A high-throughput multi-hop WSN for structural health monitoring

2.1.1. Hardware of M-RSN, D-RRN

The hardware of M-RSN, presented in Fig. 1(a), mainly contains core processing module, multiple wireless communication modules, universal serial bus (USB) communication module. Fig. 2(a) shows that the dimension of M-RSN [20] is 180 mm×115 mm×30 mm.

Fig. 1Hardware architectures of M-RSN and D-RRN

a) Hardware architecture of M-RSN

b) Hardware architecture of D-RRN

Once M-RSN starts up the supplying power, core processing module initializes and controls the operation of multiple wireless communication modules and USB communication modules. A wireless communication module is in charge of starting network operation by sending command to all sensor node and broadcasts synchronization command periodically to all sensor nodes for synchronizing network. The other wireless communication modules receive data from sensor nodes and transfer sensor data to core processing module. In the end, the USB communication modules in M-RSN get sensor data from the core processing module and upload them to PC monitoring.

The core processing module is the core module of M-RSN which controls the operation of multiple wireless communication modules and USB communication module. It adopts a FPGA chip (EP3C16Q240C8N) from ALTERA Company as the core processor. This chip contains 240 physical pins which can be programmed by EP3C16Q240C8N, and these pins provide enough efficiency of parallel data transmission for multiple wireless communication modules and USB communication module. In this paper, the frequency of external crystal oscillator for M-RSN is 50 MHz and 64 KB Random Access Memory (RAM) is provided for programming the software of the core processing module. Also, the core processing module allows two different configurations to be done by Quartus II software, knows as JTAG mode and Active Serial mode.

The USB module connects to core processing module and PC. This module adopts CY7C68013A chip from Cypress Semiconductor Corporation for uploading data from core processing module to PC. This chip is a low-power version of the EZ-USB FX2LP which is a highly integrated, low-power USB2.0 controller by intergating the USB2.0 transceiver, enhanced 8051 microcontroller and programmable peripheral interface together. The architecture of CY7C68013A can produce maximum data rate of 53 Mbytes per second in theory. The high transmission rate meets the demand for transmitting all data completely from multiple wireless communication modules to PC. The FPGA chip connects to CY7C68013A with SLOE, SLRD and SLWR pins which indicate the state of reading or writing data. The data bus FD [7:0] and address bus FIFOADDR [7:0] are used to data transmission between FPGA chip and CY7C68013A chip. In the paper, the enhanced 8051 controller in CY7C68013A is implemented in slave First-input-First-Output (FIFO) mode. In this mode, FPGA in core processing module can read and write the FIFO buffer in CY7C68013A in the same way as operating an ordinary FIFO.

The multiple wireless communication modules adopt a 2.4 GHz CC2420 radio chip from Texas Instruments (TI) for low-power wireless communication by IEEE 802.15.4 standard. The wireless communication module connects to core processing module with SFD, FIFO, FIFOP and CCA pins which can indicate the state of receiving or sending data. Additionally, the VREG_EN pin can start voltage regulator of CC2420 and produce 1.8 V to put it into proper condition, and RESETn pin can reset wireless communication module. Also, the analog Serial Peripheral Interface (SPI) (pins SI, SO, SCLK and CSn) between wireless communication modules and core processing module are used to write, read buffered data and read status information. In this paper, two kinds of wireless communication modules operate in M-RSN. One wireless communication module with 32.768 KHz oscillator is only used as synchronization manage module (SMM). The other wireless communication modules with 50 MHz oscillator are only used for receiving radio data from sensor nodes. The SMM records timestamp by oscillator’s counting, stores the timestamp into synchronization command and sends synchronization command to sensor nodes. Therefore, the frequency of oscillator in SMM is closely related to time synchronous data acquisition for sensor nodes. In this paper, the oscillator’s frequency of SMM is same as the oscillator’s frequency of all sensor nodes, which is more efficient for calculating and eliminating the clock time differences between SMM and all sensor nodes. If the oscillator’s frequency of SMM and oscillator’s frequency of sensor node are different, each sensor node has to calculate the ratio of the two frequencies for eliminating clock difference after getting the timestamp from synchronization command in each time synchronization processing. However, the calculation of ratio value will increase complexity of synchronization and produce more energy consumption in sensor node. Therefore, in this paper, the frequency of oscillator in SMM is 32.768 KHz which is also adopted in all sensor nodes.

The hardware architecture of D-RRN, as shown presented in Fig. 1(b), mainly contains core processing module and 2 wireless communication modules. As shown Fig. 2(b), the dimension of the D-RRN is 25 mm×30 mm×30 mm. The EP1C3T100 FPGA chip from ALTERA Company Cyclone I with 65 user I/O pins is chosen as the core controller for core processing module. The EP1C3T100 FPGA chip configures available I/O pins to regulate two wireless communication modules in different channels by analog SPI interface (pins SI, SO, SCLK and CSn). The channel of one wireless communication module is set to provide wireless communication links for the sensor nodes and the other wireless communication module provides wireless communication links for the M-RSN. Like wireless communication module in M-RSN, the two wireless communication modules in D-RRN also adopt CC2420 chip.

Fig. 2(c) shows the wireless strain sensor node adopted which are developed by the authors’ group [5]. The node contains strain gauge signal conditioning board and core board. In the core board, a low-power microcontroller MSP430F1611 is selected as processing controller and CC2420 chip is selected as radio transceiver. 4 channel converters with 12-bit analog-to-digital conversions (ADC) precision in MSP430F1611 MCU allow the node to simultaneously sample analog signals from multiple strain sensors. The signal conditioning board is composed of bridge circuit, amplifier and output circuit for strain signals conditioning. In the paper, although only one sampling channel is connected with the strain gauge, the data packet includes all data from the four sampling channels which are provided by the sensor node. The data of the other three sampling channels are set to zero.

Fig. 2Hardware implementations of the M-RSN, D-RRN and strain sensor node

a) M-RSN with multiple radios

b) D-RRN with two radios

c) Wireless strain sensor node

2.1.2. Software of M-RSN, D-RRN

The software system of M-RSN is divided into sequential processing subsystem and parallel processing subsystem, as shown in Fig. 3(a). Sequential processing subsystem includes logic control layer (LCL) and data communication layer (DCL), and parallel processing subsystem features data processing layer (DPL).

DPL includes data packet processing state machine (DPPSM), radio management state machine (RMSM) and FIFO in different channels. The RMSM initializes the separate wireless communication modules, responds to the interruption from the wireless communication modules simultaneously and distributes data packet to corresponding DPPDM. After DPPDM stores the processing data packets from RMSM to the corresponding FIFOs, all data in FIFOs are formed into a complete stream and then uploaded to LCL. LCL consists of dataflow generation module (DGM), FIFO scanning state machine (FSSM) and FIFO management state machine (FMSM). By merging the data from the FIFOs and forming the data flow in DPL, the DGM reduces the time spent on receiving data from different wireless communication modules and elevates the low efficiency of wireless communication modules. The USB management state machine (UMSM) in DCL connects the USB communication module to core processing module for managing the interface between USB and FPGA. The USB chip set in slavefifo mode controls the DGM to transfer bulks of data to the cache FIFO in the USB chip. The SMM controlled by the FMSM broadcasts the SC with timestamp $T_{\mathrm{0,1}}$ for starting network. And it sends time synchronization command $T_{0,i}$$\text{(}i\ge \text{2)}$ in $T_{sync}$ period to sensor nodes via D-RRNs after switches to each channel subsequently. The details processing of SMM is described in Section 2.2.

The software structure of the D-RRN, as described in Fig. 3(b), includes data processing layer (DPL) and data communication layer (DCL). The DPL, as the parallel processing subsystem, coordinates two transceivers to work separately in two channels. Each transceiver consists of 2 FIFOs, 2 data packet processing state machines (SPPSMs) and 2 radio management state machines (RMSMs). The SPPSMs are in charge of relaying synchronous data and sensor data to FIFO1 and FIFO2 respectively, while RMSMs forward the radio data from DCL to SPPSMs. The synchronous data or the sensor data can be transmitted between FIFO1 and FIFO2. To avoid affect the relaying performance of D-RRN, the data from the M-RSN or sensor nodes are not analyzed due to time-consuming data length calculation and data correction examination. Considering that the FPGA chip in M-RSN has higher capability than FPGA controller in D-RRN, data length calculation, analysis and correction are all implemented in M-RSN.

Fig. 3Software structure of the M-RSN and D-RRN

a) Software structure of the M-RSN

b) Software structure of the D-RRN

4.1.1. Evaluation experiment setup

The experiment to evaluate the maximum DTR and wireless link quality of the proposed network lasts for 2 hours. The experimental setup, as shown in Fig. 7(a), consists of 7 Telosb nodes, 7 D-RRNs and a M-RSN, equipped with 2.4 G non-directional antennas of 5 dBm. In the experiment, the transmission power level of each Telosb node is configured at 0 dBm. To mitigate the effects of interference among different channels, the Telosb nodes are set to communicate in channel 13, 15, 17, 19, 21, 23, 25 provided by IEEE 802.15.4 protocol respectively, and the D-RRNs work in channel 12, 14, 16, 18, 20, 22, 24 respectively to forward data to the M-RSN. The D-RRNs are arranged in the middle of the Telosb nodes and the M-RSN, as demonstrated in Fig. 7(b). Different arrange distances between leaf nodes and the M-RSN are adopted in the evaluation, including 30 m, 40 m, 50 m, 60 m, 70 m, 80 m, 90 m and 100 m.

Fig. 7DTR experimental setup

a) Experimental deployment in open field

b) The framework of DTR experiment

4.1.2. Software implementation setup

To achieve the maximum DTR evaluation, high data transmitting strategy in Telosb node is developed first, shown in Fig. 8. In the initialization process of the Telosb node, as shown in Fig. 8(a), complicated MAC protocol, route protocol, address recognition and in-line security operations which may influence the realization of largest network data throughput are disabled. The Telosb nodes are initialized at the same setup as reported in [38]. After initialization of the Telosb nodes, a fix data payload is directly buffered into transmit FIFO of the CC2420 chip of Telosb nodes. The fixed data payload consists of first 2-byte of 0x7E and 0x7F, 1-byte of payload length, 1-byte of sequent number, a fixed 112-byte of 0xFF bytes and the end byte 0x7E. The first 2 bytes and the end byte are used for checking the correct receiving of the packet. After the data payload filling is completed, the Telosb nodes start continuous data transmission, as demonstrated in Fig. 8(b).

In order to attain the upmost DTR of the M-RSN, the program of the M-RSN is modified. In this evaluation, the DGM in M-RSN, which is engaged in merging a large amount of data packet from the DPL and forming the data flow for data upload and data storage, is disabled. On the other hand, the program disables the HDLC and CRC mechanisms and underlines simplified checking mechanism of data integrity and correction in M-RSN. As shown in Fig. 9, after initializing DCL, LCL and DPL, the M-RSN waits for data receiving interruptions from all radio modules. When data receiving interruption is triggered, the FIFOP pin in CC2420 chip goes high and the M-RSN receives data from RXFIFO. In each data reception, the M-RSN checks the data integrity and correction according to the first two bytes 0x7E and 0x7F, 1-byte of payload length and the end byte 0x7E in the data payload. If the validity of the data integrity and correction is successful, the M-RSN counts the value of packet number and uploads it to the central computer.

Fig. 8Data continuous transmitting process in the Telosb node

a) The initialization of the Telosb node

b) The data packet filled and transmitting processing

Fig. 9The data counting process of M-RSN

4.1.3. Evaluation results and discussion

Table 1 shows the relationship between the achieved DTR and the number of radios at different distances. As shown in Table 1, when only one channel is used, the maximum DTR at the distance of 30 m is 136.5 Kbps which is close to the maximum DTR of 140 Kbps [38]. It is obvious that the maximum DTR rises steadily as the number of channel increases. The maximum DTR can reach 861.6 Kbps at the distance of 80 m when 7 radios are used together. With the distance of 30 m and one hop, the DTR can still reach 909.4 Kbps. Compared to 909.3 Kbps at a close communication distance when all the 7 transceivers are operated together [20], the value of 909.4 Kbps means the successful improvement of DTR when hop is added. As the distance increases from 30 m to 100 m, the maximum DTR for one channel drops slightly from 136.5 Kbps to 120.7 Kbps, DTR for 7 channels falls from 909.4 Kbps to 803.2 Kbps due to accumulative drop of the maximum DTR in each channel. However, the maximum DTR are still greatly improved comparing to the singe-radio node with 140 Kbps DTR based WSN.

In the experiment, the BER and PLR of the network are calculated by Eq. (2) and Eq. (3) respectively. The average BER vs. distance is presented respectively in Fig. 10(a). The BER is close to 0 % within the distance of 80 m with one hop, in comparison with the BER which jumps up suddenly over 50 m without any hop [27]. The BER results demonstrate that almost all packets produced in nodes are successfully transferred to the FIFO of CC2420 transceiver, and prove the high correctness of wireless data packet transmission in the proposed network.

Fig. 10PLR in different channels and BER vs. distance in an open field

a) BER vs. distance in an open field

b) PLR in different channels vs. distance in an open field

Fig. 10(b) presents the average PLR in different channels vs. distance. It can be noticed that the average PLR in different channels experiences an upward trend as the distance increases. Within the distance of 80 m and with one hop, the average PLR in different channels increases slightly and is less than 3 % which has been greatly improved compared to reported 68 %-93 % at the distance of 40 m without any hop [28]. After slight increasing, the rise of the average PLR in different channels is particularly noticeable above 80 m, indicating that average PLR in different channels has high wireless link quality within the radio range of 80 m. Furthermore, as mentioned in Section 2.3, the average PLR of network with all channels and the similarity of PLR among all channels is evaluated by $\mu$ and $\sigma$ respectively. In the Table 2, $\mu$ is below 2.5 % within the distance of 80 m although it rises gradually, and the corresponding $\sigma$ is below 0.6. The value of $\mu$ represents the low average of the wireless packet loss, revealing the high wireless link quality of network. Meanwhile, $\sigma$ evaluates the differences of wireless packet loss in each channel. Due to wireless communication in all channels operates in same outdoor environment, it is reasonable that wireless packet loss is almost same in each channel. If the functionality of multiple radios in the sink node is not parallel or the sink node fails to coordinate all multiple radios well, the wireless packet loss is much higher in some channels than other channels which means high $\sigma$ will be measured in the network. In this experiment, the low measured value of $\sigma$ shows that the wireless packet loss is distributed evenly in different channels of proposed network, which indicates that wireless link quality is similar in different channels.

4.2.1. Experiment system setup

Structure strain changes under external or mechanical loads are important testing parameters in SHM. The aircraft static strain test, on-board real-time strain measurement during test flight and certification of a new aircraft are critical in aircraft structure testing. It is important to measure the load strain at key locations during tests, measure strain distributions during compression testing of stiffened panels with damage to validate analytical model predictions.

This experiment consists of 4 sensor nodes clusters in different channels, in which cluster 1, 2, 3, 4 and 5 are in charge of strain monitoring of UAV composite wing structure and cluster 6, 7 are dedicated for the strain monitoring of wing box structure. Each cluster contains 4 strain sensor nodes. Fig. 11(a) demonstrates the arrangements of strain sensor nodes on the wing box structure. The structure includes 5 lines of bolt holes with an interval distance of 280 mm and 6 T shape stiffeners with a interval distance of 130 mm. As shown in Fig. 11(a), a central monitoring area without bolt holes and shape stiffeners is chosen for strain monitoring and its size is 280×130 mm². In the structure, the strain gauges connected to No. 11, 12, 13, 14, 21, 22, 23, 24, 31, 32, 33, 34, 41, 42, 43 and 44 sensor nodes are placed on the monitoring area. Although each sensor node interfaces with four sampling channels with 10-bit sampling resolution, only one sampling channel is connected to a strain gauge and the sampling data in other three channels are set to zero.

The UAV composite wing structure and the arrangement of nodes are shown in Fig. 11(b). No. 61, 62, 63, 64, 71, 72, 73, 74 strain gauges and No. 1, 2, 3, 4, 5, 6, 7, 8 fiber bragg grating sensors are placed on the beam in UAV composite wing structure. Each fiber bragg grating sensor is deployed besides each strain gauges for strain data comparison. As spectral filter, fiber bragg grating sensor is based on the principle of Bragg reflection. The gratings are printed in the core of optical fiber in a parallel and close line. The gratings alternate regions of high and low refractive indices when a periodic ultraviolet light is imposed on optical fibers. The wavelength is known as the bragg wavelength ${\lambda }_B$ and is given by ${\lambda }_B=2n\mathrm{\Lambda }$ where $n$ is the average refractive index and $\mathrm{\Lambda }$ is the grating period. When load is applied to the UAV composite wing structure, 8 grating is strained and the peak reflected wavelength is changed. In this experiment, change of reflected wavelength without temperature compensation is given by:

where ${\lambda }_B$ is the initial reference wavelength, $\mathrm{\Delta }{\lambda }_B$ is the wavelength shift, $n_{eff}$ is the effective index of the refraction of the core, $v$ is poisson’s ratio of fibre, $p_1$ and $p_2$ are gauge constants of the strain optic tensor, and $F_G$ represents in total the gauge factor of fibre. As fiber bragg grating sensor, Micron os3610 is chosen with the size of 19 mm×250 mm. The strain ranges from –2500 με to +2500 με and the strain sensitivity is 1.2 pm/με. The gauge factor $F_G$ equals to 0.8 [39]. A high speed optical sensing interrogator unit (SM130-700 Micron Optics, scanning rate 1kHz, 8 optical channels, center wavelength of 1550 nm, wavelength ranges from 1510 nm to 1590 nm) is used to demodulated the bragg wavelength shift $\mathrm{\Delta }{\lambda }_B$, and then strain $\epsilon$ is calculated by Eq. (6). In this way, the fiber bragg grating works as strain sensor.

Fig. 11Strain monitoring experiment setup for aircraft structure

a) Experiment arrangement of the wing box

b) Experiment arrangement of the UAV composite wing

c) Monitoring center setup

Two loading equipments (CHHBL2) are used to change the strain distribution in UAV composite wing structure and wing box structure respectively. When concentrated load applied on structure is changed, the strain distribution will be changed. Two loads are exerted by screw rod and increased with screw rod rotated. Initially, screw rod is rotated for two cycles to serve as preload on wing box or UAV composite wing. Then, the load increases gradually with rotating the screw rod. Once the strain displayed on the computer is stable, it will be sampled for 5 seconds and averaged. The averaged strain value is used as final strain value. In Fig. 11(c), D-RRNs, M-RSN, SM130-700 optical sensing interrogator unit and PC monitoring are deployed together for the distributed strain monitoring of the wing box structure and UAV composite wing structure.

4.2.2. Results and discussion

The distributed strain test result of 28 curves from 28 strain nodes in 7 different channels is shown in Fig. 12(a). Each curve demonstrates strain value of a sensor node for one strain gauge attached on the UAV composite wing structure or the wing box structure. Each plot value is an average value of 50 sampling point data. Initially, no load is applied on the two structures, so the strain values from all sensor nodes are zero. At time 60 s, two loads are respectively applied on the two structures simultaneously and all the 28 strain nodes have simultaneous outputs. Since different strains are produced at different positions on the two structures, the strain value output of each node is different. When time arrives at 90 s, all loads are removed from two structures. Under this circumstance, each sensor node in the network is able to monitor the loading changes at the same time. In the processing, the change data are successfully forwarded to the M-RSN via D-RRNs and displayed in the central computer system.

Besides the strain monitoring experiment above, in UAV composite wing structure, another strain monitoring experiment is carried out by 8 fiber bragg grating sensors connected to the SM130-700 optical sensing interrogator unit. The loading process is repeated and the strain measurement is implemented again. Then, as shown in Fig. 12(b), 8 strain testing results from SM130-700 optical sensing interrogator unit are compared to the strain results of No. 61, 62, 63, 64, 71, 72, 73 and 74 sensor nodes obtained by designed multi-hop network. It is concluded that the strain data obtained from the designed multi-hop network are almost the same as that of optical demodulator, verifying the accuracy of the designed multi-hop network. All monitoring experiments last for 24 hours continually without failure.

Fig. 12Strain monitoring results for aircraft structure

a) Strain results from 28 nodes in 7 channels

b) The comparison of strain results between strain gauges and fiber bragg grating sensor gauge

Additionally, the wireless link quality within 24 hours of all sensor nodes deployed in two structures is shown in the Fig. 13. The PLR in different clusters is below 0.1 % and the BER of network is below 0.05 % respectively, indicating high wireless link quality of the designed network in distributed strain experiment.

These results further prove the multi-radio simultaneous communication ability and high wireless link quality of the developed multi-hop monitoring network applied in the UAV composite wing structure and the wing box structure.

Fig. 13The wireless link quality of all sensor nodes in wing box and UAV composite wing

a) BER of all sensor nodes vs. time

b) PLR in different channels vs. time in two structures