Massively parallelized GA based optimal path planning for single and dual crane lifting in complex industrial environments

Abstract

Heavy lifting is an important task in petrochemical and pharmaceutical plants. It is frequently conducted during the time of plant construction, maintenance shutdown, and new equipment installation. Mobile cranes are lifting machines widely used in a variety of industries. The two primary issues that industries concern are safety and productivity. Accidents may happen in work sites of mobile cranes due to various reasons such as lack of operation knowledge, lack of safety awareness, lack of information about the environment, inadequate guidance, and wrong calculations in lifting. These factors may also influence the productivity by wasting time, energy and resources in unnecessary operations or stoppages. Computer-aided Lift Planning (CALP) for mobile cranes is an effective and efficient tool highly desired by industries.

This research aims to develop a new CALP system for automatic lift planning in complex industrial environments such as petrochemical and pharmaceutical plants, and construction sites. The research focuses on the lifting path planning problems for single and cooperative dual mobile cranes in these complex environments. The lifting path planning takes inputs such as plant environments, mechanical and positioning information of cranes, and start & end lifting configurations to generate optimal lifting paths by evaluating costs and risks involved. In this research, the single-crane and dual-crane lifting path planning are both formulated as multi-objective nonlinear optimization problems with multiple implicit constraints. The objective is to optimize the energy costs, time costs and safety factors of the lifting paths under constraints such as collision avoidance, coordination, and operational limitations. To solve the optimization problems, two master-slave parallel genetic algorithm based path planners are designed and developed on Graphic Processing Units (GPUs) using CUDA programming. The genetic algorithms in the planners are customized for the lifting path planning problems with their efficiency and search abilities improved. In order to handle complex environments, an image-based collision detection algorithm is developed to support the planners. The image-space parallel collision detection algorithm constructs multi-level depth maps for industrial environments and takes advantage of GPU parallel computing. Based on this algorithm, a hybrid C-space collision detection strategy is introduced to trade off the pre-processing and planning time for the planners. To reduce the computation time for continuous collision detection for the lifting target in dual-crane lifting path planning, triangle swept spheres are introduced to model the swept volumes. Finally, a lift planner cum crane simulator system is developed based on the collision detection algorithm and the path planners enhanced by a lexicographical goal programming strategy. This system can serve the purposes of automatic lift planning, interactive lift planning, and training, and thus improve the safety and productivity of lifting operations.