From Preventive to Predictive Maintenance

Part 1 — How machine learning and artificial intelligence are changing the way we take care of things.

Planned Preventative Maintenance (PPM) has long been the dominant paradigm for realizing the full life span of any major piece of equipment, system, or other assets — even our bodies. We rely on metrics-based schedules to conduct routine checks, clean components, make repairs, and replace parts. But with recent advances in machine learning and artificial intelligence — the technologies at the core of the Sparta Science System — we can move toward a whole new, more efficient paradigm that is individualized to the asset in question and how it is being used.

In this first installment of our three-part series on predictive maintenance, we will define what it is, how it differs from prior systems, and the potential impact it can have. Our second post will outline how and why the military is making the switch to predictive maintenance for some of its most valuable equipment. In our concluding post, we’ll examine how the concept of predictive maintenance can be applied not just to equipment and infrastructure but to protecting our most valuable asset of all — the human body.

How We Have Traditionally Thought About Maintenance

To maximize the return on investment in a piece of equipment or infrastructure, it must be well tended to. But what does that entail? Until recently, PPM was the dominant system for taking care of an asset to prevent breakdowns and realize the benefits of its full lifespan.

PPM involves adhering to metrics-based schedules of cleaning, repairs, and replacements. These metrics can be measured in time, such as the recommendation for most cars to have their oil changed every six months, or usage, such as changing its tires after 25,000 miles. These metrics come from manufacturer recommendations, which in turn are typically based on data from early production cycles and testing. They reflect best estimates of typical usage — considered both in quantity and conditions. PPM’s central tenet is to adhere to the schedule, regardless of the asset’s actual condition.

In theory, PPM should work to maintain most assets. But in reality, each piece of equipment is unique and ends up being used in dramatically different conditions and in different ways, many of which may not have been contemplated by manufacturers. As a result, generalized recommendations often result in too much maintenance, wasting valuable resources, or too little maintenance, which can result in unanticipated equipment downtime or can even prematurely cut short the equipment’s lifespan. Until recently, however, standardized best practices were all we had. Now, that has changed.

AI and the Move Toward Predictive Maintenance

With exponential increases in the amount of data we can collect from our machinery and the rise of machine learning and artificial intelligence (AI), we’re witnessing a significant shift away from a “one-size-fits-all” approach to maintenance toward a predictive maintenance approach. A predictive maintenance paradigm leverages machine learning and AI to process the copious data, identify patterns of problem indicators, and use them to generate actionable, individualized care recommendations.

Unlike PPM, which relies only on historical data — and in limited quantities — predictive maintenance is based on a high volume of continuous data inputs, much of which is collected in real-time. Machine learning and AI can analyze the data collected from machinery and equipment at unprecedented scales, identifying patterns that can be used to make predictions and recommendations for action. As more and more data is collected from a specific organization and even asset, these insights evolve, becoming even more targeted and accurate.

With predictive maintenance, therefore, we can foresee points of failure and take corrective action before they occur, but not before they become necessary. In this way, there is less surprise downtime and maintenance procedures become the most cost- and resource-effective. In some estimations, predictive maintenance increases productivity by 25%, reduces breakdowns by 70%, and lowers maintenance costs by 25% on average.

Predictive Maintenance vs. Preventative Maintenance: A Case Study

Implementing the technology and data infrastructure for predictive maintenance represents an upfront investment of both time and resources, but one that industries with sophisticated machinery are increasingly willing to make. Not only is it more efficient, but it is also more reliable — an imperative in situations where equipment downtime and malfunctions can have both costly and dangerous repercussions.

Take, for example, KONE, a Finnish company that builds and operates some of the tallest elevators and escalators in the world. Traditionally, KONE’s elevators had been maintained on a preplanned calendar basis or ad hoc, when problems arose. This approach did not take into account actual usage or customer needs, both of which varied tremendously from location to location. They decided to make the shift to a predictive maintenance schedule.

To monitor and manage its maintenance status, KONE partnered with IBM. All of its elevators and escalators were built with IoT-connected sensors, which enabled the continuous transmission of data to the IBM Watson Internet of Things Platform. This data served as important indicators of how the equipment was functioning, and when processed by Watson’s machine learning and AI capabilities, could be translated into early warning signals for when something was predicted to soon go wrong. This allowed KONE to act on a potential problem before it was reported by building operators or customers, but not before corrective action was actually necessary. The result has been less equipment downtime, fewer faults, and more general information about equipment performance and usage so KONE can improve future products. And as the system continues to collect data, its artificial intelligence grows stronger and more accurate.

Just as KONE partnered with IBM, Sparta Science’s technology sends raw data collected from its force plate to AWS Sage Maker and Apple’s Core ML to be analyzed and translated into meaningful metrics. From here, Sparta’s system is able to give individual, actionable insights regarding performance, injury risk, and corrective training. As Sparta’s technology gets to know an individual, it can adapt workouts based on progress and readiness.

From technology in elevators that automatically notifies technicians of potential problems to sophisticated AI software being implemented by the military, predictive maintenance is a growing field with the potential to transform how we take care of equipment, products, and even our bodies. Find out how predictive maintenance is beginning to impact the military in Part 2 of the Predictive Maintenance series.