Hi Carlo,
Please have a look into Polling Channel Watchdog functionality (SAP Notes 1808222 and 1985336). This functionality is based on periodic checking of polling communication channels by scheduled background job and if the corresponding polling channel is not in error status (is seen as green in Communication Channel Monitor), but doesn't execute polling anymore, the job may set the channel to error state and trigger alert notification. This will help you to identify situations when the channel stopped effective polling.
As for root cause analysis of the issue. What Adapter Framework Scheduler version do you use (what is the value of property scheduler.relocMode of Java service XPI Service: AF Core)? If your system is on the appropriate release / Support Package level, it is recommended to switch to a new AF Scheduler (by setting this property's value to some negative integer). Please refer to SAP PI/PO Performance Check Guide (see SAP Note 894509) and SAP Note 1355715 for more details. In AF Scheduler jobs monitor (a part of PIMON in newer releases of PI/PO, can be accessed by navigating to http(s)://<server>:<HTTP(S) port>/AdapterFramework/scheduler/scheduler.jsp in older releases of PI), you can monitor polling job for the affected communication channel and see when and on which server nodes it has been executed. If it doesn't get triggered by AF Scheduler for a long time on any of server nodes of your PI cluster (last execution time on all server nodes is far time ago in comparison to configured polling interval, number of executions doesn't change on any of server nodes), I'd suggest to look into AF Scheduler configuration (mentioned above). If it does get scheduled and triggered by AF Scheduler and is seen as executed on server nodes (check last/next execution times and number of executions per server node - you should see them changing), then I'd suggest to collect thread dumps and analyze polling channel thread in details.
Regards,
Vadim