Contextualization means adding relevant information to data to make it more useful. In the world of IIoT, data is not in short supply. PLC's, SCADA Systems, Condition Monitoring Systems and Data Historians all provide tremendous insights into process and equipment health. However, without a firm understanding of the context of this data, simply throwing it into a data lake and hoping for machine learning algorithms to "provide something useful", is a stretch. Adding context requires the knowledge of what sensor collected the data, what it is capable of and insights into what the data could identify.
As an Example, if we want to be able to predict a pump failure, we must know what type of vibration readings we are receiving. Simple, overall values do not identify a part or failure mode, it can only detect that it is vibrating more or less in general. Likewise, if a sensor is sampling at 40Khz, the overall value will not be comparable to a sensor that is sampling at 5Khz.
Orchestration takes data from multiple sources that are otherwise siloed, and brings it together for meaningful and accurate analysis. Additionally, related metadata provides the ability to identify "Like in Kind" data to be included in learning models for analytics. Orchestration requires knowledge of the problem(s) and potential solutions to those problems.
As an example, if we want to be able to predict a pump failure, we must know what data is needed, what is available and where to obtain it. Additionally, understanding the underlying failure modes is essential in determining the various types of data to consider. Suction Pressure, Discharge Pressure, Differential Pressure, Temperature, Vibration, Flow, Amperage and Voltage (just to name a few) all have tremendous use for identifying potential issues. Each to their own, their may be some value. But, when Contextualized and Orchestrated, we can not only identify defects, we can start to understand how the defect was introduced and eliminate it.
Analysis has taken on a new hope and in some cases a new meaning in the condition monitoring world as of late. The promise of machine learning is very exciting, but in most cases, not a good starting point and fails to answer the burning question of "What is wrong with it" at a prescriptive level. Additionally, several data sets with the same failure data must be available to train a model.
Or Does it Really?
Machines fail in known ways. These failure modes can be identified by highly trained experts with domain knowledge of their discipline. But, at the end of the day, for most sensor data (Including Vibration) it is just math or known correlations. We at K-IIoT understand this math and science behind failure recognition and would love to share it with your data science team or machine learning platform to provide true assisted learning models that are capable of delivering the hope of machine learning.