Automated root cause analysis in manufacturing

Seasoned quality engineers have all dealt with the pain of finding the root cause of costly warranty issues.

Discovering the root cause of a defect that already managed to leave the plant, passing existing inspection and quality assurance processes, is not a simple process. 

Finding the root cause of a problem takes a combination of:

  • Specialized process knowledge
  • Well-established techniques such as fishbone diagrams, the 5 whys, and pareto charts
  • Extensive manual inspection of parts and machines on the line
And most of all… the process takes time.

Everyone in manufacturing knows that time = money.

Every wasted hour, minute — or even second  means lost revenue. The more time passes, the greater the risk of failing to fulfill a customer order, and at the very worst, shutting down a customer’s production. This is a nightmare scenario that could be disastrous. It only takes a few lost contracts to completely devastate a plant, especially in today’s competitive climate.

It’s no wonder manufacturers are stressed out!

Industry 4.0 has made almost everything in manufacturing faster through automation. And we now have the data that could automate root cause analysis… right?

Automated root cause analysis can't be done with data alone

These days, there is more data from the manufacturing process available than ever before. This data can offer invaluable insights into the root cause of issues, as well as the scope of a warranty problem that has been discovered.

But, there is a catch… 

Automated root cause analysis can’t be done with data alone.

Even if data is being collected at all relevant points in the process, there are still many steps to get any actionable information out of it. At a high level, this includes:

  1. Running around the shop floor to gather data from siloed machines, databases, and software
  2. Pulling the data together into a single spreadsheet (like Excel) or business analytics program (Like Power BI)
  3. Sorting, cleaning and transforming the data to make sure it all lines up together and can be understood
  4. Running analysis on the data
  5. Validating the results of the analysis

It almost seems like data availability has made the job of quality, manufacturing, and process engineers and managers even harder, since they are now expected to be experts in data analysis on top of traditional methods of finding the root cause.

Data scientists are not the solution

Some plants with highly autonomous processes are already leveraging their own corporate data science teams or using outside data science consultants to analyze their data using advanced techniques like machine learning models. But does this mean root cause analysis has been automated?

In a word: no. Machine learning models can crunch huge volumes of data and produce analysis at lightning speed. But the models need to be built first, and this takes time. Machine learning does not necessarily mean “automated”.

In order to build an accurate machine learning model, data scientists need to understand the problem first, which can take some education from process, manufacturing, and quality engineers.

Then, data scientists need to be taught which metrics come from which machine signals, and which group of signals belong to each station. They must know specification and control limits, and what test station or inspection data qualifies as a pass or fail. Then, they must build or select a model, run the analysis, validate the results, and further refine the model.

It takes a lot of time to make this happen. The steps to build a custom root cause analysis machine learning model can last weeks.

One manufacturer told us that they underwent a similar root cause analysis project building a machine learning model to analyze their data. The project cost $50,000, and the data scientists took a month to find the root cause of the problem. By that time, the operations team on the shop floor had arrived at a fix for the problem by their own “old school” methods.

So far, data availability and the work of data scientists has not helped us achieve automated root cause analysis. So what will?

Automated root cause analysis needs to be done right on the shop floor

To improve root cause analysis, quality teams need to get insights faster. Data and machine learning are part of the solution, but the process of getting the data to analysis needs to be automated to make this possible.

Several steps in the process need to be made faster:

  • Collecting data from around the shop floor each time an issue arises
  • Cleaning and transforming the data
  • Adding context to understand the data
  • Selecting or building a new machine learning model
  • Training, validating, and testing the model

Luckily, technology is available today that can automated these processes inside a single platform. To automate root cause analysis, a platform must be able to:

  • Ingest different sources of quality data in real time
  • Clean and transform the data automatically
  • Contain manufacturing-specific machine learning models that can be used to solve any problem
  • Allow for configurable analysis
  • Be designed for use by engineers on the shop floor

The last point here is key. An automated root cause analysis platform needs to be used directly by the engineers leading the investigation. With data ingestion, machine learning and analysis running in the background, quality engineers don’t need to become data science or machine learning experts to benefit from the technology. Since their data is always up to date in the platform, they can run any number of different root cause investigations on failed parts or underperforming aspects of the process.

For a step-by-step explanation of such a platform, watch how to automate root cause analysis with LinePulse. 

Share on social:

Automate root cause analysis and predict defects in real time

How is that possible?