Statistical analysis empowers process manufacturers to
spend less time preparing data and more time acting on the right issues.
Many process manufacturers
today are fixing their gaze on two modern focal points—overall equipment
effectiveness (OEE) and sustainability. OEE is driven by improved reliability,
quality, and production yield, and sustainability is driven by efficient supply
chains, minimal energy usage, and reduced emissions. While improving each of
these key metrics requires time and effort, advanced analytics tools involving
statistics, process control, and monitoring—collectively referred to as
statistical analysis—make this more achievable.
Statistical analysis enables
teams to standardize their approach to data and decision-making by detecting
anomalies early and often to minimize waste and limit the cost of poor
production quality. These tools can also identify ideal operating ranges to
maximize yield and reduce raw material and energy usage.
When applied properly,
statistical analysis empowers manufacturing teams to spend less time preparing
data and more time acting on the right issues, helping meet production and
sustainability goals. In a recent webinar, we discussed four methods for
operationalizing engineering statistics using Seeq, which are summarized below.
1. Build Statistical Process Control (SPC) Charts
The greatest challenge
organizations encounter when creating SPC charts is implementing automatic
updates across numerous product operations. Whether this entails multiple modes
of operations or multiple product types, the calculation of statistical
boundaries is highly variable, making the segmenting, or “slicing-and-dicing,”
of data a headache for data analysts and process experts.
Seeq simplifies this process
by empowering users to perform these calculations using its live data source
connectivity. SPC charts are created in a dynamic fashion, with the ability to
add context, segment data sets, apply correct limits to correct operating
modes, and create aggregated signals for monitoring control parameters.
As a best practice, users
should check for normalcy amongst the data prior to creating charts. Seeq’s
histogram tool provides data visualization, making it easy to check the
frequency of distribution for sample values. With different modes of operation,
users can leverage capsules to embed the grade code context into the histogram
to gain a better understanding of its distribution.
Once normalcy is verified
amongst the data set, users can begin calculating statistical boundaries based
on periods when the process is in control. One way to do this is by specifying
in-control time periods and using that data to calculate the averages and
standard deviations.
Users can create standard deviation limits for their SPC
charts.
Leveraging these calculations,
users can create control limits using Seeq Formula, and deploy them in
near-real time to monitor current process performance.
2. Create Run Rules & Identify Violations
The next step is
operationalizing SPC run rules, which can identify anomalies and alert teams
when data is outside of control limits. This provides production teams with a
method for interpreting near real-time data in a uniform way.
The best way to create run
rules is with the User Defined Formula Functions Editor Add-on in Seeq. This
Add-on enables users to create custom formulas so they can scale run rules more
easily across product operations, and create better uniformity in
decision-making processes.
Add-on tools can be found in
the Seeq Add-on Gallery for use in Seeq Workbench and Seeq Data Lab, with
installation instructions and user guides also available.
Users can view run rule violations within an Organizer
Topic.
Once run rules are defined,
users can view violations within a Seeq Organizer Topic where several different
trends, tables, and histograms from the Workbench analysis can be compiled into
a single dashboard that automatically updates. Manufacturers can monitor
current run progress in addition to the historical performance by viewing the
Topic.
3. Process Capability Analysis: Cpk and Ppk
In many organizations,
capability analysis is performed infrequently due to the considerable time
required by process engineers and data analysts to make associated
calculations. But Seeq asset groups can be set up to automatically calculate
Cpks and Ppks, making frequent analyses more feasible.
With the Cpk calculated, users can view a graphical
representation within Workbench.
Seeq asset groups can be used
to correlate assets with different product types. Users can perform a Cpk calculation
on one product grade, then rapidly scale it across multiple product grades. The
use of rolling conditions ensures the data is updated in near-real time for
continuous analysis.
Users can perform Cpk
calculations on one product grade, then rapidly scale it across multiple
product grades.
With the most up-to-date information, teams can leverage
the data to inform business decisions.
4. Analysis of Variance (ANOVA)
Users can also create
statistical analyses and plots, such as an ANOVA to compare variances across
means, in Seeq Data Lab. Traditionally, data preparation for statistical
analysis was static. Configuring it properly was extremely time-consuming, and
updating analyses as new information became available was near-impossible.
Seeq’s connection to disparate
data sources and automated data alignment simplifies the data preparation
process so users can more easily view time periods of interest. It also
empowers engineers, data scientists, and production groups to collaborate more
easily across functional teams and among multiple data repositories. This
results in less time spent gathering and preparing the data, providing more
time spent pursuing valuable business objectives.
Seeq Data Lab, which brings
Python and Seeq together using Jupyter Notebooks, provides the option to use
the custom SPy or Seeq Python library to push, pull, and manipulate data. Using
Data Lab, users can build out statistical analyses using their desired Python
libraries, and can create their own Add-on tools to simplify commonly used
workflows and create seamless collaboration.
To begin, users open a Jupyter
Notebook and import the custom SPy library to pull in data and manipulate it.
Spy.search is used to find data, and spy.pull is used to pull in signal and
condition data during the desired timeframe. Asset groups can then be created
in Workbench to organize and contextualize data prior to processing in Data
Lab.
Once data is loaded into Data
Lab, the data frame can be formatted, in this case using the Pandas library,
for use in creating an ANOVA plot and table. By creating a simple Add-on in
Workbench, users can operationalize the workflow and send it to an Organizer
Topic.
ANOVA Plot viewed in Organizer Topic.
(Courtesy of Seeq and
Author Katie Pintar)
