systems theory//engineering patterns//context levels

Context levels are six layers at which any engineering method lives (physics, signal, hardware, algorithm, system and business), each depending on those below it and delivering to those above, and they are used to give every method an address: what it needs from below, what it hands up, where it runs and what data it demands. The point of the address is that errors cross levels: a badly mounted accelerometer (hardware) looks like a fault (algorithm) and triggers a pointless shutdown (business), and no amount of work at the algorithm level fixes it.

Context levels are six layers at which any engineering method lives (physics, signal, hardware, algorithm, system and business), each depending on those below it and delivering to those above, and they are used to give every method an address: what it needs from below, what it hands up, where it runs and what data it demands. The point of the address is that errors cross levels: a badly mounted accelerometer (hardware) looks like a fault (algorithm) and triggers a pointless shutdown (business), and no amount of work at the algorithm level fixes it.

A pump bearing starting to fail, followed through all six, shows the shape.

Level The failing bearing

Physics An outer-race defect strikes once per passing ball, at a rate set by geometry and speed (bearing defect frequency)

Signal High-frequency bursts at that rate, buried in the vibration of everything else

Hardware An accelerometer bolted to the housing, a 24-bit converter at tens of kHz, a gateway that cannot send it all

Algorithm An envelope spectrum and a threshold

System An alarm and a work order that someone has to believe

Business Replacing it at the next stop is cheap; letting it burst in the middle of the season is expensive

Read from physics upward, each level constrains the next. The defect frequency fixes what the signal must contain; the signal fixes the sampling rate and the sensor's mounting; the hardware fixes what the algorithm can compute on site and what must travel; the algorithm's false-alarm rate fixes whether the system's alarm is believed; and the business cost of a missed or a false alarm fixes where the threshold should sit, which reaches all the way back down.

A method is only meaningful in its context, from the physics to the business. Before judging an algorithm, check the levels under it: if the physics leaves no trace in the signal, or the hardware throws the trace away, the algorithm has nothing to work on.

The business level is also a loop, only slow: measure the fleet, estimate its health, decide what to inspect, act with a work order. Observability and delay apply there with other units, and a monthly report is a sensor with 30 days of latency that cannot correct anything that changes in a week (closed-loop system).

Levels give the maturity of a technique its meaning. A method can be mature as an algorithm and immature as a system, because nobody has yet built the alarm handling, the maintenance and the certification around it (technology maturity).

The ladder is also a debugging order. A surprising output is checked from the bottom up (is the physics what we think, is the sensor mounted and sampled right, is the clock right) before the model is retrained, the field version of observability before the algorithm.

The levels are one of the frames collected in engineering patterns.