Damage detection of roads using piezo sensors

You'll need to tell us a bit more.

What sort of piezo electric sensor ?

I know of force, displacement, acceleration, pressure, microphones, acoustic emitters and more.

Why don't you explain what your project is, what you have done so dar and where you are stuck ?

MK

We have taken some readings and waveforms with this piezo sensor on a road using digital oscilloscope....but we don't know how to interpret those readings....our target is to detect how damaged the road is from inside by using these readings we received....we need basically all guidance about instrument, sensor, method....or if we get any good research paper it will be a great help....thank you

Interesting project. There are lots of causes of road damage, but the result is usually a bumpier road. When vehicles go over a bump, their suspension and tires oscillate. You may be able to detect this oscillation and its amplitude with this sensor. It will take some experimentation to get the sensor coupled to the road properly without damaging the road and more experimentation to make sense of the data. You may find some frequencies correlate better to road damage. This is a good application for machine learning, but you probably need multiple installations.

I'm guessing that this is a college project.

You would help yourself a lot if you write down in advance exactly what you are trying to do - this would help you start to work out how you might do it.

From what you've said so far, I think you are attempting to measure the quality of the surface of the road by making acceleration measurements inside a vehicle driving along the road.

This is a staggeringly difficult problem to solve analytically.

Let's assume the vehicle has 4 wheels and the input displacement from the road surface is filtered by the suspension system which will include tyres, springs, dampers etc. The vehicle body, and your accelerometer will experience pitch, roll,yaw  and vertical acceleration accordng to the sum of these inputs. The input from a given road feature will often stimulate the system twice (as front then rear wheels pass over it).

The classical way to solve this problem is to make a mathematcial whole vehicle model and use this as a tool to help relate road surface to accelerometer signal. Frequently at this point the conclusion is that the motion analysis is so complex that it isn't possible to do it in real time, often more measurment data like multiple accelerometers and gyros is required.

Of course you are attempting this with a piezo electric sounder disk which will make a not very good accelerometer.

Jan Cumps has blogged on E14 about doing this - it will help you unbderstand the lmitations of the sounder disk accelerometer if you read his blogs.

Long experience tells me that you will not be able to make a good model of the vehicle (such projects require several skilled PHD+ level engineers with adequate resources in terms of computing and software tools).

You could still do something.

I would avoid the machine learing approach - it will eat up a lot of time and you will have enormous dificulty in gathering training data.

What might work for you is recording the acceleration data over time (simplest is to record peak acceleration seen in each second but better to sample faster if you can) and the position of the vehicle (from GPS I expect). Then survey the road and see how the potholes correlate with the data you have collected. My guess is that you'll be able to see at least the worst potholes from the data. This might be good enough.

Your accelerometer will need a bandwidth from about 0.5Hz to 25Hz  - so you'll need a buffer amplifier (see Jan's blogs) and  a filter.

AN oscilloscope is the wrong instrument for this work. You need to be able to record raw data at maybe 100samples per second and correlate it with GPS data. This isn't too hard - well within Arduino capability.

You don't say what your team's skills are so I can't be more specific.

Once more, I urge you, write a plan, think about what might go wrong and how you can deal with it. Limit the scope of the project to stuff you can actually do.

People on E14 can do much more to help if they have  a bit more to work on.

MK

If you are instrumenting the vehicle instead of the road, there are papers on this. I was a reviewer of a Master's thesis on a similar topic a decade ago. It was aimed more at determining when the vehicle would need maintenance, but the instrumentation would be similar.

 AniketVis , I can point you to some of the posts that Michael discusses above. They will help you with 3 things:

• turn the piezo signal into a safe signal, and condition / buffer it (front end) :  Piezo disk as Vibration Sensor: input buffer and filter 
• a possible data acquisition system that allows you to sample the signal fast and for a long time
• the calculations to turn the samples into acceleration info and frequency info (fft)

michaelkellett said:

What might work for you is recording the acceleration data over time (simplest is to record peak acceleration seen in each second but better to sample faster if you can) and the position of the vehicle (from GPS I expect). Then survey the road and see how the potholes correlate with the data you have collected. My guess is that you'll be able to see at least the worst potholes from the data. This might be good enough.

If it is for a school project, maybe there's a more economical way:

• mount one or more sensors on a bicycle
• choose a few stretches of calm roads or bike paths that are in good - fair - bad condition
• drive over them at some fairly constant speed, and collect the data. Repeat 10s of times, store the data
• derive frequency and acceleration info. These are achievable goals and show 2 interesting aspects of what's happening with the sensors (I learned from martinvalencia  that these are used in industry to see changes in vibration trends). 
• learn from the data, and differences based on road profiles
• check if this can be applied on a different stretch of road, and if the algorithms correctly recognise the road conditions

You'd have to do this under similar temperature conditions, because both roads and tires behave significantly different under warm and cold circumstances.

michaelkellett , you mention that AI is a difficult path. But maybe the data collected in the fashion above can become training data?

michaelkellett said:

What might work for you is recording the acceleration data over time (simplest is to record peak acceleration seen in each second but better to sample faster if you can) and the position of the vehicle (from GPS I expect). Then survey the road and see how the potholes correlate with the data you have collected. My guess is that you'll be able to see at least the worst potholes from the data. This might be good enough.

If it is for a school project, maybe there's a more economical way:

• mount one or more sensors on a bicycle
• choose a few stretches of calm roads or bike paths that are in good - fair - bad condition
• drive over them at some fairly constant speed, and collect the data. Repeat 10s of times, store the data
• derive frequency and acceleration info. These are achievable goals and show 2 interesting aspects of what's happening with the sensors (I learned from martinvalencia  that these are used in industry to see changes in vibration trends). 
• learn from the data, and differences based on road profiles
• check if this can be applied on a different stretch of road, and if the algorithms correctly recognise the road conditions

You'd have to do this under similar temperature conditions, because both roads and tires behave significantly different under warm and cold circumstances.

michaelkellett , you mention that AI is a difficult path. But maybe the data collected in the fashion above can become training data?

If you want something in-between then perhaps mount a '5th wheel sensor' on the back of the car... 

http://www.gmheng.com/fifth_wheel.php

Bicycle would be much cheaper and easier !

Even simpler would be to add the accelerometer to one of those wheel measuring things road surveyors use for distance.

www.ebay.co.uk/.../262890794649

It warned against AI.Machine Learning because pretty much every small project I've seen that tries it ends spending most of its resource wrestling with the AI and far too little looking at the essential basics like getting the data out of sensors and understanding the system dynamics.

MK