SBIR Phase I: Artificially Intelligent Solution to Maximize Value Creation and Upcycling Potential of Aluminum Scrap

这个小企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是增加美国废料回收商的收入和处理潜力。在这个项目中设计的人工智能算法将使国内废料加工商能够实时适应不断变化的材料消费环境，从而在材料供应链中变得更具竞争力。在国际大宗商品贸易不稳定的情况下，美国废料加工商必须减少对出口低价值废料的依赖，以维持盈利的商业模式。此外，美国废品必须执行最佳的处理时间表，以防止国内废品过剩，同时向消费者提供功能等同于新产品的回收产品。后者提供了一种环保的废料转化为产品的选择，减少了生产所需的能源和有害二氧化碳的排放量。铝废料回收已经实行了几十年；然而，大多数消费后的废料被降级回收，留下了未开发的收入和环境效益。有色金属自动碎料的平均售价为0.33美元/磅。低于2017年的实际价值，相当于近10亿美元的机会成本。人工智能分拣系统将使废料处理商获得更高的利润率，并达到环保目标。拟议的项目将完全自动化废料分类。多分类能力的出现鼓励了初步研究的需要，以确定如何操作基于传感器的分类器的最佳。人工智能可以满足这一需求。该项目的智力优势在于开发了一种能够实时优化废料分类的人工智能算法。研究目标包括：&amp；#8239；(1)识别废品回收过程中所有符合&amp；#8239的数据源；可以,# 8239;影响智能决策(2)&#8239；设计一个数据库来托管已识别的数据源，如成分、市场、库存和销售数据；(3)设计一个定制的人工智能算法；废,# 8239;回收行业实时制定分类标准。为了实现这些目标，将设计一个动态物质流模型来分析和托管所有相关的传感器、市场和实验数据流，以最大限度地减少数据预处理要求。同时，将对铝废料进行表征，以调查成分波动的频率和程度。承载所有支持性数据流的数据库将被设计为无缝地存储和集成所有相关数据流。最后，使用80/20的训练/测试数据分割和5倍交叉验证，将选择和优化机器学习算法，为建议的排序标准提供最低的错误率。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is increased revenue and processing potential for scrap recyclers in the US. The artificially-intelligent algorithm designed in this project will enable domestic scrap processors to become more competitive within the material supply chain by giving them the ability to adapt, in real-time, to an ever-changing material consumption climate. With unstable international commodity trade, US scrap processors must reduce reliance on exporting low-value scrap to maintain profitable business models. Additionally, US scrap must exercise optimal processing schedules to prevent scrap surplus domestically while providing consumers with recycled products that are functionally equivalent to new products. The latter offers an environmentally-friendly scrap-to-product option that reduces the energy required for production and the amount of harmful CO2 released. Aluminum scrap recycling has been practiced for decades; however, the majority of post-consumer scrap is downcycled leaving revenue and environmental benefits untapped. Non-ferrous auto-shred was, on average, sold for $0.33/lb. less than its actual value in 2017, which equates to nearly $1 billion in opportunity cost. Artificially intelligent sorting systems will enable scrap processors to reach higher profit margins and meet environmental goals.The proposed project will completely automate scrap sortation. The advent of multi-sort capability encourages the need for preliminary research to identify how to operate sensor-based sorters optimally. Artificial intelligence can meet this need. The intellectual merit of this project is the development of an artificially-intelligent algorithm that is capable of optimizing scrap sortation in real-time. The research objectives include: (1) identify all data sources in the scrap recycling process that  can  influence intelligent decision making (2) design a database to host identified data sources such as compositional, market, inventory, and sales data and (3) design a customized artificially-intelligent algorithm for the  scrap  recycling industry to develop sorting criteria in real-time. To meet these objectives, a dynamic material flow model will be designed to analyze and host all relevant sensor, market, and experimental data streams to minimize data pre-processing requirements. Concurrently, aluminum scrap will be characterized to investigate how frequently and to what degree composition fluctuates. The database that hosts all supportive data streams will be designed to store and integrate all relevant data streams seamlessly. Finally, using an 80/20 training/testing data split with 5-fold cross-validation, the machine learning algorithm will be selected and optimized to provide the lowest error rate for suggested sorting criteria.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.