Unplanned downtime in manufacturing is one of the largest causes of lost productivity, causing delays, unhappy customers and lost revenue. In fact, the problem costs industrial manufacturers an estimated $50 billion each year, according to recent studies.
Short of a crystal ball, what can manufacturers do to reduce unplanned downtime? Predictive maintenance has made significant strides in recent years and represents a strong solution to this persistent manufacturing challenge.
The concept of predictive maintenance is not new, but the Industrial Internet of Things (IIoT) provides advanced predictive maintenance that combines sensors and machine learning to better predict when equipment may fail. These solutions hold tremendous potential. To increase adoption, we need to advance the technology to a point where manufacturers can make reliable decisions based on the predictions provided.
The Future Relies On Specifics
Most solutions on the market today don’t have the “right” data, lack high prediction accuracy, and don’t provide sufficient lead time to act on those predictions.
To provide accurate prognostics, you need large amounts of specific data about when faults happen or run-to-failure information. If manufacturers don’t already have this data, collecting it is a time-consuming, expensive task. While current solutions offer prognostics to an extent, in my experience, they often do not deliver on it. Additionally, existing solutions typically only give a few weeks’ lead time to act on their predictions. To fully trust a predictive maintenance solution, manufacturers need longer-term results.
Moving From Poor Predictions To True Prognostics
Predictive maintenance has evolved significantly over time. The next step in that evolution is newly advanced sensors and algorithms that enable manufacturers to predict future equipment faults more accurately and with more lead time. We are seeing the introduction of sensors that can detect different phenomena that can inform prognostics.
Recently, the scientists at our Palo Alto Research Center (PARC) worked on a Department of Energy project with Con Edison and General Electric (GE) to improve the maintenance and management of their power grid infrastructure. The team embedded low-cost fiber-optic sensors in critical components within GE model network transformers, where they measured key internal parameters and events of interest. Improved sensors can enable the creation of new capabilities that provide manufacturers with a more accurate picture of their overall system health.
To address data science shortcomings, a successful platform incorporates other forms of artificial intelligence, such as model-based reasoning. This means creating physics-based digital models of equipment that capture the subtleties of real-world environments and equipment fault modes and augmenting those models with machine learning based on the data gathered by the sensors. This hybrid approach can enable more accurate prognostics and makes predictive maintenance more accessible to manufacturers who lack the massive amount of historical data required by today’s solutions.
This model-based approach is something our teams at PARC found successful on another recent project with the East Japan Railway Company (JR East) in developing customized fault detection and diagnosis pilot software using model-based system analyses combined with advanced machine learning.
Implementing An IIoT Solution
In my conversations with innovators and operators, they often ask a version of the same question: How do I get started and take advantage of these new technologies? There’s no one-size-fits-all answer to this question, but I always give the same advice: Get the infrastructure right and have a minimum viable product (MVP) mindset.
From an infrastructure standpoint, preparing internal networks for new sensors and services is often a stumbling block for many operators. IIoT sensors generate vast amounts of data, and most operational technology (OT) networks do not have the bandwidth to support IIoT deployments. Simple things like investing in data historians are a good first step, but longer term, firms should strongly consider investing in cloud readiness and IT, OT and IIoT integration. This upfront investment can enable faster implementation of innovative models, analytics and software that touch on production and operations.
Having an MVP mindset means taking the quickest route to value. Focusing on the problems plant managers and operators are looking to solve and defining the shortest path to testing and deployment is key to avoiding institutional fatigue with IIoT. In predictive maintenance, working with vendors that have prebuilt models that allow for cold starts is one way, but not the only way of accomplishing this.
The value of correctly implemented predictive maintenance technology can be incredibly high. For example, in 2016, Schlumberger announced that a new predictive analytics program it used to forecast equipment issues for fracturing pumps saved more than $8 million in less than a year. At the same time, it estimated that a similar program would save more than $30 million over three years.
Imagine the value of a solution that is even more accurate and provides more lead time?
If we can solve the problem of prognostics accuracy, we can deliver less downtime and more productivity and profitability, significantly impacting manufacturing across the globe. Zero unplanned downtime would be a boon for business globally. Though it will take time and a disciplined approach, there is much to gain from applying IIoT predictive maintenance to manufacturing.
Originally posted on: Forbes.com
