EFFECTIVE IMPLEMENTATION OF INDUSTRY 4.0 APPROACHES FOR PROCESS MONITORING IN THE BATCH MANUFACTURING OF SPECIALTY CHEMICALS

Batch processes are widely used in industrial sectors that produce high-value products due to their ease of setup and operational flexibility. Although they are effective for the manufacturing of relatively small amounts of high value-added products, controlling them to maintain consistently high product quality is more challenging than in continuous processing. The advancements brought about by Industry 4.0 have enabled the monitoring of numerous process variables. However, observing just one variable at a time can overwhelm the process supervisor with information. Multivariate statistical methodologies can alleviate this issue by reducing the complexity of the problem while handling noise, multicollinearity, and missing data. The aim of this PhD project is to develop multivariate statistical methodologies that allow to transfer into the industrial practice the Industry 4.0 approach for batch process monitoring. Within this dissertation, process monitoring is intended with a twofold meaning. On the one hand, process monitoring is required to early detect batches with an off-spec end-point product quality with the aim of minimizing the amount of produced off-spec batches. On the other hand, process monitoring is carried out for detecting anomalies in the process operating conditions, with the aim of troubleshooting the process, even if the end-point product is on specification. First, a conventional data analytics technique is coupled to engineering knowledge for troubleshooting a semi-batch industrial process that is a bottleneck for the downstream sections. Data analytics is found decisive to identify an anomaly in the reactor safety interlock system that caused an increase in the time duration of batches in certain conditions. The interlock system is reconfigured and it is assessed that the intervention resulted in a 29% reduction in batch length, an 8% overall cycle time reduction and an 11% reduction in nitrogen consumption, entailing significant energy savings. The development of the data analytics model for this case study highlighted how batch alignment is an important and complex preprocessing step affecting the performance of the analysis. A novel methodology for carrying out batch alignment in an automated manner is therefore developed. This methodology aims at preprocessing data for maximizing the performance of the model under development thanks to a surrogate optimization framework. The proposed method is completely process-agnostic, which enhances applicability to complex batch processes. An industrial fed-batch process for the manufacturing of a specialty chemical, and a simulated fed-batch process for the manufacturing of penicillin are used as test beds, and demonstrate that the proposed methodology has a superior performance than models built using other synchronization strategies. When process monitoring is carried out with the aim of detecting anomalies in the process operating conditions one can use a methodology that does not need batch alignment, namely the assumption-free monitoring methodology proposed a few years ago in the literature. However, effective implementation of this methodology is challenging due to the lack of sufficient documentation and of a clear fault detection and diagnosis procedure. Hence, a set of detailed guidelines enabling the direct implementation of the methodology is developed together with a novel procedure for fault detection and diagnosis within this monitoring approach. Five datasets comprising batch processes from different industrial sectors are used for testing the methodology implemented according to the proposed guidelines, and proved that the proposed approach outperforms traditional process monitoring methodologies. In conclusion, with this Dissertation an example of how process monitoring techniques are useful for improving the performance of industrial processes is provided, returning tangible results from an industrial point of view.