On one hand, artificial intelligence (AI) and machine learning (ML) have made huge advancements and offer tremendous potential to save maintenance costs and increase uptime by identifying problems before they escalate. On the other hand, the relevant data these ML models need is often hard to gather and aggregate into disparate systems. Legacy assets plod along without realising the full promise of digital transformation and Industry 4.0.
But what if there were an easy way to access the right data—in the right place at the right time?
The move toward predictive maintenance
Industrial downtime delivers a severe blow to the bottom line. Seven years ago, Gartner pegged the dollar amount at a staggering $300,000/hour and that number could only have gotten worse. As a result, effective asset management has become a central component of every manufacturer’s long-term maintenance strategy.
Companies have traditionally relied on preventative maintenance, replacing, or checking parts on critical machines according to a fixed schedule, just like changing the oil in your car every 3,000 miles. There are three problems with this approach: first, preventive maintenance means downtime—even if planned. Second, you might be replacing parts before extracting their full value. Third, preventive maintenance is expensive.
Predictive maintenance is a much more nuanced and cost-effective approach. A recent survey found that predictive maintenance implementations yielded a positive ROI in 83% of the cases. The promise of this method lies in data analytics. Essentially, a whole set of machine “readings” feed into ML algorithms to model a baseline assessment of what normal looks like. From there, any reading that deviates from this baseline alerts the system for proactive action. This method is a low-touch approach: It allows you to get the most out of your assets without interference until and unless the equipment signals otherwise.
The other significant advantage of predictive maintenance is that it surfaces problems ahead of a potential breakdown, so crews have enough time to respond.
The sensor challenge
While predictive maintenance promises an exciting frontier for a more efficient plant, it requires a lot of data: temperature—an overheated machine is a sure signal of impending disaster—vibrations, humidity, and more.
The data for ML models come from sensors on assets or from workers who walk the production floor taking readings at set intervals. But most legacy assets are not outfitted with sensors and to add them at scale is an expensive proposition if you have hundreds, even thousands of assets. And, as many of these legacy assets will remain operational for years to come, waiting to upgrade them is also not an attractive option.
Traditionally, one solution has been dispatching employees on rounds and readings routes. But sending workers on such missions can be expensive and inconsistent, often relying on institutional knowledge that doesn’t get passed on when people retire or move jobs. This work can also be dangerous, particularly hazardous areas of a plant that may require specialized safety equipment to access. And sporadic or inaccurate data are not useful enough for effective predictive maintenance.
The alternative: Bring the sensor to the asset
It is time we rethought the approach to predictive maintenance. You don’t need every asset to be bristling with sensors or employees checking hazardous areas. Mobile edge devices, especially on robots, can do the job efficiently and at scale. In addition to keeping your employees safer, robots free up personnel to spend more of their time doing higher value work.
This dynamic sensing approach uses a set of sensors, customized to your application, as a payload on an agile mobile robot. The robot can go wherever a person can, accessing all your equipment and taking the right measurements at the intervals you set. A quadruped robot can be outfitted with a variety of payloads and integrated with preferred sensors and software tools using onboard API. Thermal scanning, radiation detection, gas detection, acoustic modelling, vibration analysis, gauge reading—the quadruped can measure and process whatever data matter most to ensure uptime.
Equally important, the robot feeds the data gathered into asset management software from established industry leaders, processing the information at the edge. Integrated with a machine learning analytics system, an agile mobile robot can make an instantaneous decision about the health of the asset. If the mobile edge robot detects a problem, it can automatically trigger processes for repair, such as creating a work order and notifying a site worker responsible for the equipment.
These robots continuously collect and analyse asset performance data to improve reliability and operational efficiency while also keeping employees safe.
Bridging the gap
The manufacturing industry has been embracing the path to digital transformation. But before it adopts more advanced technologies such as AI and machine learning, it continues to wrangle with the problem of legacy equipment. How do we make legacy equipment “talk?” How do we extract the meaningful data we need; at the frequency we need? A dynamic sensing network with the custom sensor payloads integrated with the right software on a legged robot can be the answer. It helps us embrace advanced technologies in the right spirit—efficiently and safely.
Stop worrying about the complexities and minutia of asset management. Customized robotic solutions with the right sensors for your facility as payloads deliver the best of all worlds: predictive maintenance on legacy assets with a low-touch approach.
Source: Boston Dynamics