Low-touch machine learning is the catalyst to scale APM’s potential. A widespread integration of machine learning in APM marks a transition from estimated engineering and statistical models towards measuring actual patterns of asset behaviour. Manufacturing facilities staff can now readily extract value from existing design and operations data to better manage and optimise asset performance. This low-touch machine learning method embraces changes in asset behaviour and empowers real-time APM value creation.
Evolution of APM
The foundation for realising a new APM vision already exists — but until the late 2000s, APM software didn’t communicate accurate, reliable or timely diagnoses or clear recommendations. Around then, however, we began to see multiple parallel technology innovations merge into state-of-the-art APM methodology.
Best-in-class systems could now incorporate the detection of precise patterns of normal and failure behaviours, and perform the computational isolation of key indicators of degradation. Especially important was the 2006 debut of Amazon Web Services for scalable cloud computing. Advances in structured and unstructured databases and operational data pools were tested and improved at the enterprise level during this period.
At the same time, smart sensors saw a shift in performance, size, reliability and price. Added to this was a dramatic improvement in the computational and analytical capability of machine learning, called “deep belief networks” or “deep learning.” The result was a quantum leap in capability that enabled machine learning to become the dominant analytical method in all IT fields around the world.
Between 2007 and 2010, the process industry workforce moved from experimentation with the industrial Internet of things (IIoT) to demands for smart devices and consumer-style applications at work. Industrial software and technology began to update offerings with user interfaces incorporating low-touch, readily navigable applications and displays. At the same time, cross-industry initiatives led to the development of open standards for connecting disparate systems and work process interoperations.
During this period too, the movement from fail-fix to reliability centred maintenance techniques provided incremental improvements. However, the cost, complexity, time and staffing skillset requirements constrained deployments.
Today, there’s a growing realisation that maintenance alone cannot solve the unexpected asset breakdown problem. Market-leading companies understand that they have gone as far as possible with traditional preventive maintenance techniques. Predictive maintenance represents the next frontier.
The era of low-touch machine learning
Data-intensive and complex environments in manufacturing industries are prime candidates to deploy the new advances in reliability management. Deployed coherently, with appropriate automation, machine learning enables greater agility and flexibility to incorporate current, historical and projected conditions from process sensors, as well as from mechanical and process events. Systems become automatic and agile, flexible models emerge that learn and adapt to real data conditions, and incorporate all the nuances of real asset behaviour.
Data capacities and computational capabilities are so great that internal staff can now perform the active and accurate management of individual processes and mechanical assets. This management capability can now be applied to combinations of assets — plant-wide, system-wide or across multiple locations.
The pivot in APM’s capabilities arrives at an important time. Manufacturers are under economic pressure and razor-thin operational margins are pushing process industry executives to look to APM for additional return on investment. Low-touch machine learning APM is ready to deliver that value.
With that in mind, here are five machine learning best practices that drive state-of-the-art reliability management that is applicable to any asset in any industry at any level, from a single location to a country wide system.
Data collection and preparation
During the last two decades, every attempt at massive data analysis from diverse sources of plant data collected from sensors has run into serious issues concerning collection, timeliness, validation, cleansing, normalisation, synchronisation and structure issues. Often, such data preparation can consume 50–80% of the time to execute and repeat data mining and analysis.
However, that process is essential to ensure appropriate and accurate data that allow the end users to trust in the ensuing analytical results. New advances in APM have automated the bulk of the data preparation process to ensure trust and to reveal previously undiscovered opportunities with minimal user preparation.
Condition-based monitoring
Once data is trustworthy, condition-based monitoring (CBM) can be applied. The plant conditions vary constantly, according to the mechanical performance of assets, feedstock variations in quality, weather conditions and production timeline and demand changes. Static models cannot work under such duress. In addition, focusing CBM on mechanical equipment behaviour can reveal only a small fraction of the true issues causing degradation and failure.1 Leading organisations recognise that legacy CBM is now inadequate, as it typically ignores the salient process-induced conditions causing the bulk of the breakdowns. New advances in APM deliver comprehensive monitoring of all the mechanical and upstream and downstream process conditions that can lead to failure.
Work management history
The history of work provides the bread crumb trail of past solutions to failure prevention and/or remediation. Problem identification, coding and a standard approach to problem resolution provide an important baseline for the exact failure point in the lifecycle of an asset.
OEM data that may live in a big data solution can provide insight into process issues and outliers that are specific to the configuration and engineering within the plant process. Forward-thinking organisations understand the importance of this data and how it contributes to hyper-accurate predictions of production degradation that ultimately leads to asset failure.
Predictive and prescriptive analytics
Clean data and CBM enable in-place predictive analytics: a process to interpret past behaviour and, based on that analysis, predict future outcomes. In contrast, using engineering and statistical models to estimate the future readings of sensors, and interpret variances from actual readings, is a technique fraught with errors and false positives. Top performers use inline, real-time analysis of the patterns of normal and failure behaviours of process equipment and machines.
When performed correctly, predictive analytics can accurately portray asset lifecycle and asset reliability and focus on the early root cause of degradation. The insights available provide accurate, critical lead times. This allows time for decisions that can eliminate damage and maintenance or, at least, provide preparation time to reduce the time-to-repair and mitigate the consequences.
Best-in-class APM provides prescriptive advice based on root cause analysis and presents information on the approach that will proactively avoid process conditions that cause damage, and/or advise on the precise maintenance required. As a result, predictive and prescriptive capabilities enable asset lifecycle reliability, and they facilitate decisions regarding when and how to maximise production while proactively avoiding asset and output risks.
Operational integrity improves when organisations implement strategies to detect root causes as early as possible
Pool and fleet analytics
The next level of analytics allows patterns discovered on one asset in a pool or fleet to be shared, enabling the same safety and shutdown protection for all equipment. Once deployed, companies can rapidly scale solutions from a unit to multiple sites. From all local systems, information rollup from disparate sites into one larger model provides asset performance comparisons across sites and plants, creating common baselines that highlight areas for improvement.
Conclusion
The manufacturing world has changed. Now, previous maintenance practices can be improved to recognise all issues affecting asset degradation. Operational integrity improves when organisations implement strategies to detect root causes as early as possible, providing extended lead times for good decisions to avoid unplanned downtime. For every process industry organisation, low-touch machine learning APM is ready today to eliminate catastrophic failures on assets, improve reliability, lift net product output and increase profitability.
Reference
- Webinar, “Improve Reliability of Process Assets with Prescriptive Analytics: Get Results Today,” featuring ARC Advisory Group (14 June 2017).