TECHNOLOGY
“ Predictive maintenance can minimize down time , cut production hours lost to maintenance , and reduce the cost of spare parts and supplies ”
hypothesis , explicit or implicit , i . e . a known signal to look for ,” said Somers . “ Some of these ( decision trees and neural networks ) can be trained from multiple failures of the same type when the data is clean and consistent but in the world of maintenance , data is rarely clean and by the time you have a statistically significant sample of failures you really have a much bigger problem . Mostly the problem is that issues look like one-offs , so trying to find an ‘ unknown unknown ’ signal without generating lot of false alarms is much more of a challenging problem .”
The data dilemma So can anything really be done ?
“ Well , yes and no . Maths is maths and data is data ,” said Somers . “ If there isn ’ t enough data being monitored then there ’ s a limit . However most conventional systems which have some form of condition monitoring often do not suffer from lack of data . Often on top of historians and IoT data there are maintenance databases full of event logs , unstructured and semistructured data including mechanics logs ; there are supply chain / ERP databases , scheduling data and shift
12 June 2015