Select Page

The promise of Industry 4.0 and the Smart Factory has attracted billions of dollars in R&D investments by established vendors such as GE (Predix), SAP (Leonardo) and Siemens (MindSphere).  Industry analysts have forecast that in the era of digitalization, the winners will be the nimble factories able to use big data for real-time decision making.   The future of traditional factories unable to keep up with the massive pace of change is in doubt.

The binary view often promulgated by so-called experts is hiding a more complex reality. Although the Smart Factory is the future, the road ahead will be slow and uneven.  Existing systems may be recalibrated and updated, but it is unrealistic that we will witness a rip-and-replace of existing machine assets and legacy technology infrastructure.

Accurate Predictive Maintenance is the Holy Grail of the Smart Factory

There are four approaches to asset maintenance:  

  • Reactive Maintenance (RM) or breakdown maintenance is performed after machine failure has already occurred as it is the most expensive to fix.

  • Preventive Maintenance (PM) is performed based on a schedule (time or machine usage) to prolong the life of an asset.

  • Predictive Maintenance (PdM) is performed based on the condition of the machine.

  • Machine Learning Predictive Maintenance (ML PdM) is the use of advanced algorithms on big data to detect machine failure.

For solution providers, one of the biggest opportunities that has been identified in the Smart Factory is Predictive Maintenance.  The reason is obvious:  factories today are overspending on Preventative Maintenance. Instead of performing repairs when needed, unnecessary repairs are prematurely made to reduce the possibility of machine downtime.  In one study, 45% of asset maintenance is Preventive, 40% is Reactive and only 15% is Predictive.  As you can see below, the earlier asset degradation is identified, the earlier repairs can be performed and the lower the cost to repair.

 Data from the US Department of Energy indicates that Predictive Maintenance is cost effective. Implementing a functional Predictive Maintenance program can yield some astonishing results: 10X ROI, 25%-30% reduction in maintenance costs, 70-75% fall in breakdowns and 35%-45%uptime increase.

What’s Holding Back PdM?

If Predictive Maintenance is more effective and cheaper than RM and PM, why is adoption less widespread? The answer is that Predictive Maintenance techniques are not scalable in most facilities.  Although there are numerous Predictive Maintenance solutions with varying costs and levels of effectiveness, there is no one solution that is applicable to an entire facility.

Condition Monitoring is a traditional PdM tool use for asset maintenance. It uses rule-based human selected control limits that are applied to a small number of sensors.   If the data breaches the predefined thresholds, then an alert is generated.   If the sensor data stays within the thresholds, no degradation or fault is predicted.  Much of the underlying technology for Condition Monitoring is antiquated and only a small number of sensors can be monitored in real time.

A further limiting factor is that many industrial plants are unable to access the data that is needed for Condition Monitoring.  The pre-requisite for Condition Monitoring is access to clean sensor data that can be evaluated in real time.   Data ownership is not a simple issue.  Even if the data is stored, at a practical level, there are often third party machine vendors that are the gatekeepers and often restrict operational access to the data.

In recent years, more “smart” Predictive Maintenance solutions such as acoustic or vibration sensors have entered the market.  The challenge for the factory owner is that these are limited solutions fit mainly for smaller machines.   Furthermore, new PdM monitoring techniques require training and ongoing time commitment from existing maintenance and reliability technicians.

In summary, the limitation to greater adoption rates of PdM is the historic lack of data access and the inadequacy of legacy Condition Monitoring tools.

Towards the Hybrid Model

In the Smart Factory, there will be a re-alignment of asset maintenance to Machine Learning solutions.  Specifically, there will be less of a need for Preventive Maintenance and Reactive Maintenance.

In the Smart Factory, we expect to see more intelligent Preventive Maintenance.  Machine Learning will be used to guide maintenance staff so that the intervals of maintenance are better regulated.   For instance, certain assets could be serviced more frequently based on machine specific changes in performance that are detected.

Reactive maintenance will never disappear because there are always events outside of the control of facility owners.   However, the early detection of machine degradation is likely to reduce the instances of unplanned asset failure and the corresponding costs associated with repair.

Until the emergence of a pure Smart Factory, we expect to see migration to scalable and economically viable digitalization technologies.   In the Hybrid Smart Factory, the goal will be to create the optimal mix of Reactive, Preventive and Predictive Asset Management.

The Impact of Machine Learning for Asset Maintenance

When Machine Learning algorithms detect asset degradation and breakdown ahead of time, resources can be shifted away from over-Reliability investments such as unnecessary maintenance and double and triple redundancies.   As asset Reliability investments are rationalized and the detection of degradation is improved, the frequency of unscheduled downtime is reduced.   This lowers Total Maintenance cost as a Percentage of Replacement Asset Value (RAV).

Presenso’s Automated Machine Learning for Asset Maintenance solution for asset maintenance applies the latest advances in Artificial Intelligence to detect abnormal data patterns in an entire factory, thereby lowing the cost of Reliability Maintenance.  The Presenso cloud-based solution does not require new hardware or sensors and can be deployed remotely, without the need for human input from facility workers.

 For part two of this blog article, please click here.
Like this article? Please share it using these icons....Share on FacebookShare on Google+Tweet about this on TwitterShare on LinkedIn
Avi Nowitz

Avi Nowitz

Avi Nowitz writes about the financial impact of Machine Learning on the industrial sector. Avi started his career on Wall Street as an equity analyst, providing institutional investor research on manufacturing companies. After completing his MBA in Finance from New York University, he joined the consulting division of Peppers and Rogers Group where his international clients included Bertelsmann, Ford Motor Company and Doğuş Holdings. For more than a decade, Avi led the Microsoft Consulting Practice at New Age Media and managed the execution of initiatives in EMEA and APAC.

[:en]

[:]