According to Accelerate Your Data-Driven Transformation, a commissioned study conducted by Forrester Consulting on behalf of RapidMiner, 49% of manufacturers believe that artificial intelligence, machine learning, and advanced analytics is going to be the single most important factor in their competitiveness in the next 2-3 years. Of all the potential artificial intelligence and machine learning use cases for manufacturers, one of the most compelling is predictive maintenance.
Smart machines embedded with IoT sensors and armed with reams of data on optimal operations can save manufacturers time and money by lowering the amount of downtime for those machines. Anticipating, and acting on, potential out-of-service issues means that plants are able to run at peak efficiency.
A recent study by McKinsey says that “predictive maintenance typically reduces machine downtime by 30-50% and increases machine life by 20-40%”. That’s a compelling business case.
In this post, we’ll explain how you can harness the power of predictive maintenance to improve your manufacturing operations. But before we jump into the details, let’s establish a clear definition of what exactly predictive maintenance is.
What is predictive maintenance?
Predictive maintenance uses historical, available data to foresee when equipment failure is likely to occur so that you can proactively address that potential failure with maintenance. This maximizes efficiency and reduces downtime.
Predictive maintenance vs preventive maintenance: What’s the difference?
Let’s be clear: predictive maintenance is different than preventive maintenance.
Preventive maintenance generally involves inspecting a machine or piece of equipment and doing some sort of maintenance on it. It’s usually performed on the basis of time (for example, seasonal HVAC inspections) or usage (for example, rotating a vehicle’s tires at 25000 miles).
Predictive maintenance, on the other hand, can be continuously monitored and acted on when conditions fall out of optimal parameters.
Why predictive maintenance now?
What’s driving this move to predictive maintenance is the Industry 4.0 revolution. The Internet of Things (IoT) is one of the main enablers of Industry 4.0, as it allows machine-to-machine connection and communication where it was not possible or practical before. IoT sensors that are attached to industrial machinery generate data which is then captured, collected, and analyzed.
The collected data is the jumping off point for predictive maintenance. The data that’s needed for predictive maintenance is time-series data, meaning it’s collected at specific, discrete times. With that information in hand, you can start to build out machine learning models to predict when machines are likely to fail.
How to do it
The key techniques or models for using machine learning for predictive maintenance are classification and regression models. In classification, you can predict a possibility of failure in a certain number of steps. This method can be accurate with a limited data set.
A regression model would show how much time is left before the next possible failure (also called remaining useful life, or RUL). This type of model generally needs more data than a classification model would, but it can also be more accurate about when the failure will happen.
Depending on the industry, the need, and the sensors installed, there are many ways that the data needed for predictive maintenance can be generated.
There are digital sensors available that can essentially mimic the five human senses:
- Sight: Using infrared cameras, sensors can ‘see’ thermal anomalies, and the data generated can then help technicians pinpoint areas that need maintenance.
- Sound: Sonic and ultrasonic technologies can pinpoint potential trouble spots by ‘listening’ for friction that’s outside of normal operating parameters.
- Touch: The data equivalent of human touch would be vibration analysis, which can detect vibration in machinery in both normal and sub-optimal conditions.
- Smell: Some startups are developing ‘olfactory’ sensors that can be used to detect and analyze odor data. This could be used to detect a gas leak, for example.
- Taste: Similar in concept to odor, taste sensors are being developed that can quickly identify and classify liquids, which is useful for industries like pharmaceutical manufacturing or beer brewing, for example.
Sensors need to be in place in order to generate the data needed for machine learning. And there are lots of real-world applications in current use.
3 examples of predictive maintenance
Here are three specific examples of how RapidMiner has helped implement predictive maintenance to improve manufacturing operations.
1. Decreasing device failures & total downtime
In the airline industry—especially now that it’s under increasing pressure due to the global pandemic—it’s imperative for airline operators to understand when and where an airplane component might fail, anticipate maintenance needs in order to reduce costly downtime, avoid unplanned out-of-service intervals, and optimize in-flight and ground service crew schedules.
Airlines invest heavily in MRO—maintenance, repair, and overhaul—and generally have used preventive maintenance as the guiding principle to remain safe and airworthy. As always, MRO is a cost- and labor-intensive process. With preventive maintenance, airline service crews routinely perform checks on systems and make repairs and upgrades at predetermined intervals, such as specific hours of flight-time.
However, with over 1,000 airplanes to be maintained, Lufthansa needed to move beyond preventive maintenance and into predictive maintenance. As the largest German airline and the second largest in Europe, Lufthansa generates immense quantities of data that required analysis and assessment. But leveraging that data with machine learning allowed them to accurately predict and prevent failures.
Using RapidMiner’s data science platform, Lufthansa is able to make accurate predictions as to when a device or component might fail. This lets Lufthansa managers optimize aircraft fleet maintenance to make the most effective and timely use of service crews, as well as lower the amount of overall fleet downtime.
The predictive models also revealed where other maintenance obstacles might occur as well as identifying the root causes. As a result, total downtime was reduced by 20% and device failures and their subsequent costs were also reduced.
2. Lowering costs & increasing efficiency
A global shipping organization was looking to lower their costs associated with time in the shipyard and also to increase the efficiency of their spare parts storage. By moving from preventive to predictive maintenance, the shipper would be able to proactively address any maintenance issues before they became costly or caused unsafe conditions.
Predictive maintenance is also useful in determining and avoiding equipment breakdowns and failures while at sea. One of the values of predictive maintenance over standard operating preventive maintenance is that monitoring can be done while the equipment is in operation, meaning that there’s less downtime as you don’t need to stop everything to do inspections.
Using RapidMiner to enable machine learning for predictive maintenance, the organization was able to build models from a diverse dataset across many different systems including their error logs and messages, their on-board engine sensor data, their route schedule, the vessel’s age, level of crew experience, the weather history across the proposed routes, and their maintenance reports.
Leveraging the power of predictive maintenance, the shipper was able to reduce the times that their ships were spending in the dockyard, optimize their spare parts storage worldwide so that they had the right spare parts in the right places, and furthered the development of a resilient organization that was nimble enough to react quickly to rapidly changing conditions.
3. Reducing risk & improving operational performance
A local distribution company (LDC) in the United States natural gas industry needed to predict which parts of its pipeline were at greatest risk of failure and consequence to nearby residents and businesses. Fortunately, the company had a large volume of data available for use in building predictive models including:
- Data about the design, make, and layouts of the pipeline itself
- Geo-spatial data, which includes maps of the service area
- Data on the overall performance of the pipeline including leak surveys and pressure upsets
- History of pipeline failure
But the data is not always ready for use in building models. In some cases, the company collects data to satisfy regulatory requirements, but the regulations don’t ask the company to do much with it. So, they made heavy use of RapidMiner’s data prep functions to get the data in shape for modeling.
When trying to predict potentially catastrophic gas leaks, the company needs to understand and make very deliberate choices when weighing the trade-offs between false positives. This could result in digging a hole to find a leak that isn’t there (costly, but not a big deal) or predictive misses (not anticipating and fixing a leak, which could be very dangerous).
Predictive maintenance also helps the company with more day-to-day issues, such as optimizing the allocation of repair budget to the parts of the pipeline that need it most. This allows the company’s asset integrity management department to have a bigger impact without needing additional resources.
Using machine learning for predictive maintenance, this organization was able to become more efficient with repair operations, prevent and fix more leaks, lower risks of catastrophic damage, and improve their bottom line.
Predictive maintenance, as you can tell from reading above, relies on the collection and analysis of large amounts of data. In many cases, it also relies on a change in mindset of the organization, as moving from preventive to predictive maintenance may require new processes and procedures.
If you’d like help figuring out how to do this for your organization, reach out for a free, no-obligation AI assessment. We’ll use our expertise to guide you in finding the most impactful use cases for machine learning and AI in your business.
Getting a machine learning project off the ground is hard
With various stakeholders, differing background knowledge among team members, and administrative hurdles, many projects die before they have a chance to fly. The solution to this problem is to build a solid project foundation from the very first stages to set yourself up for success.
The process outlined in this guide will help make that easier.
Get a complimentary copy of the 2020 Forrester Wave: Multimodal Predictive Analytics And Machine Learning Solutions