i4.0 today

Megan Ray Nichols is a STEM Writer. She enjoys writing easy to understand science and technology articles on her blog, Schooled By Science.

ABSTRACT

Traditional approaches to maintenance can often struggle to balance effectiveness against cost. Preventative maintenance can be effective but relies on unnecessary, better-safe-than-sorry maintenance checks. Reactive strategies save on upkeep at first but quickly become expensive when machines start to fail. Machine learning for predictive maintenance provides another option.

Predictive maintenance approaches use algorithms trained on machine data from IoT sensors to predict when machines will need maintenance. This allows businesses to reduce total maintenance costs while still preventing downtime and machine failure.

Below, we’ll cover the benefits of machine learning for predictive maintenance, as well as how factories can implement this technology.

The Value of Machine Learning for Predictive Maintenance
Most typical maintenance approaches are preventative or reactive.

With a preventative strategy, maintenance is performed at a regular interval to help catch machine failure and ensure good operation. Depending on the kind of equipment being maintained, these visits might happen very frequently — for example, a heating unit may be inspected before each winter, while a robot arm may need a combination of daily and monthly maintenance checks.

By design, many of these visits will be unnecessary and won’t uncover any serious issues. This maintenance approach can catch many problems, but it’s inefficient and expensive — both because maintenance costs money, and because some checks will require machine downtime.

With reactive maintenance, you solve problems as they arise. A reactive approach can save businesses on maintenance costs, but will also make them more likely to suffer from extended downtime and the costs of machine failure. This is especially true when critical equipment fails because rush-ordering replacement parts directly from the manufacturer can easily get expensive.

Predictive maintenance, by contrast, forecasts when maintenance will be necessary, allowing businesses to cut down on maintenance costs while still avoiding downtime and damage. A predictive approach to maintenance uses machine learning algorithms to find patterns in operational data — like machine temperature, timing or vibrations — pulled from industrial IoT sensors. With a predictive maintenance system in place, it’s possible to more accurately predict machine failure and efficiently schedule maintenance.

Some predictive maintenance systems can also be used to improve machine performance by optimizing operational conditions. Others are linked directly to the machine being monitored and can shut it down if operating conditions are moving into dangerous territory or suggest that failure is imminent.

On average, these systems have been found to reduce maintenance time by 20 to 50 percent and decrease maintenance costs by 5 to 10 percent overall.

Examples of Predictive Maintenance Applications

Predictive maintenance systems can be simple or complex as needed. They may rely on a network of IoT sensors, or just one or two. In most predictive maintenance applications, IoT tools are used to provide some combination of vibration, sonic or ultrasonic analysis, thermal imaging or oil and lubricant analysis.

Vibration sensors are used to track deviations from normal machine operations, while sonic and ultrasonic sensors can “listen” for electrical discharges and gas leaks. For example, vibration and oil-based predictive maintenance machine learning tools can be used to detect the contamination of gear mesh boxes by oil, preventing damage like pitting and spalling with the right maintenance.

Thermal imaging sensors render infrared action as visible light, allowing predictive maintenance machine learning tools to effectively “see” heat. Like vibration sensors, thermal imaging is mostly used to detect unusual operating conditions and identify potentially damaging issues, like loose terminal connections.

A range of other sensor types can also be useful. For example, Stress Engineering used predictive maintenance to successfully predict fault statesin positive displacement pumps with more than 90 percent accuracy. The team was able to achieve this with sensors that measure suction and discharge pressure data.

Integrating Predictive Maintenance Machine Learning Tools

Begin by determining the operational variables you will measure. Depending on the equipment you use, there may be an all-in-one predictive maintenance solution that you can use — Hitachi, for example, develops a predictive diagnostics platform specifically for wind turbines.

You can also put together your own with the right combination of tech. You will need three major components — IoT sensors that collect data, a solution that aggregates and stores that data and a predictive maintenance algorithm.

Once you’ve identified the tech you want to use, begin planning how you’ll implement it. With a predictive maintenance system, it’s a good idea to start small. A pilot program can help you determine which variables will provide the most value when tracked. It will also give you a chance to familiarize yourself with your predictive maintenance platform of choice.
Now, you can install the IoT devices that will measure these variables. You may need a combination of IoT devices, depending on which variables you want to track. Once everything is in place, activate the devices and connect them to your software platform. The sensors will begin collecting and transferring data, which will then be analyzed by your platform or algorithm of choice.

Expect a grace period before your predictive maintenance system begins returning usable results. ML algorithms perform best with large data sets and it may be some time before your IoT devices have collected enough data for the algorithm to identify patterns that indicate sub-optimal performance or potential machine failure. During this period, you should keep your standard maintenance schedule in place.

Implementing ML-Based Predictive Maintenance

Predictive maintenance can help a facility keep its equipment running well with more efficient maintenance schedules. Businesses wanting to integrate a predictive maintenance solution at one of their facilities have a range of options available — including all-in-one maintenance solutions, as well as building their own.

Megan Ray Nichols is a STEM Writer. She enjoys writing easy to understand science and technology articles on her blog, Schooled By Science. When she isn’t writing, Megan enjoys watching movies and hiking with friends.