Belt Conveyors are essential for transporting dry bulk material in different industries. Such structures require permanent inspections, traditionally executed by human operators based on cognition. To improve working conditions and process standardization, we propose a novel procedure to inspect conveyor structures with a ground robot composed by a mobile platform, a robotic arm, and a sensor-set. Based on field experience, we introduce ROSI, a new robotic device designed for long-term operations in harsh outdoor environments. The mobile robot has a hybrid locomotion system, using wheels to reduce energy consumption while covering long distances, and also flippers with tracks to improve mobility during obstacle negotiation. A mechanical passive switch allows decoupling tracks’ traction, reducing components wear and energy consumption without raising mechanical complexity. Aiming the robot-assisted operation, control strategies help to (i) command both the mobile platform and a robotic manipulator considering the system whole-body model, (ii) adjust the contact force for touching the conveyor structure during vibration inspection, and (iii) climb stairs while automatically adjusting the flippers. Machine Learning algorithms detect conveyors’ dirt build-ups, roller failures, and bearing faults by processing visual, thermal and sound data as inspection functionalities. The algorithms training and validation use a dataset collected from running conveyors at Vale, presenting detection accuracy superior to 90%. Field test results in a mining site demonstrate the robot capabilities to stand for the harsh operating conditions while executing all the required inspection tasks, stating ROSI as a disruptive solution for Belt Conveyor inspections and other general industrial operations.

带式输送机对于运输不同行业的干散装物料至关重要。这种结构需要永久性检查，传统上由人工操作员根据认知来执行。为了改善工作条件和流程标准化，我们提出了一种使用由移动平台、机械臂和传感器组组成的地面机器人来检查输送机结构的新程序。基于现场经验，我们推出了ROSI，一种专为在恶劣的户外环境中长期运行而设计的新型机器人设备。移动机器人具有混合运动系统，在长距离行驶时使用轮子来降低能耗，还使用带履带的脚蹼来提高越障时的机动性。机械无源开关允许分离轨道的牵引力，在不增加机械复杂性的情况下减少部件磨损和能源消耗。针对机器人辅助操作，控制策略有助于 (i) 考虑系统全身模型来控制移动平台和机器人机械手，(ii) 在振动检查期间调整接触输送机结构的接触力，以及 (iii) ) 爬楼梯的同时自动调整脚蹼。机器学习算法通过处理视觉、热和声音数据作为检查功能来检测输送机的污垢堆积、滚筒故障和轴承故障。算法训练和验证使用从 Vale 运行的传送带收集的数据集，检测准确率超过 90%。