If something can go wrong it will…
Sometimes scans go wrong, ideally we should have a way to edit out bad frames natively in the scanner software, but as of this writing it is still not possible natively in RS5.2. The closest thing to this dream is Revoscan Flayer but for now it is still a proof of concept that does help identify the frames for manual deletion but is in no way ready for use by the large public and still only support a couple of scanner models.
With bad registred frames one ends up with a fragmented point cloud that is unusable and sometimes you just cannot repeat the scan.
When that happens here are some ways to segregate the parts and realign them that might just save a scan.
…
What if i use SOR?
One tool that can cleanup a lot of noise in CloudCompare is SOR (Statistical Outlier Removal) specially when standard deviation is set to 2.5 (for revopoint scanners work great)
But SOR and Noise filters work mainly to cleanup noisy data that appears like rough points and does not help at all with misregistration data like this example below
…
Solution
The solution to this is something called segmentation, where the software tries to find what does not fit allowing the user to split the data into separate parts that can be later realigned
Segmentation tool creates a Scalar field classification, meaning one can split by histogram selection (ex. color) 
The process is a bit convoluted, the segmentation tool that works is:
plugins > facet/fracture detection > extract facets (fast marching)
…
The resulting extraction is not what is useful…
The tool while extracting creates a scalar field in the original point cloud and that one is the one we can use.
…
Fractured point cloud after realignment
Would a video tutorial be better for You?
If there are enough likes to this post i’ll make a tutorial video and add it here.