A Scalable Vertical ASRS Architecture for SMEs with AI-Based Maintenance Support and Retrieval Optimization

Abstract

Vertical automated storage and retrieval systems (ASRSs) are a promising technology to improve space utilization and order-picking consistency, but the adoption of ASRSs by small and medium enterprises (SMEs) is limited by acquisition cost, integration effort, unscheduled stoppages, and inefficient tray sequencing. In this paper, we propose an integrated and repeatable simulation framework arranged by five interconnected components: system design, cost and scalability, safety credibility, AI-supported condition monitoring, and priority-aware retrieval scheduling. A parametric full system CAPEX/TCO model and a controller-stack scale study address the first objective; rack and trolley terms are still replaceable by supplier-dependent ones. The second objective is formulated as a maintenance alert categorization rather than a remaining useful life estimate. We use a Random Forest with 180 trees and a stratified five-fold cross-validation on 4,500 physically bounded synthetic samples. The third objective uses a distance-age-priority score, selected during validation, and compares it to FIFO and nearest-tray strategies using 2,000 paired test batches. At five identical units, the hybrid fail-safe stack saves 12.0% of controller CAPEX compared to PLC control. The classifier reaches an accuracy of 91.7%, a macro-F1 of 0.917 and Fault recall of 92.1%. The scheduler reduces journey distance by 59.7% and priority-weighted completion time by 8.9% against FIFO. These results offer preliminary design evidence, but no field validation, full system cost proof, or functional safety certification.