Making sense from good data

A well-executed program needs planning before implementation, and that means updating and preparing historical data for the predictive model.

Machine learning is a profitable venture for manufacturers, but only if the data is good – avoid garbage-in, garbage-out. Historical data used to start the predictive model must be clean, and new data collected from machines must also be clean before it will be useful in decision-making.

A well-executed program needs planning before implementation, and that means updating and preparing historical data for the predictive model. However, data quality can still rear its ugly head after implementation. The output of a predictive model feeds the next production, so if new data has errors, it cascades at each step and errors grow exponentially, across the entire process. So, where and how to start means keeping a few steps in mind.

Those pulling data for machine learning should have a clear idea of the specific roles and functions before applying machine learning to production systems, clearly distinguishing each function and ensuring the new tools align to the roles and people they will support.

Next, real profitability won’t come from having a great algorithm but from having large amounts of clean data that algorithm can learn from.

Understanding core competencies and unique values of varied data types will enable embedding machine learning in innovative ways.

After, machine learning algorithms and tools must incorporate a deep, fundamental understanding of the physical behavior of the assets being managed, as well as the context of their use. Real value creation in machine learning comes from its application to unique data.

There is ample opportunity to marry physical and data-driven models to provide better outcomes; recognize the important and distinct roles of both types of analytics, and the different ways they are created and managed.

Finally, it’s not always easy to take this from concept to production floor without tweaks, such as making sure the design is a repeatable and scalable workflow around these algorithms. Once set, the goal is to have as seamless of a transition as possible, gathering analytics and delivering good data for good decision making.

With the right implementation of good data in, good data out, machine learning algorithms, applications, and platforms are helping manufacturers find new business models, fine-tune product quality, and optimize manufacturing operations to the shop floor level. Send me an email and let me know how machine learning is helping your overall production.

Elizabeth Engler Modic, Editor
emodic@gie.net

June 2018
Explore the June 2018 Issue

Check out more from this issue and find your next story to read.