AI alone won’t fix reliability — context will
Key Highlights
- Reliability teams depend on effective data management and contextual analytics to transform sensor data into actionable insights.
- Machine condition monitoring spans from simple vibration sensors to advanced AI-driven diagnostics, each suited for different asset criticalities and organizational needs.
- Context matters: understanding machine specifics, failure modes, and operational conditions enhances diagnostic accuracy and early fault detection.
- AI tools are most effective when integrated with structured, domain-specific analytics, automating workflows without replacing expert judgment.
- Scalable solutions, starting with affordable wireless sensors, can grow into comprehensive systems, delivering measurable ROI and operational benefits.
Reliability teams rely heavily on data to deliver improved outcomes and build the foundation for operational excellence across their organizations. Today’s process manufacturers are collecting more data than ever before because the value of that data is well known. However, in the rush to collect as much data as possible, many organizations do not stop to consider exactly how they will manage it and turn it into actionable information.
As a result, while today’s industrial organizations have lots of data — especially from sensors, monitoring systems and IIoT technologies — many still struggle to turn that data into decisions. This challenge can make companies hesitant to adopt the most effective technologies to evolve their condition monitoring programs. Their leaders consider their lean teams and tight budgets and wonder how they could possibly use their limited resources to effectively collect, interpret, contextualize and act upon increasing amounts of data (Figure 1).
Fortunately, the value of modern analytics is not derived from data volume or user sophistication alone, but also from the integration of data context, user workflows and the appropriate application of analytics to successfully meet organizational needs. Regardless of an organization’s starting point, there is always an effective and right-sized approach to holistic reliability based on effective data management and analysis.
Not all data is created equal
One of the most challenging elements of implementing a machinery health analytics program is that the concept of analytics exists on a spectrum, with each segment of that spectrum playing a specific role. Reliability analytics spans a wide range of complexity:
- Basic scalar measurements
- Waveform and spectral analysis
- Targeted diagnostics
- Advanced AI-driven insights
Each of these levels builds on capabilities of previous levels and varies in complexity. Each level can provide a different value based on the criticality of an individual asset, as well as the organization’s available expertise and operational goals. Ultimately, there is no one-size-fits-all approach to implementing analytics. Reliability teams must consider their facility or enterprise’s unique situation, and then select the right level of analytics for the right application in each circumstance (Figure 2).
Basic monitoring: useful, but limited
The most basic form of machine condition monitoring many organizations undertake is overall vibration monitoring. With basic monitoring, teams add simple sensors to assets to collect the overall vibration produced by that equipment. These sensors can detect significant issues, such as severe misalignment or advanced bearing damage.
Basic vibration monitoring offers a few key strengths that typically make it one of the first stops on an organization’s holistic condition monitoring journey. The simple sensors used for basic monitoring are low cost and easy to deploy, even with minimal expertise. This makes them suitable for organizations with limited budget and personnel resources. They offer a level of protection that can be critical, especially for organizations that do not have the personnel necessary for regular maintenance rounds.
However, basic sensors also come with their own limitations. Typically, these simple sensors offer very late-stage detection of issues. By the time the sensor identifies a problem, mechanical damage has often already occurred. Moreover, an alert from a basic vibration monitor typically necessitates immediate action, which often means reacting to problems on the asset’s schedule, rather than the organization’s schedule.
In addition, these sensors typically offer minimal to no diagnostic insight, and the available sensing options do not go far beyond what a human could identify through manual inspection. Ultimately, basic vibration sensors are best suited for low-criticality assets, replaceable equipment and systems not critical to production.
Spectral analysis: moving toward insight
The next level of machine condition monitoring incorporates vibration waveform and spectrum analysis. In this case, sensors will deliver a vibration waveform and transform it into a spectrum, showing the amplitude in the vibration energy versus the frequency.
With spectrum data, teams begin to get more meaningful information. The spectrum empowers teams to spot specific patterns tied to identifiable mechanical behaviors. With the most advanced wireless vibration monitors, spectral analysis can monitor energy in specific frequency bands. This empowers users to zero in on families of specific failure modes, delivering significantly earlier fault detection.
However, while spectral analysis delivers significantly more actionable information, it still requires interpretation. The results from these sensors are not typically fully diagnostic, though they do provide far greater prewarning of failure and higher early detection potential than basic sensors.
Contextual, asset-specific diagnostics
The next level of machine condition monitoring is far more customized to specific, individual assets. These advanced sensing solutions focus on specific frequency bands associated with the equipment being monitored. Sensors are tied to internal and/or external machinery health monitoring software, wherein users define the equipment, bearing types, couplings, motor speeds and more. All this context surrounding the specific equipment being monitored is used to determine which frequencies are given the most attention.
The condition monitoring software measures the amplitude of vibration of specific frequencies — based on decades of industry expertise and knowledge of the physics behind typical machine failures — to deliver more direct detection of specific failure modes. Ultimately, such a solution delivers more diagnostic accuracy and earlier warning potential, transitioning from data collection to true condition intelligence.
Automated advanced sensing and analysis
For the most advanced sensing technology, signatures of specific failure conditions can be identified using special filtering that is applied to the raw vibration data. This very advanced and automated analysis cuts through the complexity of manual analysis and provides a simple, reliable indication of asset health via a single trend. Technicians of any experience level can quickly and easily quantify asset health — using a green, yellow, red severity indication system — and receive actionable diagnosis and next steps to help them resolve the problem.
The ability to link rich vibration data with asset context is critical to unlocking the advanced analytics that save teams time and energy because raw data lacks meaning without context. When analytics tools have access to machine details, operating conditions and component design, the tools can more easily distinguish real signals from background noise, enabling faster detection, more accurate diagnosis and reduced false positives.
Context matters because machine fault conditions vary in severity, with some requiring more immediate attention than others. For example, lubrication, bearing and gear defects are faults that can stop the machine from turning and require more immediate corrective action. Other faults — such as imbalance, misalignment and looseness — can be addressed over a longer timeframe.
Some vibration faults are directionally oriented — like resonance, misalignment and torsional vibration — and diagnosis benefits from knowing in what direction the vibration is being detected relative to the shaft. Rate of change in amplitude over time also can provide some context to not only severity, but also to how quickly the fault is progressing, and that can correlate to the amount of risk and timeframe for needed corrective action. Detecting a fault is obviously beneficial, but it is far more beneficial to determine the severity, when action is needed, how much risk the fault represents and ultimately how long the asset can run to achieve a production schedule or maximize the useful asset or component life.
This type of enhanced knowledge leads to more efficient workflows. Reliability teams can intervene earlier to detect the development of asset failures, providing more time to plan repairs. Teams can approach a maintenance task prepared with the right personnel and equipment, empowering them to solve problems quickly and shorten outages (Figure 3).
Introducing AI in analysis
The modern machine condition monitoring landscape has become even more powerful in recent years with the rise of AI technology. Advanced AI software can analyze vibration data quickly, and it can correlate patterns across time and failure modes to deliver incredibly in-depth insight and actionable information, shortening the time between identification and repair. However, it is important not to fall victim to the pitfall of AI-first analytics.
AI is an emerging technology, and as such, it is most effective when built on top of structured, contextual analytics, not when replacing them. Some new condition monitoring solutions rely on AI analyzing raw data without deep contextual modeling. While this approach can identify patterns, it often struggles with precise diagnosis. In turn, imprecise diagnosis can lead to false alerts, missed or misdiagnosed root causes, and limited trust from users.
The teams most effectively using AI for reliability today are leveraging tools that expert automation solutions providers are integrating into their existing holistic condition monitoring solutions. These tools enhance rather than replace structured analytics.
AI should automate effective workflows, not replace them
AI excels at eliminating low-value and repetitive tasks, monitoring large datasets continuously and identifying correlations. For organizations without a deep bench of expert personnel, this can free workers to focus their skills on more critical tasks.
Automation solutions are already incorporating AI tools to help flag anomalies and support failure diagnosis. Some of today’s most powerful asset monitoring devices are built with on-board AI tools to help drive deeper and more effective analysis right at the edge. By leveraging asset monitoring tools with built-in AI edge analytics, reliability teams can automatically apply embedded analytics to alert personnel to common faults for a wide range of assets, like fans, motors, gearboxes, pumps and other rotating machinery.
As AI usage continues to increase, forward-thinking organizations are rapidly learning that they do not need to manage complex, fragile, bolt-on AI solutions to deliver reliability insights. The best and most advanced tools are instead being built into the automation solutions they already use, seamlessly integrated to deliver better value, longer lifecycles and increased ease of use.
In contrast, AI found in some condition monitoring offerings today attempts to compensate for inferior sensor data. Acquisition systems with limited frequency range coverage or low sampling rate may not even be able to detect some of the most common failures that are characterized by high frequency or short duration signatures. This includes bearing lubrication or wear problems. Because AI tools rely on good quality data and poor sensing can result in erroneous or inadequate analytics, coupling even the most powerful AI tools with proven condition monitoring technologies is critical.
More advanced solutions are attainable
One of the key reasons organizations turn to less robust, quick implementation solutions for reliability management is the misconception that they are equivalent to more mature traditional systems. Low-cost, simple solutions can provide fast and easy installation and can seem far less intimidating than the larger-scale holistic solutions that deliver lasting, long-term benefits.
However, most alternative solutions are not built on the decades of expertise necessary to deliver a truly effective condition monitoring program. Generic pattern recognition can identify real change of data streams, but it takes additional analysis to clearly identify the cause of the changes. Reliance on generic pattern recognition can lead to late detection, higher repair costs and increased or extended operational disruptions.
In contrast, the context-rich analytics available in a fit-for-purpose advanced condition monitoring solution enable earlier intervention, better planning and reduced lifecycle cost. The true value of advanced analytics is unlocked by matching data signatures with known faults through knowledge of first principle physics so that automation of failure mode identification is accurate.
Moreover, when organizations invest in a condition monitoring solution from an expert automation solutions provider with decades of experience across multiple industries, they are likely to see a longer operating lifecycle for their solution. The providers that have stood the test of time are far more likely to be available for support and to deliver value-add updates many years into the future (Figure 4).
Yet, though the benefits of a fit-for-purpose, holistic system are obvious, many organizations mistakenly believe they are out of reach. However, the most advanced condition monitoring solutions are highly scalable, making it possible to start small with pilot projects, and then increase the program’s footprint over time as success and return on investment (ROI) are proven out.
Practical, affordable solutions like wireless vibration monitors and AI-enabled online monitors can be implemented as asset-specific deployments that deliver immediate value. Then, as an organization grows its program, devices designed for seamless integration can be automatically and effortlessly incorporated as the program grows. These solutions include more advanced sensors, machinery health software, enterprise-level reliability software and more, with each layer only added as the previous one has demonstrated effective ROI.
That ROI is not difficult to achieve because effective analytics can be tied directly to business outcomes. Context-driven analytics enable planned maintenance, shorter outages and higher asset availability. They also reduce unplanned downtime, and both maintenance and operational waste, while simultaneously improving production consistency and quality and workforce productivity. Each of these benefits translates directly to increased revenue and improved reputation — benefits that can be scaled in parallel with reliability program growth to drive continual improvement of the company’s bottom line.
The future of industrial analytics is context-driven
The evolution of analytics will continue toward increased automation, widespread use of AI and more real-time insight delivered directly to personnel to help them make better decisions, both in the field and in the boardroom. However, that increased innovation will be built on an existing foundation of context, effective workflows and domain expertise built into trusted automation software.
The organizations best able to align innovation with a context-rich foundation will most effectively turn data into insight, insight into action and action into competitive advantage.
About the Author
David Kapolnek David Kapolnek
David Kapolnek is Director of Product Management for Emerson’s Reliability Solutions business. In this role since 2020, Kapolnek has led development and deployment of information-based software tools and services that help manufacturers improve effectiveness and efficiency of their work practices. David has held product leadership roles in numerous organizations, with a focus on facilitation of high value user workflows through hardware and software-based systems. He earned a bachelor's degree in ceramic engineering from the University of Illinois, a master’s degree in materials science from UC Berkley, a Ph.D. in wide bandgap semiconductors from UC Santa Barbara, and an MBA from Cornell Johnson Graduate School of Management.