IMPLEMENT PREDICTIVE MAINTENANCE FOR BEST RESULTS
itself : firstly there are many situations in the most modern plants , where putting sensors over certain pieces of equipment is completely uneconomic . For example , when it comes to low value , non-critical parts of a factory , running equipment to failure might well be the most economic thing to do .
“ Secondly , there are many situations where predictive maintenance just doesn ’ t work very well . It can easily end up generating a lot of ‘ false positive ’ alarms and ‘ false negative ’ unalarmed failures particularly in the case of machinery with an unknown pattern of degradation , which is unfortunately , most machinery given
ACCORDING TO THE PROCESS INDUSTRY FORUM THE BENEFITS OF IMPLEMENTING A PREDICTIVE MAINTENANCE PROGRAM ARE CLEAR :
• Savings of 30-40 percent
• 10-fold return on investment ( ROI )
• Maintenance cost reduction of 25-30 percent
• Breakdown elimination of up to 75 percent
• Reduction in downtime from 35-45 percent
• Increase in production of 20-25 percent . that it isn ’ t designed to fail predictably in the first place . Expensive and complex machinery breaks without any well-defined or known signal , means that parts are fixed for no reason or the predictive maintenance is effectively useless for the most costly ‘ one-offs ’.
“ New technology and buzzwords are not necessarily the answer to resolving this : When you ‘ take the hood ’ off some of the latest and greatest predictive maintenance packages , they are in reality either a recipe-builder modelling platform with standard predictive analytics algorithms and techniques or a library of known signals for equipment that fails in a known and predictable way . Anecdotally there are significant disconnects between the amount of marketing dollars that some companies pour into this area versus the capability and satisfaction ,” he said .
Upon speaking to a number of plant managers , the general consensus was that some off the shelf technologies only reacted to predictable and contrived circumstances , and didn ’ t understand the working patterns of the machines they were monitoring . So what is the explanation for this ?
“ Well , simply put nearly all predictive analytics algorithms require a
11