I-Corps: Method and Tool for Time-Domain Processing of Densely Sampled Signals

这个i-Corps项目的更广泛/商业潜力将主要在半导体制造业。这个行业迫切需要一种从大量密集采样的传感器读数中提取有用信息的工具。最近可用传感器读数的数量以及收集这些读数的采样率的急剧增加特别强调了这一需求。预计该工具在该行业具有双重用途：提供获取信息的途径以进行决策，并作为减少信息损失的数据减少工具。此外，这种工具预计将在任何使用非旋转设备的行业中有用，包括石油和天然气开采、炼油厂、化工加工工业等。预计许多此类行业收集的数据量将继续增加，进一步刺激开发和投资此类特征提取和数据简化工具的需求。这个i-Corps项目将专注于探索和改进技术创新的潜力，该技术创新将能够从非旋转设备和过程产生的信号中全自动地提取信息特征。这样的工具将能够将原始数据缩减为比原始信号小得多的特征集，从而使得在大量密集采样的传感器信号中进行数据管理和数据挖掘变得更加容易。建立在该工具基础上的方法是最近发展起来的。它首先将信号分割成一系列的稳态和暂态部分，然后使用每个数据段来构建一些基于统计和基于动态的签名。最终目标是创建一种工具，在最少或希望没有人工参与的情况下实现所有这一切，而不是目前使用的方法，这些方法需要大量的人工输入和分析。

The broader/commercial potential of this I-Corps project will be primarily in the semiconductor manufacturing industry. This industry desperately needs a tool for extraction of useful information from a large number of densely sampled sensor readings. This need is especially emphasized by the recent dramatic increases both in the number of available sensor readings, as well as the sampling rates at which those readings are collected. This tool is expected to serve a dual purpose in this industry; providing access to information to enable decision making and serve as a data reduction tool which reduces information loss. Furthermore, such a tool is expected to be useful in any industry that employs non-rotating equipment, including the oil and gas extraction, refineries, chemical process industry etc. It is expected that the volume of data gathered in many such industries will continue to increase, further motivating the need to develop and invest in such feature extraction and data reduction tools.This I-Corps project will focus on exploring and improving potentials of a technological innovation that will enable a fully automated extraction of informative features from signals generated by non-rotating equipment and processes. Such a tool will be able to reduce the original data into a feature set that is significantly smaller in size than the original signal, thus enabling a significantly easier data management and data mining in a large number of densely sampled sensor signals. The method in the foundation of this tool has recently been developed. It starts by segmenting the signal into a series of steady-state and transient portions, after which each segment of the data is used to construct a number of statistics-based and dynamics based signatures. The ultimate goal is to create a tool that will achieve all this with minimal or, hopefully no human involvement, in contrast to the currently used approaches, which require extensive manual inputs and analyses.