Predictive maintenance is one of the few industrial AI applications with a genuine multi-company survey behind it, and that survey is more sobering than the marketing around it. PwC and Mainnovation asked 268 companies in Belgium, Germany and the Netherlands where they stood, and found 11 percent at the top maturity level. Meanwhile the names most often cited as proof, Rolls-Royce, Volkswagen, GE, have published capability announcements, targets and a well-documented shortfall rather than measured maintenance outcomes. The plants that did publish numbers are smaller, and their numbers come from practitioners speaking at conferences.
Eleven Percent of 268 Companies Reached the Top Maturity Level
The Predictive Maintenance 4.0 study was published jointly by PwC and Mainnovation in September 2018, covering 268 companies, 102 in the Netherlands, 95 in Germany and 71 in Belgium, answered by chief operating officers, plant managers and maintenance managers. PwC sells consulting into this area, worth stating, but the study discloses its method and distribution.
The maturity picture is the honest headline. Eleven percent had reached level 4, the fully predictive stage. Twenty percent were at level 3 real-time condition monitoring, 42 percent at level 2 instrumented inspections, 25 percent at level 1 visual inspections, and 2 percent were doing no predictive maintenance at all.
Among the 67 companies implementing at level 4, 95 percent said it improved one or more key maintenance value drivers. Their reported gains, and the 9 percent average uptime figure in particular, are the numbers to carry into a business case rather than the double-digit gains in supplier decks:
- 60 percent reported an uptime improvement, averaging 9 percent.
- 52 percent reported a cost reduction, averaging 12 percent.
- 46 percent reported reduced safety, health, environment or quality risk, averaging 14 percent.
- 45 percent reported extended asset lifetime, averaging 20 percent.
- 40 percent cited improved customer satisfaction, and 36 percent energy savings.
Sixty-Three Percent of Non-Adopters Found No Business Case
The same survey asked the non-adopters why not. Sixty-three percent said there was no viable business case, 23 percent cited insufficient data and 8 percent a lack of analytics capability. The dominant barrier was not missing sensors or immature technology. It was arithmetic that did not close.
A NIST pilot survey of United States manufacturing maintenance, published in June 2016 by Xiaoning Jin and colleagues, points the same way: large manufacturers had seen only modest success with diagnostics, prognostics and preventive maintenance projects, and adoption differed sharply between smaller enterprises and larger firms.
The Plants That Published Numbers Are Not the Famous Ones
Specific, attributable results do exist, and they come from mid-size operations whose reliability engineers presented at industry conferences and were written up by the trade publication Plant Services in October 2021. All four below are practitioner accounts relayed by trade press: named, attributable and specific, which puts them well ahead of an anonymous manufacturer in a supplier case study, and none of them independently audited.
- Noranda Alumina in Gramercy, Louisiana: a 60 percent decline in bearing changes in the second year of its programme, roughly $900,000 saved on bearing purchases, and a grease route completion rate of 92 percent, reported by reliability engineer Russell Goodwin, who put four hours of downtime at about $1 million of lost production.
- The Frito-Lay plant in Fayetteville, Tennessee: year-to-date equipment downtime of 0.75 percent and unplanned downtime of 2.88 percent from vibration analysis, ultrasound and infrared, reported by reliability engineering manager Carlos Calloway.
- SMRT Trains in Singapore: one million mean kilometres between failure in August 2019, using a predictive decision support system built on Bentley Systems AssetWise Linear Analytics, with about 20 maintenance train deployments avoided each year, reported by senior engineer Jessie Nguyen.
- San Diego Gas and Electric: iPredict, built with PA Consulting and Toumetis on Amazon SageMaker, plus drone inspection detecting more than 40 distinct asset and damage conditions in imagery, described by digital strategy lead Gabe Mika.
Rolls-Royce Announced a Capability and Published No Metrics
Rolls-Royce’s IntelligentEngine programme is cited constantly as evidence for predictive maintenance in aviation. Its press release of 16 July 2018 contains no quantitative outcome metrics at all: no engine counts, no data volumes, no maintenance interval improvements, no time-on-wing gains, no cost savings. Its concrete claims are that the latest Engine Health Monitoring system measures thousands of parameters more than previous versions, and that an engine can return up to 200 hours of specific information on request. Axel Voege, head of digital operations in Germany, said the company can monitor line replaceable units and predict when they need replacement rather than respond to failure.
That is a capability announcement, and a reasonable one. It is not a measured result. No Rolls-Royce publication with audited predictive maintenance outcomes could be located, so the savings figures attributed to the programme elsewhere are recycled supplier or consultancy material rather than company data.
Volkswagen’s Targets, GE Digital’s Shortfall, and a Vendor’s Own Estimate
Volkswagen’s Industrial Cloud release of 29 April 2020 is likewise a plan rather than an outcome. It reports three plants connected in 2019, Chemnitz, Wolfsburg and Polkowice, with 15 more intended during 2020 against a group total of 124 plants. The stated productivity goal is 30 percent across 2016 to 2025, with 200 million euros projected from the first 15 applications by the end of 2025. Predictive maintenance appears as a target application, and no measured result for it is given.
The counter-example is better documented than the successes. GE’s Predix platform was projected by then chief executive Jeff Immelt to produce $15 billion of GE Digital revenue by 2020. CIO Dive reported in August 2017, citing Reuters, that analysts called that target highly unlikely, that the shortfall was seen as at least partly responsible for a 25 percent fall in GE’s share price that year, that operations were paused for two months in May 2017 to address platform problems, and that an attempt to build proprietary cloud storage fell flat, forcing moves to Amazon Web Services and later Microsoft Azure.
Where only a supplier has published a figure, the supplier belongs in the sentence. Metso reported that its Metso Metrics for Mining service, running on Rockwell Automation’s industrial internet platform, caught a crusher anomaly at an Australian iron ore mine and averted a catastrophic event with potential costs of more than $1 million. The mine is not named and the estimate is Metso’s own.
Working the Survey Averages Through, as an Illustration
What follows is arithmetic, not a result, and none of it describes a real plant. It shows how the only credible multi-company averages behave when finance asks what a programme is worth.
Take a site spending $10 million a year on maintenance. Apply the average cost reduction of 12 percent that 52 percent of level 4 adopters reported to PwC and Mainnovation, and the annual saving is $1.2 million. Against that sit sensors, integration work, data engineering, model upkeep and training for the people who act on the alerts, spread over the years it takes to move from instrumented inspections to a working programme.
The credible saving, then, is a single-digit to low double-digit improvement on existing maintenance economics: real money at scale, nothing like a step change. That is also exactly the size of benefit that fails to clear a hurdle rate at a smaller site, which is the likeliest reading of the 63 percent who saw no business case. Start from the disclosed averages, name the source of every figure, and treat the celebrated aerospace and automotive examples as ambition until those companies publish numbers.
Sources: PwC and Mainnovation · Plant Services · Rolls-Royce · Volkswagen Group · CIO Dive · NIST
