(746e) Steady-State Data Reconciliation of a Direct Compression Tableting Line
- Conference: AIChE Annual Meeting
- Year: 2017
- Proceeding: 2017 Annual Meeting
- Group: Pharmaceutical Discovery, Development and Manufacturing Forum
- Time: Thursday, November 2, 2017 - 4:45pm-5:05pm
The objective of this work is to develop a model-based framework, which for a given sensor network configuration and measurement uncertainties, will predict the most likely state of the process. As a consequence, real time process decisions, whether in control or exceptional events management can be based on the most reliable estimate of the process state. The presentation focuses on real time measurement and monitoring of the direct compression continuous tableting line. Specifically, we report on the use of data reconciliation and gross error detection (GED) as tools for mitigating the effects of measurement errors and sensor malfunctions.
Data reconciliation (DR) has been widely used in other industries (i.e. oil and gas) to accomplish this goal. Data reconciliation is model based and thus consists of mechanistic (e.g., material balance) as well as empirical relationships between variables including properties . Through DR, we can deal with random errors and have an estimation of unmeasured variables. However, in order to use steady-state data reconciliation (SSDR), redundancy with no gross error in the measurement is required. In order to detect gross errors in the system, we need to apply the global test and measurement test. These tests are similar to outlier detection . DR also offers an opportunity to detect gross errors.
In this paper, we develop and demonstrate a DR framework for a direct compression (DC) line in real-time at steady-state. The DC line sensor network includes the on-line use of load cells for the weight of the feeders, the near-infrared (NIR) spectroscopy sensors for the powder blend composition, the X-ray sensor for the total powder flowrate and a soft-sensor model for the tablet hardness and weight. Steady-state data reconciliation (SSDR) is one of the simplest forms and offers short computational times. The results are compared to the results from the multivariate models (MVM), which are also use for the prediction of process states and error detection . Since the DC line model at steady-state is linear, the results are similar . The SSDR and MVM are implemented in MATLAB 2015b and the data is collected in the Emerson DeltaV server. The operator can see the real-time results in the process historian view and take a decision according to the result.
 S. L. Lee, T. F. O. Connor, X. Yang, C. N. Cruz, L. X. Yu, and J. Woodcock, âModernizing Pharmaceutical Manufacturingâ¯: from Batch to Continuous Production,â no. 1, pp. 1â25, 2015.
 M. Ierapetritou, F. Muzzio, and G. V. Reklaitis, âPerspectives on the Continuous Manufacturing of Powder-Based Pharmaceutical Processes,â AIChE J., vol. 62, no. 6, pp. 1846â1862, 2016.
 S. Narasimhan and C. Jordache, Data reconciliation and Gross Error detection. Houston, TX.: Gulf Publishing Company, 2000.
 C. Knopf, Introduction to Data Reconciliation and Gross Error Detection. .
 J. A. Romagnoli and S. M.C., Data Processing and Reconciliation for Chemical Process Operations. San Diego, CA, 200AD.
 O. Cencic and R. Fruhwirth, âA general framework for data reconciliation-Part I: Linear constraints,â Comput. Chem. Eng., vol. 75, pp. 196â208, 2015.