Types of Advanced Analytics

Before we delve into advanced corporate analytics, let's categorize the types of analytics that production companies have at their disposal. They are listed from the least to most complex, and categorized as ОТ (operational technology/ engineering) analytics, as IT analytics, or both.

  • 1. Online EIDI analytics (ОТ). Some examples are performance calculations, totalized values, statistics, soft sensors, or whether an early warning condition exists. These are Excel-like calculations that are configured and executed in the EIDI system. They are processed when input data is received, so they are time deterministic. They generally are the easiest to configure and deploy because they are part of the EIDI. Most of these calculations use EIDI data as input.
  • 2. Statistical quality control (SQC) charts (ОТ). These charts are populated by EIDI data, assuming the specific EIDI system supports this feature. The user simply has to configure the desired SQC chart type and the configuration parameters for that specific chart. Examples of configuration parameters are upper and lower control limits or specifying alarm rules. For example, when x number of values fall into a particular zone or are above/below a limit line, this constitutes an SQC alarm. SQC alarms are typically used for monitoring product quality, but they can be used for monitoring other inputs, such as soft sensors. Some EIDI systems have real-time SQC capability, where SQC alarms are automatically calculated, without the user having to view the charts manually.
  • 3. Microsoft Excel (ОТ or IT). Most people at plant facilities are very comfortable with using Excel spreadsheets for various applications. Financial people prefer Excel over traditional EIDI displays and trends. Engineers can import EIDI data into an Excel spreadsheet, where they analyze groups of data variables, get statistical information on historical data, visualize the data in Excel-provided charts, or expose the data to Excel's many mathematical formulas. Software vendors often include some of their software as an add-in menu to Excel, where it can use spreadsheet data to perform calculations, such as linear regression, or to model the data.
  • 4. MATLAB and software languages such as R and Python (IT).

These are powerful and often-used tools for process data analytics. These tools have libraries that contain powerful algorithms for advanced calculations that can be used for all layers of process data analytics, including descriptive, diagnostic, predictive, and prescriptive analytics. The tools help turn process data into meaningful insights by applying techniques such as machine learning. As an example, these analytics can be used to identify process or equipment behavior and make patterns visible. These analytics can predict expected performance. The EIDI provides its operational data as input, typically highly granular historical process and asset data. It may even be filtered by context, such as a specific product grade or during a certain section of the manufacturing process. For machinelearning analytics, the predicted results may be sent back to the EIDI system to assist operations personnel in making better decisions.

  • 5. Process or equipment models and simulations (ОТ). These tools are typically used to simulate either process or equipment behavior under given conditions. Process models can be based on engineering principles, either static or dynamic, linear or non-linear. Alternatively, they can be agnostic to the physical process or environment and rely solely on the historical data they consume to make decisions and predict future behavior. These predictive tools are important to effective operations and have been used for decades. Some of these tools also provide closed-loop or advisory (open-loop) control of a process. Many companies are creating digital twins of the plant or refinery, which is used for analytics and software development. Later in the chapter we discuss digital twins.
  • 6. Offline and big data analytics (IT). These are the newest and most complex analytics. They may use operations and process data for their inputs and often use multiple sources of data (e.g., operational, quality, financial, and messaging) as input. Some examples of these are advanced visualization tools (e.g., Tableau, Spotfire, Power BI, Qlik) or modeling/machine-learning based tools, such as Azure's Machine Learning Studio. Companies may also develop their own proprietary analytics or use a combination of them. The analytics may reside inside the company's corporate firewall, or they may be cloud-based, such as Microsoft's Azure or Amazon Web Services' suite of products. Because they are not connected to an EIDI system, they require either published data sets of contextualized EIDI data, or input may come as raw or prepared data sets from a data lake.
 
Source
< Prev   CONTENTS   Source   Next >