Geometric Process Control (GPC)

Geometric Process Control technology is based on the worlds most capable and extensively developed implementation of the Parallel Coordinate Transformation solving the age-old problem of how to plot a graph with more than three axes. We test our standard CVE (C-Visual Explorer) software product with 1,000 axes or variables but can go well beyond that for some of the problems we work with in the process industries.

So what is Geometric Process Control (GPC)?

Perhaps the deepest insight from this display is that the selected data define a closed shape in multi-dimensional space which, in the process industries, is called an Operating Envelope. This is the foundation of PPCL’s technology of Geometric Process Control, implemented in PPCL’s real-time CPM product which can control a process into a chosen multi-variable Operating Envelope or use the Envelope to predict fault conditions in time for them to be avoided or mitigated. This is delivering to the process industry its long-held ambition for much better and more meaningful operator alarms and provides a path to the ultimate goal of ‘predictive alarming’.

In manufacturing industry Geometric Process Control is seen as the successor to the highly mathematical Multi-variable Statistical Process Control (MSPC) which in turn is the long-running attempt to widen the applicability of the very successful univariate Statistical Process Control (SPC) of the 1950’s to today’s more complex and data-rich manufacturing environments.

Domain knowledge is essential and PDI can be learnt in days. Perhaps you can combine them to make similarly impactful discoveries in your industry…… or just save yourself a great deal of the time you spend now trying to extract meaning from large spreadsheets full of numbers.

And the PDI plot is its own dashboard so you won’t have to spend nearly as much time as you spend today creating dashboards and simple 2-d graphs to explain the findings you made in your data. How many of these have we all seen during the Covid pandemic?

We have been developing and implementing discoveries made in and for the process industries (we are chemical engineers) in our CVE product for over 20 years.

What are Parallel Co-ordinates?

In a parallel coordinate plot all the variable axes are vertical and parallel, and a single data point is a polyline connecting the values of all the variables. The polyline is actually a representation of a point in a high dimensionality space with the coordinates of the point being the values of the individual variables.

A large dataset plotted in parallel co-ordinates generally suffers from the well-known ‘over-plotting’ problem and conveys little information. What makes CVE and PDI so outstandingly useful is the capability to query the data using combinations of simple 2-d shapes such as lines and planes all brought together with Boolean logic to allow the creation of queries in as many dimensions as the user could wish for. We named this “Interrogative Visualization” and it provides an astonishingly powerful “no-maths” method for analysing numerical data. CVE and PDI contain sophisticated Join functions to facilitate combination of time-series and/or parameter data from other databases.

Many types of query have been found useful and implemented in CVE and PDI, but even the simplest query can provide immediate insight into cause-and-effect relationships to the user with domain-knowledge. Suppose the dataset in the example above comes from a cake-making  plant with actual values of the process variables such as weight of flour, size and number of eggs, dimensions of baking tins, mixing time, cooking time and oven temperature combined with quality variables of the product from a human cake-testing panel. The user of CVE or PDI can put specification ranges (he yellow triangles in the example) for what constitutes a good cake on the quality variables and all the data points or polylines with quality values inside their ranges are shown in yellow. The ranges within which the process variables can be used (for example, how many small eggs should the cooks use to give the same ‘good-cake’ result as the large eggs that are usually used?) are immediately visible. This is immeasurably more informative than putting in numbers for ranges on the quality variables and getting back numbers for ranges on the process variables.

The yellow triangles are the product specification limits in this cake-making example. Each polyline is one cake. Cakes that met all of the tasting panels specifications ( taste, texture, shape, wetness etc) are in yellow so looking at the manufacturing variables (how much flour, number and size of eggs, cooking time and temperature etc. one can see in yellow how to make good cakes and in black how to make cakes that tasting panel will reject.

Parallel Data Investigator (PDI)

PDI is a new derivative of PPCL’s C-Visual Explorer(CVE) product that all domain experts could use to push back the boundaries of knowledge in their domain. We have simplified CVE by removing all the process industry special functionality and developed a different business and support model to bring prices way down hence will be selling the new Parallel Data Investigator (PDI) product under the Curvaceous Software Ltd. brand. Anyone from the process industries should head back to PPCL and the full application and domain support of the process-industry-specialized  CVE and CPM products.

Outside of the process industry we believe that many users will use far fewer variables and we have taken this into account in producing the PDI (Parallel Data Investigator) derived product.

CVE and PDI enable the user to visually understand the whole of the data without statistical reductions, and empower the innate abilities of the human mind that are not brought into play with traditional purely numeric analysis. Sometimes a picture is worth a thousand numbers!

PDI Features

PDI is a workstation/laptop product supported under Windows 8 and 10 and utilizing the Parallel Coordinate Transformation.

PDI Data sources are flat files in csv format where the rows are different sets of related variable values and the columns are the variables and their labels. Files below 50 million cells (rows x columns) are recommended  as above this size some PDI functions may not be able to obtain the amount of temporary memory some PDI functions need.

PDI includes sophisticated Join and Append functions for combining multiple data files including alignment of any time-based files at different data frequencies.

PDI recognises blank or non-numeric fields as invalid or missing values and provides several methods for substituting alternative values. It also allows the user to mark data values he knows from his domain knowledge to be invalid

PDI includes the powerful query features of CVE allowing high dimensionality queries to be created in a number of steps

PDI includes visual clustering of very high dimensionality and can identify the genuine multivariable outliers

PDI includes the unique Box Variable and its associated Pareto plot for identifying variables contributing most to variability

PDI comes with a 200-page User Guide in pdf format

PDI supports use on multiple displays through the extended Windows desktop feature and includes capture of images of PDI windows in png format

PDI is available with a one-year licence in three different tiers defined by maximum numbers of variables.

Click here to buy your copy of PDI now!