Predictive Maintenance (PdM) based on machine learning (AI) and (big) data in industrial/technical services is something that has been receiving a lot of attention, as well as hype, over the past couple of years and awareness among relevant managers, in both end-user organizations and OEMs and service/maintenance contractors, is high.

Nevertheless, despite some well-publicized case studies and entry into the market of a good number of software companies, either with cloud-based solutions or tools based on platforms like Microsoft Azure, the degree of implementation is low. According to one survey, managers feel that the value-added provided by PdM is low simply because most machine failures are due to operator errors which existing systems cannot predict. And another reason may be that OEMs already have sufficient information and knowledge about their machines and have already optimized maintenance intervals and practice, therefore negating again any significant impact from PdM programs.

The validity of these reasons can be debated. For example, while operator errors may play an important role in machine failures, there is no reason why they could not eventually be incorporated into prediction algorithms (the outcome could include better training). Furthermore, it is also doubtful that OEMs, or indeed end-users, have optimized maintenance intervals and practice: The former is attested to by the fact that most end-users’ maintenance practice diverges significantly from OEM recommendations after a time period and the latter by the fact that most end-users use condition monitoring, which is expensive, and continuously invest in improving maintenance.

There may, of course, be other reasons for low levels of implementation: Generic programs may not provide sufficient added value (in terms of predictions) because of quantity or quality of data or because there is insufficient tailoring to accommodate different data patterns due to different operating conditions, even for identical machines. Cost of tailoring may be currently prohibitive, as may be some methodologies such as the “digital twin” for anything other than large, critical equipment, such as power plant turbines or paper machines. And while there may be substantial awareness about PdM, that doesn’t mean there is enough understanding of the technology, given also the lack of machine learning and data science expertise in non-digitally native companies or companies not large enough to invest significant resources.

Finally, there is the issue of productization and monetization. Many companies, OEMs or service/maintenance contractors would be loth to invest in PdM without a clear idea of how to recoup the investment, either directly, e.g. packaged into service contracts for “uptime” or through longer-term improvement of products and systems. But this is easier said than done. For one thing, competition is already strong and pricing pressure is significant. For another, the internet of things situation is still in a state of flux and not yet standardized. And, in addition, the strategies of end-users, particularly larger players, is still unclear on a whole host of issues from data confidentiality to developing own PdM systems. Who is the best “owner” of PdM (systems, processes, data) is still unclear.

The dilemma for OEMs (large and small) and, to a lesser extent, for maintenance contractors is the commoditization conundrum. If OEMs are not directly connected to their products, their data and their performance post-sale, so that they can improve and optimize the value added generated from them, they risk these products becoming, over time, low margin commodities -as value shifts from the product to the utility they provide. As we have said elsewhere, the value of knowing when and how to intervene in a process is becoming as or more important than the value of the underlying products themselves. This has strong parallels to what is currently happening in the automotive industry, where the car as a product has come under sustained assault from mobility services (coupled with the double whammy of electrification and autonomy). Car manufacturers are scrambling to become mobility service providers and a good number of them will not make it in any recognizable form.

So, at Si2 Partners, we have started preparing a new study and management report in our “Technology in Service” series to follow our Augmented Reality report which was published in early summer.

Our Predictive Maintenance in Service study will include an overview of how and how well current PdM systems actually work to improve uptime and maintenance (from algorithmic, data and implementation perspectives); How successful and straightforward implementation actually is in different environments and what it takes to actually deliver value with PdM; And, finally, what approaches are (and should) OEMs (large or smaller) taking to participate in this market in terms of productization and monetization, in light of market conditions and approaches of end-users and competitors. It will be based on interviews, surveys, literature research and case study construction.

The design of the study is still a work-in-progress and we’d like to invite readers to would like to participate in the study either through case study or through opinion and experience to contact us. Participants will have access to the results free of charge and the opportunity to join in discussions and other exchanges as well as provide input in the design of the study/report.

If you are interested please email titos.anastassacos@si2partners.com