This page acts as a landing page for our immune-inspired fault detection approach for sensor networks that utilizes node-level diagnostics. It summarizes the used resources and links to the respective repositories. The approach combines the node-level self-diagnostics of the used sensor nodes with an immune-inspired fault detection algorithm based on a modified version of the deterministic dendritic cell algorithm (dDCA).
The entire approach has been implemented in a real wireless sensor network (WSN) consisting of several sensor nodes, a cluster head, and a sink node. Information on the particular parts is presented in the following.
The sensor nodes are based on the AVR-based sensor node with XBee radio, or short ASN(x).
The nodes were programmed in C-language on the bare metal (i.e., without an operating system) using the software libraries available in the repository linked above.
The cluster head is based on a Raspberry Pi 3 Model B+.
Information on the general setup is available in the setup instructions.
The Python script implementing the modified dDCA is located in the source directory of the dca branch.
Like the cluster head, also the sink node is based on a Raspberry Pi 3 Model B+. Information on its components and setup is available in the respective RPi-based Sink Node repository.
Our approach was evaluated with a tripartite setup consisting of simulations, a lab experiment setup, and an indoor WSN deployment. Details on each experiment are given below.
To evaluate the approach’s efficiency, we analyzed its resource overhead.
In this context, we compared the memory requirements of the sensor node software with and without our detection approach enabled.
The version without the fault detection and without any node-level diagnostics is available in the dca-central-analysis branch of the ASN(x) repository in the 006-dca_centralized-without_diag directory.
An example implementation of the entire approach is provided as demo application in the default branch (master) in the source directory and is named 005-dca_centralized.
The comparison is based on the memory calculation of the avr-size tool.
In both cases, the sources were compiled with optimization for size (-Os) and link-time optimization (-lfto).
We ran several simulations where we randomly injected faults in one of three base datasets.
The base datasets were recorded from our WSN deployments and show fault-free sensor values and node-level diagnostics.
The injected faults, on the other hand, are based on fault signatures captured in our previous works and our lab experiments described below.
The base datasets, the fault signatures, and the Python scripts used for the simulation are available in the ASN(x) Analyses repository.
The datasets generated during our simulations and the respective results of our fault detection approach as well as visualizations of the used data are stored in the dca directory of the dca-central-analysis branch.
Additionally, we performed several experiments in a lab setup to evaluate the reaction of our approach to various environmental influences partly while forcing the sensor node in faulty conditions.
In this setup, the environmental conditions of the sensor node were controlled by our Embedded Testbench (short ETB).
In contrast to the indoor WSN deployment discussed below, the sensor node transmitted the data directly to the ETB via UART rather than sending it via Zigbee.
The respective source code is available in the dca-etb-exp branch in the source folder (i.e., 005-dca_centralized).
Our indoor deployment consists of seven sensor nodes (as described above) that were running software implementing extended node-level diagnostics.
As stated above, the source code is available in the source directory of the master branch named 005-dca_centralized.