Data Driven Batch and Furnace Part 1
Batch and furnace generates data and has for many years. It is one of the first areas of a glass plant that was computerized and has long dealt with numbers and science than other more “artistic” parts of the manufacturing process. Back in the 1970s engineers were looking for ways to automate furnace operation and manual batching has long since been replaced, first with relay logic and then with programmable controllers. From my own experience, eliminating manual batch weighments, eliminated chronic cord problems in the late 1980’s. Data in the melting was always available to be scrutinized , sometime to a fault. Almost any problem in the glass plant could be traced back to batch and furnace operations, rightly or wrongly. Just because there is data being generated, that doesn’t mean its driving the process. Only recently can we harness the power of modern data analysis to understand and improve the batch and furnace operation.
Lets start with the batch house. Most glass properties can be traced back to composition. If there is variation in the weighments of raw materials and deviation from the recipe, is it causing other problems? With the automation of the batch weighing alarming has been set to alert the operator of misweighments. How are these alarms set? Are they based on scale capability, operator opinion or science? I can only assume it has been down by tradition and by what seems right. We now have data analysis tools to use science to set weighment limits.
In the early 2000’s researchers produced a database of glass properties based on composition. High level number crunching has made it possible to calculate glass viscosity based glass chemistry. This is now openly available. If we look at a simplified soda lime batch and calculate the change in glass chemistry from misweighments and the subsequent changes in the viscosity curve of the final glass we can see the the tolerances allowed to prevent a forming upset. In the example I look at over weighment but the same could be done for the under.
What this shows is that a glass batch tolerates sand and feldspar misweighments more than modifiers like soda ash and limestone. Taking this data, we can say that a +20 pound limit on the sand scale will have a minimal effect on forming properties, since softening point would vary 1°C and working range by 1.2°C. This is not the case with the soda ash as it would see similar change from a +6 lb. change. We would define a 6 sigma scale process by taking the standard deviation of each scale and comparing it to the tolerances of the batch. The scale alarm limits should then be adjusted to reflect tolerance limits.
How do we use this information to practically evaluate a batch house? On a daily basis we can track the variation of each material. If the standard deviation calculated from these results in < 1/6 of the tolerable limits, we have 6 sigma control. In operation was can set scale parameters to get within these ranges. As in the above example, soda ash should needs tighter tolerances than sand, but in practice may have higher variation. We must tighten the alarm points and engineer a solution for better control. A possible solution may be changing the weightment method or just slowing the scale. In this case science and data drive the need for improvement.
In my next post I will examine ways that modern data can be used to improve or evaluate furnace operation.