关键词: 白钨矿　, 摇床　, 精矿接取　, 机器视觉　, 在线状态识别　, 实时调节

Abstract: In tungsten gravity circuits,shaking tables are widely used for cleaning of medium and fine sized particles, while concentrate collection as the terminal output stage is highly sensitive to belt drift on the deck.Under fluctuating feed con ditions and parallel multi-table operation,manual collection based on periodic inspection cannot provide continuous and objec tive process awareness,often resulting in delayed responses and inconsistent operation,thereby amplifying fluctuations in recov ery and grade.To address this issue without modifying the existing flowsheet,equipment configuration,or operating parameters, a machine-vision-guided intelligent concentrate collection system was developed.The system integrates four modules,including image acquisition,visual recognition,automatic alignment of the collection device,and centralized monitorin,to form a percep tion-execution closed loop for concentrate collection.In addition,a "collection state" is introduced as an engineering proxy for operational stability,together with state classification and criteria for unattended operation,enabling online awareness and cen tralized visualization of collection states across multiple tables.Industrial test results show that automated collection achieved a cleaning-stage recovery of 92.43%±1.35% (CV 1.46%),markedly higher than manual collection (69.89%±13.96%,CV 19.98%).The recovery gain (auto minus manual) ranged from 10.34 to 50.25 percentage points,with an average increase of 22.54 percentage points.Meanwhile,grade variability was reduced (CV from 4.09% to 1.71%).A 24 h time series of belt-po sition offset further reveals high-frequency fluctuations and abrupt shifts at short time scales,providing process-level evidence for the necessity of real-time alignment.This study offers a practical route for intelligent upgrading and operational management optimization of concentrate collection in tungsten gravity processing.