ProcIndustries Integrates Operational and Business Data

The Proclndustries team, led by Peter Argus (continuous-improvement manager) looks forward to working with corporate IT director Pat Verlaine to start implementing business analytics and identifying artificial intelligence tools that leverage the EIDI data they already possess. Peter's vision is clear. He would like to use the collected refinery data to accomplish the following:

  • • Improve equipment uptime through data analysis. If possible, determine the expected lifetime of critical refinery assets.
  • • Avoid product quality excursions.
  • • Increase product yields.
  • • Increase production flow from raw materials, during process transformations, and shipping final product to customers.
  • • Minimize consumable resources (e.g., power, water, steam, hydrogen) while maintaining acceptable yield, quality, and throughput.
  • • Minimize environmental and safety issues.
  • • Simplify production, quality, equipment, and safety and environmental reporting.

Time-series data enables companies to use continuously fed real-time sensor data with derived variables and event-framed data (see Chapter 3). A time- series derived data variable can measure standard deviation, maximum, minimum, average, totalized value, mode, and other things. These derived variables can also be time-sliced to frame and capture data within a specific production or operational event. The EIDI provides this rich set of information in real time, which assists when implementing optimization strategies, such as machine learning (ML) for predictive asset behavior. These augmented data sets become available to train predictive analytics models using machine-learning tools. In turn, these models train algorithms with added information about variability in the data. The basic raw sensor data is often insufficient to accomplish advanced analytics.

The availability of the data enables these algorithms to learn and continuously improve, and to adapt to process changes as production or equipment behavior deteriorates. Having the historical data available enables forward- looking models. By extracting contextual historical data and transforming the data into a data set that the algorithms are easily able to consume as input, predictions regarding process unit behavior are made. As a result, users and process control systems are able to make better-informed decisions that avoid production losses; inferior quality products; and reduce wasted energy, water, and steam.

Time-series data can be used to generate real-time alerts indicating impending equipment failures. Systems are configured to prescribe detailed actions that mitigate or solve problems before an unscheduled downtime event occurs (see Chapter 6). Peter Argus commented, "Today we can have both production prediction models and asset equipment warnings in real time, using much larger data sets than ever before. Predictive models can recognize precise patterns that indicate degradation and impending failure." Peter added that their EIDI had the ability to store and display future time data, meaning refinery personnel can monitor variances in real time. They can observe how equipment or a process behaves, measured against predicted values.

Chuck Smith (instrumentation and process control engineer) emphasized, "These data patterns are the key to developing predictive analytics models." In the past, Proclndustries used mathematical/statistical models based on engineering principles. They are still a very important basis for the development of these models, but the inability to recognize time patterns embedded in the operational data makes them less attractive.

Peter pointed out that they need a standard, simple way to extract the data for the increasing number of queries people and software systems will need, once people gain access to this information: "The team has to investigate how to develop a standardized tool that reduces the time for data extraction. The operations teams will be more engaged if the extraction time is short and they don't have to develop custom scripts."

 
Source
< Prev   CONTENTS   Source   Next >