Spatio-temporal AI inference engine for estimating hard disk reliability

This paper focuses on building a spatio-temporal AI inference engine for estimating hard disk reliability. Most electronic systems such as disks routinely collect reliability parameters in the field to monitor health of system. Changes function time are monitored and any observed changes compared with known failure signatures. If trajectory measured data matches that signature, operators alerted take corrective action. However, interest lies being able identify failures before they occur. The state art methodology including our prior work is train machine learning models temporal sequence capturing variations across multiple features using it predict remaining useful life devices. we show this prediction capability alone not sufficient can lead low precision uncertainty around very large. primarily due non-uniform progression feature patterns over time. Our hypothesis accuracy be improved if combine methods spatial analysis compares value key SMART devices similar model fixed window (unlike method which uses from single device much larger historical window). In paper, first describe both approaches, select various hyperparameters, then workflow these two methodologies provide comparative results. results illustrate average long-short memory networks impending next ten days was 84 percent. To improve precision, use set identified potential start applying anomaly detection those disks. helps us remove false positives tighter bound failure.