raw material and is located in Chile. Brine water is laid in huge salt lakes and dried and treated continuously for many iterations to get the final product. Australia is currently the largest producer of Lithium but in the last 2 decades we have seen Chile catching up pretty quickly. Our analysis is based on quantifying the mining activity in these two mines. The idea is to correlate the extraction of Lithium with the corresponding environmental, economic and social impact.
How we built it -> We used Google Earth Engine (GEE) to download the satellite imagery for the last 10 years for analysis and all available data (since 1984) for creating the time lapse to understand how the mining activity has evolved over time. The next step was to take 10 years of imagery and apply image processing techniques to this collected imagery. We first tried to segment the mining activity by trying out some common algorithms in Python and using cv2 (Computer Vision Library) like Canny edge detector, Otsu’s thresholding etc. However, we saw that the results were not consistent throughout the time frame selected and also performance was different for the two datasets. Thus we extended the idea of localized gradient to color space and computed the pixel wise distance between two temporally separated images by converting them to a suitable (CIEDE) colorspace. This method was robust to cloud coverage and gave consistent results for both the datasets. Mining activity was correlated with the actual production of Li and the conversion factor of depth was estimated from the increase in area of the mines. We also leveraged other publicly available datasets to understand the social, economic and environmental effects related to mining in these two major mining sites globally.
Challenges we ran into -> The first challenge was to figure out which data source is to be used for obtaining the imagery. We also did not have a clear picture of how to get the other pieces of data to make the analysis holistic. However, we were able to figure out nice ways to make reasonable assumptions to fit publicly available data to our analysis. We also had a lot of trouble working with the multiband data as provided by a lot of the data partners for the hackathon. We eventually chose working with GEE imagery because we were interested in a wider time range which was not possible with other satellite image providers.
Accomplishments that we’re proud of -> Our model can provide real-time estimation of mining resource availability. In addition, we incorporated environmental input-output model to quantify the environmental impact of mining of these specific sites. We also incorporate historic and population projection that allow us to gain a comprehensive understanding of these sites. Although we had to make a few assumptions throughout the study, lot of these assumptions were either found to be eventually correct or we could find valid reasons for approximating the data. We were happy that we had set the bar high initially and whatever we could achieve in the given time. We interacted well with each other and worked as a team toward a common goal.
What we learned -> We learned a lot of tools and techniques to work with satellite imagery. We also learnt how to perform and end-to-end analysis on geospatial data and draw meaningful conclusions and insights from the analysis. We learnt a lot about the mining and processing of Lithium. . We also were able to advocate for the importance of environmental, economical and social impact for sustainable mining. We had fruitful discussions with other teams and mentors and made a lot of friends during this hackathon.
What’s next in Hacking for Sustainable Battery Supply Chains? -> We were fortunate enough to set the bar high initially as it gave us a lot of motivation to work continuously during the hackathon. Although we could not achieve all that was planned, we are off to a good start. We would like to continue working on this project thanks to support from providers of satellite imagery. We would in particular like to focus on building automatic mine detection using Convolutional Neural Networks and other Machine Learning techniques. We would also like to improve our estimation of mining activity and the measurement of the throughput of the extracted mineral such that it is evident from the map without making a lot of assumptions. We would also like to discuss about our idea with other experts in the fields of batteries, supply chain and mining at Stanford to understand what can be further improved. Last but not the least, we would like reiterate the message of 7Rs - Recycle, Refuse, Reduce, Reuse, Repair, Regift, Recover for a more sustainable tomorrow.
Log in or sign up for Devpost to join the conversation.