SBIR Phase I: Predictive Algorithms for Water Point Failure

这个小企业创新研究（SBIR）第一阶段项目的更广泛影响/商业潜力在于为发展中国家创造一个财政上可持续和负责任的水服务市场。改善水、环境卫生和个人卫生对公共卫生的影响是巨大的，并有可能预防至少9.1%的全球疾病负担和6.3%的总死亡。目前在发展中国家提供供水服务的方法通常集中在部署、维护和监测援助项目上，时间只有几年。名义上，影响由实施者（非营利、私人和政府等）直接评估。然而，即使测量到积极的影响，这些环境服务和监测干预措施中的大多数也是短期的，而且测量结果可能具有误导性。例如，一个为期数十年的项目显然将农村地区的清洁水供应从1990年的58%提高到2015年的91%。改进的服务可以通过预防性和“及时”维护活动来实现，通过仪表和预测故障数据分析算法来实现。关键的是，这可能使供水零中断成为可能。众所周知，由于供水点故障导致的中间取水途径和清洁水会增加健康风险。这个小企业创新研究（SBIR）第一阶段项目旨在开发预测机器学习算法，用于解决由安装在发展中国家农村供水基础设施上的蜂窝报告电子传感器引起的供水点故障。本提案中提出的研究创新包括采用鲁棒机器学习分类技术的集成，使用交叉验证方法来调整模型参数和评估性能，以便开发能够提前足够好地预测故障的数据自适应系统，以便进行预防性维护，维修或更换。具体来说，我们将首先检查基于状态的维护。与基于时间的维护相比，基于状态的维护有几个优点，特别是能够在需要的地方分配有限的维护资源，而不是均匀地分配维护资源，包括可能不需要的地方。我们提出的第一阶段SBIR重点是使用我们现有的传感器硬件开发水点故障的预测算法，并应用于现有的客户。我们第一阶段SBIR的成功标准是一种预测算法，可以准确识别接近故障的水点。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is in creating a market for financially sustained and accountable water services in developing countries. The impact of improved water, sanitation, and hygiene on public health is significant, and has the potential to prevent at least 9.1% of the global disease burden and 6.3% of all deaths. Present-day approaches for delivering water services in developing countries typically focus on deploying, maintaining, and monitoring aid-projects for only a few years. Impact is nominally evaluated by implementers (non-profit, private and government alike) directly. However, even when a positive impact is measured, the majority of these environmental service and monitoring interventions are short-term, and measurements may be misleading. For example, a multi-decade project apparently increased access to clean water supplies in rural areas from 58% in 1990 to 91% in 2015. Improved services may be realized through preventative and "just in time" maintenance activities, enabled through instrumentation and predictive failure data analysis algorithms. This may, critically, enable zero-interruption in water supply. Intermediate access to water, caused by water point failure, to clean water is known to increase health risks.This Small Business Innovation Research (SBIR) Phase I project intends to develop predicative machine learning algorithms for water point failures derived from cellular reporting electronic sensors installed on rural water infrastructure in developing countries. The innovation proposed for research in this proposal consists of employing an ensemble of robust machine learning classification techniques, using cross-validation methods to tune model parameters and evaluate performance, in order to develop a data-adaptive system capable of predicting failure well enough in advance to allow preventive maintenance, repair or replacement. Specifically, we will first examine condition based maintenance. Condition based maintenance has several advantages over time based maintenance, especially the ability to allocate limited maintenance resources where they are needed, instead of spreading maintenance resources evenly, including where they may not be needed. Our proposed Phase 1 SBIR focuses on developing predictive algorithms for water point failures using our existing sensor hardware and applied to existing customers. Our success criteria for a Phase 1 SBIR is a predictive algorithm that can accurately identify water points in near-failure.