By: Katarina Struckmann, Manufacturing Data Analytics Scientist at Sight Machine
In the popular imagination, big data analysis is a magical blender: if you pour in enough data and hit blend, it produces immediately useful insights.
In practice, it’s not so simple; every step, from data collection to advanced analytics, must be carefully executed by a team of well-trained professionals.
Manufacturing data, which is notoriously complex, is particularly difficult to gather, analyze, and utilize in a meaningful way.
As a Data Scientist with Sight Machine, I’ve had the opportunity to work with leading manufacturers on their digital transformation efforts. Here are the five things I’ve learned that make the most difference in determining whether a project ultimately delivers impact:
1. Make sure you’re capturing the right data
If you ask me what the impact of the Great Pacific Garbage Patch is on the local ecosystem, but give me a data set about baroque composers, the answer isn’t in that data.
Time and again, I’ve seen manufacturers interested in solving specific issues, such as predictive maintenance, limiting scrap, or reducing downtime, proceed without the required data to address those problems. This ultimately ends up in frustration for the project sponsors and the data analysts involved.
There are two lessons I’ve taken away from these efforts:
First, build out a catalog of your data resources to determine what types of manufacturing data analytics and associated processes you can impact today. Most manufacturers need to better understand exactly what data they have at their disposal. Working with your existing data will help you build the capability to turn insights into action.
Second, define the future state of your operations and all of the associated use cases that will enable this vision. Once you’ve established this, you can begin to determine the data requirements and define a roadmap for filling in the data gaps that will enable your future state. Most manufacturers I’ve worked with don’t articulate a long-term vision or the milestones for building the data capabilities that will enable them to create a competitive advantage in the market.
2. Make sure you’re capturing good data
During initial discussions with manufacturing operational leaders, I’ll often hear something like, “don’t worry, we have the data for that.” Unfortunately, as we delve into projects, we quickly determine that the data quality isn’t what’s needed to execute the project successfully.
Typical problems include having low data granularity or sample sizes that are too small to draw accurate conclusions from. Compounding the problem, a manufacturer often doesn’t have standard operating procedures in place that enable them to develop data that is consistent enough to be useful. For example, if the standard operating procedures don’t ensure that readings are taken in a consistent manner, the data capture process will introduce extra variables, making analysis difficult or impossible.
Unfortunately, in most cases, many of these issues don’t become apparent until you are in the middle of the project. Therefore, it’s critical to build in project milestones where the team can discuss if the data being collected is appropriate for analyzing the problem in the original scope. There will likely be a need to reset the project scope and/or timeline expectations to correct how data is being captured. Another way to account for this challenge is to engage resources (consultants, software vendors, or even your internal resources from another facility) to leverage their expertise on common data challenges for each given use case.
3. Free your manufacturing data analytics team from manual data preparation
Data scientists will tell you that 80 percent of their time is spent on cleaning and blending data, while only 20 percent is focused on performing analyses and developing insights. This is unacceptable at a time when data science resources are in short supply and the need to incorporate data-driven decision-making into operational processes is continuing to accelerate.
Therefore, it’s critical for manufacturers to invest in data modeling and contextualization tools that automate data cleaning and blending activity. Building out a real-time data modeling platform will be a foundational requirement for manufacturers in the next decade.
4. Focusing on the data first will let you scale
Typically, initial discussions with manufacturers are focused on addressing one specific problem. The team involved with the project advocates for the creation of static data models that support a narrowly focused analysis or digital manufacturing application.
Unfortunately, I’ve found that this approach limits flexibility, scalability, and, ultimately, the effectiveness of the project. Because the solution isn’t architected to scale across multiple use cases, plants, or machines, the time and effort that went into the point solution is mostly wasted.
I’ve found that the most successful manufacturers don’t start with these narrow applications, but instead focus on building out a data platform that will enable them to transform the way they operate. Ensuring that the data models are generalizable and adaptable will enable the project to address multiple use cases and extend capabilities across the enterprise.
5. Ensure the results are actionable
One of the biggest challenges for data science teams in manufacturing data analytics is ensuring that analyses and insights can be operationalized. Too often, interesting insights end up as slideware or tools for specialists that don’t broadly impact the organization. I’ve found two ways to address this:
Present the results that are relevant to the user. Design independent dashboards or user interfaces for the different audiences that will use the data. Too often, not enough thought is given to how users will digest the analysis. Figure out what the operators, line managers, and plant managers need to know, and deliver the analysis to each audience in a way that is clear, unambiguous, and not cluttered with things they find irrelevant.
Ensure change management resources are involved to operationalize the output and drive buy-in with local plant operational teams. Too often, efforts fail because no one is in place to drive the opportunities for operational and cultural change identified by the analysis. To ensure this happens, it’s critical to incorporate change management resources in any analytics project. Ultimately, a lack of buy-in will doom the project, regardless of all the work done to properly capture and analyze the right data.
The road to delivering impactful data-driven manufacturing projects is filled with potential roadblocks and pitfalls. There are numerous technical and organizational factors that can derail success. At Sight Machine, we’ve aggregated many of these factors into our Digital Readiness Index, which is a great tool for elevating some of the data architecture and organizational readiness questions that can ultimately impact the project.
