Cavity pressure is in your historian. It isn't in your control loop.

A recent Springer chapter on adaptive process control using cavity pressure sensors walks through how in-mold pressure traces are being wired into closed-loop schemes that compensate shot-to-shot, cavity-to-cavity. Reading it against what we actually see on molding floors, the gap is obvious. The sensors are there. The traces are logged. The loop is almost never closed.

What's already on the machine

Walk a modern molding cell and the signal density is high: clamp tonnage, screw position and velocity, injection pressure, hold pressure, cushion, backpressure, melt and barrel zones, mold surface temps, and, on tools instrumented for it, cavity pressure per cavity. Guidewheel's 2026 injection molding benchmarks report cycle-time variability and short-stop distributions across a large installed base, and the pattern tracks what you'd expect: the machines are talking. Somebody is listening at the OEE dashboard level.

Nobody is listening at the shot level.

What "using it" actually means

There are three honest tiers of cavity pressure use:

1. Logged. It lands in the historian. It gets looked at during a defect investigation, maybe. Most molders stop here. 2. Alarmed. Peak pressure or integral goes outside a band, the shot gets flagged or diverted. Useful. Still reactive. 3. Controlled. Switchover from injection to hold is triggered by cavity pressure, not screw position or time. Hold profile adapts to the previous shot's integral. This is what the adaptive control literature is actually talking about.

Most plants we walk are at tier one with ambitions of tier two. The jump to tier three is not a sensor problem. It's an integration problem between the press controller, whatever is reading the strain gauge or piezo, and someone willing to let a model touch a setpoint.

Why it matters for the numbers that matter

A case study on parameter design for defect reduction in injection molding walks through structured DOE on process inputs and ties specific variables to specific defect modes. The lesson that generalizes: a small number of process variables dominate the defect rate, and cavity pressure characteristics (peak, integral, time-to-peak) are usually in that short list for short-shots, flash, sink, and dimensional drift.

If you are running a multi-cavity tool and cycle time varies cavity to cavity, your scrap number has a cavity-pressure explanation sitting in the historian right now. Same for the slow drift you see three hours into a long run as the tool heats up and the hold profile stops matching reality.

The practical move

Before buying another sensor or another dashboard, pull a week of cavity pressure traces for your worst-performing tool and overlay them with the scrap codes from that same week. In our experience, the correlation is embarrassing. Not because the data was hard to get. Because nobody had asked the question yet.

That's the boring work. That's also where the defect rate actually moves.