In this post we’re going to explore taking a plant-wide vs. asset-by-asset approach using real-time monitoring and offline exploration.
The plant-wide approach focuses on broad coverage of plant assets, using real-time data analytics. Data analysts and engineers leverage AI generated math models to virtually monitor assets across your plant or entire fleet to predict equipment failures and identify asset efficiency issues to improve overall plant performance.
The asset-by-asset approach focuses on specific assets using offline data analytics. Data scientists build specific math models to explore historica
l asset data with the hope of identifying something that could potentially improve an asset or process.
The real difference is coverage time:
For those just starting out, historical data via a data Historian, or other data sources, is the base dataset needed for offline asset exploration to create your math models. The data is used to explore historical events, trends, and other areas of interest. Access to the data is often via offline batch or timed access during low data transfer hours.
Real-time plant analysis also leverages historical data for developing generated baseline asset behaviors and models from historical norms of assets and processes. The Plant approach also uses the historian as a continuous real-time data feed which is used to evaluate and learn from data changes in real-time as part of the prediction and performance evaluation process.
For those looking to scale and get widespread coverage, model volume often becomes the biggest hurdle to overcome. If you take the asset-by asset approach, managing ten, or even a hundred assets models is manageable. However, as that model count continues to grow and more assets come online, models can grow exponentially which can lead to model chaos.
With a plant-wide approach you immediately start working with 100s or 1000s of models and therefore requires a good model management environment or it can be a real headache, and seriously impede the success of the program.
Another important consideration when scaling is use and reuse. When scaling a program it’s important to reduce redundancy by encouraging model sharing and learning from what others have done to solve a similar problem. This includes templating and attributes, building a hierarchical representation of assets, and tag mapping. This can be daunting, and is rarely considered when starting out or planning to scale within an organization. Whether Asset-by-asset/offline or Plant-wide/real-time, model management and model reuse is key to scaling any performance & reliability program using software.
Both plant-wide and asset-by-asset processes get results. Your desired pace and coverage are the biggest differences. Asset-by-asset takes more of an ad-hoc approach, where models are created for trying to solve a problem. Overtime you can create 100s, or even 1,000 to provide full plant coverage.
Something to consider, though, when building models manually, one by one:
Algorithmia released an independent survey data showing that 40% of companies say it takes more than a month to deploy a single Machine Learning model into production.
Real-time plant-wide coverage takes an automated approach and casts a wide net by producing 100-1000s of models to cover the majority of plant assets. This is accomplished using detailed asset libraries of asset classes, types and functions, AI/ML automation, and precision and accuracy scoring to determine the best algorithms and models for an individual asset x 1,000 assets. It’s rare that this process will cover 100% of a plant, however >80% coverage is not uncommon, and it can be accomplished in weeks – not years.
The table below represents the process used to create models using the plant-wide vs. asset-by-asset approach.
Both approaches have their benefits:
Which is best, or right for you? Perhaps both.
Fast results win favor with any initiative, but focus around unique assets with unique operating parameters can’t be ignored.
Don’t reinvent the wheel by asking your math experts to solve problems that have already been solved 1000s of times before. Focus their attention on unique asset problems.
Similarly, don’t distract your data analysts and engineers with data scientist problems at the cost of ignoring plant operations to prevent unplanned downtime and loss of efficiency.
With the right foresight and planning you can achieve fast results, 100% coverage without software management fatigue, all while preventing unplanned chaos.
You can have the best of both worlds. It just helps to understand which tools will help you get there.
Atonix DigitalTM offers insights by connecting the power of math and data through a portfolio of products, powered by our AtonixOI software platform, that help simplify asset management by optimizing performance, improving reliability, detecting emerging risks while providing operational knowledge to take actions. To learn more about Atonix Digital’s AtonixOI software platform, visit us on atonix.com.
See what Atonix can do for you with a free demo. Contact us below.