A new generation of analytics software simplifies extraction of insights and value from data historians.
Like many process plants, pharmaceutical and biopharmaceutical facilities are awash in process and manufacturing data, but often struggle to extract value and insights from this information.
The challenge of deriving process industry data insights stands in stark contrast to the commercial sector where innovative business intelligence software products are used to extract value from relational databases. Unfortunately, until recently, there has not been an equivalent burst of innovation in applications designed to create value from the time-series data associated with process manufacturing. This leads to one of the key issues: the manual and laborious spreadsheet-based approach to insight that confounds the efforts of pharma organizations to derive value from their historian data.
But this situation is changing, as some companies have recently introduced data analytic software specifically designed for creating insights and extracting value from the time-series data stored in historians such as OSIsoft PI. These companies are not only making it easier to rapidly enable investigation of historical data, but they are also taking a more modern approach to the software delivery and user experience.
Different data analytic software products have varying deployment models and features, but most follow these basic steps:
- On deployment, the data analytics software connects to the historian. Most historians reside on the corporate intranet, so it’s easy to make this connection by installing the software on a workstation or server connected to the intranet. The software then automatically locates the historian and establishes the connection.
- Automatic indexing of the tag or sensor names in the historian to make them easy to search and access the related data. This step typically takes less than an hour for tag counts of up to 250,000 tags, and occurs just once at setup. Once this step is complete, live data from the historian is available in the software for the user to investigate and analyze.
- The user searches the historian and other data sources for the data points of interest. This is typically done using a Google-like search algorithm. Search terms correspond to data names given in the historian, and any part of the name can be used in the search. For example, a dryer temperature could be found by searching on its tag name, TT-101, or its label, Dryer Inlet Temperature, or a subset of the name string.
- Users can then visualize the time series data over a designated period time. This feature has long been associated with trending packages, but new search tools using visual patterns, limits and the ability to dimensionalize data with context from other data sources improves the trending experience.
- Finally, just as the trend viewer experience is redefined by this next generation of data analytics software, so too is the user experience. Work can be stored for reuse or shared with colleagues, either as a way to capture expertise, or in real-time to enable distributed discussions across an organization. And, of course, the deployment model for the software accommodates customer requirements for on-premise, cloud, or hybrid scenarios.
Specific examples of how new data analytics solutions are being applied to address specific production issues in the pharma industry include the following reference points demonstrating the value of improved insights into existing historian data.
Reducing Dryer Cycle Time
A major pharmaceutical company needed to reduce the cycle times of a dryer unit because it was the bottleneck for the production of small molecule products. Data from the dryer, and from many other components in the plant, was stored in a historian.
The dryer unit takes liquid slurries and forces the moisture out to create powder. The plant makes runs of different products and performs a clean-in-place (CIP) operation between each product.
The CIP operation was taking too long, which created a bottleneck and slowed production. A second issue was optimizing the tradeoff between quality and production. More heat would result in higher throughput, but the extra heat could adversely affect product quality.
Analysis of asset effectiveness was difficult because each of the dryer’s states of operation had to be separately identified and measured. The data came from many months of production, and similar analyses had to be repeated across multiple lines in five different processing plants.
Data analytics software enabled production engineers to quickly identify operating states based on the digital signatures of key temperature, pressure and other sensor data. Further analyses were then used to generate aggregates to allow breakdown of operating state duration, productivity, and resulting product quality for each asset by date and time. The result was a detailed look into the tradeoff of productivity versus quality, with improved performance the result.
Predictive Maintenance Saves Money
A leading pharmaceutical company in the field of biotherapeutics needed to streamline production of its pilot-scale operations to optimize production of high-quality large molecules. Maintenance of various items of equipment was required, but performing these tasks on a calendar basis was inefficient. Some items were serviced too frequently which added to downtime and costs, while others weren’t serviced often enough which negatively impacted quality.
Using data already resident in a historian, plant maintenance personnel now use conditioned-based maintenance protocols that target the actual use of equipment and enable proper maintenance timing. With continual real-time trending in place, equipment performance is monitored continuously, and preemptive measures are taken when maintenance is required. Perhaps more importantly, maintenance intervals can be stretched out, increasing uptime.
The data analysis required to predict problems before they occur is quite complex, as it requires looking at data in detail and automatically alerting plant personnel before problems occur. This is done by creating data signatures using information from sensors monitoring equipment known to be in good running order and then tracking deviations from these signatures.
This is a relatively simple task for data analytics software connected to a historian, but is very difficult using traditional tools. Specialized maintenance management software suites are available to perform these tasks, but these products tend to be very expensive to purchase, install and maintain.
Analyzing Production Phases
A process R&D team at a major pharma firm was looking for ways to improve yields and quality. They knew the specific production phases, and were aware that operating conditions of those phases determine product quality and throughput.
The company now uses features of data analytics software called pattern search and value search to isolate these important phases. This context or dimensional data is used to isolate all the operation data for those phases. Although this task sounds simple, it was taking weeks using a spreadsheet to extract data from a historian, and then building a model within the spreadsheet to analyze the information.
The company’s engineers and scientist now use data analytics software to perform various analyses on their data to identify correlations among operating conditions, yield and quality, with the result being greater throughput.
Pharma and biopharma plants often possess all the data they need to improve operations within their data historians, but creating insight from this information can be difficult, expensive, and time-consuming using traditional data analytics tools.
Several companies now offer data analytics software specifically designed to interact with the time-series information found in historians. Because this software is designed for just one task, as opposed to a general purpose tool like a spreadsheet, it provides a faster and less expensive solution consistent by applying the innovative experiences associated with software in the consumer and IT markets.
About Michael Risse:
Michael Risse is the Vice President at Seeq Corporation, a company building innovative productivity applications for engineers and analysts that accelerate insights into industrial process data. He was formerly a consultant with big data platform and application companies, and prior to that worked with Microsoft for 20 years. Risse is a graduate of the University of Wisconsin at Madison and lives in Seattle.