Elliott Middleton, Director of Product Management, Aveva, sets out seven steps to data-driven decision making.
There is a lot of buzz surrounding industrial analytics, Artificial Intelligence (AI) and Machine Learning (ML). However, companies are not limited to taking large and costly decisions when it comes to exploring the possibilities.
There are smaller, incremental steps that may be taken along the journey of digital transformation.
Until recently, more mature process industries have pursued AI and ML initiatives, while others are only using some basic levels of automation.
The unrelenting pressure to keep production lines moving, especially at a time when businesses are responding to changes driven by the Covid-19 crisis, means there is little time or headspace to consider implementing new technology.
Automating the collection of sensor data, spotting problems and patterns can result in faster troubleshooting and major process improvement for all sizes of operation. It is possible to start with small incremental steps that add immediate value.
The Journey To Adding Value Through Data Analytics
Here are seven steps to generating incremental value with industrial analytics and ML:
- Automate data collection from sensors. This is a critical prerequisite. An infrastructure for automated data collection requires multiple sensors to feed through the data that is needed for meaningful analysis.
- Record measurements from the sensors over time. The challenge for many organisations is collecting the data and making sure that data is captured throughout the process so that the workforce has the option to draw some useful inferences at a later stage. You may not have an immediately obvious use case but if you start collecting all the data, additional insights are likely to arise.
- Accelerate Diagnostics: Once you have collected the data, diagnostics can greatly reduce the time required to pinpoint and correct operational problems—and lead you to make process changes that prevent them from happening again in the future.
- Make the Data More Accessible: As the key efficiency metrics are better understood, it is important to make them readily available to a broader group within the operational staff, so they can independently monitor and react to potential problems earlier. Many sites use a large screen display showing a dashboard of live metrics or an automated email report. Others take it a step further and deliver alerts to mobile devices.
- Add alarm history to the process history to give more context and significance to the data. Traditionally, data historians just record sensor values, they do not record the alarm state. But combining sensor and alarm data makes it easier to understand the potential impacts on quality, safety, cost or the environment.
- Add Operational Context: In some applications, differences in recipes, equipment, or personnel can further complicate identifying root causes or improvement opportunities: was the yield lower because it is a different product, operator, or production line? By including information about this kind of operational context, you can begin to consider these other factors in your analysis.
- Add machine learning: Making this detailed information available to people has tremendous value, but it still requires them to actually spend time looking at it. Although it cannot do everything, in many cases, you can automate that analysis process so that it is continuously analyzed. Some systems have unsupervised Machine Learning (ML) capabilities: simply provide the data and the system will look for and report anomalies. This style of ML has the significant advantage of simplicity: no expertise required. Other applications warrant investment in process and technology expertise to add supervised ML, commonly as early failure predictions.
- Compare like for like: Building on process analytics successes at individual sites, many organisations advance to analysis and comparison across their fleet. The more homogeneous the sites, the more direct the comparisons can be, but few industries actually have ‘cookie-cutter’ plants. So, before making these comparisons, there must be some effort to normalise information—whether that is standardising on ‘m3/day’ vs. ‘litre/minute’ or agreeing on how to calculate Overall Equipment Efficiency (OEE). In global businesses, with multiple operations and multiple sites, being able to do peer to peer comparison – in a standardised manner – is a real value add. Previously, this would have been a task of overwhelming complexity and cost for smaller businesses. Now, with the cloud, this type of comparison is much more accessible and achievable.
Eating The Elephant
Tackling sensor data analytics automation can feel overwhelming, but there is no need to eat the elephant in one sitting. It is quite possible to take single, exploratory steps on this journey as the need arises and the budget is available.
You might stop for a while, having implemented one step, and see how it works out for you. Most of the incremental steps can be self-service and do not need external expertise.
The journey to getting value from applying AI and ML to sensor data starts with a single step. As the benefits start to accrue, cost savings will mount and productivity will rise, demonstrating clear business value from intelligent sensor data analytics. It will become a no-brainer to progress further along the route to optimum data-driven decision making.
Check out these articles:
CLICK HERE FOR LATEST NEWS.
READ CURRENT AND PAST ISSUES OF IAA.
KEEP YOURSELF UPDATED, SUBSCRIBE TO IAA NOW!