(171e) Maintenance Management for the Sensor Network in Continuous Pharmaceutical Systems

Authors: 
Ganesh, S., Purdue University
Reklaitis, G. V. R., Purdue University
Nagy, Z. K., Purdue University
Rossi, F., Purdue University
The progress in the adoption of monitoring and control systems has accelerated the progress towards smart manufacturing in the pharmaceutical industry. Smart manufacturing necessitates reliable and accurate measurements from field devices combined with good quality estimates of unmeasured process variables. Thus, the sensor network which consist of equipment measurements, field devices, PAT tools, process models and data analytics is a fundamental requirement for robust process operations.

Data reconciliation exploits the redundancy in the sensor network together with process models to ensure the validity and accuracy of process measurements. The effective implementation of data reconciliation requires reliability in the raw measurements from PAT tools. However, since secondary pharmaceutical manufacturing primarily involves processing particulate materials, fouling of the inline and online PAT tools is inevitable. Corrective maintenance of fouled sensors can cause operational losses and uncertainties in manufacturing systems. Moreover, regulatory filings would require sensor network strategies for handling sensor faults where multiple sensors could fail simultaneously. For a smart pharmaceutical facility, it is imperative to anticipate and correct problems with the sensor network instead of only reacting to problems.

In this paper we investigate the impact of predictive maintenance of the sensor network in continuous tableting processes. The maintenance policy seeks to maintain measurement bias within acceptable limits such as to satisfy the conditions required for data reconciliation, using mechanistic models and material balances. The integrated framework schedules the anticipated downtime in maintaining every sensor – disconnecting the sensor or a group of sensors, cleaning the sensors and connecting it back to the sensor network. It is important to note here that the maintenance policy depends on the physical setup, materials used, the sensor network of the process and must rely on and learn from process history.

The challenges of implementing a robust and reliable sensor network are discussed, followed by demonstration of the effectiveness of the maintenance framework for a direct compaction case study. By integrating predictive maintenance with data reconciliation, the framework enhances the reliability in the measured estimates of CQAs for practical realization of Real Time Release Testing.

References

  1. Smart Process Plants, M. Bagajewicz (2010)