Geïntegreerde procesmonitoring en -regeling voor productie van actieve farmaceutische ingrediënten via tensor-gebaseerde data-gestuurde modelleermethoden

General purpose The ultimate goal of this project is to develop an advanced data driven modeling methodology for full scale plant-wide API production processes. The constructed models are intended for advanced monitoring and control of the processes, which are understood as integrated systems formed by multiple heterogeneous subsystems. Therefore the key aspect is learning the relationships between process variables along the process units and from them to product quality attributes and output parameters. This project will bring Janssen Pharmaceutica towards a cyber-physical API production process where models founded on machine-learning techniques allow monitoring the processes with uncertainty quantification, to enforce quality risk management strategies and in a subsequent step carry out control and optimization of the production processes. Concrete objectives and criteria (i) Developing a novel strategy for exploiting multirate data. The industry has adopted data driven methods for process monitoring, process safety and quality control [9]. However, many practical applications have been restricted to models based on online process data sampled at high frequency (e.g., pressures and temperatures), neglecting the multirate nature of most of the processes and the measurements [10]. Exploiting multirate data guarantees capturing important information from the underlying system that otherwise would be omitted. Hence, the first objective of this research project will be investigating wavelet analysis as a promising method for multirate data pretreatment. In the pharmaceutical industry, this is a key aspect since it enables incorporating information from PAT tools and other infrequent measurements with a high information content. (ii) Developing a modeling methodology for plant-wide multivariate data. The development of multivariate statistical process monitoring (MSPM) strategies has been boosted by the advances in data gathering and computing technologies [11]. In the last two decades it has been a productive research area with a strong interest from the process industry [12]. However, at the current state-of-the-art there are still many challenges to be addressed to develop learning (data driven) methods, particularly for batch processes, which provide models that sufficiently approximate integrated processes, give accurate predictions and enable controllability. Therefore the second objective of this research project aims to develop tensor based data driven modelling methods as these offer advantageous characteristics that can be exploited for modelling integrated plant-wide API processes. Additionally, these methods will allow to investigate the advantage of incorporating available process knowledge in the training process. (iii) Investigate uncertainty propagation in the models to develop a decision support tool. One of the main challenges of the pharmaceutical industry is the highly demanding quality conditions that need to be fulfilled to produce drugs that comply with quality standards (FDA, EMEA). Therefore deploying models and methods requires additional efforts to ensure validity and reliability of the obtained results before basing critical production decisions on them. The third objective will be to investigate methods that provide reliable quantification of the uncertainty propagated through the developed models. Then based on the most appropriated method the best models are selected to develop a multi-agent decision support tool for online monitoring of the Invokana API production process with uncertainty quantification and to support the implementation of quality risk management strategies to the production. (iv) Developing adaptable control and optimization strategies. The transformation of the production systems into CPS is a key element in the expected technological progress as part of the ongoing 4th wave of industrial revolution (the so-called Industry 4.0) [13]. Aiming to direct the innovation at Janssen's Geel site towards this technological prospect, the final objective of this project intends to exploit the constructed data driven models and tools to provide the Invokana API production process with features of a CPPS. Two aspects in the field of process engineering will be investigated to bring the intended novelty and allow the integration with future developments. These are: controllability and optimization of the integrated plant wide API production process to allow the intervention of the computational part into the physical side, and model adaptability approaches to guarantee robustness and sustained validity of the system, while reducing model maintenance costs. Valorization potential The accomplishment of these objectives will result in significant economic gains for Janssen Pharmaceutica by: (i) reducing offspec production, (ii) supporting the application of quality risk management strategies and (iii) favoring simultaneously the quality and the non-quality output parameters. An increase in any in-house knowledge at the Geel site will also help cement Janssen Pharmaceutica's position as Johnson & Johnson's leading production/launch site for small APIs. These factors provide Janssen Pharmaceutica an economic benefit, but also imply a positive socio-economic impact on Flanders by a guaranteed employment and turnover.