Elevator Predictive Maintenance: From Words To Reality – Part 2

By 22 septiembre, 2023Destacado

Vibration analysis is a long shot


Some predictive maintenance initiatives in the industry have focused on installing additional hardware — specifically, accelerometers — to capture in-shaft vibrations. In the past, vibration analysis has had some successful use-cases, such as


  • in the elevator industry, to benchmark the ride quality of a newly installed elevator, and
  • in the manufacturing sector, to help maintain rotating engines.


However, both these use-cases operate in predefined and controlled environments, which is not the case with the elevator installed base. The park is aging and subject to continued external actions from users and technicians.


As a counter-example, the car industry could have a bright future with vibration-based predictive maintenance. It has a long-standing process for noise and vibration data — NVH; every part of a car has a manufacturer’s NVH signature when leaving the factory, and a car model’s assembly remains the same over the years. None of this applies to elevators.


Vibration analysis can be quick and cheap to develop and deploy but extremely slow and expensive for providing tangible insights compared to controller data.


Don’t OEMs already have controller data?


An OEM assembles mechanical parts from various manufacturers and conceives the electronics: by definition, it has access to its own controllers’ data. Neil Green, Otis’ Chief Digital Officer, told InformationWeek that Otis has collected daily data from over 300,000 of its elevators since the 1980s via their Remote Elevator Monitoring (REM) tool. In other words, Otis invented the IoT for lifts.


So, what went wrong?


  1. Data structure. Receiving data once a day about an elevator’s current status is very different from accumulating, in real-time, all the events from a controller to feed into an AI model.
  2. Data usage. IoT data, if not associated with relevant technician data and not targeted at reshaping maintenance methods, is useless. Data alone will not produce any ROI.
  3. Universality. OEMs have access to their controllers, but they maintain dozens of makes. This might be one of the reasons Otis says it installed only 300k REMs out of a 2M unit portfolio. It is hard to change field methods when those changes can’t apply to all serviced elevators.


IoT elevator data alone is useless; a predictive AI with access to it won’t generate any tangible impact. The whole service delivery needs to be data-driven. It is like installing a 2020 Formula 1 computer into a 1990s mainstream car with no electronics; it’s nonsense.


Data-driven maintenance is Phase 0

If service is to change from being technician-oriented to being technology-enabled, it is necessary to systematize field operations before any predictive maintenance can happen.


Know your portfolio

To perform first-level analytics and feed a predictive algorithm properly, knowing the installed material is essential. A component’s cycles of wear are different by make, model, and environment, and referencing and updating the main organs — at least the controller, the door operator, and the engine — of the installed base is the first task. It sounds obvious, but today, most companies do not have proper data on their portfolio.


Know your failures and field actions

Second, you need to normalize all failures. Failures cannot be just callbacks in a call-center database. Their precise qualification will feed the algorithms:

  • In what exact state was the elevator when the technician arrived?
  • On which components did he act and how?
  • What was the status when the technician left?


Third, actions performed by technicians outside of failure situations have to be recorded precisely:

  • Exactly what procedures — visual check, operation test, repair, lubrication, adjustment, replacement, wear-level check…
  • Exactly which components.


Out of all the data necessary for predictive maintenance, our experience shows that field data can be the hardest to generate. Choosing the right level of abstraction against the thousands of different components and environments, creating the right software user experience, and efficiently changing the technician work culture are the main challenges.



This initial dataset can already provide some analytics and enhance field performance; now you can investigate useful metrics, such as


  • % callbacks that are not breakdowns,
  • uptime rate,
  • breakdown rate by material or component type,
  • breakdown rate after a specific field action,
  • first-time fix rate on a failure type, and
  • success rate of various failure responses.


And you can start to correlate field actions and failure patterns.


Notice that this hasn’t yet required any IoT — just the right software with the right user experience for the technicians. If we add elevator data, we’ll be able to tailor operations to each lift: we enter predictive, condition-based maintenance.


Now we can plug in the Formula 1 computer. But wait — what for?


The right product vision wins


Product is focus

The great thing about modern technology is that it can do almost anything. The downside is that nearly every one of those possibilities is entirely worthless, and large companies that do not have a software innovation culture struggle to grasp this.


Successful technology startups do not win by the depth of their resources, but by the precision of their focus.


Technology-driven quality kills costs all day long

The end-game is simple: organic portfolio growth, both in volume of serviced elevators and in profit generated per contract.


However, the critical path is often blurred by productivity thinking: shouldn’t one focus on predictive maintenance’s impact on costs first, as they are more visible than the indirect outcomes of a higher quality of service?


Focus on quality first, and costs will reduce themselves:

  • If your first-time fix rate increases and your breakdown rate decreases, this means fewer field hours for tackling callbacks.
  • If your maintenance operations are data-driven and standardized, it becomes easy to lower the skillset barrier for value-added interventions.
  • If your perceived quality is high and the information provided on the service is abundant, it becomes easier to sell necessary additional repairs.
  • Once the right quality is reached, it becomes easy to raise the average contract price and, therefore, the margins.


Predictive maintenance is not an add-on. It is a reconstruction.



Elevator OEMs today have a simplistic vision of predictive maintenance in their industry: plug in an IoT device, gather data, anticipate breakdowns before they happen, and send a technician. This cycle is seen only as an addition to what the technician already performs:


  • Mostly useless compliance visits. We call them “ghost visits.”
  • Breakdown response: receive a callback, get to the building, reboot the elevator, and leave.


Not only is this add-on vision hard to achieve, as pure IoT data is rarely sufficient to anticipate breakdowns, but it entirely misses the point. It generates additional unplanned field hours (the most expensive ones) instead of reducing them.


The right product vision is a reconstruction of the service delivery from scratch:


  • Rebuild the technician user experience to gather normalized field data and to standardize field actions.
  • Within local regulation boundaries, avoid unnecessary field hours with a differentiated (high-value or compliance-only) visit model.
  • Focus predictive maintenance efforts on early breakdown avoidance by applying the right preventive actions: transform expensive curative field hours into cheaper, predictable planned hours.


(To be continued ).

Source: medium.com