What we learned building force sensors, written down.
Engineering notes from our bench, one five-minute read at a time. Calibration, decoupling matrices, characterisation, inertial compensation and trusting your data stream.
The Force Sensor series
Published section by sectionA force sensor is only as good as its calibration
What a decoupling matrix actually is, why the numbers on the wire are not yet newtons, and why calibration is a contract rather than a one-off script.
Five sensors, two calibration philosophies Coming next
Why sensors carrying an IMU are calibrated by reorienting gravity, and why the 6-axis sensor still gets dead weights on levers.
Getting a stream you can trust: serial, TCP and the simulator
Framing, checksums, resync through garbage, and why every workflow should run end to end with no hardware attached.
Calibrating with gravity: IMU poses instead of dead weights
The rest gate, cardinal and oblique poses, the three-stage solve, and the rank and condition-number gate before anything is exported.
Dead-weight calibration done properly
Zero capture, lever arms, parasitic moments from lateral forces, and why seven points is the minimum for a 6×7 solve.
Proving the numbers: accuracy, crosstalk, hysteresis and friends
Every metric normalised by full scale, honest fixture coupling, and the green and red lines we actually use.
Removing payload inertia from a moving sensor
A rigid-body physics model against nine machine-learning methods, scored on payloads the model never saw.
Monitor mode: packet counters, noise floors and spectra
Dropped-packet estimates without timestamp guesswork, per-channel histograms, and finding mains pickup in seconds.
Deliverables and regression-locking
What a calibration ships as, and how 250 automated checks keep the software trustworthy.
Why publish this?
Force sensing is taught as textbook physics and practised as tribal knowledge. The gap between the two is where projects lose months. These posts are our attempt to close that gap in public: the methods, the thresholds, the failure modes and the screenshots, from sensors that are actually deployed.
Everything here comes from work by PT ANT Engineering Consultancy, developed in collaboration with KFUPM IRC-IMR. Corrections and questions are welcome via Instagram @antengconsult.