Why Widespread Predictive Maintenance Implementation Will Take Longer Than We All Think


When Benjamin Franklin first uttered these words, he was referring to fire safety and prevention.  Today these words are equally applicable to machine maintenance and uptime.  If this is true, then why isn’t Predictive Maintenance (PdM) based on machine learning being deployed everywhere. 

While there are a number of reasons that we all can list like cost, reliability, accuracy, and supportability, there are a few obstacles that may not be obvious:

  • Conflicts with established procedures
  • Interconnected manufacturing lines cause unique challenges
  • Shortage of skilled and qualified plant maintenance personnel

Each of these is discussed below.


Many products have well defined inspection schedules. Think about aviation, automobiles, and railroad rolling stock.  Currently, these schedules are based on easily measured intervals such as time, miles, or hours.

  • Automobile safety and emissions inspections occur every year in the same month while oil change intervals are still based on mileage.
  • Aircraft structural parts are inspected after a fixed number of flying hours or takeoff/landing cycles
  • Train wheels and trucks are inspected after a fixed number of rolling miles

Assume aircraft structural integrity was continuously monitored and the system detected an issue.  The plane operator would schedule an inspection and/or repair as soon as the very small crack or sign of fatigue were detected.  All good, right?  What if the following month the historical metrics indicated that a complete inspection was needed and the airplane would have to be out of service for a number of days?  While the operator would argue like crazy that the inspection was redundant, the authorities that required the complete inspection would dig their heels in and today they would win!

I believe the solution to remove this type of potential conflict will take a number of years and require each manufacturer to demonstrate that their PdM system was accurate enough to detect all visibly detectable flaws without taking the equipment out of service.  This is something that the medical device industry has done and PdM systems are now being widely used on imaging systems.


Most interconnected manufacturing lines consist of a mix of commercial off the shelf (COTS) and custom designed modules.  The COTS modules may come from different manufactures and consist of robots, conveyors, and machine tools.  The custom designed modules may come from external or internal design and manufacture businesses.  It is probable that the COTS equipment will be available with robust PdM capabilities long before the custom equipment.  Think of the burden this puts on the maintenance crew.

Because a mature PdM system will present the maintenance team with one or more probable failure timeframes, at different statistically significant intervals, the team will have to continuously decide when to stop a line for an unscheduled repair and what else, if anything, they will do during the shutdown.

Here are two simple examples for a line with two COTS and two custom made modules that assumes that any repair and any PM would take approximately the same time.  If there were a major time difference, then all bets are off:

  1. One of the two COTS modules is predicted to fail within 4 weeks with a 95% confidence level (this is as sure a bet as you can make).  The second COTS module has no failures predicted in the foreseeable future.  The two custom modules have a scheduled preventative maintenance check in six weeks.  In this case, the team would repair the COTS as soon as parts were available, but definitely within four weeks when the production schedule could accommodate the downtime.  The two PM’s would be conducted at the same time,
  2. One of the COTS will fail within four weeks, the other within eight weeks, and the two custom modules have a scheduled PM in 16 weeks.  What do you do to minimize the number and duration of line shutdowns?  Not so easy.

Even in these very simple scenarios, you can see how many possible situations can are available, especially until all modules have access to a well-developed PdM system.  The complexity of these possible outcomes may cause some plants to purchase equipment that is PdM equipped with appropriate sensors, communications equipment, and access to remote AI powered software, but not enabled until a total line can be automated at the same time.


Today, most plant maintenance professionals are skilled at commissioning, troubleshooting, repairing, and testing the equipment they support.  They also work closely with material planners to ensure that spare parts are available when needed.

In the new world of PdM, these people’s jobs will change in two very different ways:

  1. Skills related specifically to PdM

As my friend Titos Anastassacos of Si2 Partners told me:

“Facilities engineers and managers will have to understand the principles of modeling and how machine learning algorithms work (at least their philosophy).  Additionally, they will need to understand what data they have, in what form it is, and what they can do with it -so as to also decide what to collect.  Culturally, they will have to learn to work with the computer, trust (eventually) what the computer is telling them, and find ways to continuously make improvements.  Finally they will have to learn to articulate technical issues in a way that data scientists can use their observations and corrective actions to turn into predictive algorithms.”

  • Skills required because of the workplace changes resulting from PdM

Facilities engineers and managers will also need to understand statistics, finance, purchasing, and vendor communications.  They will do less hands-on troubleshooting and more explaining to plant managers why they elected the maintenance course they are recommending.  Their total compensation may be tied into the same metrics as the plant manager or line supervisor.

As you can see, until all interconnected production equipment has PdM, there appear to be a number of reasons to hold off implementation.


With new technology, everything takes longer and costs more than everyone thinks.  And gaining widespread acceptance for Predictive Maintenance (PdM) is no exception.

Here is some data to demonstrate this phenomenon with respect to Internet connected devices:

  • In 2012, IBM projected 1 trillion IoT devices by 2015
  • In 2017, CISCO projected 50 billion IoT devices by 2020
  • In 2018, GSMA projected 25 billion IoT devices by 2025
  • In 2018, CISCO projected 14.6 billion IoT devices by 2022

While the target years keep changing, the forecasted number of IoT connected devices is declining.  Part of the reason for the decline is technical difficulties like battery capability and platform availability.  On the other-hand, commercialization is also being limited by a lack of compelling use cases and user acceptance.

And, a number of months ago, AP (Associated Press) published an article that started with:

Jeff Bezos boldly predicted five years ago that drones would be carrying Amazon packages to people’s doorsteps by now.

Amazon customers are still waiting. And it’s unclear when, if ever, this particular order by the company’s founder and CEO will arrive.

As you can see, although widespread implementation of new technologies usually takes longer than we expect, the results will be worth the wait.


The journey to full PdM implementation in a large facility will be a technical, cultural, and economic challenge.  You have two choices:

  1. Wait until it is plug and play
  2. Start now

If you elect to wait, you will miss out on all the learning experiences that will make you a much better user than if you wait.  Your facilities team will grow their skills on an as-needed basis, which will make it easier to keep them motivated and with a positive mental attitude.  And you will be generating an ROI at every step of the journey.

If you start soon, you will be able to play a role in co-creating PdM solutions with your key equipment suppliers.  Your needs will be included in products you will buy in the future.  You will benefit from being an early adopter because of the new relationships you will forge with the suppliers.  You will also be able to attract the best and brightest people who will then be willing to focus their creativity to solve your problems over the long haul.

I recommend that you heed the advice provided by Francis of Assisi:Start by doing what’s necessary; then do what’s possible; and suddenly you are doing the impossible.”