During the course of 2014, the SAP Custom Development organization conducted its latest series of Innovation Days in and around Atlanta, Dallas, and Denver. Innovation Days are basically half-day workshops where a small number of representatives from various SAP customers meet with topic experts from SAP Custom Development to explore and discuss a particular subject.
In this series, we discussed the current state and the future of predictive maintenance. Featuring brief presentations provided by SAP accompanied by brainstorming activities, the intention of Innovation Days is to get all participants into a collaborative spirit and to engage in open-thought exchange in order to come up with new insights that all participants can profit from. Each day’s outcome is an experiment: Will the participants join the discussion? Will they have thoughts the others can built upon? Is the topic a good fit for this day’s discussion, or will the group drive the discussion elsewhere?
One of the first discussion points was which maintenance strategies do our customers from manufacturing, utilities, and high tech currently follow. Unsurprisingly, all agreed that reactive maintenance (that is, repairing an asset after it has failed) is still the most widely used maintenance strategy, especially with respect to noncomplex commodity assets used in processes that are interruptible without causing high costs. In other cases, especially where an interruption in production would result in high costs, scheduled maintenance in regular intervals is the predominant strategy.
Practically all participants had already begun thinking about predictive maintenance which involves the usage of statistical methods and machine learning. The key to predictive maintenance lies in integrating operational (OT) or sensory data with business data (IT) and potentially other types of data (such as geospatial or environmental data) in a single big-data platform. Analyzing this data over time, finding correlations among the different data types, and feeding the insights back into established processes are the other crucial building blocks for predictive maintenance. Any of our customers who were already attempting to implement a predictive maintenance strategy were only just in the initial stages – often working with static data. They were trying to get their feet wet, so to speak.
However, all participants could see the potential for their organizations to achieve high levels of business value from adopting a predictive maintenance strategy. Some of those participants were envisioning the natural path of optimizing scheduled maintenance; others were thinking widely about avoiding reactive maintenance entirely. The business use cases mentioned were focused on reducing equipment downtime , having spare parts at the right location just in time, and centralizing the monitoring and servicing of equipment.
On the practical side, participants seemed to agree that getting the operational data out of the assets and into a big-data platform via a standard IT infrastructure was where high investment costs were anticipated. Also of concern was the quality of the data and whether or not the above-mentioned relations and subsequent relevant insights could actually be found. As far as driving any actions from those insights, all operations leaders agreed that the relevance of human experience must not be overlooked in favor of machine data. At all times it is necessary to be able to explain which parameters have influenced maintenance recommendations, what the underlying models looked like, and what the original raw signal data stream from individual sensors looked like graphically.
This final part of the discussion revealed an influencing factor that cannot be underestimated: IT folks and OT folks are two different types of human beings. The former act in an IT-shaped environment; understand traditional, structured enterprise data; and have processes built around this data. The latter are focused on running or producing stuff. In their world, IT is a tool that must prove its value by making the running or producing of stuff easier and better. But sensor-produced data comprises raw, real-world data that does not fit neatly into IT’s relational databases. To paraphrase one participant: Trying to make sense of one OT data point is comparable to looking at an impressionist piece of art from close distance. One needs to move further away to grasp the meaning of the overall picture and drive the right conclusions. Thus, only after making sense of the data, establishing its relations to the other data, and proving the benefits is it possible to look at the application level. And what’s required at that level are light-weight, custom-tailored, role-based solutions that integrate seamlessly into existing operational processes.
At the conclusion of SAP Custom Development Innovation Days, we came away with the understanding that predictive maintenance is still in its early stages, but that our customers are looking to SAP to help them get their feet wet. Our approach here is to start with the data; let industry experts, data scientists, and big-data architects collectively investigate data available from different sources; and then identify a viable, feasible, and commercially desirable use case that allows a first step to be taken. It all often starts with IT and OT convergence and this is where SAP HANA comes into play. This first step can be a cost-effective undertaking in an initial sprint, paving the way for a complete predictive maintenance strategy that is built upon concrete, provable, and data-specific insights.