Machine Learning and Sensor Technology Unlock Predictive Maintenance Insights

Plant and building managers are facing challenges on several levels — with labor issues, supply chain issues, and parts challenges all creating downtime risk and putting pressure on their performance goals.

More than 80 percent of manufacturers experienced at least one instance of unplanned downtime during the past three years, and a single factory can lose $2.3 million annually due to unplanned stoppages. In some industries, such as automotive, those costs are even higher, with losses of tens of thousands of dollars per minute of unplanned downtime.

In the current environment, the impact of unplanned stoppages is magnified by the hiring and labor shortages many companies face. Data from the Chamber of Commerce indicates that 35 percent of all job openings in durable goods manufacturing were unfilled at the end of 2021. That figure is likely to only worsen in the coming years, with 25 percent of the U.S. manufacturing workforce over the age of 55. By 2030, there could be up to 2.1 million unfilled jobs in the sector due to a growing skills gap.

The ability to get equipment spare parts is not something that can be assumed. For instance, at one point in 2021, 60 percent of the manufacturing sector reported they had experienced delays from domestic suppliers in the previous seven days.

All those factors have driven companies to move away from their status quo and evaluate Industrial Internet of Things (IIoT) solutions to help solve those problems.

The idea behind the IIoT isn’t new — the concept has teased plant managers for decades. But technology has finally reached the point where all that theory and planning is giving way to solutions that can create real value.

Predictive Maintenance a Priority — In Theory

In a cost-pressured environment, considering predictive maintenance (PdM) — and maintenance processes as a whole — is a critical step. Maintenance costs are generally the largest single controllable expenditure in a plant, often exceeding firms’ annual net profits. Predictive maintenance is a further evolution of the preventative maintenance processes already in place for most assets.

While preventative maintenance programs are a benefit, the schedules are generally time- or usage-based, but are implemented on assets that don’t have age-related failure modes. This increases the necessary planned downtime on those machines while also increasing maintenance expenses in the immediate term.

Predictive maintenance uses data measurements to detect system degradation and take action only when needed, often leveraging a variety of sensors to monitor potential failure modes. This eliminates any excess maintenance actions and allows organizations to focus on those issues that require attention.

When properly executed, predictive maintenance programs have strong results, with data from the Department of Energy showing an average 45 percent reduction in downtime, a 75 percent elimination of breakdowns, and a 30 percent drop in maintenance costs in general. However actually deploying a full sensor-driven PdM process across a facility has its own obstacles.

Overcoming Data and Connectivity Challenges

When evaluating Industrial IoT applications, one of the core questions that continues to surface is: “Why hasn’t this already happened?” — especially when viewed in the context of the wealth of consumer-facing technologies that people constantly see in other aspects of their lives.

The challenges have been multi-faceted. Factories are historically interference-rich zones for wireless transmission, and long cable runs are often not a good alternative due to both cost and potential inconvenience. For instance, the array of different device protocols requires additional harmonization processes. Even then, network and connectivity challenges often limit plant managers because of either IT policy restrictions or limited bandwidth.

The fact that additional transmitters and gateways are generally required to ensure that reliable data is pulled from the machine assets, has historically led many plant managers to focus only on their most critical assets. But as technology has become more standardized, connectivity enablers, such as Sensata’s wireless gateways, transmitters, and receivers have made expanding data networks more practical.

After addressing the connection challenges, the next step to generating real value is to use the data to create actionable insights about the asset. If companies go into a project without a plan for how that data can be tied into their processes to improve operations, the collected data will probably be discarded — the system becomes nothing more than an expensive vanity project that never makes it beyond the pilot phase.

The fallback for many firms has been manual walkarounds and route-based maintenance. But those procedures take away resources — either in terms of personnel or fees paid to outside firms — that could be used elsewhere, and they only deliver insights on a sporadic basis. The lag between inspections allows for a significant period where issues may develop.

Leveraging AI/Machine Learning for Sensor Insights

The idea of using a temperature sensor to provide an alert when an asset may be overheating is a valuable first step toward better machine asset monitoring. But it only scratches the surface of how companies can use machine learning and AI processes to push that approach into a truly scalable system.

Take for example, a series of temperature sensors mounted on different assets. Each of those pieces of equipment will have different standard operating temperatures. And even identical assets located in different places in the building may have variations based on draft air, sunlight, or other factors. For a sensor-only installation, different alarm levels will need to be determined and programmed manually. Each of those processes puts additional demands on the team installing and managing the system, adding either consulting costs or the potential for additional headcount. In either case, adding a solution that adds costs makes it harder to justify from an ROI perspective.

But with a system driven by machine learning, the platform will actively learn what the standard temperature envelope for a particular installation is and then set different alarm models and levels on its own, based on that information.

AI-driven calculations allow plant managers to understand what maintenance actions may be needed for an asset, and how quickly those actions need to be taken. As a result, plant managers gain insights that help them optimize planned downtime schedules, parts ordering, and more. As more assets are monitored, plant managers can expand their understanding of the relationships between assets.

Setting the Stage for Future Growth

With AI-driven alarm processes in place, companies can avoid the need to choose to monitor only their ‘most critical’ assets because of limitations on the expertise and size of their maintenance team.

With today’s labor and supply chain challenges, companies have invested in automation to a greater degree than ever before. However, by focusing monitoring efforts only on ‘critical’ assets, organizations are often in the dark about the true health of the rest of the equipment in their plants. An asset that may not have previously been ‘critical’ could have since become a crucial part of an automated process.

Monitoring all assets in a facility allows plant managers and operators to truly understand their plant and the performance of all its assets — identifying potential issues before they reach a critical stage. And that is an improvement that can drive real value throughout the entire company.

This article was written by Paul Heine, IIoT Product Manager, Sensata Technologies (Swindon, UK). For more information, visit here  .

---