SBIR Phase II: Predictive Algorithms for Water Point Failure

Period of Performance: 09/01/2017 - 08/31/2019

$750K

Phase 2 SBIR

Recipient Firm

SweetSense Inc.
5548 NE 18th Ave Array
Portland, OR 97211
Firm POC, Principal Investigator

Abstract

This Small Business Innovation Research (SBIR) Phase II project will develop and apply machine learning statistical tools to Internet of Things (IoT) water delivery and water quality sensors. This will enable prediction and preemptive response to water point failures. The resilience of these environmental services is dependent upon credible and continuous indicators of reliability, leveraged by funding agencies to incentivize performance among service providers. In many locations, these service providers are utilities providing access to clean water, safe sanitation, and reliable energy. However, in some rural areas, there remains a significant gap between the intent of service providers and the impacts measured over time. Achieving the SBIR Phase II core objectives will help close the loop on effective and clean water delivery. IoT sensors and services will address one of the most critical public health gaps by enabling delivery of reliable and safe water. IoT solutions for this environment may help address these information asymmetries and enable improved decisions and response. However, given the remote and power constrained environments and the high degree of variability between fixed infrastructure including age, materials, pipe diameters, power quality, rotating equipment vendors (pumps and generators), servicing, and functionality, any IOT solution would have to either be bespoke engineering, or compensate for these site-wise complexities through analytics. Instead, our SBIR II approach is to develop universal, solar powered cellular and satellite IOT hardware for each service type, and addresses site complexities through cloud-based sensor fusion and statistical learning. In this way, we significantly reduce hardware and logistical costs, and provide value to our customers through service delivery analytics. In Phase I, we demonstrated the application of simple sensors and sophisticated machine learning to identify off-nominal service delivery across a cohort of water pumps of various designs. We developed a universal electrical borehole sensor compatible with disparate fixed infrastructure, and we demonstrated solving the problem of heterogeneous customer hardware with a homogeneous sensor platform and adaptive machine learning backend.