Sensor Development for the imote2 Smart Sensor Platform March 7, 2008 2008
Introduction Aging infrastructure requires cost effective and timely inspection and maintenance practices The condition of a structure following an extreme loading event must be quickly assessed Structural health monitoring systems must provide relevant information without data inundation The local nature of damage requires a dense array of sensors to adequately assess the structural condition Networks of densely distributed smart sensors with the ability to process data in a distributed manner have the potential to improve SHM dramatically
Smart Sensors The evolution research on sensor networks for structural health monitoring Wired sensors Wireless sensors Smart Sensors Advances in microprocessor and MEMS technology have led to the potential for highly adaptable, densely distributed smart sensor networks Features of a smart sensor: Utilize the computational power of the smart sensor nodes to realize distributed computing Limit the amount of raw data that is shared amongst sensors Selfpowered Processor Wireless Radio Scalability Data Storage Low Cost
Intel s imote2 Second generation of Intel mote Geared toward higher bit-rate applications Low-power 32-bit XScale processor (PXA271) Scalable processor speed to improve power consumption 802.15.4 radio (ChipCon 2420) Data storage 256KB SRAM 32MB External SDRAM 32MB Flash Mica2 imote 2 Microprocessor ATmega128L XScalePXA271 Clock speed (MHz) 7.373 13-416 Active Power (mw) 24 @ 3V 44 @ 13 MHz, 570 @ 416 MHz Data rate (kbps) 38.4 250 Program flash (bytes) 128 K 32 M RAM (bytes) Nonvolatile storage (bytes) 4 K 512 K 256 K + 32 M external 32 M (Program flash) Size (mm) 58 x 32 x 7 48 x 36 x 7
Sensing with the imote2 The imote2 has all of the desired characteristics except the sensor Sensor Board Basic Connector Sensor Interface Connect to main board via two connectors Interface allows flexibility in sensor applications Digital I/O options only Two options Utilize the Basic Sensor Board created by Intel User/developer must provide the sensor board(s) Main Board Advanced Connector Battery Board
Intel s Basic Sensor Board Original Basic Sensor Board features Light Sensor Temperature/Relative Humidity Sensor ST Microelectronics Digital Accelerometer 3-axes ± 2g measurement range 12-bit resolution Anti-aliasing filters with selectable cutoff frequencies ITS400 Sensor Board (Crossbow) additional features Additional temperature sensor General purpose ADC 4-channel 12-bit resolution Top Bottom
Intel s Basic Sensor Board Limitations: Low ADC resolution Limited signal processing capabilities Inaccurate sampling rate Clock jitter (sample rate fluctuation) Decimation factor Cutoff frequency (Hz) Sampling rate (Hz) 128 70 280 64 140 560 32 280 1120 8 1120 4480 Sample rate fluctuation on a single sensor Variation in sample rates amongst sensors
imote2 SHM-A Board Designed specifically for vibration based structural health monitoring applications Motivation Typical civil structures exhibit low frequency response but damage is often evident in higher modes Flexible measurement bandwidth Vibration based SHM is often limited to ambient excitation High resolution, low-noise data is required Distributed SHM requires synchronization of signals from each sensor node to a high degree of accuracy Accurate clock and timing components are required Primary design goals Provide flexible, user-selectable anti-aliasing filters and sampling rate options Utilize commercially available, low-cost MEMS components 3 axes acceleration High ADC resolution High sensitivity
imote2 SHM-A Board Components 3-axis Analog Accelerometer Single-pole RC Low-pass AA Filter f c = 50 Hz Single-pole RC Low-pass AA Filter f c = 500 Hz Gain Difference Amplifier 20 MHz Crystal 16-bit Analogto-Digital Converter SPI Interface Imote2 Top Bottom
imote2 SHM-A Board Components ST Microelectronics Analog Accelerometer 3-axes acceleration 0.66 mv/g resolution Low-power Moderate noise (50μg/ Hz) Cost: ~$12 per chip Design limitations High output impedance (110 kω) High margin or error in output resistor (±20%)
Filter Considerations Accelerometer internal resistor + external capacitor create a single-pole low-pass RC filter Internal to Accelerometer V in R C V out f c = 1 2π R C out load φ ( f ) = 1 tan ( 2 π frc) External Not effective anti-aliasing filter Very slow roll-off Non-linear phase response 0dB 3dB Rolls off at 6dB per octave after f c 0 45 Potentially large error introduced to signal by resistor error different for each channel f c 90 f c
Filter Considerations Capacitor chosen sets the cut-off Consider three potential cut-off frequencies: 50 Hz, 500 Hz, and 1500 Hz (maximum allowed) 2 Filter Gain 20 Phase 12 Maximum Phase Mismatch db 0 2 4 6 8 fc = 50 Hz fc = 500 Hz fc = 1500 Hz 20 40 60 80 100 f (Hz) Degrees 0 20 40 60 80 fc = 50 Hz fc = 500 Hz fc = 1500 Hz 20 40 60 80 100 f (Hz) Phase Difference (deg) 9 6 3 0 fc = 50 Hz fc = 500 Hz fc = 1500 Hz 20 40 60 80 100 f (Hz) Phase mismatch occurs when channels have different values of output resistance (due to resistor error) To avoid potential phase mismatch and signal distortion, use the highest possible cut-off frequency Address aliasing in another way
Digital Signal Processing Quickfilter Programmable Signal Conditioner 4-channels, single- or double-ended 16-bit analog-digital-converter (ADC) Programmable gains Built-in anti-aliasing filters Individually programmable digital FIR filters Digital SPI output Cost: ~$18 per chip Flexible signal processing User-selectable anti-aliasing filters, sampling frequency Oversampling-filtering-decimation provides sharp rolloff and linear phase response Similar to PC-based analyzers
Quickfilter Block diagram of QF4A512 PGA Gain amplifier x2, x4, x8 AA Filter 3 rd Order Bessel f c = 500kHz Analog antialiasing filter 12-bit ADC f s1 = 12.5MHz Digitized with high oversampling rate Cascaded Integrator-Comb (CIC) Filter/Decimator Filtered and decimated to f s2 *4 Sinc filter Amplitude droop Linear phase Cascaded Integrator- Halfband (CIH) Filter/Decimator Filtered and decimated to f s2 Compensates for CIC droop in frequency domain Maintains linear phase FIR Filter User Defined Many filter designs available Designed with QF software Oversampling-filtering-decimation serves two purposes Increases resolution by reducing quantization noise In conjunction with simple analog anti-aliasing filters, produces un-aliased signal
FIR Filter Design Built-in digital signal processing Software allows for many filter design options User specifies characteristics such as sampling rate and cut-off frequency Filter design is optimized and header file created Included when sensing application loaded on the imote2
Software TinyOS is the operating system for many smart sensors including the imote2 Small memory footprint Power efficient Large user community Driver was developed in TinyOS to control the functions of the ADC Allocate memory Load FIR filter coefficients Set sampling rate Do timestamping Write data
Design Validation Calibration testing performed on bench-scale uniaxial shake table Signal compared to wired sensor Excellent agreement in time and frequency domains mg 400 200 0-200 -400 0 0.5 1 1.5 2 2.5 3 Time (sec) 10 4 Time Histories Power Spectrum Reference SHM-A Board mg 2 /Hz 10 2 10 0 10-2 Reference SHM-A Board 0 5 10 15 20 25 30 35 40 f (Hz) Transfer Function 1.5 Magnitude 1 0.5 0 0 5 10 15 20 25 30 35 40 f (Hz)
Clock Accuracy Sampling rate accuracy estimation: Sample rate set to 1000.32Hz (dt = 9.997e-04sec) Record timestamp from processor every 10 th sample Subtract consecutive timestamps and divide by ten to get average timestep over ten data points μsec msec 10.000 9.999 9.998 9.997 9.996 9.995 9.994 9.993 9.992 9.991 0 100 200 300 400 500 600 700 800 900 1000 Timestamp Count Measured sampling rate: 1000.44Hz (0.012% error) Standard deviation of sample rate fluctuation: 0.11μs (0.011% error) Performance may be better than measured because timestamping itself may interfere with hardware timing
Ongoing Work Test Structure Historic bridge in Mahomet, Illinois Truss bridge built in 1912 Environmental hardening Antenna testing Multi-scale sensor board for imote2 Network optimization/fault tolerance
Conclusion Smart Sensors, which incorporate wireless sensing with computational capability, allow for distributed SHM scenarios The imote2 provides data storage and computational capability required for SHM applications A versatile accelerometer board has been designed for vibration-based SHM applications
Acknowledgement This work is sponsored in part by the National Science Foundation, Dr. Shih-Chi Liu, program manager Additional support was provided by the Vodafone Graduate Fellowship program