Predictive Maintenance and Machine Learning: Revolutionizing Reliability
Predictive Maintenance and Machine Learning: Revolutionizing Reliability
by Richard Irwin
Figure 1: PdM is one of the main benefits across all industries, particularly oil and gas.
"Machine learning is paving the way for smarter and faster ways to make data-driven decisions in predictive maintenance (PdM)."
One of the goals of reliability is to identify and manage the risks around assets that could fail and cause unnecessary and expensive downtime. Organizations know it is important to identify areas of potential failures and rate them in terms of likelihood and consequence. They also have put in place good reliability strategies and have implemented proactive, condition-based maintenance programs. But now, machine learning is helping maintenance organizations get to an elevated level of situational intelligence to guide actions and provide early warnings of impending asset failure that previously remained undetected. Machine learning is paving the way for smarter and faster ways to make data-driven decisions in predictive maintenance (PdM).
Figure 2: Machine learning can help the smart grid get even smarter
While machine learning has been researched for decades, its use in applying artificial intelligence (AI) in industrial plants and infrastructure asset operations is now advancing at a rapid pace. This influx of using machine learning is due to the growth in big data, the expansion of the Industrial Internet of Things (IIoT), the availability of computing power to number crunch this increase in data, as well as the need for superior predictive and prescriptive capabilities required to manage today’s complex assets. While machine learning has typically been linked to such industries as transportation and banking (think self-driving cars and fraud monitoring, respectively), there are many uses for machine learning and PdM within the industrial sector.This article focuses on some of the principles within machine learning and the industries primed to take advantage of its application to maximize the benefits machine learning brings to improve situational intelligence, performance and reliability.
But first, it is important to point out that there are many options and techniques available to gain more insight and make better decisions on the operation and performance of your assets. It all comes down to knowing what the best fit is for your needs and what type of data you are using. Data comes in many shapes and sizes and can consist of time series, labeled, random, intermittent, unstructured, and many more. All data holds information, it’s just a case of using the right approach to unlock it. This is where algorithms used within machine learning help decision makers.
6 Questions to Ask Before Investing in Machine Learning
It is important to understand the complexity involved with machine learning before you make a decision on what is appropriate for you and your organization. Here are some questions to consider before implementing machine learning:
- What do you want your data to provide Question your data. What do you need to know, what are you looking for exactly? What do you want your data to tell you? What aren’t you seeing that you hope the data can provide?
- Is your data clean? Make sure your data is available, ready and validated. The more data the better and the more accurate the outcomes will be.
- Do you have enough data? For accurate predictions, machine learning needs lots of historical data from which to train, then it can be applied to data in real time.
- Which machine learning platform should you choose? Choose your machine learning platform carefully and consider interoperability.
- Should you hire a data scientist and how will this individual be integrated into the organization? With machine learning, there might be a need for a data scientist or analyst, but this individual should not be locked in a dark room.
- Can you share the data output? Knowledge gained through machine learning shouldn’t be applied to just one project at a time. Its scalability means it can and should be incorporated across the whole enterprise, delivering insight into any area rich in data. Plan to get the most out of machine learning.
Figure 3: Machine learning is comprised of many different data science techniques
The Route to Deeper Understanding
Machine learning makes complex processes and data easier to comprehend and is ideal for industries that are asset and data rich. In any industry, the ability to recognize equipment failure and avoid unplanned downtime, repair costs and potential environmental damage is critical to success. This is even more relevant in today’s turbulent times. With machine learning, there are numerous opportunities to improve a situation with PdM and the ability to predict critical failures ahead of time.
PdM is one of the most relevant areas where machine learning can be applied within the industrial sector. Predictive maintenance is a failure inspection strategy that uses data and models to predict when an asset or piece of equipment will fail so proactive, corrective actions can be planned in time. PdM can cover a large area of topics, from failure prediction and failure diagnosis to recommending mitigation or maintenance actions after failure. The best maintenance is advanced forms of proactive, condition-based maintenance. With the combination of machine learning and maintenance applications leveraging IIoT data, the range of positive outcomes and reductions in costs, downtime and risk are worth the investment.
Whichever path is chosen, the benefits machine learning can offer to big data are just being brought to fruition. Opportunities are rapidly developing, with productivity advancements at the heart of the data rich industry in which you work. Here are some examples leading the way in this fast-moving digital transformation.
Electric Power – Electric utility companies are affected by aging assets, increasing energy demand and higher costs. The ability to recognize equipment failure and avoid unplanned downtime, repair costs and potential environmental damage is critical to success across all areas of the business. Machine learning is augmenting the smart grid to better leverage and gain insight from the IIoT, with an enormous number of connected assets spread across a large network. With transformers, pylons, cables, turbines, storage units and more, the potential for equipment failure is high and not without risk, so predicting failures with data and models is the new answer to keeping the network running smoothly.
Oil and Gas – In the oil and gas industry, the ability to recognize equipment failure and avoid unplanned downtime, repair costs and potential environmental damage is critical to success across all areas of the business, from well reservoir identification and drilling strategy to production and processing. In terms of maintaining reliable production, identifying equipment failures is one of the main areas where machine learning will play an important role. PdM predicts when an asset or piece of equipment will fail so maintenance can be planned well ahead of time to minimize disruption. With the combination of machine learning and maintenance applications leveraging IIoT data to deliver more accurate estimates of equipment failure, the range of positive outcomes and reductions in downtime and the associated costs means it is worth the investment.
Water Utilities – Water companies also face the same challenges of an aging infrastructure, rising costs, tighter regulations and increasing demand. They also share the same benefits that machine learning offers, such as identifying equipment failure before it happens, but not just to predict a failure, but also to identify what type of failure will occur. Other machine learning benefits in the water industry include meeting supply and demand with predictive forecasting and making smart meters “smarter” to help curb waste, such as during water shortages.
Manufacturing – Manufacturing has been the main industry mentioned alongside machine learning, and for good reason, as the benefits are very real. These benefits include reductions in operating costs, improved reliability and increased productivity — three goals that relate to the holy trinity of manufacturing. To achieve them, manufacturing also requires a digital platform to capture, store and analyze data generated by control systems and sensors on equipment connected via the IIoT. Preventive maintenance (PM) is key for improving uptime and productivity, so greater predictive accuracy of equipment failure is essential with increased demand. Furthermore, by knowing what is about to fail ahead of time, spare parts and inventory can use the data to ensure they align with the prediction. Improving production processes through a robust condition monitoring system can give unprecedented insight into overall equipment effectiveness by regularly and consistently monitoring air and oil pressures and temperatures.
Digitization and Transformation With Machine Learning
Early adopters of machine learning are already reaping the benefits of PdM in the speed of information delivery, costs and usefulness. This gives them more information and insight to make smarter decisions. Some of these early adopters are also combining machine learning with other digitization technologies, such as visualization dashboards, cloud-based IIoT data, analytics and reality modeling, for an even more model-centric, beneficial process. The result is a complete solution for operations, maintenance and engineering.
Having a PdM plan in place powered by machine learning will give you unprecedented insight into your operation and lead to significant benefits in efficiency, safety, optimization and decision-making. The digital transformation for industry is now at a tipping point, with technologies all converging at the same time. A PdM approach to reliability and asset performance means root cause analysis (RCA) could become a thing of the past. In its place will be machine learning, which takes into consideration the whole history of failures and identifies the signs of failure in advance.
Figure 4: A PdM plan provides unprecedented insight regarding your assets
Machine Learning Case Study
This case study demonstrates the application of various machine learning techniques within a processing plant.
A steel manufacturer routinely shuts down operations to perform maintenance on its assets, which is very costly. The steel output can sometimes warp or crimp during the production process as it travels through different stages. These failures only can be corrected every six months, as well as monthly for smaller fixes, during planned and very expensive maintenance that involves long periods of downtime.
The goals the steel manufacturer wants to achieve are:
- Reduce defects and locate root cause;
- Identify key variables that matter the most;
- Prioritize assets during shutdown.
It was determined that machine learning could help the steel manufacturer meet its goals.
The first part of the machine learning process was to sort the data into a self-organizing map using neural networks to organize data into 10 distinct classes based on parameters of the steel, such as thickness and weight, as it entered each manufacturing stage. Other techniques included decision trees to learn the pattern of data and identify which features were important in those patterns; asset health prioritization to provide ranking; asset health indexing to determine the health of the assets; principle component analysis to reduce the dimensionality of the data; and clustering and anomaly detection, which highlights how each stand deviates from its normal operating mode.
What developed was a method for dealing with different types of products, the ability to identify the top variables associated with production defects and a process for applying anomaly detection to equipment in an industrial plant.
It was shown that these processes could reduce the need for extensive analysis of equipment and give operators better tools and more insights to make maintenance decisions. A significant amount of time is spent locating the cause of the issues and performing maintenance. The new algorithm can be run before planning the shutdown and it can identify which stand to prioritize during shutdowns through analysis of the asset anomaly charts. Focusing on assets that are the most at risk optimizes the shutdown, as it is only conducted for a limited time.