We are drowning in data and starving for information. There are potentially invaluable knowledge and undiscovered relationships hidden in the data already stored in the PI System. The complexity and volume of data makes it formidable for naked eye or classic tools to discover and extract such multivariate relationships.
Machine Learning is an umbrella term referring to methods to learn from previous observations and apply the extracted structure for future predictions. For example, I may be able to observe and archive temperature, day of the week, time of the day, season, fuel price, and the going price of electricity in a certain region. Machine learning allows me to deduce a relationship among these variables, build a model, and use it to predict the going price of the market in presence of observations of other variables.
As another example I could make observations of various physical variables of my device such as pressure, temperature, viscosity, flow, etc. as well as my device’s status, i.e., failure or functional. All the data is stored in PI System. I can use these past observations to teach my model. Later on I apply my measurements of current physical variables to predict the chances of my device going down. I can then use this result to schedule preemptive maintenance.
Do you have other examples that building such a predictive model can help you? We are about to publish a white paper on performing machine learning on PI Data and would love to hear your opinion.