The projects below show an impression of what I have been working on:
Werkman is one of the largest producers of horseshoes in the world. It is a family business that has been around for more than a hundred years and covers 4 generations. With the emergence of the Internet of Things (IoT) and the increasingly available technologies, the idea was born to create a smart sensor that can support the farrier in his daily work by objectively recording movements.
In this project I provided prototyping and development of the necessary data analysis algorithms, made dashboards for continuous quality monitoring and took care of the integration of the algorithms with third parties.
System on a chip (SoC)
A client within the wearable market came to me with the request to have an algorithm responsible for several integration and filtering tasks that needed to run in the free space of a 512 kB Nordic chip. The free space left was estimated to be around 128 kB. Using Matlab Coder toolbox we could verify and validate the algorithm using tons of data before deploying the final code to the SoC.
This project required out of the box thinking since best practices like vectorization and moving averages which usually would take up much more than 128 kB of space had to be rethought. The end result was a highly tuned algorithm that delivers great results while staying within memory limits and consuming only little battery power.
Combining the pleasant with the technical. With a group of friends we wanted to increase our collective racing talent at the local kart track. To gain insight into our driving skills, we filmed on-board and used an iPhone with sensors. This video and data were then combined into this video.
Thanks to indoor karting track Duiven!
The data in the video was measured with an iPhone and edited in Matlab. The iPhone has a so-called IMU chip on board with 6 (or more) degrees of freedom. An IMU is an acceleration sensor and rotational speed sensor (gyroscope) in one.
Since the iPhone is not neatly mounted on the kart, but was somewhere loose in a jacket pocket, the orientation of the iPhone relative to the kart needed to be determined first.
Next, the measured data must be compensated for the offset of the sensor with respect to the center of the go-kart. Otherwise, the level of acceleration while turning is biased because the phone is relatively on the inside or outside of the turn.
The raw data from the sensor can contain aliasing because no anti-aliasing filters have been applied in the iPhone. Simplistically, aliasing happens when there are more vibrations in the signal than the amount of samples that can be taken. In order to have a usable result, the data was filtered in 3 ways. This is indicated in the video with the 3 balls. Low frequency filtering provides a good indication of the average g-forces in curves (largest ball). High frequency filtering provides a better indication of the bumps occurring in the ride (smallest ball).
In the picture you can see the grip circle of the entire ride. Points to the left or right corresponds to steering to the left or right respectively. Points to the top indicate acceleration and points to the bottom indicate braking. Since only 2 wheels are used for braking and not 4, it can be observed that cornering is more effective in changing momentum than braking in this case.
I can't say too much yet, but for another client I am working on a project where strengths are measured on athletes. This project includes force sensors, acceleration sensors and gyroscopes.
Force sensors consist of strain gauges that convert physical measurements into electrical currents. By converting these streams into numbers and applying the correct algorithms to them, the athletes can receive feedback about their training.
In this project I also combine Matlab and Matlab C-coder to come up with a good solution together with the subcontractors. In addition, this project also consists of various tools to easily perform data analysis and even to automate the production processes.
Uncertainty of mean
Since 2011 I have been working with colleagues at my previous employer to develop a piece of applied mathematics under the heading 'uncertainty of statistical parameters'. Even now I freelance, we are still writing on publications on this subject.
It is a complicated piece of mathematics to answer an apparently simple question: "If we have a measurement with a lot of noise and disturbances in the measurement signal, how random is the average, we get?" In practice you ask this question especially if the measurement in question is very expensive and you do not easily repeat a measurement 10 times.
This perhaps very abstract question has been examined very well by us and many interesting things have come out of it. Not only were we able to answer the question fully with a theoretical derivation and a practical elaboration, but we also discovered that we had a new and much sharper definition to prove whether a signal is slowly drifting or not. In practice this means that a minimum calculation or test time can be established.
Do you like to know more? We have published a journal paper on this topic and on Researchgate you will find a project page to this topic.