Intelligent alarm system management & value of information analysis applied to continuous pharmaceutical manufacturing

Anshu Gupta, Purdue University

Abstract

One of the important challenges in real time process management is the implementation of intelligent systems that can assist human operators in making control decisions. Modern technological advances have resulted in increasingly complicated processes that present considerable challenges in their design, analysis and management for successful operation. Given the size, scope, and complexity of these modern engineered systems and their interactions, it is becoming increasingly difficult for people to anticipate, diagnose and control serious abnormal events in a timely manner. Failure of the operator to exercise the appropriate mitigation actions often has an adverse effect on the product quality, process safety, occupational health and environment. Hence, there exist considerable incentives to develop intelligent systems for automating fault diagnosis and mitigation. The difficulties associated with implementing intelligent control and the opportunities for improvements are even greater in the pharmaceutical manufacturing domain due to processing challenges. Most pharmaceutical manufacturing involves particulate matter and the behaviors of particles are not well defined. Traditionally the pharmaceutical industry has employed batch manufacturing, but the recent encouragement of the industry by the FDA to modernize manufacturing, has provided the impetus to move towards continuous manufacturing. However, the paradigm shift to continuous processing hinges critically on real time online measurement of Critical Quality Attributes (CQAs) and hence on the development of rational approaches for sensor network design. Moreover, the industry is pushing for real time release of the drug products and it is of critical importance that the quality attributes can be monitored online. In addition, the number of process variable that are measured determines the strength of the diagnostic framework. Hence, various sensing schemes has been developed to monitor the process online. NIR spectroscopy along with Microwave sensing is used to monitor content uniformity of the blend and ribbon density and moisture content of the ribbons. However, the use of intensive online measurement schemes is new to the pharmaceutical industry and thus both the industry and the FDA are on a learning curve on how to deploy this technology effectively. Since, product quality is the main issue in pharmaceutical manufacturing, one way to build confidence in on-line measurement of CQA's is to use multiple sensors across the process train even for monitoring the same CQA at a given point in the train. However, cost considerations dictate that sensor network design strategies need to strike a balance between benefit and cost. At present, there has been no clear cut approach on sensor network design in the pharma manufacturing setting, which includes sensor placement and number of sensors to be used for the measurement of any given quality attribute. In this work, an approach to sensor network design for continuous tablet production is studied along with development and application of Intelligent Alarm System (IAS) framework to the manufacturing process. The designed control system consists of: i) low-level or regulatory control and ii) Exceptional Event Management (EEM) module. The regulatory control tries to keep the controlled process variable at a specified value. Limitations to regulatory control were found in dealing with certain exceptional events where it exacerbates the situation instead of controlling it. The EEM module forms the top layer and helps to mitigate exceptional events that are missed by regulatory control. The control layers along with the knowledge management system forms the complete package of IAS. Numerous abnormal events were studied by simulating the situation in the lab and faults signatures and trends were stored in the knowledge database along with the corresponding mitigation strategy. The framework developed was able to detect and diagnose various common abnormal events, including those involving a single process unit, multiple units, and controller and material faults. The study includes consideration of sensor placement for the on-line measurement of various quality attributes on the continuous manufacturing line and the benefit associated with redundant sensors. A value of information analysis is presented to calculate the additional benefit associated with multiple sensors on a given node of the sensor network. The methodology developed is general, includes identical and non-identical sensors on the network, and calculates the value for all sensor combination across the node. A data-driven model is derived to relate various quality parameters across multiple nodes. The analysis is expanded to include values from multiple nodes to provide the system wide value gained by having redundant sensors at different nodes. The results from the study are used to determine the optimal number of sensors on a node and can be used to carry out the economic and performance analysis.

Degree

Ph.D.

Advisors

Reklaitis, Purdue University.

Subject Area

Chemical engineering|Pharmaceutical sciences|Systems science

COinS