With many years of engineering experience in the process industries, we know determining the cause of defects can be very difficult. Is it a formulation problem? Are the process conditions to blame? A combination of them? Perhaps it's a contaminant in a raw material? Or maybe the equipment was not properly cleaned? These pesky defects often are given interesting names, such as "chicken tracks", "ghosts", "slime holes" and the like.
One approach you can take with our tools is to use a wide variety of potential causes; process conditions (temperatures, pressures, rates and states), materials (quantity, lots, vendors) and a sampling of cases where the defect exists and where it does not. Then you model using the possible causes to predict defect/no defect and look at the sensitivity analysis of what is driving the defect. You can eliminate those variables that do not have much effect and repeat. As you do this you zero-in on possible causes. Perhaps you will indeed find the cause or you will at least get clues as to where the cause exists. A group of process conditions around a unit operation or workcenter may tend to float to the top of causal factors.
If the above activity does indeed result in a model that locks on to the cause of the defect, you can put that model on line and see the probability of the defect occurring as it occurs and before seeing it in the resulting product. This can save a lot of money enabling you to make an adjustment or stop producing bad product. If you set a threshold on that probability, in the Intellect Server you can send an alert to the control system and thus the operator or send an email or text to product or process engineers to make them aware that the problem is likely occurring.
Benefits achieved include:
Causes of defects are hard to determine.
Use our tools in a "data discovery" process to track down causes.
Put the models on line to estimate likelihood.
Send alerts, emails and/or texts when they are probable.
A critical benefit for a difficult challenge.
Desktop Analysis Only Option: