Dynamic Field Monitoring Based on Multitask Learning in Sensor Networks

Area of Science:

• Engineering
• Data Science
• Sensor Networks

Background:

• Field monitoring is crucial for engineering supervision and timely detection of out-of-control events.
• Missing, inaccurate, or delayed sensor data due to various failures poses a significant challenge to effective field monitoring.

Purpose of the Study:

• To develop an efficient field monitoring method capable of accurate and timely detection amidst considerable missing data.
• To integrate multitask learning with a log likelihood ratio (LR)-based multivariate cumulative sum (MCUSUM) control chart for enhanced performance in sensor networks.

Main Methods:

• Utilized a log likelihood ratio (LR)-based multivariate cumulative sum (MCUSUM) control chart, considering spatial correlations.
• Integrated a multitask learning model to address the challenge of missing data within the LR-based MCUSUM framework.
• Validated the approach through both simulation and real-world case studies.

Main Results:

• The proposed multitask learning-based method demonstrates accurate and timely detection of out-of-control states.
• The approach effectively handles scenarios with a large volume of missing sensor data.
• The method proves robust in real-world engineering applications.

Conclusions:

• The developed multitask learning integrated LR-based MCUSUM control chart offers an effective strategy for field monitoring.
• This method enhances the ability to detect abnormal quality products during production, thereby reducing losses.
• It provides a reliable solution for sensor networks facing data acquisition challenges.