A Synthetic Multi-Scale Benchmark Dataset for Calendar-Aware Resource-Constrained Project Scheduling with Expert-Informed Disruption Scenarios

We provide an open benchmark dataset that is expertly crafted.A synthetic benchmark expertly designed is provided in the form of a public one. for calendar-aware resource constrained algorithms In the context of project scheduling (CA-RCPSP) condition. This data set consists of 12 problem instances in four scale levels: tiny (5–8 tasks), small (25–40 tasks), medium (120–180 tasks), and large (500–1 000 tasks). Each instance represents (i) a directed acyclic task graph (DAG) over six typed (ii) a heterogeneous resource pool including CNC machines and (iii) a resource pool consisting of a single type of resource.(iii) a resource pool consisting of a single type of resource. skilled workers, (iii) a structured work calendar (shift definitions, lunch To four to five weeks (excluding breaks and statutory public holidays), and (iv) two to four stochastic A catalogue of fourteen types of disruption specifications were used. The parameters of the scenarios were elicited from structured interviews with a Production planning engineer (senior) (15+ years, discrete CNC metal-machining) The manufacturing process is not the same as the used to limit stochastic generation ranges, but rather. To directly record information from production that is considered confidential. This expert-informed approach is based on the data set which is grounded in industrial operating. conditions, but still allowing free publication. All twelve instances have fixed random seeds per scenario. Then the seed is guaranteed to be byte-for-byte compatible with the original (base seed = 42 and scenario seed = 42 + i). reproducibility. Accompanying results from three state-of-the-art solvers a Max-Plus The software packages used are algebraic scheduler (CARM-Plus), Google OR-Tools CP-SAT and Gurobi. MIP—show that calendar-compliance rate (1.000 vs. 0.85–0.95) and The result is a system that is scalable, as both competitors fail to scale past large instances.The result is a scalable system as both systems time out when run in large instances, CARM-Plus does not. There is significant variation between methodologies