(4ai) Statistical Process Inference, Control and Engineering (SPICE)

The field of systems engineering incorporates a broad range of techniques for advanced process inference, control and design. These techniques are commonly deployed to obtain better understanding, tracking and/or manipulation of complex processes in view of economical and energy efficiency, improved safety and reduction of environmental impacts. Historically, first-principle models have been most popular for the tasks of process state estimation, inference and control. Data-driven, statistical techniques have been developed more recently and require less fundamental process knowledge. As such, they have become increasingly popular since the nineties for ill-understood processes. Today, a tremendous amount of techniques is available within the Statistical Process Control (SPC) field.

Despite the boom in SPC research, adoption of these techniques in the industries is limited. This can at least in part be accounted to the lack of oversight in the field. Often, different techniques are aimed to tackle the same problem yet without elaborate comparison or real-life testing. In other cases, the absence of guidelines for practical implementation is a significant barrier for technology transfer. Therefore, I expect to devote a large part of my research career to bringing structure and guidelines to the SPC field.

Even if the amount of SPC techniques is vast, Fault Detection and Identification is largely based on the on-line analysis of data streams without direct inference with the process. As such, they can be labelled as Passive FDI techniques. However, it is technically possible to manipulate process variables in such a way that the produced data stream is more informative about the status of the monitored process. By doing so, the outcomes of on-line inference procedures can be made more accurate. To search for optimal manipulations of process with respect to on-line process inference accuracy, Optimal Experimental Design (OED) techniques will be adopted, refined and integrated into FDI theory and applications. As the active manipulation of a process for improved FDI is a new concept, I expect to take a lead in the emerging field of Active Fault Detection and Identification (Active FDI).

From a theoretical angle, it can also be said that there is no consensus on the theory and application of non-linear data-driven techniques for process monitoring and diagnosis. At this point, the leading research groups in this niche each have their own technique and a status quo is observed on the theoretical level. A better analysis of existing techniques and further theoretical development is essential to make further advances.

Based on the above analysis, a research plan is set out along three tracks. Each of these tracks relates to the field of applied statistics in process systems modelling, inference and control. The three tracks are listed as follows, from the most theoretical track to the most application-oriented track:

• Track I will result in the establishment of proper techniques for non-linear process monitoring. It will result in the specification of new model structures and algorithms.

• Track II will deliver techniques for improved on-line fault detection and identification (FDI) by means of the development and application of Active Fault Detection and Isolation methods. This will require the integration of Optimal Experimental Design (OED) techniques within FDI theory as well as specific  adjustments to enable real-time computation. Extensive testing in both simulation and laboratory will be important.

• Track III is oriented at bridging the gap between theory and practice in Statistical Process Control (SPC). This track relies on comparative studies, validation and integration into real-life systems. Industrial partnerships are considered important to ensure impact on the field.

Together, the three tracks provide a balanced plan for research in systems engineering. The first track is largely theoretical. The second and third are more practice-oriented with the third focusing the removal of the gap between statistical theory and industrial process systems engineering.