Avishai Mandelbaum
Professor of Service Engineering, Operations-Research and Statistics – Faculty of Industrial Engineering and Management, Technion, ISRAEL
Jointly with Mor Armony, Nitzan Carmeli, Petar Momcilovic, Galit Yom-Tov
I shall describe a data-based modelling framework that supports analysis and optimization of large and complex, congestion-prone service-operations (e.g. hospitals, contact-centers, courts, banks). Drawing from many-server asymptotic regimes, all participants in the service process (e.g. customers, servers) are equally considered resources that are either busy or await each others’ availability. Models are then activity-oriented, where each activity (e.g. service in a hospital) first consumes a subset of the resources at specific states (e.g. waiting patient, available doctor, idle exam-room); and then, upon completion, produces possibly other resources at other states (e.g. served patient, available doctor, exam-room that requires cleaning). We hence refer to our models as Resource-Driven Activity Networks, or RANs for short.
The RAN framework is both parsimonious and rich. Practically, it covers features such as multiple resource types, complex interactions among resources, highly variable long processes, and multiple designs and protocols, all within time-varying environments. Theoretically, RANs cover dynamic- and static-models, closed- and open-networks, and both many- and single-server asymptotic regimes.
This first part of our RAN research starts at the fluid (average) level. Our Fluid RANs are thus deterministic models, yet they are also stochastic-aware in that activity durations are not negligible (this in contrast to conventional fluid models); in particular, duration cdf’s are model primitives. Notably, already Fluid RANs give rise to ample research challenges, that are either novel or originate from classical models (e.g. Gt/G/st, machine-repair, generalized-Jackson networks) when viewed through a RAN-lens.
Ultimately and hopefully, RANs will enable the creation of models directly from data – automatically and in real-time – and consequently the validation of the models’ value against actual service systems. (This is in contrast to still prevalent OR/OM/IE practice, where models have been too often too remote from data, and approximations are validated for merely accuracy relative to their originating models.) This research agenda has been advanced, over 15 years or so, at the Technion SEE Laboratory (SEE = Service Enterprise Engineering). SEELab data will hence be used, throughout my lecture, to both make the agenda concrete as well as to motivate the RAN framework; and the SEELab experience also provides an opportunity to briefly comment on “OR/OM/IE Research – Quo Vadis?”