Sample Predictive Analytics for plant maintenance
I am new to PA.We need to show some POC to our client in PA based on plant miantenance so our case we need to know when breakdown happens in plant which equipments under go more failure frequently so we can plan for them based on analysis. Also we can get analysis of plant breakdown so we can avoid planned maintenance and do preditcive maintenance,
So we can download data from standard tables like QEML ,etc - please cna you advice via automated analytics which alog we can use to generate some sample kind of analytics and show.
Initialy we can load via flat file but later on we can automated it via connecting on som DB and then do analysis, Please let me know if you need any more information.
Pierpaolo VEZZOSI replied
I'd like to go back to your original question about predictive maintenance.
If I understand correctly you have data about machines, you know when they breakdown and hopefully you have measurements on the machine and its environment before and at breakdown time.
If this is the situation then there are multiple approaches which you can take depending on the exact question you want to answer and the input data you have.
Some examples (by the way I would suggest you try Automated Analytics first):
+ If you want to know IF a machine is likely to undergo a breakdown and WHY, you can run a classification model (supposing you have a column in your input dataset saying if yes or no a machine failed in the past). From that information you might be able to predict in advance that machines will fail based on the newer machine data and environment data. You might also be able to understand what you can change in your system to reduce breakdowns. Read this blog showing how results of a classification can be used to take strategic decisions: Exploratory analytics with SAP Predictive Analytics 2.2: take strategic decisions based on your data. The example here is about improving your business but you can adopt the same concepts to predictive maintenance.
+ If you want to know WHEN a machine is likely to break then you can try with a timeseries analysis where you add as much extra predictive variables as possible even if I am not sure you might have a lot of environment data for the future. But keep in mind that even if you can do short term predictions, they might be good enough if you have the ability to react quickly to an alert.
+ If you want to group your machines to keep an eye on those who are more likely to break, you could do a supervised clustering using the breakdown (yes/no) variable as a target. This way you can understand the typical profile of machines at risk.
+ Finally be aware that in Automated Analytics you have tools to increase the data available for your model like the Event Log Analysis or the Sequence Analysis. Those tools help you automatically extract information 'hidden' in your dataset by calculating new aggregates or showing the sequences of events which brought to a failure (or not).
There are obviously other ways to run your predictive maintenance, these are just a few hints.
I haven't given you a solution (because there is no single answer to your question) but just a few tracks which you could explore. Maybe one of them suits your needs