Walk through a modern factory on a weekday and it is easy to see it as a set of machines: presses, robots, conveyors, ovens and test stations, each doing its own job. Spend a week with the maintenance team and a different picture appears. The plant behaves like a body. Its production areas are organs, its utilities are circulation, and its control systems keep a hundred variables steady in the way a body keeps its temperature and blood pressure steady. When one system weakens, the effects travel.
The metaphor is useful because it changes where you look. This article uses it to map how a factory's systems depend on one another, to explain why problems in the least visible systems spread the furthest, and to describe what it would take for a plant to sense its own condition rather than only its output.
Organs, circulation and nerves
A factory's production areas are the obvious organs. A press line forms parts, an assembly area joins them, a paint shop protects them and a test area proves they work. Each has its own equipment, its own specialists and its own measures of performance.
Around them runs a circulatory system that most visitors never notice. Electrical distribution feeds every motor. Compressed air drives tools, actuators and pneumatic conveyors. Cooling water carries heat away from presses, welders, furnaces and hydraulic units. HVAC holds temperature and humidity in the paint shop and clean areas within tight limits. Vacuum systems, process water and steam serve particular industries.
The nervous system is the plant's sensing and control. Thousands of sensors measure positions, temperatures, pressures and speeds. Programmable controllers close loops many times a second. Supervisory systems show operators what is happening, and manufacturing execution systems track orders, quality and genealogy. The ISA-95 standard describes this as a hierarchy of levels, from the physical process at level 0 through sensing and control at levels 1 and 2, to operations management at level 3 and business planning at level 4.
That nervous system is very good at one thing: keeping the process on target. It is much weaker at sensing the condition of the body itself. A controller knows the pressure in an air header. It usually does not know that the compressor producing that pressure has a bearing in the early stages of failure. This is the gap we described in why the physical world needs a nervous system: plenty of signals about what the process is doing, few about how the machines doing it are holding up.
Mapping the dependencies
The simplest way to see a factory as a living system is to draw who depends on whom. The map below is a simplified version for a discrete manufacturing plant.
Count the lines leaving each box. In this simplified plant, power feeds three systems directly and everything else indirectly. Compressed air feeds assembly and paint. Cooling water feeds the press line and the paint shop. Each of those feeds test and pack, which feeds shipping. A failure in the press line stops one stream. A failure in the compressed air system stops two production areas at once and starves the area downstream of both.
Real plants are more tangled than this. A large site might have several compressor houses, multiple cooling loops, dozens of air handling units and hundreds of drives, with cross-connections that grew over decades of expansion. The principle holds: the systems with the most connections are usually the ones the fewest people watch.
A worked example: one compressor, three areas
Consider what happens when a compressor in the air system above begins to fail. The illustration uses round numbers for clarity.
The plant runs three compressors, two on duty and one in standby, feeding a common header at a set pressure. One duty compressor develops a cooling problem: its oil cooler is fouling, so it runs hotter each week. Its controller compensates, and header pressure stays on target. Nothing appears wrong on any production screen.
On a hot afternoon the struggling compressor trips on high temperature. The standby starts, but it takes time to load, and in the meantime header pressure dips. Pneumatic clamps in assembly lose force, and several stations fault out. In the paint shop, air-driven pumps and atomizers waver, and a batch of parts is painted out of specification. Test and pack runs out of work within half an hour.
| Area | Effect | Recovery |
|---|---|---|
| Compressor house | One unit tripped, standby now on duty with no spare | Clean cooler, restart, return to standby arrangement |
| Assembly | Stations faulted on low clamp pressure | Reset, check part seating, restart |
| Paint | Parts out of specification during the pressure dip | Inspect, strip and repaint or scrap |
| Test and pack | Starved of parts | Resumes when upstream recovers |
| Shipping | Afternoon dispatch delayed | Overtime or a missed delivery window |
None of the production areas did anything wrong. The root cause was a slow change in a support machine that the plant's control system was designed to compensate for, and did, until it could not. Siemens' True Cost of Downtime 2024 report found that a typical large plant in the sectors it surveyed lost about 27 hours a month to unplanned downtime in 2023, down from 39 in 2019, while the cost of each hour rose. Events like this one are a large part of those hours, and they are hard to attribute because the symptom appears far from the cause.
Homeostasis and its blind spot
Biologists use the word homeostasis for the way a body holds its internal conditions steady. A factory does something similar. Control loops hold temperatures, pressures, speeds and flows on target despite changes in load and environment.
Homeostasis has a cost that is easy to miss. A system that compensates for decline also hides it. The compressor with the fouled cooler worked harder to deliver the same pressure, and the output looked perfect until the moment it failed. A cooling pump with a worn impeller runs longer hours to hold a setpoint. A fan with a loose belt is driven harder by its variable speed drive. In each case the process variable stays steady while the machine's condition gets worse.
Bodies solve this with a second kind of sensing. You feel fatigue, pain and fever, signals about the state of the body rather than the outside world. A factory that only measures its process variables is like a person who can see and hear perfectly but feels nothing. It finds out about an injury when a limb stops working.
What a plant that can feel itself looks like
Giving a plant that second sense means measuring the condition of machines directly: vibration, sound, temperature and electrical behavior at the motors, pumps, compressors, fans and gearboxes that do the work. It is a different job from process instrumentation, and it is usually done in different ways.
Three features matter most in practice.
Coverage of support systems. Plants usually protect their biggest production machines with permanent monitoring. The compressors, cooling pumps, air handlers and auxiliary drives behind them are much less likely to be watched continuously, even though, as the map shows, they have the widest reach.
Several kinds of evidence. A fouled cooler shows up first as temperature. A bearing defect shows up first in vibration and high frequency sound. Winding problems in a motor show up in its electrical behavior. Reading several signals together, as we explained in why multimodal sensing matters, turns a vague anomaly into a specific finding.
Context from the process. Condition data mean more alongside operating data. A compressor that runs hotter at full load on a hot day is normal. One that runs hotter at the same load and ambient as last month is not. Linking the condition of machines to what the process was asking of them is what lets a system tell the difference.
From connected to self-aware
Factories have spent two decades getting connected. Controllers talk to supervisory systems, supervisory systems talk to execution systems, and execution systems talk to business planning. The International Federation of Robotics reported in 2024 that more than four million industrial robots were working in factories worldwide, each one a richly instrumented machine. By most measures, the modern factory is full of data.
Most of that data describes what the plant is producing. Far less describes the health of the machines doing the producing, especially the ones in the basement, on the roof and behind the walls. Closing that gap is the purpose of manufacturing machine intelligence: giving the plant a sense of its own condition, so that a slow change in a compressor is noticed as a change in a compressor, weeks before it shows up as a delayed shipment.
Where to start
The living system view suggests a practical order of work.
- Draw the dependency map for your plant, even roughly, and count the connections leaving each system.
- List the machines inside the most connected systems: compressors, cooling pumps, main air handlers, key drives.
- For each, ask how its condition is known today. If the answer is a monthly route or a test run, it is a candidate for continuous monitoring.
- Agree in advance who acts on a finding about a support system, since those machines often sit between facilities and production teams.
A plant that knows the condition of its most connected machines does not stop having failures. It has fewer surprises, and the surprises it does have arrive with more warning and a clearer explanation.
