FPGA - Vibration Analysis

Why do you want to this ?

Will the user of the instrument look at the screen and make decisions based on  the DFT of the incoming data.

Or will you store the data and need to process it later.

Why not just buy some kit that is ready sorted and can do it all for you - there is a huge amount of work in getting the boards you discuss suitable package with batteries etc to take out into the field.

Will you want to do order tracking - if so then you'll need additional sensors and inputs.

Is this a commercial project or for education.

Is the primary aim to protect the machine, gather data or design some data acquisition hardware.

Why might you wish to use an FPGA to do the Fourier Transforms - some FPGAs are very capable but unless you know how to use them you will need to do a lot of work to get as far as you would running  a Python library on a Raspberry Pi or using C/C++  on a PC.

The accelerometer you quote has a working bandwidth of about 10kHz to you would not need a very high sampling rate - 50k s/s should be enough.

The MCC board would do fine -  Jan Cumps has reviewed that part on E14. You should be able to display results on the Pi's display - no tablet required.

MK

Thank you so much for the reply.

I am a Mechanical / Marine Engineer with some knowledge of electronics. We operate several diesel engines and often run into problems with huge financial effects. Though we strictly implement planned maintenance routines but engines do suffer frequent breakdowns. The purpose is to put in place a "Condition based monitoring" plan. The latter is my motive to develop such a monitoring system. Perhaps that answers the first question.

Idea is to capture the FINGERPRINT of a good engine and keep track while it runs, continually comparing it with the FINGERPRINT. Any anomaly ie a rise in amplitude or a new frequency popping up / frequency harmonics, would mean that engine needs to be checked for possible issues. Yes, indeed I do need to store date to view FFT / Frequency domain graphs later. 

Honestly, never used FPGAs. My colleague at work has done work on FFT related to RF Receivers. But  tying up a suitable FPGA to a LCD is under thought. Same reason seeking help from friends on this forum. 

Raspberry Pi is not powerful enough. PC is the solution but difficult to place around engine room spaces. Tablet / Dell i7 type should work for their higher resolution.

I am planning to go for the CN0540 by Analog Devices which has a similar sampling rate of 50k samples/ sec.

I shall discuss with Jan Cumps too.

At present the Project remains  within the confines of my garage. If I am successful in distinguishing between the fingerprint and subsequent running FFTs of the engines, then will go commercial. As yet, there is many a slip between the cup and the lip.

I think you might find that Pi is a good deal more powerful than the Arduino type boards the CN0540 is designed to connect to.

And if the MCC172 will work with the Pi4 or even better the Pi5 there is plenty of computing power for single channel of vibration.

Good luck.

MK

you are correct indeed. I have been told that Pi5 is not available yet? 

MCC172 is preferable as can be stacked if more sensors are needed to work in tandem.

Thanks.

Great project, although an FPGA solution will take a year or two to get through the learning curve, the build, the software development and the test/tuning phase. Even expensive commercial solutions would take significant time to apply to your application and achieve desired results. If you need more power than a Pi, you could look at PC "stick" computers, but they start at around $250.

Hi Dougw,

Thanks. That is the last resort; get a mini PC to fit into the PCBs snugly. In parallel, if I use a TABLET, the latter would also resolve my problem.

Raspberry Pi5 is an another option as pointed out above by Michael.

These are short cuts but will definitely work.

I am going to give FPGA a shot. Most ostensibly the better way ahead. 

Use of FPGA for such purposes, an example is of the CN-0540 (Analog Devices) / ADC 24 bits, around 50 K samples / sec, has ARDUINO pins which plug in to the DE-10 Nano  (FPGA Board by INTEL) . 

Still need an Operating System for the GUI of the FFT display. Battle not won !

https://wiki.analog.com/resources/eval/user-guides/circuits-from-the-lab/cn0540/de10-nano

Friends, if any of you have ideas, please shoot.

Once I succeed in getting the FFT / Frequency Domain data display working i.e. prototype ready, then it won't take long as I have the opportunity to test the set-up on premises. 

Sincerely,

How about using a DSP / TMS320F6xxx series from Texas Instruments for the purpose of FFT ?

Any one with TI DSP experience, please ?

I think you are not quite in the right gear with respect to the processing hardware to do FFT calculations.

Start with defining the problem rather than the hardware to do the maths.

What size of DFT (determined by ratio of max frequency and resolution needed)

eg 1024 length DFT with 20kHz max frequency will give you about 20Hz resolution which may not work well for big engines. So you could use a much bigger DFT or reduce the maximum frequency.

Then you need to work out how many DFTs you need to compute per second.

Then you need to look at what kind of hardware will do the trick.

For example - about the cheapest way to hit your problem would be to use an ST STM32H7xxx series processor - some have 16 bit ADCs, easily fast enough for your application. Being ARM Cortex cores there is FFT code for free. The STM32H7xx will clock at 400 - 500 MHz. They are MUCH easier to use than FPGA or DPS type parts and the tools are cheaper and much easier to get help with.

You'll need to do some research to work out how fast any given chip or board can do the processing you need.

Of course if the chip isn't fast enough you may need to step up to something faster.

My advice would be to

define the signal processing task

get a prototype working using parts chosen to be cheap and easy to work with.

then refine the design.

Most people start off by using  a bought in ADC board that will just plug straight into a computer (almost always a PC).

I've known people start off with a PicoScope (or similar) and a PC - that way you hardly need to think about hardware design issues at all and can get right into the important bit of measuring some real data. (But you always have to define the processing task to scope the hardware you need first.)

MK

• I've written a flow for the mcc board, that lets you experiment with sample rates and sizes. For FFT and acceleration force calculations. A tool to test the theory in practice, and play with the parameters.

Products like this are essentially a usb soundcard with signal conditioning for an ICP sensor

Finding a widget that converts an ICP sensor to audio line level may be one option then just use a spectrum analyzer app on a tablet.