Manufacturing industries are looking for Asset based analytics solution which can allow them to have real time view of asset Performance data. Analytics solution can help organization to keep watch operational efficiency and maintenance cost associated with it. Real time condition and performance monitoring solution of asset can lead to an improvement in the Plant asset up time.
We tried to develop Proof of concept on condition based monitoring for asset with solution architecture as below and tried to leverage the OSISoft PI as Data infrastructure and R as Data analytics platform.
The asset data was in excel format and also some of the data was coming from Kepware OPC Server so we used both PI OPC DA interface and used script to load excel data into PI Data archive directly. We further used PI OLEDB and PI ODBC to connect PI Server with R studio using R ODBC. Once data started coming into R studio we used R for data cleansing as lot of raw data had issue of missing data, bad data etc. So we considered asset Pump and taken different attributes of Pump such as Pump Speed, Pump Torque , Pump Efficiency, Pump Pressure and Pump flow rate and we had last 2 years of worth data for analysis. Based on this raw data analytics was performed on this data and we came with health index of Pump as shown below based on analysis of all this different pump attributes.
The multivariate data array is used for computations using R. Data from an initial period of operation of the asset has been used for training. Subsequent data is used for evaluating the condition of the asset. In this initial exercise, the MEWMA framework has been used for purposes of simplicity. Control chart using Hotelling T2 has been used as an indicator of asset health.
Now we wanted to push this data back to PI Server so we again used PI ODBC connection to write back data in real time to PI comp2 table and we could also show the analysed data into PI Process book and PI Vision.
Benefit from this Proof of concept
This solution helped us to have real time view of Condition and performance of asset and we were able to trigger the notification to maintenance team to take actions in advance before asset fails. This was proof of concept using PI and R to perform advanced analytics in real time and to prove that this works.Now in next phase we are trying to do this Proof of concept on larger asset data set.
I would appreciate comments from PI Square community on this and also suggestion to further improve this Proof of concept.