Multi-type consigned spare parts replenishment decisions of manufacturers considering on-site storage location constraint and demand forecast accuracy

On-site spare parts warehouses are widely used in smart manufacturing to mitigate production downtime. However, on-site spare parts replenishment under a consignment policy is challenged by discrete storage-location limits and uncertain intermittent demand. Existing studies either assume known demand distributions or separate forecasting from inventory optimization. This study develops a chance-constrained integer programming model, which extends the newsvendor-type and capacity-constrained models to embed historical forecast accuracy and discrete storage-location occupancy into multi-type spare parts replenishment problems. A Normal Approximation Method is adopted to reformulate the storage chance constraint, and a Difference-Oriented Adaptive Searching Algorithm (DOASA) is devised to solve the problem. Using real data from a large automotive assembly plant, we demonstrate that DOASA achieves near-optimal solutions with a 0.81% cost deviation from GUROBI in small-scale cases while reducing total cost significantly compared with other heuristics in large-scale cases. The case study reveals threshold and diminishing-return effects in both budget and storage capacity decisions. Improving forecast accuracy from 0.60 to 0.85 decreases total cost by approximately 29.3%, whereas ignoring forecast reliability incurs a 24.5% cost penalty. High forecast accuracy would weaken the binding effect of the budget and storage. The proposed framework can serve as a decision support tool for managers optimizing on-site spare parts control.