Predictive Analytics is Here to Scale

Predictive analytics solutions have already proved that they can offer value to the process industry, but what does it take to scale digital solutions across dozens of plants? We recently held a webinar on the subject, discussing how to succeed at implementing predictive analytics solutions at scale across numerous plants. You can view the full webinar here

How to prepare a plant for a digital solution

There are two primary tasks that plants need to complete in order to successfully scale up their digital solution:

• Mapping the needs and resources of every plant in the organization
• Building out the correct implementation process

If this part of the process is handled well, you’ll see a successful scale-up, but if managed poorly, you’ll encounter repeated hiccups and disappointments.

Drawing on our experience providing predictive analytics solutions to numerous manufacturing plants over the last few years, we’re happy to share what we consider to be the best strategies to implementing any predictive analytics solution.

Stage 1: Mapping the needs of the plant

Every plant faces different needs, even within the same organization. These varying circumstances place different demands on the predictive analytics solution.

The first step is to map the critical needs of each plant. These tend to fall into one of several categories:

Part failures – these can refer to complete failure of one or more parts, or part degradation which causes the part to operate at a sub-optimal level. The failure of small, inexpensive parts can have as big an impact on plant productivity as issues with large, high-cost items.

Problems with process and optimization – typical for batch manufacturers, especially when trying to identify and reproduce a golden batch.

Maintenance and planning – this is often a primary concern, but not one that predictive analytics is optimized to solve. Plant managers want to predict when a part is expected to fail so that they can save money by delaying replacing it. But no predictive analytics solutions can assure you that a part will last until a certain date.

Energy consumption and related issues – this is a concern for some plants that are otherwise stable.

Plant-specific issues can affect some plants that don’t fall into any of the above categories.

Mapping the needs of a plant across these 5 categories enables you to focus your solution in the right place in order to show the best results. Organizations that scale successfully are the ones that map needs correctly.

Stage 2: Mapping the plant resources

The next stage is to map the plant resources. There are two main resources that affect scaling a predictive analytics system:

• Data
• Personnel

Mapping the data

Data is vital for predictive analytics to succeed. Most plants have a large number of data tags, process data, and equipment data, and it is usually sufficient for a scaled predictive analytics solution, but sometimes plants are lacking specific data types:

Deep digital data. If you only have binary values, like saving just the high and low temperature measurements, you might not have deep enough data.

Tags in the data historian. You need to save tags in the data historian as well as just sending them to the DCS. It takes at least a few months to have sufficient history for a digital solution.

Utilization data is needed to prove whether the predictive analytics solution achieved its goal.

Historical failure data is not needed for an effective predictive analytics solution. SAM GUARD bypasses the need for historical cases and instead uses human knowledge combined with machine learning to provide optimal results.

Mapping the personnel

To scale an effective predictive analytics solution, you need to ensure both enthusiastic management support, and that the right users are ready to manage the solution. Management support is needed to encourage workers with an already busy workload to take on the extra task of implementing a new solution. At the same time, if the right users aren’t ready to manage the solution, it will surely fail.

Mapping the data and the personnel in each plant reveals whether or not your plant is ready to scale a predictive analytics solution.

Stage 3: Building the right implementation process

After mapping plant needs and resources, it is time to implement your chosen solution. The business model for most predictive analytics solutions is an annual subscription. Since you will not be locked in to a long-term contract, we recommend running the solution for a year instead of doing a pilot. A year is long enough for you to add the relevant data, and experience the paradigm shift of predictive analytics.

You will no longer be responding to “fire drills,” because your problems will become predictable rather than urgent, and you’ll respond to them in an orderly manner rather than in an expensive and pressured one.

When you use a predictive analytics solution, your definition of the problem also changes. With static thresholds, you only face an issue if the measurements are above or below a certain value. With predictive analytics, you’ll receive alerts about things that are not yet critical, but will become critical if left untended. You’ll begin responding to very different types of problems.

Conclusion

Predictive analytics solutions are ready to scale across large networks of plants, but the implementation process must be approached in the right way to be effective. Plants should prepare by mapping their needs and resources, to be sure that they are addressing the right problems and that they have the data, the management support, and the users ready to act on scaling the solution. Implementing the solution for a 12 month period, and encouraging a paradigm shift on the part of the users, helps set the stage for success in scaling predictive analytics solutions in the process industry.

For more in depth information on implementing predictive analytics solutions at scale, please view our recent webinar.