I have faced the same problem before. The approach is to monitor the average TRX duration per TRE in the global Network. you can then set a threasold acceptable to maybe 10 Seconds and above being ok. Any TRE with less than ten seconds then has to be monitored. First let me know if you have the NPA tool running on Metrica then i can present you with the script to run a check of the performance of all TREs in the Network. From the results obtained, you can then move on to the OMCR and check the TREs identified by USD. here you will be able to notice the actual faulty TREs which may then be replaced.