PREDICTIVE MAINTENANCE USING LONG SHORT-TERM MEMORY ALGORITHM ON SENSOR DATA

Think of a complex system with very expensiveparts. We can't risk running into failure as it will beextremely costly to repair highly damaged parts. Butmore importantly, it's a safety issue. This is whynumerous organizations attempt to avoid failurebeforehand by performing regular inspections on theirequipment. One big challenge is to determine when to domaintenance. Since we don't know when failure willoccur, we have to be conservative in our planning. LTSMcan be used to predict the remaining useful life. But if weschedule maintenance very early, we will end up wastingmachine life that is still usable, and this will add up toour costs. However, if we can predict when machinefailure will occur, we can schedule maintenance rightbefore it. Recurrent Neural Networks can predict whenthis machine failure is bound to happen. Predictivemaintenance lets us estimate time to failure. It alsopinpoints problems in complex machinery and helps usidentify what parts need to be fixed. This way, we canminimize downtime and maximize equipment lifetime.