A few years ago, Flintmore got the opportunity to put our IMU-based motion tracking to the test in the harshest of applications, marine power. It is well known that everything put in the sea eventually breaks, however, anything complicated, especially if it is new, breaks quickly.
Two off-the-shelf MEMS IMU units, hooked up to long-life batteries within marine-resistant casings were deployed aboard an experimental wave energy device in the North Sea, completely sealed. For three months these IMUs gathered data uninterrupted, independent of any other power source and external data gathering.
Once the device was towed ashore, these IMU units were retrieved, opened up and the data analyzed. Both had performed brilliantly and were still working as new. Surface level analysis revealed waveforms that we could correlate to historical weather patterns at the location in question. Whilst this gave us increased confidence in the collected data, the usefulness of this surface information is limited.
Utilising our experience with IMU data analysis we fed the data into models that produced position and orientation from the IMU.
IMU analysis is all about constraints to minimise the effect of compound errors. Recent research into IMU analysis has demonstrated that if start and end positions are known then velocity and trajectory can be reconstructed with extreme accuracy just from a low-cost MEMS accelerometer, an extremely impressive feat. Position estimation from accelerometers usually results in the target object flying off into the far distance, integrated errors build up too quickly, preventing this whilst still getting accurate information from which conclusions and responses can be formulated is the key challenge.
To counter this inertial navigation systems utilise extremely precise sensors on an inertial table. Even these high-cost precision instruments can build up errors measured in kilometres over a matter of hours. We were working with low-cost MEMS accelerometers and gyros, the same as can be found in modern consumer electronics meaning our application of limits to constrain the solution is even more important.
As the ocean surface is constantly moving, particularly in this case, the exact position could never be known or verified. GPS can give a good estimate, but the accuracy available to commercial users still leaves a large margin of uncertainty so it cannot provide a hard constraint on position.
Sensor fusion in this case provides cross constraints for other sensors, each one compensating for the uncertainty of the other. As the device was entirely isolated from outside signals, in this case, we had no GPS to provide outer bounds on position.
IMU units often include magnetometers, this was the case with the ones we deployed, whilst it doesn’t give a position, in the absence of an external magnetic field, it gives a good cross-reference against the gyroscope for the pointing of the IMU.
In this case, the wave energy device was made of steel with a large generator inside creating a magnetic field, this limited the use of the magnetometer. However, to trace the movement of the device, estimate response to waves, and the pitch and roll of the device, the lack of a clear point to ‘North’ was not an insurmountable setback. Changes in the magnetic field can still be used to verify pointing in fusion with the gyroscope.
The accelerometer provides one useful piece of information for cross-reference; the gravity vector. In relatively calm seas, this means that the orientation can be cross-referenced against the gravity vector, which, for all intents and purposes, points straight down.
With the necessary components to estimate orientation accurately; the next challenge to tackle was position. As already mentioned, no access to GPS means that we have no way to externally verify the position of the device. With no access to a data-fusion method, and with no known fixed points in the data to constrain the solution, theoretical constraints were our next choice in this situation.
When performing data analysis it is always key to consider what important information needs to be derived from the data. There is no point in us trying to work out the position of the device after three days, it is both nigh on impossible with the data we had available, but also what use would it be? The short-term positional changes, the response to waves and gusts, this is what is of interest, we can discard long term drift as we are less interested in this. The fact that the device is tethered to the seabed also minimises this effect.
Taking these aims into account we can begin to work on a solution. Because of the tether and the fact that long-term changes in position are not important, we can disregard these changes, essentially, we can apply a constraint that the average horizontal velocity over an hour is zero. This may not be the actual case due to slack and flex in the tether but allows us to isolate the information we actually want from the data.
The sea level variation is also limited to barometric variations coupled with the variation due to wave crest and trough. We can also apply this constraint, such that, practically speaking, the average vertical velocity over an hour is going to be close to zero.
In this case, the clever application of these constraints allowed us to capture the response to waves with high fidelity allowing us to derive key bits of information that were made available to the client giving valuable insight into the behaviour of the machine. The total cost of the equipment was essentially nothing in engineering terms. Whilst better results could have been achieved by the deployment of a high-precision inertial navigation unit, the insight offered at such a low cost is an achievement we are very proud of.
We have continued to develop this technology, a successor forms key parts of our fall detection algorithms for our next-generation intelligent fall detector project.
Do you have a similar problem or project to tackle? Get in contact today and we can discuss how Flintmore can help you.
The geometry of the device in the sample footage has been altered to preserve the anonymity of the project.
