As an important hub for logistics connection and production rhythm control in the steel production process, the warehouse area is the basis for the unmanned and intelligent construction of the factory. The operation of the warehouse area in the traditional steel rolling production workshop is mainly done manually, with low automation level, large labour quota, harsh operating environment, high safety production and occupational health risks, standardized operation and low production efficiency, which seriously restrict the intelligent and unmanned steel rolling production line. development bottleneck.
The realization of unmanned crane operation and intelligent scheduling of the warehouse area will fundamentally change the operation mode of the warehouse area, which is of great significance for improving production efficiency, reducing production costs and improving product quality.
Electromagnetic spreaders are the most common way to realize unmanned lifting of long products. However, it is difficult to stack rods and wires strictly according to the stacking position, and it is easy to roll and shift. At this time, according to the positioning method of the steel coil warehouse, it is easy to cause the “virtual hanging” phenomenon that the centre of gravity of the steel is outside the edge of the spreader due to insufficient positioning accuracy. As a result, lifting failure or steel falling accidents during hoisting and transportation may result. Therefore, it is necessary to carry out key technical research on the unmanned crane lifting of long products.
On the other hand, the intelligent warehouse management system should be able to adapt to the characteristics of long material lifting and transportation, have the ability to automatically adjust and adapt to changes in storage locations, and should meet the global dynamic optimal scheduling of large-scale logistics warehouses for cranes, ground logistics equipment, and shipping equipment. At the same time, for the in-factory reshipment and storage, transhipment and sales functions of the logistics warehouse area, it is necessary to meet the needs of fully automatic unloading and automatic loading under unmanned conditions and to achieve docking coordination with the production and sales systems.
For this reason, the Engineering Technology Research Institute of Beijing University of Science and Technology (hereinafter referred to as “Beijing Science and Technology Research Institute”) has innovatively developed a warehouse area environment perception and three-dimensional reconstruction, deep integration of machine vision and crane control, and a warehouse area for the bar and wire storage area. Multi-agent cooperative optimization of intelligent warehouse management scheduling, warehouse area centralized control and intelligent factory collaborative optimization is the iconic second-generation unmanned cranes and intelligent warehouse management technologies, and realize the integration of bar smart warehouse and high-line smart warehouse is the first demonstration application.
1. Three-dimensional reconstruction and environmental perception technology of multi-source information in the long product storage area
Aiming at the characteristics and problems of a large field of view, dense storage of materials in the storage area, and various vehicles in the metallurgical long product storage area, the vehicle and vehicle-mounted material identification, storage area map construction and feature recognition technologies have been developed, and the bar, bar, and The precise positioning of high wire coils, transport vehicles and onboard materials provides comprehensive and accurate material spatial location information for the precise control of unmanned cranes.
(1) Vehicle and onboard material recognition based on non-repeated scanning.
A fixed detection system based on non-repetitive scanning was built, and a vehicle feature recognition model based on probability density, a rod position and shape recognition model based on projected dimensionality reduction grid features, and a circle fitting based on principal component direction dimensionality reduction were developed. The wire coil position recognition model realizes accurate recognition of vehicles such as vehicles and onboard cargo.
(2) Map construction and feature recognition of bar storage area based on binocular vision.
A three-dimensional detection model of the bar storage area based on binocular stereo vision, a multi-pose measurement model of the bar storage area image based on visual and inertial information, and a bar position recognition model based on the mean curvature feature of the dimensionality-reduced grid are proposed to realize Accurate identification of the stacking of bar bundles in the bar storage area.
(3) Map construction and feature recognition of the high-line reservoir area based on line scanning.
A dynamic scanning 3D point cloud map construction model and a point cloud segmentation model for high-line coils in the high-line storage area are proposed, and the accurate identification of high-line coil stacks in the high-line storage area is realized.
2. The precise positioning control technology of unmanned overhead crane integrated with machine vision storage location recognition
In view of the high positioning accuracy requirements, easy extrusion deformation, easy position change and high lifting stability requirements faced by the long material lifting process, the innovative development of precise positioning control and operating process spreaders based on machine vision warehouse location recognition Key technologies such as anti-swing control, multi-segment path cooperative transition control, four-axis linkage control, and multi-vehicle dynamic collision avoidance control have broken through the technical bottleneck of unmanned cranes for long products and realized high-efficiency, high-stability and unlimited Faulty operation.
(1) Precise positioning control based on machine vision location recognition.
Establish accurate and safe positioning control technology based on machine vision warehouse location recognition, and realize lifting primary + secondary positioning control based on warehouse management goals and scanning correction, lifting weighing based on weighing and scanning verification in the entire positioning control process +Position checking control, primary + secondary positioning control of lifting based on warehouse management target and scanning correction, position checking control of lifting based on visual scanning.
(2) Spreader anti-swing control technology during operation. establish
The open-loop speed setting control and closed-loop acceleration control model of vertical anti-sway, adjust the acceleration of the crane in real time by setting the multi-section speed curve between the start and end positions, and according to the measured shaking angle of the crane, the actual position and speed of the crane, etc., to maintain The spreader swings within a certain limit range.
(3) Four-axis linkage and multi-segment path cooperative control technology.
Through the development of four-axis linkage control technology for cranes, trolleys, lifting and rotating, and multi-section path cooperative transition control technology, the coordinated optimization of lifting, lowering and avoiding dangerous areas can be realized, the running time of cranes can be effectively shortened, and the overall efficiency of the warehouse area can be improved. operating efficiency.
3. Multi-agent-based efficient optimal scheduling technology for warehouse logistics
Under the condition of complex space-time constraints, various logistics equipment are coordinated and dispatched, and a multi-agent-based efficient production optimization scheduling technology for logistics in warehouse areas is established. At the same time, dynamic production tasks, delivery plans and various real-time statuses of the scheduling policy are dynamically updated.
(1) A dynamic scheduling model for cranes based on multi-vehicle collaborative optimization.
Establish a multi-task multi-crane cluster scheduling model, consider the cooperative work of various delivery equipment and cranes, and realize the dynamic scheduling of cranes for equipment state change disturbances, and improve the operating efficiency of cranes in the warehouse area.
(2) The intelligent optimization model of the storage area in and out of the warehouse.
Comprehensively consider factors such as real-time inventory in the warehouse area, production plan, shipping plan, lifting capacity, etc., and establish an intelligent decision-making model for material storage and delivery stacks to reduce the number of inverted stacks in the warehouse area, reduce the degree of irregularity of stacks, and reduce logistics waiting time and improve work efficiency in the reservoir area.
(3) Dynamic optimization model of crane path in the unmanned reservoir area.
Aiming at the complex environment and dynamic changes in the actual warehouse area of steel mills, a path-planning model for unmanned cranes based on reinforcement learning is developed. Path Planning Problems in Deterministic Environments.
4. Centralized control of long product intelligent warehouse area based on digital twin and industrial 5G
Developed key technologies such as digital twin and cluster dispatching simulation systems based on key elements of reservoir operation, logistics design and analysis and optimization in the reservoir area, industrial 5G network oriented to online control and monitoring, and security monitoring in reservoir area based on visual AI. A new smart centralized control mode for long product warehouse areas featuring unmanned operation on site, efficient and coordinated operation of logistics equipment, remote centralized visualization, and digital monitoring.
(1) Cluster-integrated intelligent centralized control centre in the reservoir area.
The centralized and flat control mode is adopted to realize the real-time centralized monitoring of dispatch management, equipment spot inspection and receipt and delivery in the warehouse, to centralize and quickly deal with on-site abnormalities and failures, to streamline and integrate on-site operation positions, and to improve the efficiency of multi-professional collaborative work.
(2) Digital twin modelling and analysis system based on key elements of reservoir operation.
Use digital twin technology to truly display the scene in the warehouse area, drive the 3D scene in real-time through production data, and present the information concerned by production operations, process technology and management personnel to the corresponding personnel in a timely, accurate and clear manner, and improve the implementation of the warehouse area, process analysis and Efficiency in the implementation of management decisions.
(3) 5G network technology for online control and monitoring.
Combining with the new generation of 5G network technology, it realizes the full coverage of 5G signals in the high-line warehouse area, and uses the characteristics of 5G low latency and large bandwidth to realize the wireless and highly reliable transmission of surveillance video, 3D point cloud data, and PLC signals, and complete the vehicle-to-ground Communication to ensure the unmanned driving requirements of the crane.
(4) Reservoir area security monitoring based on visual AI.
Using artificial intelligence identification technology to complete the intelligent supervision and automatic recording of abnormalities in the reservoir area through the monitoring terminal + intelligent AI terminal, and carry out early warning prompts and quick reference.
3. Main innovative achievements
1) Three-dimensional reconstruction of multi-source information and environmental perception technology and equipment in the long product storage area.
Aiming at the characteristics and problems of a large field of view, dense storage of materials in the storage area, and various vehicles in the metallurgical long product storage area, the identification of vehicles and onboard materials based on non-repetitive scanning, and the construction and characteristics of bar storage area maps based on binocular vision have been developed. Key technologies such as identification, map construction and feature recognition of the high-line warehouse area based on line scanning have realized the precise positioning of bars, high-wire coils, transport vehicles, and vehicle-mounted materials in the long product warehouse area, and completed the digital reorganization of the entire warehouse area. Structure and precise perception.
2) Unmanned crane machine vision location recognition technology and safe and precise control equipment.
In view of the characteristics of easy extrusion deformation, easy position change, high positioning accuracy and high hoisting stability in the lifting process of unmanned cranes for long products, we innovatively developed precise positioning control based on machine vision warehouse location recognition, based on fuzzy Key technologies such as spreader anti-sway control with variable gain control, multi-segment path cooperative transition control, four-axis linkage control, and multi-vehicle dynamic anti-collision control have broken through the technical bottleneck of positioning control technology for unmanned cranes for long products, ensuring the high efficiency of intelligent storage areas. High efficiency, high stability and safe operation.
3) Optimizing efficient production scheduling of warehouse logistics based on multi-agents.
Aiming at the characteristics of multi-crane and ground equipment collaborative interaction in the long product warehouse area, dynamic changes in the stack position, frequent material in and out of the warehouse, and complex and changeable warehouse environment, a crane based on multi-car collaborative optimization under complex time-space constraints was established. The cluster dynamic scheduling model, the intelligent decision-making model of material in and out storage stacks based on multi-objective optimization, and the crane path planning model based on reinforcement learning have realized the overall optimization of indicators such as the working efficiency of the crane, the amount of stacking in the storage area, and the pace of logistics.
4) Intelligent centralized control of the reservoir area integrating industrial 5G and digital twin technology.
Aiming at the demand for remote centralized control construction in the long product intelligent storage area, a digital twin and cluster scheduling simulation system based on key elements of storage area operation, logistics design and analysis optimization in the storage area, industrial 5G network for online control and monitoring, and visual AI-based Key technologies such as security monitoring in the storage area have built a new smart centralized control model for the long product storage area featuring unmanned on-site operation in the storage area, efficient and coordinated operation of logistics equipment, remote centralized visualization, and digital monitoring.
4. Application situation and effect
This achievement has been demonstrated and applied in Zhuhai Yueyufeng Iron and Steel Co., Ltd.’s bar intelligent warehouse, Maanshan Iron and Steel Co., Ltd.’s special steel high-wire intelligent warehouse, and Hebei Logistics Group’s bar and wire multi-functional logistics warehouse. Human automatic lifting, and efficient, stable and intelligent collaborative scheduling. In June 2021, the overall online operation was completed, and the 7*24-hour fully unmanned automatic off-line transfer, car shipment, train shipment and other functions in the warehouse area were successfully completed. The entire warehouse area only needs 1~2 people for remote monitoring, which solves the problem of traditional The problem of manual operation in heavy and harsh environments under the reservoir mode has significantly improved the operational efficiency of the reservoir and effectively released production capacity by breaking through the bottleneck of production in the reservoir.