Chatterjee: Automation Standards, Statistics and Process Validation

Oct. 5, 2011
Revived interest in SPC only heightens the need for new data architecture models

As organizations begin the task of transformation to meet the demands of process validation guidance, it won’t be long before someone asks, “Is there an easier way to get the data we need?” 

One of the key challenges of the new guidance is gathering the necessary data to meet a baseline level of understanding of the process to be validated, as it equates to Stage 1 of the guidance. For legacy products, process understanding usually derives from performing additional characterization studies or from analyzing current and retrospective process data from the manufacturing shop floor. Either method requires a stalwart constitution since there is the risk you may not like what you see, once you look. This explains why interest in implementing baseline best practices, such as Statistical Process Control (SPC), is on the rise on the shop floor. 

SPC was pioneered by Walter Shewhart in the 1920s, but it wasn’t until W. Edwards Deming started broadly applying it in the U.S. military’s World War II effort did it gain prominence. SPC is the application of statistical methods on the factory floor for the purposes of monitoring and controlling the process. Historically, SPC was applied using manual documentation methods, through sampling and testing plans to track the stability of the process. Together with alert and action limits, it was possible to provide rules to operators about when to notify their supervisors or subject matter experts (SME) if they saw process drift. This practice greatly reduces the interpretive source of variation often seen in historical data analysis created by operators making process adjustment judgment calls in the absence of true process understanding. 

Today most shop floor equipment is capable of capturing process data at the very least, and in many cases capturing in-line testing data and exporting them to a database for analysis. While this greatly simplifies the data gathering challenges of SPC, it shows the need for a clear information architecture to handle and analyze the acquired data. 

Luckily, in the information world there is no shortage of standards to turn to. Here are a few key models to keep in mind when considering your information architecture:

1.    SCOR (Supply Chain Operations Reference): Supply Chain View of Business Processes

SCOR is an architecture for business processes that categorizes a supply chain into five components (plan, source, make, deliver and return). SCOR is used to benchmark performance measurements in manufacturing. Not all supply chain processes are part of the model: quality, maintenance, product innovation, IT and administration processes are not included in SCOR.

In our world, where continuous improvement programs and initiatives are already part of the current infrastructure, the SCOR model provides a framework for gathering and analyzing data to maximize the performance of the supply chain. While not an explicit component of the new process validation guidance, it is a direct byproduct of process predictability. Process predictability is a key component to schedule accuracy, which lays the foundation for managing process velocity on the shop floor and inventory costs, considered to be key metrics of most Lean initiatives.

2.    ISA-95 (International Society of Automation): Plant-to-Enterprise Architecture of Systems

Probably one of the most important standards to consider, ISA-95 refers to information systems architecture. The standard defines the boundaries of enterprise systems according to how the data is managed, in terms of real-time processes. The reference model described by ISA-95 offers manufacturers a guideline to decide system boundaries, to design interfaces between systems and establish ownership of the relevant data. Many of the Level 2 and Level 3 processes in SCOR can be put into different levels of the ISA-95 model.

The ISA guidance in broken into five parts:
•    ANSI/ISA 95:00.01-2000-Part 1—Models and Terminology
•    ANSI/ISA 95:00:02-2001-Part 2—Object Model Attributes
•    ANSI/ISA 95:00:03-2005-Part 3—Models of Manufacturing Operations Management
•    ISA 95.04-Part 4 defines object models and attributes of manufacturing operations management
•    ISA 94.05-B2M-Part 5—Business to Manufacturing Transactions

ISA-95 breaks information systems into five levels. Level 0 is the production process or equipment control level. Levels 1 and 2 are the batch production control levels. Level 3 is the manufacturing operations management level. This is typically where Manufacturing Execution Systems (MES) reside. Level 4 is the business logistics level where most quality and enterprise resource planning (ERP) systems reside.

SPC would reside in Levels 1 or 2 depending on whether we are using a Supervisory Control and Data Acquisition (SCADA) system to gather data or using operators as the control mechanism.

3.    DiRA: Reference Architecture Framework for Discrete Manufacturing Industries

DiRA is a Microsoft initiative to integrate data from advanced digital infrastructure systems to benefit business processes. DiRA defines six pillars of capabilities, including user experience, role-based productivity and insights, social business, dynamic value networks, smart connected devices and secure, scalable infrastructure. These capabilities cover a range of technologies from human movement sensors developed for gaming machines, social media technologies to the latest advances in cloud computing and device convergence.

Typically the process of discovery, innovation and socialization within a global enterprise is unstructured. The DiRA model is intended to help global manufacturers better leverage technologies by unlocking the business value of their data and relating it to business processes.

Standards and Salmon?
Standards such as these provide a decision-making framework in terms of capital and technology investment, to support and manage data within the organization. As an industry, applying standards is something we are very familiar with but are not always good at. I would cite the industry’s insipid adoption of Quality by Design (QbD) and Process Analytical Technology (PAT) as evidence that some frameworks have only had a marginal impact, so far, on our industry.

The danger in not considering standards when automating data acquisition and analysis is twofold. In the best case there is the potential for only realizing a portion of the true horsepower that the data can provide for the business. In the worst case, the absence of a clear experimental framework from acquired data may lead to skewed or false interpretation of process behavior. 

As with all statistics, it is important to understand the strengths and weaknesses of the tests being applied to reach the correct answer with a minimum of risk. A recent Scientific American article explored the meaningfulness of the “p-value”. The author cited the widespread adoption of an alpha risk level of 0.05 as the de facto level for significance for testing the null hypothesis. The study summarized the dangers of overanalyzing data by comparing too many variables in an effort to understand the underlying process. The study presented salmon with pictures of people expressing emotions, then measured activity within the salmon’s brain [1]. When activity was present, different regions of the salmon’s brain showed electrical activity. Statistically, the result was significant with a p-value of less than 0.001; however, as the researchers argued, there are so many possible patterns that a statistically significant result was virtually guaranteed. The result was therefore worthless. In fact, it was impossible that the fish could have reacted to human emotions.

The salmon was dead to begin with!

The irony of the new process validation guidance is that enforcement of the new guidance may actually drive the industry into the arms of continuous improvement and business performance initiatives. If we can leverage best practices and standards we will avoid creating broader technical confusion as we move toward the agency’s vision of greater scientific process and product understanding.

1.    Seiffe, Charles. The Mind Reading Salmon, Scientific American Magazine, August 12, 2011.

About the Author

Bikash Chatterjee | President