Author : Rahul Nandgave and Dr. A. R. Buchade

There is an exorbitant amount of unstructured data which is available on the internet and is increasing at an exponential rate every day. The term ‘Big Data’ is used to represent such data. There is a need to handle such vast amounts of data efficiently and securely. RAID (Redundant Array of Independent Disks), NAS (Network Access Storage), SAN (Storage Access Network) etc. are some of the storage solutions that are available today and are supported by many companies that provide storage solutions. Significant loss of data as well as financial loss can be faced by companies in case of failures of storage solutions. Prediction of suchfailures at real time may help organizations for predictive maintenance and reducing the replacement downtime of such storage solutions. The LSTM Networks can be used to train our model which will predict the failure of storage devices based on data generated by S.M.A.R.T(Self-Monitoring and Reporting Technology) Parameters. Here we will provide an architecture inspired by an LSTM Network that will be able to predict the failure in a hard disk with lower false alarm rates and higher precision and recall.

Keywords: LSTM, SMART Parameters, Predictive Maintenance, Failure Detection, Deep Neural Network,