I have a high aspect flexible aircraft wing of 2 meters in which I want to place 6 gyroscopes along it to measure its deflection for research purposes. I want to be able to collect all the data effectively at 100Hz frequency from all the gyroscopes ( at the same time) to feed an estimator . It is not an easy task to do because I need communication protocol to be fast, robust to noise generated from BLDC motor, works for long distances and cheap.
Please see specs below :
- The longest distance between the control unit and any IMU will not exceed 2 meters.
- The Data collected from all the IMU’s should be relatively at the same time.
- The communication protocol that to be used should be highly robust to noise.
- The protocol to be used can be adapted with available microcontrollers.
- Data should be collected at 100 Hz frequency in control unit (T sampling = 10 ms).
There are alot of IMU sensors which can be used from adafruit, sparkfun or silicon labs. Currently i have two candidates thunderboard sense 2 and Razor sparkfun IMU in which both can be used as a sensor and a microcontroller at same time since they have arm processor and can be programmed.
Any one can suggest a suitable way to connect and interface with these sensors?
Any one can suggest a cyber physical system in which we can connect these sensors in a specific architecture in which we can gather data with interrupts respecting the above specs?
Thank You.
Why do you want to use a gyroscope? Strain gauges are usually used. They are used for measuring deflection, vibration, load, torque, pressure, stress, and strain. They are much more accurate and reliable than a gyroscope.
The technique I am developing avoids strain gauges and fiber
Bragg gratings (FBG) sensing due to their structural strain–
displacement modeling requirements, although their use as complementary
aiding sensors is still feasible. Using strains for now is out of the scope of my research.
Apart from the question about the type of sensors, I can think of two basically different structures:
If each sensor has its own microcontroller attached, then the collection of data by the central unit can be delayed with respect to the actual measurements. So you can use any kind of bus. With a length of 2 m, I guess a twisted pair connection would provide enough protection against induced noise; if necessary, you could use differential signals (like CAN bus). All you need is a mechanism to synchronize the measurements; it would be helpful if the bus protocol provides a broadcast mechanism (a shared address beside the individual addresses).
If the sensors don't provide a mechanism for triggering measurements and storing the results so that you have to read all results simultaneously, then you would need a star configuration. With 6 sensors, your central unit should have at least 6 identical interfaces, and a mechanism for transferring received data to the memory simultaneously. Microcontrollers like SAMC21 come to mind, providing between 6 and 8 serial interfaces and a DMA (direct memory access) unit.
The interfaces can be configured as SPI, I2C, or UART; unfortunately not as CAN. I guess the best immunity against noise would be provided by UART with converters to/from V24 levels (+/-12 V).
Evaluation boards are available: ATSAMC21-XPRO and ATSAMC21N-XPRO
On second thought, DMA isn't even necessary because the core is fast enough to read 6 receive buffers during the time it takes to receive one byte per interface.
Joerg Fricke Thank you for answering my question.
For the first architecture : How can we synchronize the measurements? and is it guranteed for any bus usage that we can sample all data coming at 100 Hz frequency in the control unit. The challenge of distance that it reduces the frequency to operate on. So, I don't think a protocol like I2C could work.
For the second architecture; Your idea sounds very interesting. However, in my setup there will exist some gyros near the control unit and other far away from it. Placing the SAMC21 is a very good idea in the controll unit but this doesn't solve the problem of obtaining the data at the same time since there will be synchronization issues due to the distance variation. Besides UART is Asynchronous communication protocol.
Please find a picture attached that will explain the setup.
Thank you sir, your comments are helpful.
Ali Srour
Generally, how accurate has the simultaneity in your application to be? I'm a bit confused because a data rate of 100 Hz limits the spectrum of mechanical processes to 50 Hz. If indeed no higher frequency is to be expected then I would guess that an inaccuracy of the phase of, say, 1° would be tolerable. 1° at 50 Hz boils down to about 56 us.
We can assume a propagation velocity on insulated wires of about half the velocity of light in vacuum. For a line of 2 m length, this results in a (one-way) delay of about 13.3 ns. The difference to 56 us is more than 3 orders of magnitude; or, to put another way, in this context, you should start worrying about the delay at a length of 4 km!
I2C and SPI are certainly meant neither for noisy environments nor for communication over distances longer than the dimensions of usual desktop devices. As I said, in the first case, CAN bus seems to be a good choice.
I don't know the bandwidth of possible sensors but my blind guess is that you should expect signals well above 50 Hz. In this case, the data rate has to be higher but, as I outlined above, even at 5 or 10 kHz, the delay introduced by 2 m wire resp. a difference of 2 m should be no problem.
If you, for whatever reasons, are really aiming at a simultaneity better than 13 ns, the only solution that comes to mind is to employ programable delay lines like MC100EP195B at each sensor, and to program each delay line in such a way that the delay introduced by the wire is compensated. The time steps of the MC100EP195B are 10 ps.
PS. In the star configuration, the least expensive solution is to connect each sensor with a 2 m line, of course, provided there is space enough for loops of excess lines.
Try looking at the XPS protocol.
Monitoring the deformations of a structure is a critical problem and of the utmost importance. A structural engineer would prefer to have accurate and deterministic data with long-time validated methods. Other than the measurement/acquisition chain, how accurate is the gyro you want to use? Which one do you plan to use?
I would base this DAQ on Arduino driven by inner 100Hz interrupt and collecting data in parallel (as 8-bit wide port) on 6-lines input bus via SPI (clock, /CS signal then is just duplicated to all devices).
You can add some large SPI memory for data collection or run lower, high speed serial priority serial port to Raspberry Pi. Raspberry may have some irregularities (time jitter) when you try to give it interrupt at 100Hz ( if it acceptable you can use it directly).
If you have reasonable EMI performance of the power systems I guess SPI system should work on 2 meters ( some 47 ohm or so series resistors recommended on SPI lines to reduce ringing).
Extra robustness against electrical interference can be achieved by using shielded cables (some of them quite thin and you need only 4 wires inside). The shield that is connected to thin Faraday cage shield made around each unit and central processor, while having only single grounding point to the rest of the system.
Also to make it more robust one can use differential digital receivers/transmitters like SN75C1167NSR and twisted pairs cable.
Other way, to increase EMI robustness you can use 6 of SPI isolator with dc/dc converter ( thus isolating grounds of your sensors):
https://www.silabs.com/documents/public/data-sheets/Si88x2x.pdf
Using the CAN bus might be a viable option:
* You can send a message to the sensors that is intended to trigger a synchronous
data acquisition on all devices.
* Following this, the sensing devices will transmit the data acquired. Thus it won't
matter that the data transmission is sequential.
* On the "dark" side: this will require sensor nodes with a microcontroller at each
node (best a microcontroller with integrated CAN controller), plus an adequate
power supply for the sensor nodes.
Anyway - CAN plus power supply requires no more than 4 wires.
Hi,
I agree with U. Dreher. We had similar problem. We solved it by using CAN-bus for communication and control. There were 100Hz control messages and 100Hz measurement' sample rate running simultaneously. The disadvantage is the need of additional PCB with mC for every gyro to establish communication. The advantage is the ease of use of CAN in collecting messages with data.
I hope it helped.
Michele Raucci Well thank you for your feedback about accurracy. I am planning on experimenting with multiple gyros such as :
- Thunderboard sense 2 from silicon labs
- BMO055 from adafruit
- BMO085 from adafruit
- or any other sensor that may work.
The choice of the sensor will be after the choice of the protocol of communication. However, we already have some sensors in hand.
Nikolay Pavlov Thank you Nikolay for your answer that was really helpful.
In the SPI protocol you mean to use only one pin for chip select, in that case we read from all sensors at the same time ?
If that was the case then I have some concerns concerning about the ability of SPI to collect the data from the sensors at 0 meters and the one that is at 2 meters especially the synchronization issues. I know as Dr. Joerg mentioned earlier that SPI is not suitable for this application. Your idea is great and suitable for me, can you further clarify with some links or illustration.
U. Dreher Sebastian Brol I agree with you both about using CAN. After all , it is the industry protocol standard and proved to be very robust. The real draw back is the complexity associated with using an Electronic Control Unit ( ECU) at each node which will include for example an (arduino + CAN controller + sensor). CAN bus solves all the problems, but still for this project it has the complexity of adding too much hardware parts.
Sebastian Brol Can you please send me a link for your project setup and perhaps code. Your project sounds very interesting and I would like to learn more about it.
Ali Srour I guess it must be ok for moderate clock. E.g. if you are using 5MHz the length of signal in cable will be 30 meter (half speed of light in the cable) and considering back and forth delay the clock domain has radius of 15 meters . One can use e.g. 1 or 2 MHz for peace of mind, should still be OK for 100Hz (one can transfer almost 1e6/100 = 1e4 bits at 1MHz/100Hz). Alternatively, you can poll sensor sequentially but it complicates things e.g. as adding the time shift between them. For sequential polling however i2c would be easier at that (rather low still) sample rate (there are nice i2c isolators too considering EMI hardening).
Also, to assure extreme EMI performance low cost plastic optic fiber and transmitters receivers like
https://www.digikey.ie/product-detail/en/toshiba-semiconductor-and-storage/TORX1355-F/TORX1355-F-ND/7100238
https://www.digikey.ie/product-detail/en/toshiba-semiconductor-and-storage/TOTX1350-F/TOTX1350-F-ND/7100231
Hello all,
I just want to thank you and share with you my setup which can potentially help others. Finally, I have udecided to use the CAN bus with Teensy 3.2 microcontroller and CAN transcieviers MCP2562 as shown in the pictures. The transmission worked for high rates and long distances. I was able to send multiple bytes data from the BNO055 sensor ( GYRO- ACCEL- MAG ) with very high speed. All of this was done utilizing the Flex CAN library.
