Mass Balances and Data Reconciliation

There are several requirements for developing a methodology to implement a data reconciliation system. First, the algorithm that is able to balance and reconcile the plant data must be robust and must perform correctly against any process topology or configuration. In the past, several mathematical and statistical tools have been developed to solve this reconciliation problem. However, a mathematical algorithm without an information infrastructure is of little value. Second, the right infrastructure to connect the object-oriented model to the real-time process data must be in place. A database that allows storage and manipulation of elements is necessary. The system has to adapt to changes in the process topology because, for example, a meter going out of service is enough to change the mass balance of a process network. Additionally, this object-oriented database should be able to communicate in real time with the EIDI system. Third, the EIDI system acts as a repository of process data, both raw and reconciled, that allows the distribution of data to all plant staff, from the operators to the plant manager.

The typical problems with process data in industrial plants are:

  • • An overwhelming amount of data;
  • • Low confidence in the available data;
  • • Lack of consistency (the data do not make sense);
  • • Data violating known constraints (mass and energy balances); and
  • • Poor data quality creating a decision-making fog at all levels of an organization, which may result in a financial penalty (fine).

In the past, these problems have been addressed using traditional methods to capture the required information. However, a key issue with these traditional methods is that human errors can occur during manual data entry, and operational events may be ignored.

Online process stream analyzers provide chemical assays for a given time period. The laboratory provides sample data for the streams. As such, these measurements provide redundancy, which require data reconciliation to find the measurement errors and close the mass balance. Having data redundancy enables us to infer other streams' properties, which are not measured. The unmeasured variables are estimated using a least squares reconciliation algorithm. The model (plant structure) can run hourly, by shift, or daily to produce reconciled reports. In addition, the reconciled data is sent to the EIDI system where it is available to operations support staff and knowledge workers. Figure 7.9 describes the daily procedure to reconcile process data. The unbalanced data is collected from the EIDI. Once the data is in the system, a group of analysis rules is executed to detect gross errors that can negatively affect results. These


Daily procedure for data reconciliation. (From Bascur and Linares 2005.)

gross errors are eliminated before the final reconciliation run, which provide a unified balance for the whole complex.

An oil refinery mass balance model includes the crude receipts, storage tank inventories, and the shipping and loading rack receipts. Once the mass flow, stream, and inventory compositions are reconciled by the system, many calculations and yield reports can be performed. It is important to realize that in the process of solving the network problem, the measurements are validated and gross errors are detected prior to providing a solution.

Once the data are reconciled, they can be sent to the enterprise resource planning (ERP) and other business information systems, where they can be distributed to users. The infrastructure of the data reconciliation system has to adapt to any changes in process flow or changes to the measuring system because both the process topology and the data are not static.

The reconciled data can then be used to improve the process to continuously improve yield performance by looking at the overall refinery performance. The data can be reused to improve process planning and to determine the optimal setpoints for steady-state optimization of the plant (Narasinham and Jordache 2000; Bascur and Linares 2005). Hodouin (2011) also described additional uses for this type of methodology.

< Prev   CONTENTS   Source   Next >