Understanding the history to improve the future

More than anything ( read robotics, 3D printing, AR/VR, Simulation etc. etc.) Industry 4.0 is about the integration. Integrating anything and everything around the value chain axis ( Supplier to Consumer), Business axis ( Shop floor to the board level ) and the Product axis ( from design to retire ). These are fundamental for the other technologies like IoT, 3D printing, Robotcs & AGVs, Digital Twin, xR ( AR/VR/MR) to generate value.

When you start integrating all the axis into a common Industrial (big) data model, you will see the significant portion of your data is temporal or time series data. Time series data is nothing but a metrics ( or measurement ) with a timestamp.

Usually time series data is generated from Industrial sensors, fleet systems and often many business transactions ( like USD conversion rate, Price index etc.) are also considered as time series information and archived for long term play back of business and manufacturing scenarios.

Time Series data is being stored traditionally for playing back past events to find out improvement opportunities, problem areas and sometimes as record keeping for compliance purpose.Over the years, Time Series becomes the basis of predicting the future leveraging statistics, data mining, Machine Learning, Deep Learning and AI.

Industry 4.0 solutions are well equipped to handle this massive volume of data leveraging cloud computing and peta byte scale cloud storage solutions. They effectively store your time series data generated by the entire manufacturing eco-system of yours efficiently and in an inexpensive way.

We often call these capabilities of Industry 4.0 solutions as "Data Historian". They are implemented on-premise or cloud or hybrid combination of cloud and on-premise.

No matter where you store your time series data, the value can only be generated if we could make sense out of the data fast and effortlessly.

Following is my guide to answers you must look for while considering building a time series storage and retrieval solution for your Industry 4.0 journey .

At minimum the storage and the analytics layer on top of the time series historian should be able to answer below 18 questions in order to be effectively use the history to predict the future

1. Have we ever seen a pattern that looks just like this??
2. Are there any repeated patterns in my data?
3. What are the three most unusual days in this three month long dataset??
4. Is there any pattern that is common to these two time series??
5. How do these two time series differ in terms of structural alignment??
6. Find the most conserved pattern that happens at least once every two days in this two week long dataset.
7. If you had to summarize this long time series with just two shorter examples, what would they be??
8. Are there any patterns that appear as time reversed versions of themselves in my data??
9. When does the structural change happen in this time series??
10. How can I compare these time series of different time span??
11. Are there any patterns that repeat in my data, but at two distinct time span??
12. Have we ever seen a multidimensional pattern that looks just like this??
13. How do I quickly search this long dataset for this pattern, if an approximate search is acceptable??
14. How can I optimize similarity search in a long time series??
15. What is most likely to happen next??
16. What is the right span for repeating patterns in this dataset??
17. How do I find repeating patterns faster in my years worth of time series data
18. Have we ever seen a pattern that looks just like this, but possibly at a different time span??

Once the Time Series is stored properly and the retrieval is possible preferably using open APIs, many of the questions can be answered using several Time Series algorithms available in Python. Venturing into a technical solution for time series makes "perfect" sense only when you can answer the above leveraging the technical platform you envisage to deploy.

I will try to create examples for each of the 18 questions on Python leveraging Azure as an example Data Historian and post the same in my GitHub repo.

keep watching this space and follow #prangyapov for many such interesting topics.

[The views in this blog is author's own view and it does not necessarily reflect the view of his employer, Wipro Limited]