This Lab was part of the Hands-on Labs at PI Users Conference in 2016 in San Francisco and in Berlin. The Lab manual is here; the manual is intended for an instructor led interactive workshop – so you may find the written content short on some of the explanations.
You can access the VM used for the lab via Home Page - OSIsoft Learning and, look under the Virtual Learning Environment - TechCon 2016 Labs.
Below is an extract from the Intro section of the Lab manual:
In this Lab, we will walk through a statistical approach to predict equipment failure. We will use sensor data taken from 100 engines prior to engine failure. We will use the R scripting language to develop a predictive modelling equation that can be run in real time to predict engine failure before it actually happens.
In a deployment with about 100 engines which are similar, sensor data such as rpm, burner fuel/air ratio, pressure at fan inlet, and twenty other measurements plus settings for each engine – for a total of about 2000 tags – are available. On average, an engine fails after 206 cycles, but it varies widely - from about 130 to 360 cycles.
Using an open source tool such as R for machine learning, you will create a multivariate model (Principal Components) to predict engine failures within approximately a 15 cycle window before they fail. The lab will walk through the end-to-end data science process – preparing the dataset, visually exploring it, partitioning the data for training and testing, model development and validating the models using previously unseen data, and finally deploying the model with AF asset analytics for predictive maintenance.
The lab consists of:
• Part I - Extract data from the PI System for a set of Engines’ operation parameters into a text file using PI Integrator for Business Analytics (BA)
• Part II - Load the exported dataset from PI System into R, and utilize the functionality in R to perform machine learning on the dataset for prediction of engine failures.
• Part III - Convert the output prediction logic from R into an equivalent equation in PI Analytics for continuous execution and prediction (online scoring).
The data used for the lab has been adapted from: A. Saxena, K. Goebel, D. Simon, and N. Eklund, “Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation”, in the Proceedings of the Ist International Conference on Prognostics and Health Management (PHM08), Denver CO, Oct 2008. https://ti.arc.nasa.gov/publications/154/download
Another machine learning use case from TechCon 2017 re. anomaly detection in HVAC air handler operations is here.
And, the machine learning use case from PI World 2018 re. golden batch run in a feed dryer is here.