Proper planning and strategy will determine the quality of your production intelligence and the success of your smart-manufacturing goals.
By Andrew Ellis, manager, Global Technical Consultants, Information Software, Rockwell Automation
The goal is the same for many industrial firms: Use analytics to improve performance and solve production challenges. But how do you get there? How do you make the transition from raw data to useful production intelligence?
It’s a question that many companies are trying to answer. They likely already collect data, and probably lots of it. However, too often, they’re unsure of what to do with that data, where it comes from, if they can trust it, and how it can help them achieve their smart-manufacturing goals.
This is why a detailed data-integration plan is so important. It defines what you want to achieve with your analytics, and details how to achieve it. It also answers critical questions up front to help you avoid future problems — from unreliable data to data overload.
Map Out Your Plan
Begin your data-integration plan by mapping it out from your current state to your future state. This process should outline:
- Your current control and information architecture.
- The analytics or smart-manufacturing goals you want to achieve.
- Future plans for your control and information architecture.
In addition, you’ll want to answer important questions such as:
- What data sources do you currently have?
- What key performance indicators (KPIs) do you want to monitor and why are they important?
- What data are you automatically and manually collecting?
- Will you want to deliver production intelligence to workers via mobile devices in the future?
- Do you plan to move data storage or applications to the cloud eventually?
All this will help you create a well-defined plan and avoid challenges later. For example, identifying KPIs up front allows you to build analytics that deliver them. Otherwise, workers can end up having to analyze hundreds — or even millions — of rows of data to find them.
Knowledge about data, networks and industrial Internet of Things (IIoT) technologies is critical during this process, but so is time. If you’re short on either of these, find a partner that can help you build out and execute your plan.
Pick a Data-Storage Option
Live data has a purpose, but it won’t solve many analytical or long-term problems in your operations. That’s why you need a data-storage system.
Many KPIs such as overall equipment effectiveness (OEE) require historical data. However, advanced analytics that use capabilities such as machine learning also need historical data. They build a data model over time to learn how a machine or process works, and then use that model to predict potential issues so workers can intervene before a problem arises. Some systems can even use this data to trigger automatic adjustments.
A manufacturing execution system (MES) database might suffice for data storage in discrete operations where you’re collecting transactional data. However, a historian will be necessary if you’re collecting process data or a substantial amount of time-series data.
Again, it’s important to consider your future needs. You may only require storage for a few data points on a single machine today, but could that increase to several thousand data points across an entire line in the future? If so, invest in a historian now.
Define Your Data Requirements
The quality of your analytics depends on the accuracy and availability of your data. Every company will have its own unique data requirements. You can build yours around the three Rs of data:
Reliability: How good do you need your data to be to trust it? You should make sure that your data is not only error-free but also within acceptable tolerances. For example, will you need real-time raw data, or will interpolated value or last good value meet your reporting needs? Remember: It only takes one or two suspect data points to quickly erode confidence in all your data.
Redundancy: How available do you need your data? If you want to minimize data loss, you should use a historian with redundant interface nodes. If you need highly available data, you should use a “collective” of redundant historian servers.
Retention: How long do you need to keep your data? You may only want to keep it for months or years for analytical purposes. But you may want to keep it for longer if you’re in a highly regulated industry. Some companies, for example, retain data for decades. This can help them prove that their production processes adhered to specific standards or customer requirements. And it can help protect them against potential legal action.
Historians allow you to create rules in your data to meet your specific data requirements. This might involve creating an average of one data point over time or interpolating other data points. When you make such rules, it’s important to inform workers about them so they understand how the data was determined, what rules were followed and are aware of any potential bias.
Start with a Baseline
Before you implement your analytics software, getting access to preliminary data can be useful.
By using a historian and a simple trending tool, or conducting some calculations in an Excel spreadsheet, you can get an initial look at your data. This can help verify that your performance is where you expect it.
The data can serve as a baseline to compare future analytics. If any of the data you see is unexpected or contains gaps, you can adjust your strategy at that point, before you’ve expended too much effort or have invested in analytics solutions.
It’s important to remember that you probably won’t be able to identify all your analytics needs during the planning stage.
That’s why it’s crucial to implement targeted analytics software for the production intelligence you know you need while also building in flexibility for future analytics you don’t yet know you’ll need. New ad-hoc analytics software can give you this flexibility. It can give a worker access to any existing data source so they can explore data, troubleshoot and solve problems on the fly — all without the help of a data expert.
Consider a key performance indicator (KPI) such as energy usage, which has become a high priority in many companies. While numerous producers are monitoring energy usage at a single location or enterprise wide, once you have access to ad-hoc analytics, you can start to dig deeper.
You might want to start measuring things like energy intensity, or OEEE — overall equipment energy effectiveness. This can help you understand usage across machines, lines or processes. It lets you measure performance based on how much energy it takes per product, pound, batch and more.
Other considerations, such as industrial security and reducing islands of information, can be addressed either here or as part of a larger connected strategy. But, by creating a data-integration plan and addressing key decisions upfront, you can better prepare your organization to take advantage of — not fall victim to — your wealth of data.
Watch a demonstration video about FactoryTalk® Analytics and Project Scio.
The Journal From Rockwell Automation and Our PartnerNetwork™ is published by Putman Media, Inc.