Before the defect appears: what quality prediction really looks like
Final inspection is too late. How to read the early signs of a quality problem in process data.
If a defect was caught at inspection, that is the good outcome. The problem is that by the time it was caught, the material, the time and the energy had already gone into it. Quality prediction is an attempt to change that order — instead of checking after the part is finished, catching it while it is still being made.
From inspection to prediction
Traditional quality control puts its weight on inspecting the finished product. But by the time a defect turns up at inspection it is already too late. And the finished part alone rarely explains why it happened, because the real cause usually sits further upstream in the process.
Quality is already written into the process
Defects rarely appear out of nowhere. Temperature wavers slightly, pressure drifts outside its usual range, a material property shifts a little — the early signs are left in process data first. Quality prediction is the work of reading those signs and warning before the part is finished.
Pinning down a cause takes context
Knowing that “this batch has a high defect rate” is not something you can act on. To narrow the cause you need to know which machine, which material lot, and under what conditions it was made — connected together. With an ontology tying process, equipment, material and inspection results into one structure, the conditions correlated with defects can be identified in the data.
From warning to action
The point of quality prediction is not a defect-rate chart but action that reduces defects. When guidance arrives with its evidence — “current conditions resemble a period that produced defects; check the temperature” — the floor can intervene before the part is finished. Here too, an answer needs its evidence attached before anyone will act on it.
Where to start
Pick the single defect type that occurs most often, or costs the most. Gather the process data likely to relate to it, connect that to final inspection results, and look at which conditions came first. Getting one right is faster than trying to predict every defect from the start.
In closing
Quality prediction, too, runs on well-connected data. How scattered process data gets tied into one body of knowledge is covered in the industrial data integration whitepaper, and the groundwork is in five principles for turning data into an asset. If you are wondering how it would apply to your process, get in touch through contact.