Data fusion techniques for structural health monitoring wireless nodes

Motivation

Nowadays, composite structures are massively used in various industrial sectors (civil engineering, aeronautics, etc.) to enable them to offer better performance in comparison to bulk materials (higher Young modulus, lower weights ...). However, these composite materials might also be subject to significant destructive effects (crash, destruction, collapse, etc.), especially because of the high expectations they are facing together with harsh environments and tough conditions they are intended to be used in. There is a real seek for remote monitoring of superstructures using composite materials. This trend is backed up by the Internet of Things and thus constitutes a massive and durable shift. Sensors must be able to achieve reliable, fast, continuous and non-destructive measurements over the whole structure. Additionally, the generated data have to be processed and merged in order to enable remote monitoring at low energetical costs [1,2].

This thesis takes place in the SEISMIC (Structural hEalth monItoring Systems based on MICrowave sensors: hardware and software developments) Cominlabs project.  The aim of this proposal is to develop distributed sensor networks composed of innovative microwave sensors and the associated data analysis algorithms for real-time non-destructive health monitoring of large-scale composite structures. The microwave-based sensors of different natures (reflectometer and microwave transmission line, developed in the other thesis of this project) will be either embedded inside the monitored structure or deposited at its surface. Moreover, some classical SHM (Structural Health Monitoring) sensors as IMU (Inertial Motion Unit) will also be used for complementary information about the structure evolution. The aim of this thesis is therefore to design data fusion techniques for SHM and implement them in energy-constrained sensor nodes that can be accessed remotely by radio communication.

PhD program

1. Inverse problem resolution

The fact that those two sensor topologies are distributed (over a transmission line or over an array) enables the monitoring of large areas, but thus requires specific data processing to extract the relevant information. The latter will be defined once the coupling between the sensors and the physical characteristics of the structure to monitor will be established, allowing the design of a model that will serve as a reference for signal processing [3], [4].

2. Data Fusion techniques

Three types of data-fusion techniques have to be  investigated

• data of the same type that are situated at different positions of the structure. If the data are measured simultaneously, they can be processed either locally or together by a central unit, taking into account the cost of wireless signal delivery, the computational burden of local processing and the benefit of optimized global processing. In general, most of the work are processed locally and only pertinent information is transmitted to reduce energy cost. If the data are processed at different time steps, they need to be fused, taking into account the variability of the environmental conditions.
• data of the same type with different qualities at different locations. The problem of fusing data with different quality will be investigated, when some sensor is of better quality (but more expensive so in limited numbers) merged with possibly cheaper and more widely available sensors. A possible problem is for example the merging of strain gages, accelerometers and IMU, out of the same mechanical system but with different quality. The main idea is to use the evaluation of uncertainty of each data type as data fusion weight.
• data of different nature but related to the same kind of defaults. For example, cracks in a structure can impact both mechanical sensors (IMU) and the scattering data of embedded transmission lines. The fusion of such data will be investigated based on the mathematical model of each sensor system, in a probabilistic framework taking into account the uncertainty of each data source (background research area of I4S team).

3. Sensor node integration and energy optimization

Data fusion algorithms have then to be integrated on real sensor nodes embedding a microcontroller and a low power FPGA for accelerated co-processing. A key issue to reach a complete, real-time SHM system is the energy optimization of both the SHM algorithm and the communications between nodes to transmit required information for data fusion.

In a first step, only a centralized processing will be considered, i.e. a node can access the complete information of every sensor. However, in real deployments, sensor nodes will probably be disseminated over the structure and will be subject to severe energy constraints. Therefore, the second step will consider limited transmission capabilities of leaf nodes to the central node. In other words, the objective is to tune sensing and transmission rate such that the SHM algorithm can still work correctly, even if accuracy is slightly degraded, while respecting both energy and low processing capability constraints of embedded microcontrollers. The Granit team owns a strong experience on designing sensor wireless platforms. A variety of material solutions including energy harvesting [5,6] or FPGA co-processing, are already available and ready to use.

4. Proof of concept

To evidence the outcomes of this thesis, the developed nodes will be deployed on the DECID2 platform. This equipment, deployed on the site of Université Gustave Eiffel (formerly IFSTTAR) in Nantes, consists of an experimental footbridge made from composite materials (fiberglass) on which a set of sensors is installed. This real bridge, yet not used for transport purposes, is then subjected to deformations obtained by the positioning of loads. The resulting deformations are monitored by a set of integrated sensors (optic fibers and ultrasonic-based) that will be used in the SEISMIC project. They will serve not only for providing data that will be merged and processed by the developed algorithms, but also as references to assess the RF-based sensors to be developed within the SEISMIC project.