The Impact of the IIoT on Predictive Maintenance, and What its Future Holds

By Peter Vowell

The full impact of product connectivity has yet to be realized by the industrial world. When people think of the internet of things (IoT), many picture consumer devices - smart sneakers, connected doorbells, maybe even home security. The IoT has farther-reaching implications, however. The ability to stay constantly connected is creating big changes in equipment maintenance, allowing tools and processes to constantly report on their own status. The current marketplace of predictive maintenance is being transformed by these changes, but it will take more than a few fancy sensors to create something truly revolutionary. The idea of predictive maintenance isn’t a new concept. It was being suggested by engineers back in the early nineties, to prevent machine failure due to oil contamination. While the idea may not be new, industry capabilities have vastly changed. New connectivity equipment has combined with low-cost sensors to provide accurate quality monitoring. Processing power has become increasingly easy to obtain, allowing for comprehension of the vast array of information sensors produce. Machine learning has risen to new heights, enabling complex adaptability suitable for an equally complex world. The result of these advancements presents itself as intelligent and adaptive predictive maintenance and is in huge part due to the IIoT.

Connected monitoring equipment requires two main things: sensors, and a method of data transmission. Bluetooth solutions have made big strides in the last six years, largely due to the introduction of Bluetooth Low Power (LE). If you choose to take the path of the Bluetooth-enabled sensor, it’s important to understand the differences between normal Bluetooth and the LE counterpart. While lower power consumption sounds good on the surface, there is a tradeoff. Bluetooth will transmit data continuously, making sure your applications have the latest sensor data at hand. Bluetooth LE transmits small amounts of data packets periodically, to conserve energy. If your monitoring motor rotation time, LE will probably be too slow for your needs. If you are looking at water quality, transmitting every ten seconds or so shouldn’t be an issue. Choosing the right solution for a given situation is one of the most challenging portions of preventative maintenance. The other challenge comes from the sensors themselves. It’s surprising how many equipment providers don’t want to discuss this part of the recipe - they make it seem as if the data will just produce itself. You will need sensors, you will need a lot of them, and you will need the time to set them up. Setup is possibly the most daunting portion of predictive maintenance, precisely because of the initial time investment it requires. However, the time it takes to install is incomparable to the time and finances that will be saved by taking the IoT maintenance route. An intensive study performed by Jones Lang Lasalle indicated a return on investment (ROI) of an astounding 545 percent over a 25 year period. This seems incredibly high but becomes reasonable if you start thinking about the cost of replacing a failed tool. Even enhancing tool lifetime by two to three years can save a company thousands, making connected preventative maintenance an obvious choice.

It’s hard to talk about the advancements in IoT without talking about “big data”. Taken in a literal sense, big data is just a large amount of information, which doesn’t tell much of a story. Big data is just any raw data set that pertains to a specific system or group. The number of sensors in monitoring systems continues to increase, allowing for a clearer picture of device operation. It also means the monitoring equipment is producing more data per minute, and this data is useless unless it goes through processing and analysis. Because so much of this technology is new and thus in an early adoption phase, it can be hard to judge how well it will function in specific use cases. In the instance of Caterpillar however, connected preventative maintenance clearly produced significant results. The Marine Division of Caterpillar often serves customers whose profits depend heavily on their own fuel use. By setting up an array of sensors focused on fuel consumption, customers were able to determine that generators ran more efficiently at lower power outputs. Thus, they implemented a system that used more generators but actually consumed significantly less fuel. This predictive measurement allowed for the implementation of a non-obvious solution that led to monetary savings, as well as having a positive environmental impact.

Machine learning is where the future advancements will take place for predictive maintenance. So far, we have discussed sensors and data processing. The sensors produce data, but can’t relay any information about what that data means. Big data processing allows for efficient compilation of that information, but it still needs to be analyzed manually. Machine learning utilizes intelligent algorithms to comprehend and analyze incoming data sets, and then indicate anomalies and points of interest. This may sound simple when stated so concisely, but it’s a daunting task with no clear correct answer. The popular approach is to allow a “learning period”, where an analysis program learns what is normal, and then compares everything else against this benchmark. If an instance occurs that strays outside pre-defined tolerances, then an alert is created indicating that a maintenance issue may need attention. An approach like this has a couple clear flaws however. If the tool isn’t functioning at one hundred percent efficiency during the machine learning period, then the golden standard every data set is compared against isn’t the best solution. Secondly, this method tends to produce an extraordinary number of false positives. Take for example a motor that has a vibration sensor monitoring for excessive tremor. If the motor vibrates too much, a maintenance request will trigger, so the tool receives attention before it malfunctions completely. The location of the motor could cause constant false positives, however. If the motor is on a boat, waves crashing into the side of the boat could cause an alarm to trigger even though the operation is normal. Currently, there is no perfect machine learning option available to the preventative maintenance market. The future it promises, however, appears bright on the horizon and could lead to much more practical solutions for the industry. Increasing sensitivity in sensors will only enhance the experience, and low-power wi-fi and Bluetooth solutions continue to decrease the energy consumption of IoT systems. While there may be no definitive solution currently at hand, the next five years will yield a wealth of new products and maintenance solutions.