Instrument makers: point-by-point calibration costs bench hours

Temperature compensation is calibrated point by point in a chamber

What happens. Every device goes into a climate chamber, is held at several temperatures until it settles, and the correction values are written to its memory. The chamber is the bottleneck of the line, and the same procedure is repeated for each device, although the devices behave similarly.

What it costs. Test-bench hours per device, chamber capacity that limits the output, and a calibration time that grows with each added temperature point or each new variant.

Why the usual fix fails. More chambers multiply the cost. Fewer points shorten the time and raise the residual error. A table of corrections from one reference device does not match the scatter of the individual device.

How a fixed-price remote diagnosis checks it. From calibration records of a few hundred devices and the raw sensor signals I check how much of the behaviour a physical model with few parameters can explain and how many points per device remain necessary. The fixed-price remote diagnosis answers whether the calibration can be shortened without losing accuracy in the field.

Background: Replacing hand-tuned parameter tables by a model

Temperature errors limit the uncertainty of coordinate measuring

What happens. A coordinate measuring machine and the part grow or shrink with the temperature in the hall. The measured value changes through the day, though the part has not changed. The room is kept at a set temperature, but the temperature at the machine, the part and the scales is never exactly that.

What it costs. A measurement uncertainty that is larger than the machine could achieve, waiting time to let the part settle in the room, and a cleanroom or a conditioned room which is costly to run.

Why the usual fix fails. A better room control reduces swings, but cannot reach the part. A compensation with one temperature sensor assumes one temperature for the whole machine. Waiting hours for the thermal equilibrium costs measuring time.

How a fixed-price remote diagnosis checks it. From temperature sensor data on the machine and the part, and repeated measurements of a reference part over the day, I check how much of the error a model of the thermal behaviour can remove. The fixed-price remote diagnosis answers whether a compensation with a model can reach a lower uncertainty, or where the machine design sets the limit.

Background: Computing unmeasured states for control

Power electronics and control loops inside instruments need rare engineers

What happens. Modern instruments have fast control loops inside: power stages, heaters, lasers, detector cooling. They must be stable, quiet and small, on a microcontroller. The people who combine the electronics and the control design are rare, and the developer of the instrument often does the loops on the side.

What it costs. Development time lost in tuning loops, noise or drift in the measured value from a mediocre loop and a delay for the next instrument generation.

Why the usual fix fails. A hire with both skills takes long to find. Using standard controller blocks from the library gets a loop working, not a good one. Outsourcing the whole device development is too much for one loop.

How a fixed-price remote diagnosis checks it. From the schematic, loop recordings and the specification of the instrument I check which control task can be separated, modelled and handed over with the code and the description your team can maintain. The fixed-price remote diagnosis shows what a delivered controller module would look like.

Background: Predictive control on embedded hardware

Inspection tools need models of the sensor and signal chain

What happens. A tool that travels through a pipe or along a surface measures indirectly: magnetic, ultrasonic or other signals. The signal is changed by the sensor, the electronics, the speed and the material. The interpretation of a signal as a defect depends on all of that, and engineers fit each case by experience.

What it costs. Defects that are missed or reported falsely, repeated inspection runs, and a long time to qualify each new sensor or tool version.

Why the usual fix fails. Recording more data does not tell which part of the signal is the defect and which is the tool. A rule on the signal amplitude works at one speed. Training a detection method on examples from one tool fails on the next.

How a fixed-price remote diagnosis checks it. From recorded runs and the known defects I check how well a model of the sensor and the signal chain explains the signals, and where an estimate of the defect size can be made more robust. The fixed-price remote diagnosis answers what such a model would add to your evaluation.

Background: Computing unmeasured states for control

Noisy body-point data must become a robust biomechanical result

What happens. Cameras or sensors on the body give positions of joints and limbs. The points jump, vanish behind other parts and carry noise. From them, a system must compute angles, speeds or forces that a coach or a doctor can use, and the same movement gives slightly different numbers each time.

What it costs. Results that nobody trusts, repeated measurement sessions, and manual cleaning of the data by a specialist before every analysis.

Why the usual fix fails. Smoothing the points removes noise and also the fast movements that matter. Fixed thresholds break with another person or another camera angle. A learned method without the physical constraints of the body returns impossible values now and then.

How a fixed-price remote diagnosis checks it. From recordings of your sessions I check how a model of the body, with its limits, can estimate the movement from noisy points and how stable the result is. The fixed-price remote diagnosis answers whether an estimate with a physical model gives a more robust output than the present filter.

Background: Computing unmeasured states for control