Hi Toh,

Proactive maintenance is definitely an interesting use-case, and to consider how data can be converted into insights to drive pro-active maintenance.  However, this is a use-case that can only be tackled up to a certain level in a generic manner (e.g. so that you can apply it to any 'product' - whether that be a high-power amplifier in a DTT network, a video encoder in a video headend, a playout server, a CMTS in an HFC network or a car, or production machine so to speak).  Each specific product, or even each specific product from each specific vendor would require some very specific expert knowledge (which metrics are involved, and which ones are not).  You could consider self-learning, but then you would have to have a lot of historical data (i.e. data sets for that specific product, which show all the key operational metrics over the past months (and optionally labels of when maintenance was needed - you do need a grounded truth, either labeled data or pre-defined expert knowledge) - so that the solution can try to learn about what type of data patterns across the metrics reveal the need for maintenance.

As Toon indicated, there are various generic capabilities in DataMiner, which can help you for this use case (preventive maintenance) and others (malfunctions, misconfigurations, etc.).  See also Advanced analytics features in the Alarm Console | DataMiner Docs.

Maybe noteworthy to highlight is the Relational Anomaly Detection - where DataMiner looks at the behavior of a group of metrics (e.g. battery voltages of all cells as a group), and identifies any changes in behavior as a group.

Also - we could maybe help you out also if we can have some more information about your specific use case.  What kind of products are you talking about, do you have already a sense wat metrics could be involved and what kind of behavior could indicate the need for maintenance?  A more detailed description of the use case could help our experts to provide feedback and/or solutions.  If you are interested, we'll be happy to get you in touch with those experts to discuss.

Have you checked out proactive cap detection? It's not a neural network, but it seems to fit the bill when it comes to your use case.