AI-enabled efficient fiber-optic sensor/actuation system using deep neural networks (DNNs) and information fusion
Abstract
Fiber-optic (FO) acoustic emission (AE) sensors are an emerging technology and, compared to their electronic counterparts, they have many advantages such as small size, light weight, immunity to electromagnetic interference, and multiplexing capability. The overall goal of this project is to develop and demonstrate an AI-enabled real-time structural health monitoring (SHM) framework for tailorable diagnosis, and probabilistic prognosis for detection of damage in Naval infrastructure. The major innovations of this proposal are: i) Probabilistic physical model and DNN-based damage diagnosis using FO sensing/actuation system with rigorous uncertainty quantification, ii) intelligent FO sensors/actuators control for adaptive acoustic emission/ultrasonic data acquisition, and iii) signal processing with information fusion for accurate and efficient prognosis based on deep learning techniques.Although the DNNs can usually deliver the outstanding accuracy, for Naval CSSs with complexity in degradation mechanisms, structures and materials, in-service environment, and limitations in operation conditions, it is very challenging to integrate efficient processing of DNNs, costeffective and accurate sensing/actuation, and data processing hardware/software into one system, and expand the deployment of AI-enabled SHM sensor networks in this domain.If successful, this project will address the key technical challenges in the development of a realtime in-situ SHM system for complex Naval CSSs without sacrificing accuracy. The proposed system will comprise a powerful signal processing unit based on advanced Artificial Neural Network (ANN) and deep learning techniques with inputs from the AE/ultrasonic sensor system and also from a network of other fiber-optic sensor systems that measure the temperature and strain of the structure simultaneously, for reliable diagnosis and prognosis of the structural health. The AI module including DNN units can intelligently assign the laser resources to activate targeted set of sensors that can provide the most valuable information for the SHM purpose.This work will also pave the way to enable advanced technologies that can facilitate moving SHM computation and decision-making closer to the source by embedding computation near or within the sensors and the memories. It will enable us to detect structural flaws and defects in a large area before they become a safety concern, which will significantly improve Navys ability to conduct systems maintenance and repairs with increased safety confidence levels and at reduced costs.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 17, 2020
- Source ID
- N000142012649
Entities
People
- Yiming Deng
Organizations
- Michigan State University
- Office of Naval Research
- United States Navy