Machine learning is a powerful tool. It can help you predict the outcome of your decisions.
Last year a client discussed a #coal-fired #boiler #fouling problem. The client’s boiler would periodically get fouled with ash, reducing heat transfer from the burning coal to the water (to make steam). This reduced the performance of the boiler, meaning they were getting less steam for each tonne of coal burnt.
The heat rate can be observed via plant instruments. So they know when there is a problem. The issue is that there are multiple potential causes for a drop in heat rate. The main two are boiler fouling and poor quality coal.
So they have a conundrum - do we take the boiler offline for cleaning (with lost production) or not? Or do we wait for coal lab results (which are always many hours in the past).
If they do a clean, they will conclusively know if it was fouling, because they will see an immediate increase in performance.
The decision to clean has been always an intuitive one. This is an excellent candidate for machine learning #classification.
Classification allows us to train models of the plant which predict if the boiler is fouled or not (with a probability estimate). This takes the guesswork out of the decision and improves plant performance.