GPS and swell height - no mention of anchors.

[Buck Turgidson]

Ive posted this before but its worth another look.
waves

 

[[194224]]

............................ The data can be integrated to get distance moved...........

The data can be integrated twice to get distance moved

 

[boomerangben]

Perfect is the enemy of good. OP just wanted a way to measure swell while on board. While interesting, it's unhelpful to assume that his requirement is a perfect representation of the environment. We can get close enough, most of the time with a 9 axis sensor, and ML removes most of the legwork in that. Yes, ML can cause issues for those who don't understand it. For those that do, it'll easily handle this problem, and a 9 axis sensor will probably provide enough feedback to determine direction given sufficient training data.

I have no idea what ML is, but unless it knows the RAOs for the vessel for a variety of headings, wave frequencies and vessel load conditions, how can it determine wave height? It’s not a case of perfect being enemy of good, it’s a case of omission of fundamental physical properties leading to rubbish results.

edited to say: using accelerometer/gyros to determine movement is straightforward, converting that into wave height requires a singular solution ie, a set of movements can only be produced by one particular wave. It maybe that different waves (Height/period/length and direction) may produce the same vessel response.

 

[Refueler]

I'm a little skepticak about the Accelerometer use .....

I have a number of Drones that have extremely sensitive, sophisticated Flight Controllers. These give the ability for the Drone to fly itself great distances to set routes / heights etc and land anywhere set into the plan or back at launch point. Or be flown via Mk1 eyeball or via Viewscreen as FPV

They have :
GPS
Compass
Barometer
6 axis accelerometers

+ of course usually the video TX .... Control RX.

My point is that if the accelerometers were sufficient - why would the FC have Compass and Barometer ? They are not back up - they are the defining items for their info. The accelerometers provide the acceleration info (+ and -) and its direction relative to the forward motion line ... basically to provide the correction to flight control to maintain the flight line.

Similar is in our MEMS Gyro Stabilisers we fit to various models.

Just thinking out loud here ....

 

[LittleSister]

look for the swell, basically do the same, this time for the swell.
The first swell is the easier seen, higher and shorter. than the second ect.
Thee swell is a bit harder to see but you get better with practice. . .

sometimes it’s hard to see accros the crests of the swell, so you sight along the crest. And the direction is +- 90. . .

Watching the wind direction and swell in the trades was a very important . . .

You were doing it all wrong, trying to see the swell.

The swell, and the combination of different swells and their scales, timings and directions, were, apparently, what the traditional navigators of the South Seas used in deciding and maintaining courses to steer when offshore. (And also sensing proximity/direction of islands.) They didn't rely on seeing the swell, but instead knelt in the bows of the ship and used their testicles to sense the various swells! (Works in the dark and fog, too!)

It's most remiss of the RYA not to include this technique in the Yachtmaster Offshore exam syllabus!

This did take a lifetime to learn (along with much other knowledge), hence the role of navigator among the Pacific Islanders ran in families. Of course, the reliance on testicles ruled out women being navigators.

Even in the unlikely event that we could develop such skills, the heavily constrained (laterally and vertically) and strongly tidal seas around the UK, unlike the open Pacific, presumably would mean that swells and their directions were disrupted such that they couldn't be used for steering in these parts.

Interesting to think, though, of how culturally specific are the aspects of nature that we think are important, and the skills, knowledge and technologies we apply to beetle our way around the place.

 

[KeelsonGraham]

With its 9-axis IMU Raymrine’s EV-100 system can give you a value for ‘heave’. Not quite sure what that is but it could approximate to wave height.

 

[srm]

Even in the unlikely event that we could develop such skills, the heavily constrained (laterally and vertically) and strongly tidal seas around the UK, unlike the open Pacific, presumably would mean that swells couldn't be used for steering in these parts.

According to local tradition Shetland fishermen going back beyond 100 years ago, used similar senses to navigate their open Sixareens. They used the "Modder Dia" mother wave, for direction when so far west that they had "rowed Foula down" rowing until the high island to the west of the group was below the horizon.

 

[srm]

I'm a little skepticak about the Accelerometer use .....

Again, going back to the 70's I was deploying and servicing two "Waverider" buoys in the approaches to Sullom Voe, Shetland. According to the manual they used accelerometers to measure wave height and transmitted a continuous output to shore based, analogue, recorders. There was a serious warning not to allow the buoy to rotate rapidly during deployment and recovery as this could tangle the fine suspension wires of the sensors. The UK oil industry association (or suchlike) also ran a large data buoy (or Waverider when the big buoy broke loose) on the edge of the continental shelf west of Shetland at the time.

Sorry, no idea of the maths involved in converting acceleration to wave height but simple tests with the buoy suspended on a bungee cord by a crane showed fairly accurate results.

The Waverider buoy was the industry standard for wave measurement at the time and, much to my surprise, it is still being made, though much more advanced now with different models including a directional Waverider.
Datawell - Home of the Waverider - Datawell

 

[AntarcticPilot]

Perfect is the enemy of good. OP just wanted a way to measure swell while on board. While interesting, it's unhelpful to assume that his requirement is a perfect representation of the environment. We can get close enough, most of the time with a 9 axis sensor, and ML removes most of the legwork in that. Yes, ML can cause issues for those who don't understand it. For those that do, it'll easily handle this problem, and a 9 axis sensor will probably provide enough feedback to determine direction given sufficient training data.

ML can't overcome physical reality, and a sensor at a single position can't determine the direction of wave trains; there is insufficient information. It might manage for short period waves with a strong circular motion at the surface, but it won't for swell, where there is only vertical motion.

 

[boomerangben]

According to local tradition Shetland fishermen going back beyond 100 years ago, used similar senses to navigate their open Sixareens. They used the "Modder Dia" mother wave, for direction when so far west that they had "rowed Foula down" rowing until the high island to the west of the group was below the horizon.

They must have needed testicles too - rowing that far offshore in the North Atlantic!

 

[boomerangben]

With its 9-axis IMU Raymrine’s EV-100 system can give you a value for ‘heave’. Not quite sure what that is but it could approximate to wave height.

Heave is simply pure up and down movement. It may but unlikely match the wave height.

 

[srm]

They must have needed testicles too - rowing that far offshore in the North Atlantic!

Yes, when first in Shetland I met an old retired fisherman. As a boy his first job had been on a sixareen (six oared open boat) fishing west of Foula. He said the best thing that could happen was a fair wind so they would set the lug sail. If the wind was ahead they would row.

 

[lustyd]

I have no idea what ML is, but unless it knows the RAOs for the vessel for a variety of headings, wave frequencies and vessel load conditions, how can it determine wave height? It’s not a case of perfect being enemy of good, it’s a case of omission of fundamental physical properties leading to rubbish results.

edited to say: using accelerometer/gyros to determine movement is straightforward, converting that into wave height requires a singular solution ie, a set of movements can only be produced by one particular wave. It maybe that different waves (Height/period/length and direction) may produce the same vessel response.

Machine Learning. And it doesn't need to know those things specifically, it just needs well tagged data to train a model which will give the appropriate results. It's how almost all of these things work these days and is producing much better results.

ML can't overcome physical reality, and a sensor at a single position can't determine the direction of wave trains; there is insufficient information. It might manage for short period waves with a strong circular motion at the surface, but it won't for swell, where there is only vertical motion.

I feel that your understanding of the subject may be lacking here. The Garmin solution I mentioned earlier can accurately determine a bunch of things from a single position, having been trained with machine learning and well organised data. It's now quite common to use a watch on one wrist to determine length of stride/left right balance, height of bounce etc. while running. All of which are just as complex as this problem, they just need appropriate data to start from which in this instance would be recordings of the motion under varying sea conditions, with those sea conditions tagged against the recordnings. The more data, the more accurate it will be, but it absolutely could overcome a lot of your perceived problems.

 

[lustyd]

There was a serious warning not to allow the buoy to rotate rapidly during deployment and recovery as this could tangle the fine suspension wires of the sensors

Not an issue with modern solid state sensors, which also include gyros in all directions to measure the rotation, compasses in all directions to orient the data, and accelerometers in all directions so that none of the above matters to the measurements.

 

[srm]

and a sensor at a single position can't determine the direction of wave trains;

Perhaps it is possible:

Directional Waverider DWR-G
Measuring waves with GPS
The DWR-G wave buoy measures waves with help of the Global Positioning System (GPS) only. It features a patented algorithm and custom-made GPS receiver. With a single stand-alone GPS receiver it can measure directional waves, up to 100 s periods, without any calibration ever, and even in the middle of the ocean.

• Sensor
  wave motion sensor based on GPS.
• Measures
  – wave height for wave periods of 1.6 to 100 seconds
  1 cm precision in all directions (free floating)
  – wave direction
  – water temperature

Directional Waverider DWR-G - Datawell

 

[AntarcticPilot]

Machine Learning. And it doesn't need to know those things specifically, it just needs well tagged data to train a model which will give the appropriate results. It's how almost all of these things work these days and is producing much better results.


I feel that your understanding of the subject may be lacking here. The Garmin solution I mentioned earlier can accurately determine a bunch of things from a single position, having been trained with machine learning and well organised data. It's now quite common to use a watch on one wrist to determine length of stride/left right balance, height of bounce etc. while running. All of which are just as complex as this problem, they just need appropriate data to start from which in this instance would be recordings of the motion under varying sea conditions, with those sea conditions tagged against the recordnings. The more data, the more accurate it will be, but it absolutely could overcome a lot of your perceived problems.

I often see several papers a week that depend on Machine Learning and may have a clearer idea of its limits than you do. It is amazing what it can do, but wrongly applied it can produce utterly spurious results. I rather like Flawed Data! A pre-requisite is a LARGE training dataset, which includes ground truth for examples of all possible scenarios. Indeed, most papers and proposals I see are mainly concerned with acquiring such training data and incorporating it in the ML model (both non-trivial when considering distributed, mobile applications). It also requires fairly well-constrained models for the outputs. Using your own example, all it is looking for is pace-related data, which is a fairly small and well-defined list of parameters, in a situation where gathering the necessary training data is fairly straightforward..

 

[boomerangben]

Machine Learning. And it doesn't need to know those things specifically, it just needs well tagged data to train a model which will give the appropriate results. It's how almost all of these things work these days and is producing much better results.


I feel that your understanding of the subject may be lacking here. The Garmin solution I mentioned earlier can accurately determine a bunch of things from a single position, having been trained with machine learning and well organised data. It's now quite common to use a watch on one wrist to determine length of stride/left right balance, height of bounce etc. while running. All of which are just as complex as this problem, they just need appropriate data to start from which in this instance would be recordings of the motion under varying sea conditions, with those sea conditions tagged against the recordnings. The more data, the more accurate it will be, but it absolutely could overcome a lot of your perceived problems.

Absolutely possible then. But as you point out, you need to take the boat out in known conditions to enable the machine learning to learn in the first place, which means knowing the seastate etc. or you could get computer generated RAOs for your hull shape. It’s certainly possible but is it viable for us humble boaters?

 

[AntarcticPilot]

Perhaps it is possible:

Directional Waverider DWR-G
Measuring waves with GPS
The DWR-G wave buoy measures waves with help of the Global Positioning System (GPS) only. It features a patented algorithm and custom-made GPS receiver. With a single stand-alone GPS receiver it can measure directional waves, up to 100 s periods, without any calibration ever, and even in the middle of the ocean.

• Sensor
  wave motion sensor based on GPS.
• Measures
  – wave height for wave periods of 1.6 to 100 seconds
  1 cm precision in all directions (free floating)
  – wave direction
  – water temperature

Directional Waverider DWR-G - Datawell

That's interesting. I presume it must be breaking the measurements down into rotational components, and using that to determine directions and amplitudes. I can see that working for short wavelengths , but not for long swell waves where the motion is predominantly vertical.

 

[lustyd]

Absolutely possible then. But as you point out, you need to take the boat out in known conditions to enable the machine learning to learn in the first place, which means knowing the seastate etc. or you could get computer generated RAOs for your hull shape. It’s certainly possible but is it viable for us humble boaters?

I didn't mean take that specific boat out. You take lots of boats out in lots of conditions and tag the data well (which may include recording other data like wind etc.). The end user would just be able to install the sensor and use the model, as with Garmin watches. I didn't have to teach my watch anything, although it does learn and improve over time.

 

[lustyd]

I often see several papers a week that depend on Machine Learning and may have a clearer idea of its limits than you do

Maybe, although I do do this for a living at one of the largest tech companies on the planet so I feel like my finger is somewhat near the pulse on the subject.

Using your own example, all it is looking for is pace-related data, which is a fairly small and well-defined list of parameters

On the contrary, it's looking for about 10 different outputs and had hundreds of input parameters. This is where ML really shines, and humans don't, you just need the right person in the chair to get the ML working, and those people are vastly outnumbered by people pretending to know about ML.