Over the years, the humans in charge of maintaining industrial machinery have developed a variety of strategies for setting and following schedules of care for that machinery. They track the number of hours a machine has run and check that against the expected lifespan of its critical components. And still, equipment fails, idling production and projects.
Now, some businesses are giving their maintenance teams an additional tool to help their maintenance schedules by keeping equipment running and catching problems before they can shut down a line, or a plant: artificial intelligence (AI). They are using pattern recognition software or embedding AI sensors into key components and tying them to machine learning systems that can be trained to discern when a part is giving off signs of distress that could mean that it needs repair or replacement. Preventive maintenance is becoming predictive maintenance.
While AI-driven maintenance is still in its infancy, its potential seems large. Studies have found that machine learning can predict a part failure with a high degree of accuracy. In 2016, Manufacturing Business Technology asserted that predictive maintenance would help companies save $630 billion in costs over the next 15 years (1). The higher the capital cost of your assets, the more you should be getting comfortable with predictive maintenance.
One approach to predictive maintenance is to use pattern recognition, something already used in many aspects of our daily lives. If you are in the habit of shopping only at the stores closest to your house and one day a large purchase is made with your credit card number at a store 100 miles from home, you will likely get a call from your credit card card company asking if you were the buyer. Your credit card company has recognized that that is not part of your usual pattern. When pattern recognition is applied to machine maintenance, software might read the existing activity and error logs on your machinery and looks for patterns in them. When signs of trouble are spotted, your maintenance team gets a heads up and can get to work. Your line stoppages become only those for planned maintenance and not malfunctions.
As industrial machinery becomes more computerized and connected--the so-called Internet of Things--there are opportunities to move beyond passive monitoring into predictive maintenance and quality control. Issues are resolved when the equipment needs it, and not simply because it is the third Friday in the first quarter, or whenever an arbitrary calendar has called for it. The data science company deepsense.ai recently gave this example: An automaker implemented a predictive maintenance system on a hydraulic press and it forecast failures with a 92% accuracy rate (2).
Data scientists are rising to the challenge, often through competitions sponsored by manufacturers. Bosch, the giant German manufacturer of auto parts and consumer appliances, posted an online challenge to identify a way of predicting which of the auto parts it made would fail quality control based on things that had happened to that part earlier on the assembly line (3). More than 1,300 data science teams from all over the world participated in the competition, solving the challenge and earning a top prize of $15,000. Since machinery often fails based on a chain of events, the work could help companies to also better understand maintenance across a wide range of other situations.
What can your operation do to get on the predictive maintenance bandwagon? Information Age advises starting by gathering together all the data you currently have on your machinery and the environment in which it operates (4). That means data from previous service records and the manufacturer’s service recommendations, but also from building management systems and inspections by your maintenance staff. You may also need to gather data on geography or weather if those can affect your machines.
Even for small operations, this is likely to be a mountain of data that could swamp an internal data science team, if you have one. There are a growing number of AI predictive maintenance companies to outsource the analysis to, or consider throwing the work to a Kaggle competition, as Bosch did. Going with an outside company can help you to tap into analysis that goes well beyond your own machines. Augury, a New York- and Israel-based predictive maintenance technology specialist, has its HVAC maintenance systems installed in more than 2,000 facilities across the United States and Canada. Data gathered from one location goes into a “malfunction dictionary” that can be used for all.
When you get results, share them widely with your C-suite and maintenance teams, and think about having real-time results displayed in an easy-to-read dashboard so that maintenance staff can monitor it, even when they don’t need to intervene. Then use the data to devise a maintenance action plan for the equipment that will take it down for service when that will least disrupt operations.
The goal here is not to replace your maintenance workers; they will always be needed. Rather, preventative maintenance through predictive maintenance can help them focus their skills where and when they are needed most.
Sources
3. Bosch. "Production Line Performance." January 2017.
4. Information Age. "Preparing for AI With Predictive Maintenance." Nick Ismail. July 2017.
Additional Resources
The information contained in this article is intended for general information purposes only and is based on information available as of the initial date of publication. No representation is made that the information or references are complete or remain current. This article is not a substitute for review of current applicable government regulations, industry standards, or other standards specific to your business and/or activities and should not be construed as legal advice or opinion. Readers with specific questions should refer to the applicable standards or consult with an attorney.
SIGN UP FOR EMAIL
Get more great content like this sent to your inbox.
1994-2023, W.W. Grainger, Inc. All Rights Reserved.