Developing a smart Six Sigma program in Digital Manufacturing

Digital Manufacturing is plain automation..until you know how to extract the full value and evolve it into Smart Manufacturing

Real time process monitoring is one of the key functionalities of Digital manufacturing. In addition to allowing you to monitor your manufacturing process for optimality , it also provides you an opportunity to upgrade your quality control program that is still using a conventional approach, into a process integrated approach that leverages your newly found Digital manufacturing capabilities.

Conventional quality control process

Conventionally, manufacturing processes most frequently carry quality inspections of features for parts chosen randomly after some cooling time. In mass production scenarios, this long cooling time means that a large amount of scrap materials can be produced before defects  in the process are recognized. This makes this approach more reactive rather than proactive.

A process integrated quality control- leveraging digital manufacturing

To avoid the major issue identified in the conventional process above, a process-integrated quality inspection is necessary to obtain information about the quality of forged part immediately after the forming process. This has to be achieved by linking measured quantities during the forging process and quality features of the forged part. At a high level, a process integrated quality control delivers the following advantages:

  • Early sorting out of scrap
  • Fast detection of failures during the process
  • Reducing the scrap quota
  • 100% components inspection
  • Decreasing manual inspection

The Methodology

To define the methodology, I will use the forging process example. As an undergrad engineering student 15 years ago, forging was my favorite workshop subject so I can still recall elements of that process. Ignore the jargons-they are process characteristics of forging process and you don’t need to know them to understand the approach.


If you are familiar with Six Sigma methodologies, you know that a correlation between characteristic quantities and quality attributes of the forged part allows the judgement of the component quality, by means of a computer based system. These computer based systems need manual inputs and configurations. If you have ever taken a Six Sigma coursework in college, chances are high that you ended up using a software called Minitab. If you recall the process, you take historical data points and use the tool to determine the correlation. The process is manual, generally, and because the data point being used is historical, it is reactive.

In the “Smart” approach, we do not wait for data to become available/pool to analyze it (at least not when the smart process has already been implemented). Historical data is used to do the initial training of a Machine Learning algorithm and the subsequent quality failure dataset will be available for the algorithm to keep “learning”. By leveraging Machine learning algorithms, we can develop an approach of automatic evaluation of acceptable parts and scraps. To develop a process model correlating the attributes of the part and the characteristic quantity, rule-based systems (if-then-rules) need to be used.

One must take into account that the characteristic quantities interact with each other. These interactions clarify that the characteristics quantities must not be interpreted separately. The experimental determination of these inter-relationships between characteristic quantities and quality attributes has to take these interactions into account by using an appropriate experimental method. If the right algorithm is used, this process can be taken care of by the algorithm on a continuous basis, as it taps into newer and newer datasets.

Based on my analysis of literature and use cases, the model to produce the correlation is best offered by intelligent systems such as artificial neural networks. A neural network based system is adapted to its tasks not like common data processing systems by programming, but by teaching. Modifications of selected processes parameters may be done rapidly, when necessary. In order to determine which specific Machine learning algorithm will work best for you, you will probably have to run some proof of concept studies using various approaches.

How will the scrap get minimized?

As you can see in the illustration above – the process integrated approach leverages real time data generated by sensors to execute a “real time, process integrated quality control”. So as an example, if the algorithm determines that the Ram Path value detected by the sensor is not with a desired range, it will flag it immediately and the manufacturing process can be paused momentarily to test if the finished products are meeting the quality specifications or not.

Get your people and process elements right

The approach defines above is just theory- if you don’t have the right people to drive the initial implementation and the ongoing execution. Similarly, how you define the process (choosing the right measured variables, characteristic quantities etc.) is also critical, since non optimal selection of variables will not yield the results you desire.

If the variable selection process and the model selected does not align with process characteristics, the outcome of process integrated quality control implementation can be worse than the conventional approach.

Invest time in selecting the right people and developing the right process (assuming you have already invested in digital technology). Bon Voyage !

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s