Control of Systems with MEMS Sensors and Actuators via Data Mining Techniques

这项跨学科研究的目标是使用数据挖掘技术分析大量的MEMS传感器数据，以发现MEMS执行器上的动作之间的关系及其对系统状态的影响。这些关系(以规则的形式捕获)然后被用来构建飞机控制的反馈回路。大多数系统(如三角翼飞机)的输入输出关系是高度非线性的。传统的数据挖掘方法丢弃了数据集中的许多重要信息，不能提供足够的传递函数信息，不适合系统控制。该项目开发了一种可扩展的多变量数据挖掘技术，可以在广泛的条件下(动态、时间、空间等)发现完整的传感器-执行器关系和预测模型。研究内容包括收集动态系统行为数据，扩展用于总结时态规则的数据挖掘算法，开发驱动模式的规则选择策略，以及开展风洞实验以验证该方法。这项工作有可能大大提高数据挖掘的最新水平，因为这个问题具有许多在其他应用中不常见的特征(实时反馈、时空性质)。数据挖掘技术的成功有望推动MEMS传感器和驱动技术在系统监控和其他工程问题上的应用。

The goal of this interdisciplinary research is to analyze the vast amounts of MEMS sensor data using datamining techniques to discover relationships among actions at MEMS actuators and their impact on the system state. These relationships (captured in the form of rules) are then used to build a feedback loop for aircraft control. The input-output relationships for most systems (e.g., the delta wing aircraft) are highly non-linear. Traditional datamining approaches discard much important information from the datasets and cannot provide sufficient transfer function information, which makes them unsuitable for system control. This project develops a scalable multivariate datamining technique that discovers full sensor-actuator relationships and predictive models under a wide range of conditions (dynamic, temporal, spatial, etc.). The research includes collecting data for dynamic system behavior, extending the datamining algorithms for summarizing temporal rules, developing the rule selection strategy for actuation schema, and developing wind tunnel experiments to validate the approach. This work has the potential to advance the state-of-the-art in data mining substantially, as this problem has many features (real time feedback, spatio-temporal nature) that are not commonly found in other applications. The success of data mining techniques is expected to advance the MEMS sensor and actuation technology in system monitoring and control and in other engineering problems.