The market has been quite receptive to know about digital transformation journey in all the functions of an organization and I believe the market globally is now at the inflexion point to accept the benefits of the journey basis the credible return on investment numbers. One of the most prominent advantages of applying IoT transformation in operations especially in asset heavy organizations is undoubtedly the Predictive Analytics in maintenance or more popularly known as Predictive Maintenance.
To put it simply, the predictive maintenance is a type of maintenance which is triggered basis the health of an equipment or per the operating limits set by the user. This is a clear shift from the usual preventive or scheduled maintenance that forms an integral part of old-school maintenance style where the plants were shut down for maintenance activities. The scheduled maintenance got into the mainstream of maintaining the equipment since earlier there was no way to know whether an equipment or a machine needs maintenance unless it broke down. This led to ‘overdoing’ the jobs that weren’t necessary at that point in time like for example replacing the parts that still had ‘life’ in them or as simple as oiling the parts that were already oiled. On the other side of spectrum, sometimes the parts were left untouched because they ‘looked’ alright but were either worn off or were about to break.
In many instances that I have seen, I have come to this conclusion that predictive maintenance brings in the ‘right balance’ of decision making whether to perform any maintenance activity on an equipment or leave it as is. This approach requires intelligence emanating from machines wherein the intelligence is analyzed to trigger an alarm to the maintenance engineers that there is something ‘wrong’ in the machine. This causes the engineer to check the performance of the machine and hence look out for possible problems. The predictive maintenance works on an algorithm that may have several parameters to check upon. These parameters could be noise, vibration, temperature, pressure, run time, power usage etc. Before the algorithm is written and programmed, the sensors for each of the parameters should be on the body of the equipment and a data lake should have been set which would send the sensor data to the data lake which in turn is connected to a data and analytics platform. The noise sensor, for instance, could be placed on the motor to check for any unusual noise caused by ball bearings. One of the questions that is generally asked is if we can leverage the historical data or is there any historical data that is available which can be used to our advantage. The answer is yes! The historical data in the form of past breakdowns and the specific reasons behind the breakdowns can be leveraged to build the predictive algorithm more accurately.
Now, how exactly the predictive algorithm works? To answer this question, one needs to understand the ‘logic’ that stiches the algorithm. The predictive maintenance algorithm, if based on noise, vibration, temperature, power factor (KWh value) and run time values, then the logic would look like the following:
A). Check for run time – if between a certain permissible limit, then go to B and if beyond the permissible limit then go straight to F.
B). Check for Noise levels – if not OK then go to C
C). Check for Vibration levels – if not OK then go to D
D). Check for Temperature – if not OK then go to E
E). Check for Power Factor values (KWh) – if not OK then go to F
F). Trigger an alarm via email and SMS to maintenance engineers and production supervisors & managers.
The advanced analytics in the form of Prescriptive Maintenance can also be applied in the algorithm whereby the algorithm not only triggers an alarm for the relevant personnel to check what is wrong but also to give a suggestion about what exactly is a problem and hence what needs to be done to fix it. This insight obviously requires deep learning mechanism wherein the entire modeling happens on a large set of data and the algorithm undergoes rigorous testing. Furthermore, the algorithm should also have a capability to learn from third party databases and from the unstructured data available on web.
The cost of the ecosystem to set up the predictive maintenance has gone down drastically in past 5 – 10 years. If there are legacy machines on the shop floor, then the cost of sensors, edge devices and the connectivity to the cloud computing platform that has a data & analytics layer on it has come down to a level that justifies the investment with a payback time in few months or sometimes even weeks. And if the machine is a latest one with the sensors embedded in it already then, the payback could be even lesser. The return on investment (RoI) on predictive algorithm can be calculated by looking at the number of breakdowns and the cost of each breakdown. By estimate, each breakdown in an industrial manufacturing set up costs anywhere between $30,000 and $50,000 per hour. (Source: https://www.wipfli.com/insights/blogs/manufacturing-tomorrow-blog/170628—the-real-cost-of-unplanned-downtime). Therefore, by a simple calculation it is possible to look at the cost of unplanned downtime and hence justify the investment on going the predictive way.
Finally, the organizations today are converging their IT and OT functions and the overlap has significantly reduced in recent times in digitally matured organizations. This leads to the fact that there is a need to upskill or reskill the staff in digital knowledge and hence build a data culture within the organization. The Predictive Maintenance that runs on Predictive Analytics, once deployed, is not the end all and be all; rather the algorithm needs to be continuously nurtured, improved and fed so that the algorithm can learn new patterns and hence become more intelligent and be suited to the changing needs of the user and of the end customer. Therefore, the organizations today must think about the digital upskilling strategy from data & analytics outsourcing to developing a Center of Excellence within the organization with a mix of data scientists and people with operations background.
Author: Rahul Sharma, Director, PwC Middle East
DISCLAIMER: The views expressed are solely of the author and DigiAnalysys does not necessarily subscribe to it.