Available online at www.sciencedirect.com
ScienceDirect Procedia CIRP 22 (2014) 23 – 32
3rd International Conference on Through-life Engineering Services
Fast Path Finding System with GPGPU Computing for Replacement Tasks in Plant Maintenance Y. Nonakaa,*, A. Enomotoa, N. Fujiia, J. Kolibabkab, J. Raschb, S. Schulteb, M. Engelhardtb, J. Kanekoc, T. Ichijoc, K. Shibutac a
Hitachi, Ltd., Yokohama Research Laboratory, Yokohama, Japan b Hitachi Power Europe GmbH, Duisburg, Germany c Saitama University, Saitama, Japan
* Corresponding author. Tel.: +81-50-3135-2022; fax: +81-50-3135-3412. E-mail address:
[email protected]
Abstract Plant constructors maintain the performance of plant by health diagnosis and replacements of deteriorated components. The maintenance business becomes one of the important revenue sources for the constructors because profit rate of the maintenance business sometimes becomes much higher than that of new plant construction. To respond this business requirement, a fast automatic path finding system for replacement tasks has been proposed. The system realizes interactive planning operations for a large size of plant 3D-CAD model data. That is, the algorithm finds optimal carry-out/carry-in paths for the replacement tasks automatically, by a collision configuration posture map of the replacement component at each point in a carry-out/carry-in path. Since the creation of those maps take a huge number of calculation steps, it is processed by a parallel computing process on GPGPU hardware. And some experimental results revealed the proposed algorithm is over 200 times faster than a conventional serial computing process. Finally, the proposed algorithm realized the interactive planning operation. Today, this research result has been utilized in multiple power plant sites.
© ©2014 2014 The The Authors. Authors. Published Published by by Elsevier Elsevier B.V. B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/). Peer-review under responsibility of the Programme Chair of EPSRC Centre for Innovative Manufacturing in Through-life Engineering Peer-review Services. under responsibility of the Programme Chair of the 3rd InternationalThrough-life Engineering Conference Keywords: maintenance, replacement, carry-out, carry-in, path, 3D-CAD, GPGPU, octant, voxcel, collision, power plant, nuclear, decomission, radiation
1. Introduction To retain plant performance, plant maintenance such as health diagnosis, as-built construction field survey, and component replacement is required. The maintenance business is becoming one of the important revenue sources for the constructors because its profit rate tends to be much higher than that of new plant construction in some cases. Furthermore, maintenance business is becoming even more important to constructors in the context of the growing number of deteriorated plants in advanced countries. Since the maintenance is executed by subcontractors, its engineering such as planning and preparing instruction documents must be fast and accurate even if a target plant is large-scaled and has a complex structure. Especially, carry-out
and carry-in path planning for replacement components takes a heavy engineering, but conventional paper works with 2 dimensional drawings face limits of its effectiveness in terms of speed and accuracy. To solve this problem, development of digital assistance technologies with 3D-CAD (3 Dimensional Computer-Aided Design) has begun to support the carryout/carry-in planning work. This research focuses on this technology area, and presents a new path finding system with a plant 3D-CAD model. This paper consists of seven sections. Related works are discussed in section 2. The background, requirements, and approaches are defined in section 3. Then the fast path finding system overview is described in section 4, and a component posture adjustment method with parallel computing is proposed in section 5. Business application results are
2212-8271 © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/). Peer-review under responsibility of the Programme Chair of the 3rd InternationalThrough-life Engineering Conference doi:10.1016/j.procir.2014.07.120
24
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
introduced in section 6 followed by a conclusion of the research in section 7. 2. Related works In recent construction project of buildings including plants, BIM (Building Information Modeling) [1]-[3] technology is introduced to visualize and manage the project. This technology provides us a simulation of construction process of supplied components and its visualization with 3D animation. If carry-out/carry-in path is given, BIM system or 3D-CAD system [4]-[9] checks interference between the component and the plant structure along the path. However, there is a problem that this path is not guaranteed to be optimum as it is usually designed manually. Another problem is that manual path designing and collision checking takes long time, for example five hours per path. To reduce workload of path planning, automatic path finding technology needs to be applied in plant maintenance. Automatic path finding has been researched for the last decades. Early researches on path finding were achieved in the field of graph theory. Graph theory has been developed to solve mathematical problem such as Konigsberg bridge problem. Graph theory formulates various problems using graph with nodes and edges, where a node indicates a certain condition of the problem (e.g., position of carry-out/carry-in component in three dimensional spaces) and an edge indicates connectivity between two nodes. One of the most popular problems in graph theory is the shortest path problem. This is a problem to find the shortest path from the start node to the goal node. To solve the shortest path problem, Dijkstra proposed a fundamental method [10]. Many researchers have implemented improvements on this method, mostly focusing on the adjustment of data structure. For example, Fibonacci heap is able to shorten calculation time of Dijkstra method. A* method [11], which is a modification of Dijkstra method, enables more efficient search by using heuristic information to prioritize nodes. These methods are utilized in various commercialized systems such as car navigation systems, robot motion planning systems and so on. The shortest path problem is extended to K-shortest path problem. While the goal of normal shortest path problem is to find a single path with the minimum cost or distance, finding K-paths with minimum cost or distance is required in Kshortest path problem. Finding multiple paths for carryout/carry-in is also required in plant maintenance because there might be an obstacle not included in CAD model. Although reduction of calculation cost for this problem has been proposed by Matsuura [12], there is another subject to apply this algorithm to plant maintenance. The subject is that K-paths with minimum distance or cost often are extremely similar and go along the same space. In such cases, all K-paths turn to be unavailable. Graph theory has successfully provided a lot of practical methods to find paths. To apply graph theory to path finding problem in 3D space, its geometric data needs to be converted
to a graph network in advance. Methods of converting geometric data to graph structure have been researched mainly in the area of robotics. Lozano and Wesley [13] proposed an idea of configuration space to build graph structure from geometric data. A configuration space is a set of free spaces where there is no interference between robot and other structures. To convert configuration space into a graph, its dividing methods are proposed. For example, Brooks [14] utilized cell decomposition approach to divide a plane into non-overlapped free areas. To handle path finding in three dimensional spaces, a data structure called octree was proposed. To build octree, whole geometric area is divided into eight boxes called voxels. If a voxel includes a part of plant structure, the voxel is divided into eight voxels again. Through the conversion from 3D geometric data to a graph network, each voxel is represented by a node and a connection between two neighbored voxels is represented by an edge. After this conversion, graph theory becomes adaptive to automatic path finding in 3 dimensional spaces. One of the technical challenges for the conversion method is to reduce its calculation cost. The calculation cost depends on complexity of geometric structure because a number of interference checks increases. To handle complex plant 3DCAD models, which includes millions of parts, acceleration of the conversion process is one of the critical issues. Automatic path finding in a three dimensional space is called as "piano movers problem" [15]. The problem is solved by graph theory and building methods of configuration space. Path finding in three dimensional configuration spaces requires a heavy calculation even if graph is given. Therefore, most of recent path finding algorithms adopt heuristic approaches [16] or probabilistic approaches to reduce calculation cost [17]-[19]. Subjects of applying “piano movers problem” to the complex plant structure are as follows: i. To reduce calculation cost to build configuration space ii. To find practical path for easy carry-out/carry-in operations iii. To finding multiple paths as candidates in case there is mismatch between 3D model and actual plant due to time off-set between planning and execution. Another technical subject to carry a component in plant is its posture optimization. For example, if overhead cranes are used to carry the component, position and the posture of the component are constrained by the overhead crane kinematics and the operation rule for worker's safety. However, most of studies about path finding with overhead crane do not consider the posture because the suspended components are assumed as a ball [20]. Some studies propose utilization of multi-body dynamics [21], but the position and the posture were not computed against collisions to the adjacent plant structure. Regarding planning trajectories of the position and the posture, multi-body dynamics has been employed in various areas. For instance, car dynamics models are installed in an urban traffic model in intersections to reproduce realistic traffic simulation [22]. In robotic area including AGV
25
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
(Automated Guided Vehicle), most of researches adopt path finding methods in configuration [23][24]. This research focuses on these technology areas mentioned above, and proposes new methods to realize an automatic fast path finding system for plant maintenance. 3. Background, requirements, and approaches 3.1. Background
14 Billion toe, Primary Energy Africa 12 Middle East East Europe 10 Russia Asia/Oceania Latin America 8 6 Europe
4 2
North America
0
’70
’80
’90
’00
’10
Fig. 1. World energy consumption trend.
Fig. 1 shows a world energy consumption trend [25]. In the figure, toe means tone of oil equivalent. From past several decades, world energy consumption grows continuously, especially Asia, Middle East and East Europe consumptions expand typically. That is, power plant maintenance businesses spread globally, so systematic maintenance approaches are requested. Additionally, since power plant lifecycle is about 40 years in general, replacement demand grows in developed countries. Marketing & Sales
Front-end Engineering
EPC Execution
Service & Maintenance
Accurate Engineering and Rapid Response to Customer
Constructability and Maintainability Engineering for On-time On-budget Fig. 2. Key challenges in whole plant lifecycle.
For plant maintenance subject definition, key challenges of a whole plant lifecycle should be discussed. Fig. 2 indicates key challenges in whole plant lifecycle. In this figure, a plant lifecycle is divided into four phases; Marketing & Sales, Front-end Engineering, EPC (Engineering, Procurement and Construction) Execution, and Service & Maintenance. In the Marketing & Sales phase, stakeholders establish schemes, and feasibilities are studied with some assumptions of engineering. In the Front-end Engineering, customer specs
and constraints are configured into plant layout design and plant component specs. For these two phases, accurate engineering and rapid responses to customer are key challenges. In the EPC Execution phase, the detail design and the construction schedule are executed, and the interface management is implemented. The interface management is to make agreements with related stakeholders about roles, responsibilities, timelines, deliverables, and so on, for critical interface identification in early EPC phase [26]. In the Service & Maintenance phase, health diagnosis, as-built construction field survey, component replacement, and some are executed. For these two phases, constructability and maintainability are key challenges to the engineering. In the components replacement work this research focuses, a short time replacement is requested to achieve the plant high operation rate. But the replacement takes a heavy engineering workload. That is, due to the plant structure complexity and system size, it is very difficult to find an appropriate carryout/carry-in path without collisions between the component and adjacent plant structure. Therefore, an early phase path planning with plant drawings has been desired. To respond this aim, 3D-CAD digital assistance technologies have begun to support the planning. Our research focuses on this technology area. 3.2. Requirements Fig. 3 indicates the carry-out/carry-in planning work as our research target. Left hand shows a conventional work, and right hand shows the target situation. In the conventional, carry-out/carry-in paths are planned by human examination with 3D-CAD visual check. Normally, one path planning in 3D-CAD takes one day, and trial and error work in the real construction scene takes several days. Conventional
Targget Target Automatic Installation Path Planning
Visual Check ONLY Path
Human Examination tio on on
Interactive Review
•Heuristic-based plan, no optimality y •Long planning timee Subcontractor Fig. 3. Research target.
Here, the update subject of the plant 3D-CAD model should be discussed respond to the real plant condition changes, but this would be treated in another report. In case of this research sample plant, a special engineer keeps the 3DCAD model to the real plant condition. Fig. 4 shows a sample
26
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
plant 3D-CAD model of this research. This model consists of around one hundred thousands of steel structure parts, pipe system parts, and other parts.
From those candidate paths a supervisor at the plant site decides the optimal carry-out/carry-in path depending on the maintenance situation. Therefore, the automatic planning system must calculate those paths in a very short time to establish an interactive operation between the system and the supervisor. After all, to achieve the target situation shown in the right hand side of Fig. 3, requirements of the carry-out/carry-in planning with the plant 3D-CAD model are following items. (1) Automatic plural variant paths finding with a few turning points (2) Automatic posture adjustment to avoid collisions (3) High speed calculation for an interactive path planning 3.3. Approaches
Fig. 4. Sample plant 3D-CAD model.
Also, an optimal path for a carry-out/carry-in work is rapidly changed respond to the plant maintenance work condition such as type of available construction machines, number of available workers, maintenance progress, and so on. That is, in real situation, there are so many uncertainties between a maintenance work plan and the real situation, such as worker’s availability uncertainties, hoist’s availability uncertainties, situation uncertainties of placed materials, and so on. Herewith, in the maintenance work planning phase, no one can quantify those uncertainties, and no one can make the best carry-out/carry-in path. Therefore, according to those strong requests of real business, the system must produce plural variant paths as candidates for multiple situations. Also, in the most plant structures, there are some large space areas called ‘erection shafts’ for lifting up or down components, so those plural variant paths will take different election shafts. And in the detail motion of a carry-out/carry-in work, components are suspended by overhead cranes because most components weigh more than a ton. Workers change directions of the suspended components at each turning point on the path without collisions to the adjacent plant structures. And fewer turning points are preferable from the view point of cost of the crane operations because the rotations of a suspended component require a long operational time and many travel rails for overhead cranes. That is, those plural variant paths should have a few turning points with large workspaces to rotate a component without collisions.
To solve those requirements above, this research takes following approaches; a. Octant modeling approach for the free space recognition in a plant b. Graph network analysis approach for finding plural variant paths with a few turning points c. Parallel computing modeling approach with GPGPU (General-Purpose computing on Graphics Processing Units) for the high speed calculation Regards to ‘a’ and ‘c’ above, another paper reported detail methods and application results [27]. This paper reports the overview of this system and focuses on the item ‘c’. 4. System overview 4.1. Hardware configuration Computer A for Designers
Plant 3D-CAD Computer C for Workers 3D-Viewer Work Instruction Printer/Viewer
Project Data Base Plant X Project Folder 3D-CAD Data
Computer B for Planners with GPGPU Board
STL Data Movie Data
Path Planning System
Work Instruction Files
Work Instructions
Fig. 5. Hardware configuration.
Fig. 5 shows the hardware configuration of the system. It consists of Computer A for designers, Computer B for planners with GPGPU board, Computer C for workers, and Project data base. Compute A presents a plant 3D-CAD system and produces the plant layout design, structure design, P&ID (Piping and Instrumentation Diagram), components spec list, and so on. Those data are saved as the original file format and STL (STereo Lithography) data file format for the
27
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
path planning system. Because there are many plant 3D-CAD systems, the input data file format of the path planning system is STL produced from many CAD systems as a standard. Computer B owes the path planning system, and produces some movie data and work instruction files as the results of path planning. Computer C presents a 3D-viewer and a work instruction printer/viewer for workers in the plant. Project data base archives 3D-CAD data, STL data, movie data, and work instruction files, and presents those data by way of a network. Fig. 6 shows the parallel computing architecture with GPGPU to establish an interactive operation between the system and an operator. Plural variant paths finding and a component posture adjustment take a huge amount of calculation steps, so a calculation model design for GPGPU application is a technical challenge. In the GPGPU chip set, there are hundreds of process units. And once a CPU requests a process to GPGPU for multiple computation objects, the process is mapped to multiple process units in GPGPU in parallel. And results from those process units are returned a memory area. Therefore, if many similar processes are requested in the same time such as an image processing, an engineering mechanics analysis with the finite element method and so on, GPGPU performs a high calculation speed. But, if the algorithm is not suitable for parallel computing such as a monolithic calculation steps, GPGPU can not perform efficiently. CPU Chip Set
Process Request
GPGPU Chip Set
Process Parallel Mapping
which is a scalar parameter as a distance from the centre of the voxcel to the nearest point of the adjacent plant structure. START A. Free Work Space Blocks Recognition by an Octant Modeling Approach B. Start Point and End Point Setting as Input Data C. Plural Variant Paths Finding by a Graph Network Analysis D. Component Suspended Posture Adjustment by a Kinetic modeling Approach E. Path Performance Result Sheet Report for an Optimized Path Selection END Fig. 7. Overall flow chart
Then, free work space blocks are extracted from those voxcels, which contains no potions of the plant structure internally. After that the connectivity between those voxcels are analyzed, and a network model is created as the result. In the network model, each node has some attributes such as the reference ID correspond to the voxcel, the size of the voxcel, the space margin of the voxcel, and so on. Fig. 8 is an image of extracted free space blocks from the plant 3D-CAD model, and the network model. Octant Voxcels
Extracted Free Work Space Block
Memory Process Result
Process Unit
Fig. 6. GPGPU architecture.
4.2. Algorithm overview Fig. 7 shows the overall flow chart of the system. This system treats STL files as the input data of a plant 3D-CAD. Firstly the system recognizes free work space blocks in the plant structure by an octant modeling approach (Step A). In this step, the plant structure is divided into cubic areas named voxcels, which sizes are categorized into some sizes by the octant process. The smallest size of the voxcel is defined as a setting parameter, which is small enough to review the carryout/carry-in component size for the planning accuracy, and which is large enough to finish the calculation in a practical time. Also, each voxcel gets an attribute of space margin,
Network Model Fig. 8. Free work space block recognition.
The next step of the overall flow chart is a start point and an end point setting for carry-out/carry-in work in the plant structure (Step B, Fig. 7). As Fig. 9, the start point and the end point are addressed to two nodes of the network model. Then, with a sphere model which diameter is the same as the smallest representative dimension of the carry-out/carry-in
28
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
component, plural variant paths such as Route 1, Route 2 and Route 3 in Fig. 9, are found by a graph network analysis (Step C, Fig. 7). Since each node has some attributes, each path has different performance in some measures like the number of nodes, the average collision margin, the total distance, and so on. Path2
Path1
5. Parallel computing for component posture adjustment 5.1. Subject definition In this paper, a carry-in work of an L-shape pipe in a plant structure by an overhead crane is sampled. In Fig. 11, the situation is represented, the blue one is the pipe, two yellow lines are the suspension chain, and red line is the carry-in path.
Goal
Start
Suspension Chain
Path3 Collision Check & Posture Adjustment
Carry-in Pipe
Carry-in Path Fig. 9. Plural variant paths in network model.
By the previous steps, the carry-out/carry-in component is assumed as the sphere, collisions between the 3D-CAD model of carry-out/carry-in component and those of adjacent plant structures are checked on all the way of each path. And the suspended posture of the carry-out/carry-in component is adjusted by a kinetic modeling without collisions to the adjacent plant structures (Step D, Fig. 7). Finally, the performance result of all paths is reported, and the operator selects an optimized path depending on the situation of the replacement condition (Step E, Fig. 7). Fig. 10 indicates an example of the performance result sheet, and Route 1 is selected in the situation.
Perspective Num. of Turning Point
1 3
Path 2 6
3 8
Ave. Collision Margin
2.2
1.8
2.6
Distance
162.0 161.6 168.5
Fig. 10. Path performance result.
Fig. 11. Sample condition of single suspended carry-in pipe.
Fig. 12 shows a ground plan of the sample carry-in pipe. In this figure, the carry-in work is planned for the suspended pipe between point P50 and p60. And the start point is P10, and the end point P60. Workers change the direction of the pipe at each turning point such as P20, P30, P40, and P50 without collisions to the adjacent plant structures. That is, the straight motions such as P10-P20, P20-P30, P30-P40, P40-P50, and P50-P60 keep the same suspended posture. P10
P60 P50
P40 P30
In Fig. 7, Step A is only executed in the first application and in the update timings of the plant 3D-CAD model, and other steps are executed every time when the carry-out/carryin paths are planned. The parallel computing with GPGPU has applied to Step A and Step D, and a severe high speed calculation is requested to Step D in the viewpoint of the interactive path planning. Therefore this paper reports the parallel computing approach for the latter.
P20
Fig. 12. Ground plan of a sample carry-in pipe.
Therefore, the system must decide optimized suspended postures for the start point and those turning points for each on the way to the next point. Fig. 13 shows a single view drawing of the sample carry-in pipe situation from the turning point P50 to the end point P60 of Fig. 12. The blue one represents the suspended carry-in pipe, and the pipe is passing through the portal shape plant structure. In this situation, the collision check between the pipe and the plant structure must be
29
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
checked at discretization points between P50 and P60, such as P51, P52, P53, and P54. With the restriction of the overhead crane, the pipe posture is defined with two attitude angles ( , ) of the relative coordinate system on the carry-in path, which origin is placed at the gravity centre of the carry-in pipe. It means, the number p of collision calculation is exploded, which is the number of discretization combinations ( , ) at a discretization point multiplied by the number of discretization points between the start point and the end point.
P54 P53 α P52 P50
P51 P46 P45
Fig. 13. Sample condition of carry-in pipe posture adjustment.
When a real carry-in pipe situation is assumed, the total distance between the start point and the end point is about 100m, the granularity of the path is 10cm, and the granularity of the attitude angle is 5 degree. This means over 5 Millions of collision calculations are requested to find the carry-in path with the optimized suspended postures. Therefore, to establish an interactive path planning, high speed calculation is the subject.
5.2. Approach To find the optimized suspended postures rapidly, this research utilizes the overhead crane restriction. That is, only turning points change the posture to avoid collisions to the next point, and straight motions keep the posture to the next point. Therefore, the straight motion and the rotational motion are optimized independently. According to this finding, this research proposes a collision configuration map for each discretization point, as shown in Fig. 14. This map consists of the collision check results of all discretization combinations ( , ) at a discretization point on the path, and each collision check procedure is mapped to a process core module of the GPGPU independently. In Fig. 14, a scatter diagram of collision conditions is created by the GPGPU parallel computing. The white circles indicate postures without collisions, and the black circles indicate postures with collision. By this map, the solution space for the optimized suspended posture is provided. With this parallel computing for collision checks, the optimized suspended postures are found rapidly. Fig. 15 shows the overall algorithm of the optimized suspended postures finding. For each candidate of carryout/carry-in path, the start point of the path and the first discretization point are selected as the initialization (Step a and b). With the condition, the collision configuration map is created by the GPGPU parallel computing (Step c). And for all discretization points of the path, those collision configuration maps are created (Step d, e, f, h, and i).
START a. Select start point of the path b. Select first discretization point between the point and the next point
Ǫ
Posture with Collision
c. Create collision configuration map d. Select next discretization point
Posture without Collision
h. Select next discretization point e. Next turning point or next turning point or end point ? no or end point yes f. End point ? yes
ǩ
no
i. Select next turning point or end point
g. Select optimized posture for start point and each turning point END
Fig. 15. Algorithm of posture optimization.
GPGPU
Process Core Module
Fig. 14. Collision configuration map.
In Fig. 16, a group of posture candidates on the straight motion P50-P60 of Fig. 13 is extracted by the overlay of all collision configuration maps (CCMP50, CCMP51, CCMP52, CCMP53, and CCMP54). This overlay process is executed for
30
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
each straight motion such as P10-P20, P20-P30, P30-P40, P40-P50, and P50-P60 shown in Fig. 12, and a group of posture candidates is extracted for each straight motion. From the series of groups, the optimized suspended posture is selected for each straight motion with the posture connectivity from the start point P10 to the end point P60.
CCMP54 β CCMP53 β CCMP52 β CCMP51 β CCMP50 β
that of the developed method with GPGPU. Hardware specifications of CPU, RAM and GPU in this experiment are Intel Core i7-3930k 3.2GHz, 32.0GB RAM and NVIDIA Tesla C2075 respectively. At each sampling point on a path, 9216 postures are generated by the combination of three rotational angles as the candidates of the suspended posture. While single CPU calculates whether there is a collision at each candidate in serial, GPGPU calculates it in parallel. This result shows that the developed method is approximately two hundred times faster than the conventional method. 6. Business application results
α
Optimized posture candidates from CCMP50 to CCMP54 without collision
α
α
6.1. Application to component replacement service The developed technology has been applied in a maintenance service of a power plant.
α
α
Fig. 16. Posture adjustment approach.
5.3. Application result
Calculation Time [sec]
104 Single CPU Process
103
Fig. 18. Carry-in work animation for pipe replacement service.
102
101
1 109
Developed Method
1010
1011
Target Plant Size (Number of Polygons) Fig. 17. Calculation time comparison.
Fig. 17 shows the calculation time comparison for finding an optimum suspended posture. In this experiment, calculation time for collision check was evaluated with twelve paths in three plant models. The horizontal axis indicates target plant size which is represented by the number of polygons involved in collision check, and the vertical axis is calculation time. Blue dots in the graph represent calculation time with a conventional method with single CPU, and red dots represent
Fig. 19. Real pipe replacement service by research result.
Fig. 18 shows a carry-in work animation for the pipe replacement service by this research result. The blue pipe is
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32
the new pipe, and the vertical blue line is the calculated path. After an optimal path selection, the system produces a 3Danimation of carry-in pipe with the coordinates for the start point, turning points, and the end point. In this animation, the dynamics of an overhead crane suspension dynamics is represented for each time-frame, and the linkage from a straight motion to the next straight motion is also represented. Also, the system produces some instruction sheets in which the optimal path situation is described on some maps of the plant. Fig. 19 shows the real pipe replacement service by the research result. The red pipe is the new pipe, and it is suspended by a single wired crane. With this technology, installation frequency reduction by optimal large pack installation was established, and that leads to the maintenance cost reduction.
The developed technology has been under application to a nuclear power plant decommission process. Fig. 20 shows a trial result of the automatic carry-out path creation minimizing radiation exposure. In the grey construction, the color sphere indicates a radiation map which was created as an imaginary, and blue polylines and a red polyline are carry-out paths minimizing radiation exposure. This situation was established by adding the radiation information to each free work space block shown in Fig. 8 as an attribute of the block. Fig. 21 shows a trial result of the automatic carry-out path creation with two point suspension. In the grey construction, the blue component is the carry-out one, and blue polylines and red polylines are carry-out paths avoiding collisions. This figure indicates the situation from a horizontal motion to a vertical motion with overhead cranes with two point suspension.
6.2. Ongoing application to nuclear plant decommission
7. Conclusion This paper reported a fast path finding system with GPGPU computing for replacement tasks in plant maintenance. To find an optimal path, a new automatic finding method for plural variant paths and a new automatic posture adjustment method for collision avoidances were proposed. Also, to establish an interactive path planning, a high speed calculation method with GPGPU was proposed. And proposed technologies were verified and perform over 200 times faster than a conventional technology in a carryout/carry-in path finding. And their effectiveness was also confirmed in some real plant maintenance works. Today, this research result has been utilized in multiple power plant sites.
Radiation Map
Start
References Goal
Carry-out Path Minimizing Radiation Exposure
Fig. 20. Trial result of carry-out path finding to radiation.
Carry-out Component with Two Point Suspension
Fig. 21. Trial result of carry-out path with two point suspension.
[1]http://usa.autodesk.com/building-information-modeling [2]http://www.bentley.com/en-GB/Solutions/Buildings/About+BIM.htm [3]Continuously Growing Nuclear Business, 2010, Toshiba Review, 65/12, http://www.toshiba.co.jp/tech/review/2010/12/index.htm. [4]http://www.3ds.com/, Dassault systems [5]http://www.informatix.co.jp/top/profile_e.html, informatrix [6]http://www.autodesk.com/suites/plant-design-suite/overview, Autodesk [7]http://www.aveva.com/en/Products_and_Services/AVEVA_for_Plant/AV EVA_for_EPCs.aspx, Aveva [8]http://www.bentley.com/en-US/Products/microstation+product+line, Bentley [9]http://www.intergraph.com/ppm/3dmv.aspx, Intergraph [10]Dijkstra, E.W. (1959). A note on two problems in connexion with graphs. In Numerische Mathematik, 1 (1959), S. 269 ̚ 271. [11]Hart, P. E.; Nilsson, N. J.; Raphael, B. (1968). "A Formal Basis for the Heuristic Determination of Minimum Cost Paths". IEEE Transactions on Systems Science and Cybernetics SSC4 (2): pp. 100–107. PDF [12]Hiroshi Matsuura, A proposal of k shortest simple path algorithm to minimize computational complexity, IEICE Tech. Rep., vol. 110, no. 341, IN2010-107, pp. 57-62, Dec. 2010. [13]Tomás Lozano-Pérez, Michael A. Wesley, An Algorithm for Planning Collision-Free Paths Among Polyhedral Obstacles, Communications of the ACM, Volume 22 Issue 10, Oct. 1979, Pages 560-570 [14]RODNEY A. BROOKS, 1983, Solving the Find-Path Problem by Good Representation of Free Space, IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS, VOL. SMC-13, NO. 3, MARCH/APRIL
31
32
Y. Nonaka et al. / Procedia CIRP 22 (2014) 23 – 32 [15]Jacob T Schwartz, 1983, On the “piano movers” problem. II. General techniques for computing topological properties of real algebraic manifolds, Advances in Applied Mathematics, 4/3: 298–351 [16]Ratliff, N., Bagnell, J. A., 2006, Zinkevich, M. A., Maximum Margin Planning, International Conference on Machine Learning. [17]Svestka, P., Latombe, J. C., Overmars Kavraki, L.E., 1996, Probabilistic roadmaps for path planning in high-dimensional configuration spaces, IEEE Transactions on Robotics and Automation, 12/4:566-580. [18]LaValle, S. M., and Kuffner, J. J., 1999, Randomized kinodynamic planning, In Proceedings IEEE International Conference on Robotics and Automation, 473-479. [19]Diankov,R., and Kuffner., J., 2007, Randomized Statistical Path Planning, In Proc. of the 2007 IEEE/RSJ International Conference on Intelligent Robots and Systems, TuA1.1. [20] Nagai, S., Kanesige, A., Ueki, S., 2011, Three-Dimensional Obstacle Avoidance Online Path-Planning Method for Autonomous Mobile Overhead Crane, In Proc. of 2011 International Conference on Mechatronics and Automation, 7-10. [21] Gross, M., James, D., 2008, Elevation Cable Modelling for Interactive Simulation of Cranes,. European SIGGRAPH Symposium on Computer Animation,
[22] Felez, J., Maroto, J., Cabanellas, M. J., Mera, M., J., 2013, Simulation: Trans. of the Society for Modeling and Simulation International, 89(9), 1099-1114. [23] Diankov, R., Kuffner, J., 2007, Proc. of the 2007 iEEE/RSJ Int. COnf on Intelligent Robots and Systems, Snn Diego, CA, USA, Oct 29-Nov 2, TuA1.1. 1-6. [24] Fragkopoulos, C., Graer, A., 2011, Dynamic efficent collision checking method of robot amr paths in configration space, 2001 iEE/ASME Int. Conf. on Advanced Intelligent Mechatronics, Budapest, Hungary, July 37, 784-789. [25] METI, 2013, Energy White Paper, 2013, http://www.enecho.meti.go.jp/topics/hakusho/2013energyhtml [26] Caglar, J., et al., 2007, Interface Management, Effective information exchange through improved communication, ABB Value Paper Series, ABB Inc.,5. [27] Enomoto, A., et al., 2014, Multiple path finding system for replacement tasks, 9th CIRP Conference on Intelligent Computation in Manufacturing Engineering 2014, Capri (Naples), Italy.
