Generally a digital twin is a set of mathematical formulas that represents the physical plant in question. As such, it already integrates all your diversified sources of data into a single coherent model of the process. You can use the digital twin to compute the result of an action or you can computationally invert the model to compute the action you need to take in order to get a result that you want. A third use-case for a digital twin is to compute the ramifications of an "issue." This is done by fixing the value of those process variables that you know by physical measurement, i.e. the "issue," and then computing the other variables that result from this via your model. This is the way you can use a digital twin to locate the root cause of a problem. An issue cannot really be "invisible" however. It must be made known to the model in terms of some numerical values. Perhaps a visible issue has a unknown root cause and the model can be used to find it.
Supposing that the digital twin was built using only healthy data, you can analyze the deviation between the real plant and the digital twin to see where the divergences are and when in time they occured. That will usually give you the full chain of causation, at least the part that is visible in the data. I have successfully used this approach numerous times both for maintenance measures and insurance claims on process industry plants.

this can be solved by using Hadoop and Sqoop tools from Apeachi open source project. Bascially what i am saying in plan english is that you need to combine structure data and with unstructure data and there are ways to integrate them today.
If you need more details i can point you to the articles or some videos.