This paper is accepted for Pacific Conference on Manufacturing '96
October 29-31, 1996, Seoul, Korea.
DESIGN EXPERT SYSTEM FOR AUTO-ROUTING OF SHIP PIPES
Sang-Seob Kang
CAD/CAM Development Team
Hyundai Heavy Industry
Ulsan, Korea
Se-Hyun Myung
Department of Automation and Design Engineering
Korea Advanced Institute of Science and Technology
Seoul, Korea
Soon-Hung Han
Department of Mechanical Engineering
Korea Advanced Institute of Science and Technology
Daejeon, Korea
ABSTRACT
Finding the optimum route of ship pipes is a complicated and time-consuming process. Experience of designers is the main tool in this process. To reduce design man-hours and human errors a design expert system shell and a geometric modeling kernel is integrated to automate the design process. A framework of the intelligent CAD system for pipe auto-routing is suggested. The CADDS 5 of Computervision is used as the overall CAD environment, the Nexpert Object of Neuron Data is used as the expert system shell, and the CADDS 5 ISSM is used to build user interface through which geometric models of pipes are created and modified. Existing algorithms for routing problems have been analyzed. Most of them are to solve 2-D circuit routing problems. Design of ship piping system, especially within the engine room, is a complicated, large scale 3-D routing problem. Methods of expert systems have been implemented to find routes of ship pipes on the main deck of a bulk carrier.
Key Words : Pipe auto-routing , Expert system , Geometric modeling , Intelligent CAD, Ship design
1. INTRODUCTION
1.1 Motivation of Research
The ship piping system is as important a system in a ship as a blood vessel system is in a human body. Its required performance should be optimized based on engineering, spatial, regulatory, and economic analyses. Piping systems can also be found in the AEC (architecture, engineering, and construction) area, where technology-oriented man-power plays an important role. Previous development efforts for automating the design process are focused on building design-oriented databases. If heuristic design knowledge can also be systematically applied, a good portion of the human decision making process can be replaced by computing power.
Fig. 1 shows four design steps of the ship piping system: basic design, functional design, detail design, and production design. First, the detail design step involves the completion of most of the design information. Arrangement plans for pipes and equipments are produced. Geometric and non-geometric attributes are determined. Second, production information required at the production design step can be extracted from the detail design output. Detail design process requires a high level of experience and knowledge in addition to time consuming iterations of calculations.
Because the piping design process requires a variety of knowledge, the expert system can further reduce design man-hours, shorten design period, and standardize design plans. This paper explains an expert system for piping design, which optimizes paths of ship pipes while placing them away from on-board equipment which might interfere. Also the system places piping parts along the pipe paths. The knowledge-base is constructed based on printed design knowledge and the empirical knowledge of human experts. The system is modeled with the following objectives, (1) to minimize user input and user decision, (2) to systemize the knowledge-base for easy addition of further knowledge, (3) to make the system easy to use, (4) to be used in real shipyard design processes.
Fig. 1 PROCESS OF PIPING SYSTEM DESIGN
Fig. 2 CONFIGURATION OF AUTO-ROUTING SYSTEM
The scope of work is confined to the implementation of a prototype system to design an upper deck piping system of bulk
carriers. The design result from the inference process can be visualized using display functions of a commercial CAD system.
Together with this interface to CAD system, the system acts as an intelligent CAD system to automate routing of ship pipes.
1.2 Related Works
Previous works can be grouped into three categories: (1) auto-routing of semiconductor and printed circuit board, (2) auto-wiring within a car or a space craft, (3) auto-routing of pipes within a building or a ship. The third category is different from the other two in that one pipe way has many pipe lines with different sectional dimensions and shapes. These pipe lines should be bent in 3D space and should be secured with proper supports. Little previous work could be identified in this category.
For auto-routing in micro-electronics and auto-wiring of automotives, lines of uniform sectional shapes should be arranged while minimizing the route lengths and satisfying constraints. Lee Maze algorithm defines a path as an ordered list of planar elements with the same sizes. It minimizes an objective function by dividing the work into global routing and detail routing (Rubin, 1974; Brown, 1990). Geometric modeling capability is required to arrange 3D cable ways. An expert system shell with native geometric modeling capability, I-CAD, has been used for auto-wiring within automotives (Lee, 1994).
Auto-routing within a hydraulic manifold block does not have any obstacles except the paths themselves (Chambon et al., 1991). The problem is similar to 3D wiring and requires little domain knowledge compared to ship pipes. Auto-routing within an industrial plant uses Lee Maze algorithm to find the optimum path (Mitsuta et al., 1986). As Lee Maze algorithm is meant for semiconductor design, it is expanded to support 3D pipe lines. The knowledge reduction process speeds up the reasoning process, and the knowledge-base is made of rules. The routing space is subdivided based on obstacles for efficient path finding.
Pipes within a ship or a building have different sectional sizes and shapes along a pipe route. 2D path finding algorithms
should be expanded to 3D space algorithms. For instance, the Sprout method (or Vector method) uses the depth-first
search in finding pipe routes (Van Der Tak et al., 1976). This goal-oriented approach requires too many user inputs. For a
second example Dijkstra's algorithm finds the optimum path for a weighted graph G = (V,E,W), where V is a set of nodes, E
is a set of arcs, and W is a set of weighting factors. The MATES system uses the Dijkstra's algorithm, and the user should
supply predefined candidate nodes and arcs (Funaoka et al., 1985). It requires a lot of user inputs, and the user should have
enough experience in piping design. MATES is not a stand-alone system. It is a multi-functional system with an integrated
database covering design to production. HICAS is another example of shipbuilding software system which has piping
subsystem called HICAS-P and HICADEC-P (Gotoh et al., 1982; Inoue et al., 1988).
2. PIPING DESIGN OF A SHIP
2.1 The Problem Characteristics
The auto-routing system automatically determines the topology and geometry of the ship piping system, while maximizing the performance of equipment constituting the piping system, and at the same time meeting spatial, legal, and cost constraints. Piping design is a creative process and does not have unique solution space. Its characteristics can be summarized as follows:
(1) Rule of thumb : It requires practical knowledge of experienced designers, in addition to regulations and design codes recorded in documents.
(2) It should be designed to have good (aesthetic) appearance, for example grouping pipes into bundles, in addition to the constraints on performance, space, or legislation. The resulting design may be against a quantitative objective function.
(3) Assessment of design results is a subjective matter due to heterogeneous design preferences and conventions of designers.
(4) Exact solution cannot be found, but feasible solutions or optimal solutions are sought.
2.2 Piping Design Process
Design is a progressive process getting more detail information on the final product using the 'principle of iteration' and the 'principle of the least commitment' (Suh, 1990). The followings show abstraction levels on design specification and modeling of piping design (Martin, 1995). The subject of this paper falls under the detail design step.
(1) Basic design: Performance of the piping system and constituting equipment are determined according to requirement specifications that come from the ship owner and ship classification society. Overall topological information and principal parts are also determined.
(2) Functional design : Starting from performance data and overall topology settled in the basic design, performance of the system is re-examined, and topologies and attributes of all components are determined.
In this step cost factors are considered and design is adjusted to more refined ship structure.
(3) Detail design : Physical (graphic & non-graphic) information of all the piping system components is determined, and basic information related to manufacturing and shop arrangement to be used in the production step is determined.
(4) Production design : All the information needed in the manufacturing stage is derived from the drawings from detail design
step.
2.3 Modeling Process in Detail Piping Design
Principal steps of detail piping modeling process can be summarized as follows:
(1) Preparation : The project library of the ship to be designed is composed from standard parts libraries. For the arrangement of piping elements, back drawings are prepared from structural drawings of the ship hull CAD system. Specifications of piping elements are prepared from the diagram of piping system.
(2) Schematic piping & unit piping : Planning and modeling are performed using prepared data and applying application routines. A compromise between structural elements and pipe fittings is found while focusing on grouping pipes and minimizing the number of bends. At the same time, standardized piping units and project-specific piping units are designed.
(3) General piping : Drawings prepared separately are joined into a module, and fine detail design is performed by adding manufacturing information to be required in the production design step.
(4) Dimensioning & labeling : Drawing is completed by inputting dimensions, BOM (bill of material), and formatting
information.
3. AN EXPERT SYSTEM FOR PIPING DESIGN OF A SHIP
3.1 Representation of Knowledge in Knowledge-based Design Expert System
The expert system mimics human experts by representing and processing domain knowledge. As an application-oriented knowledge engineering of artificial intelligence (AI), it stores knowledge in a computer and solves problems through a reasoning process. Design knowledge processing falls under the same technology category. Structure of the pipe design expert system is shown in Fig. 2.
To handle a design problem effectively, objects manipulated in the design stage should be classified. Solutions can be found using relationships and associations of these objects. Objects manipulated in the pipe routing process are pipe-path, pipe-element, space-element, and some constraints. Top level objects used in the piping process and their relationship is shown in Fig. 3. This object diagram is prepared according to the OMT method (Rumbaugh et al., 1991). The objects are derived by analyzing the routing process. The types of relationships among objects are hierarchical relationship 'a_kind_of', inheritance relationship 'is_a', assembly relationship 'consists_of', connection relationship 'connected_with', and reference relationship 'reference_to'.
Space elements represent spaces and obstacles where pipe elements should be placed. This information comes from the structural design CAD system. Data of space element should be extracted from a CAD model
Fig. 3 OBJECT DIAGRAM OF PIPE-PATH AND PIPE-ELEMENT BY OMT NOTATION
Fig.4 OBJECT DIAGRAM OF SPACE ELEMENT BY OMT METHOD
through an interface module or a data bridge between the expert system and the DBMS system. Similar to pipe elements, space elements are modeled as show in Fig. 4 (Kim, 1994; Lee et al., 1994). Constraints reduce the solution space of pipe routing problem. Constraints are implemented as frame, rule, or methods within objects.
Modeling of knowledge objects with the knowledge-base is performed as follows: defining knowledge objects, finding their relationships, adding attributes and methods to objects, and then iterating the steps. These knowledge objects are then classified hierarchically to model the real-life objects and heuristic knowledge. Design objects considered in this stage are primary systems such as a compressed air system, a hydraulic oil system, a hold cleaning & bilge discharge system, a fire & wash deck system, and an electric cable pipe system. Knowledge contained in regulations of classification societies and port authorities, design practices, and experience should be applied. Following are some examples of such knowledge:
- Pipes should be arranged to allow easy installation and maintenance of equipment.
- Pipe lines should be straight and grouped into bundles as much as possible.
- For curved parts, an elbow is not recommended, 45 or 90 degree angle bend is prefered.
- A minimum work space should be reserved for easy installation and maintenance of pipe parts.
- Pipe ways should not disturb other shipboard traffic.
3.2 Abstraction Hierarchy of Design Knowledge
Pipe paths and pipe elements are implemented as objects, constraints implemented as rules, and algorithms implemented as sub-routines. They should be abstracted into levels in order to make the modification and addition of knowledge easy. Design knowledge is modularized and hierarchically structured. Only the active design knowledge-base is loaded and it is unloaded when the knowledge-base is not needed anymore. The knowledge-base in the system consists of three parts.
(1) The meta-control knowledge makes main decisions and controls activated design knowledge pieces.
(2) The global designer (typical section design) finds the optimal arrangement of main pipes over a 2-dimensional sectional plan.
(3) The detail designer expands the 2D sectional plan into 3-dimensional space along the ship length.
The process of inferencing using the knowledge-base can be summarized as follows:
(1) The design environment is set up by meta-control knowledge, (2) A typical section designer makes iterations to find
optimal arrangement of main pipes which have high priority of placement. This typical 2-D section designer is constrained by
structures like hatch covers, main equipment and outfittings. (3) 3-D pipe routes are designed using knowledge about
secondary structures and outfittings in priority sequence by a typical sectional designer. The objective function is the length of
pipe route and number of bends. (4) Design result is visualized by a CAD system. Fig. 5 shows this process.
Fig. 5 ROUTING PROCESS OF MAIN PIPES
Fig. 6 DESIGN RESULTS FROM THE PIPING EXPERT SYSTEM
Pipe design results which are designed with the above knowledge are then visualized through the interface module between
the expert system and the CAD system. The user can evaluate visualized results. Unacceptable design can be modified by
adjusting design factors in the knowledge-base. Three different knowledge-bases constructed in this research store 167 rules
and 106 supporting methods.
4. EXPERIMENTS WITH PIPING DESIGN EXPERT SYSTEM
4.1 System Configuration
Structure of the design expert system is shown in Fig. 2. It is composed of a knowledge-base and a CAD interface. The knowledge-base is made of regulations, design practices, and empirical laws. The CAD interface allows visualization of design results. The interface module between the expert system shell and the general-purpose CAD system is built by calling API (application programming interface) routines of both systems. The pipe design system is programmed in C language which is provided by a SUN SPARC workstation with Solaris 2.3 operating system. Nexpert Object 3.1 is used as the expert system development shell, and CADDS 5 for the CAD environment.
The designed knowledge-base is a prototype system because it does not cover the full spectrum of design knowledge. To
verify the designed knowledge-base the piping design expert system has been tested against an existing ship. The tested ship
is a bulk carrier of 160,000 tons with LBP of 281m, breadth of 46m, and draft of 16m.
4.2 Analysis of Design Results
After the interface module visualizes the information from the inference engine shown in Fig. 6, design results have been compared with real drawing of an existing ships. The number of bends is the same and the pipe lengths are not very different. These are the expected results because the piping systems under consideration are the main lines of the main piping systems which have high priorities. If auxiliary piping systems and branch lines which have low priorities had been designed, the results would be somewhat different.
Although the positions of the main lines are somewhat different, it is hard to grade two designs since the solution of the piping
design is not unique. It is evaluated that both design results, existing ship and expert system result, are reasonable. If the
objective function is expanded to include maintainability and available space around pipes in addition to pipe length and
number of bends, it will be a problem of multi-objective function, and the space of optimal solution becomes smaller.
5. CONCLUDING REMARKS
A prototype expert system for the piping design of ships has been developed. Emphasis is put on the modeling of the design knowledge-base to extract knowledge from documented design regulations, design know-how of human experts, and design histories of a shipyard. To verify the developed knowledge-base model, an interface module between the expert system and a commercial CAD system has been developed. The final system minimizes user decision making, systematizes the knowledge-base, ease expanding the design knowledge-base. After the verification process, the following observations were made.
(1) Previous CAD systems for piping design were useful to automate calculating equations and drafting drawings. The pipe design expert system proposed in this paper introduces expert system methods to utilize empirical knowledge of the piping design domain, and then to help design the process.
REFERENCES
Brown, A.D., Zwolinski, M., (1990), "Lee Router Modified for Global Routing", CAD, Vol.22, No.5, pp. 296-300.
Chambon, R., Tollenaere, M., (1991), "Automated AI-Based Mechanical Design of Hydraulic Manifold Blocks", CAD, Vol.23, No.3, pp. 213-222.
Funaoka, K., Wumeda, H., Fukahori, S., Yano, H., Ooahiba, T., Iida, A., Fujita, K., (1985), "New Integrated Ship Design System MATES", Computer Applications in the Automation of Shipyard Operation and Ship Design ¥´, ICCAS '85, pp. A5.1-A5.12.
Gotoh, K., Ishizuka, T., Izumi, Y., Ueda, M., (1982), "A New Computer Aided Streamlined System for Ship Piping Design", Computer Applications in the Automation of Shipyard Operation and Ship Design ¥³, ICCAS '82, pp. 257-268.
Inoue, T., Shirakami, H., Shishida, K., Moriya, Y., (1988), "HICADEC : Integrated CAD/CAM System with 3D Interactive Data Processing Architectures for CIMS in Shipbuilding", Computer Applications in the Automation of Shipyard Operation and Ship Design ¥µ, ICCAS '88, pp. A7.1-A7.12.
Kim, Joon-Ho, (1994), "Knowledge-based Modeling for Arrangement Design Expert System", M.S. Thesis, Chungnam University, Korea.
Lee, Jong-Gab, et al., (1994), "An Expert System (MADES) to Help Engine-room Arrangement Design of Ships", Report of Korea Research Institute of Ships and Ocean Engineering.
Lee, Soo-Hong, (1994), "An Agent-based System for Automotive Wiring", Transactions of Korea Society of Automotive Engineers, Vol.1, No. 3, pp.83-94.
Martin, D. J., (1995), "AP 217 Ship's Piping", ISO TC184/SC4/WG3 N455 (T12).
Mitsuta, T., Kobayashi, Y., Wada, Y., Kiguchi, T., (1986), "A Knowledge-Based Approach to Routing Problems in Industrial Plant Design", Int. Conf. Expert Systems & Applications, pp. 237-256.
Rubin, F., (1974), "The Lee Path Connection Algorithm", IEEE Transactions on Computers, Vol. C-23, No.9, pp. 907-914.
Rumbaugh, J., Blaha, M., Premrlani, W., Eddy, F., Lorensen, W., (1991), Object-Oriented Modeling and Design, Englewood Cliffs : Prentice Hall.
Suh, Nam P., (1990), The Principles of Design, New York : Oxford Univ. Press.
Van Der Tak, C., Koopmans, J.J.G., (1976), "The Optimum Routing of Pipes in a Ship's Engine Room", Computer
Applications in the Automation of Shipyard Operation and Ship Design II, ICCAS '76, pp. 333-341.