外文翻译船舶专业.doc

Ship Design Optimization
This contribution is devoted to exploiting the analogy between a modern manufacturing plant and a heterogeneous parallel computer to construct a HPCN decision support tool for ship designers. The application is a HPCN one because of the scale of shipbuilding - a large container vessel is constructed by assembling about 1.5 million atomic components in a production hierarchy. The role of the decision support tool is to rapidly evaluate the manufacturing consequences of design changes. The implementation as a distributed multi-agent application running on top of PVM is described
1 Analogies between Manufacturing and HPCN
There are a number of analogies between the manufacture of complex products such as ships, aircraft and cars and the execution of a parallel program. The manufacture of a ship is carried out according to a production plan which ensures that all the components come together at the right time at the right place. A parallel computer application should ensure that the appropriate data is available on the appropriate processor in a timely fashion.
It is not surprising, therefore, that manufacturing is plagued by indeterminacy exactly as are parallel programs executing on multi-processor hardware. This has caused a number of researchers in production engineering to seek inspiration in other areas where managing complexity and unpredictability is important. A number of new paradigms, such as Holonic Manufacturing and Fractal Factories have emerged [1,2] which contain ideas rather reminiscent of those to be found in the field of Multi- Agent Systems [3, 4].
Manufacturing tasks are analogous to operations carried out on data, within the context of planning, scheduling and control. Also, complex products are assembled at physically distributed workshops or production facilities, so the components must be transported between them. This is analogous to communication of data between processors in a parallel computer, which thus also makes clear the analogy between workshops and processors.
The remainder of this paper reports an attempt to exploit this analogy to build a parallel application for optimizing ship design with regard to manufacturing issues.

2 Shipbuilding at Odense Steel Shipyard
Odense Steel Shipyard is situated in the town of Munkebo on the island of Funen. It is recognized as being one of the most modern and highly automated in the world. It
specializes in building VLCC's (supertankers) and very large container ships. The yard was the first in the world to build a double hulled supertanker and is currently building an order of 15 of the largest container ships ever built for the Maersk line. These container ships are about 340 metres long and can carry about 7000 containers at a top speed of 28 knots with a crew of 12.
Odense Steel Shipyard is more like a ship factory than a traditional shipyard. The ship design is broken down into manufacturing modules which are assembled and processed in a number of workshops devoted to, for example, cutting, welding and surface treatment. At any one time, up to 3 identical ships are being built and a new ship is launched about every 100 days.
The yard survives in the very competitive world of shipbuilding by extensive application of information technology and robots, so there are currently about 40 robots at the yard engaged in various production activities. The yard has a commitment to research as well, so that there are about 10 industrial Ph.D. students working there, who are enrolled at various engineering schools in Denmark.

3 Tomorrow's Manufacturing Systems
The penetration of Information Technology into our lives will also have its effect in manufacturing industry. For example, the Internet is expected to become the dominant trading medium for goods. This means that the customer can come into direct digital contact with the manufacturer.
The direct digital contact with customers will enable them to participate in the design process so that they get a product over which they have some influence. The element of unpredictability introduced by taking into account customer desires increases the need for flexibility in the manufacturing process, especially in the light of the tendency towards globalization of production. Intelligent robot systems, such as AMROSE, rely on the digital CAD model as the primary source of information about the work piece and the work cell [5,6].This information is used to construct task performing, collision avoiding trajectories for the robots, which because of the high precision of the shipbuilding process, can be corrected for small deviations of the actual world from the virtual one using very simple sensor systems. The trajectories are generated by numerically solving the constrained equations of motion for a model of the robot moving in an artificial force field designed to attract the tool centre to the goal and repell it from obstacles, such as the work piece and parts of itself. Finally, there are limits to what one can get a robot to do, so the actual manufacturing will be performed as a collaboration between human and mechatronic agents.
Most industrial products, such as the windmill housing component shown in Fig. 1, are designed electronically in a variety of CAD systems.

Fig. 1. Showing the CAD model for the housing of a windmill. The model, made using Bentley Microstation, includes both the work-piece and task-curve geometries.
4 Today's Manufacturing Systems
The above scenario should be compared to today's realities enforced by traditional production engineering philosophy based on the ideas of mass production introduced about 100 years ago by Henry Ford. A typical production line has the same structure as a serial computer program, so that the whole process is driven by production requirements. This rigidity is reflected on the types of top-down planning and control systems used in manufacturing industry, which are badly suited to both complexity and unpredictability.
In fact, the manufacturing environment has always been characterized by unpredictability. Today's manufacturing systems are based on idealized models where unpredictability is not taken into account but handled using complex and expensive logistics and buffering systems.
Manufacturers are also becoming aware that one of the results of the top-down serial approach is an alienation of human workers. For example, some of the car manufacturers have experimented with having teams of human workers responsible for a particular car rather than performing repetitive operations in a production line. This model in fact better reflects the concurrency of the manufacturing process than the assembly line.
5 A Decision Support Tool for Ship Design Optimization
Large ships are, together with aircraft, some of the most complex things ever built. A container ship consists of about 1.5 million atomic components which are assembled in a hierarchy of increasingly complex components. Thus any support tool for the manufacturing process can be expected to be a large HPCN application.
Ships are designed with both functionality and ease of construction in mind, as
well as issues such as economy, safety, insurance issues, maintenance and even decommissioning. Once a functional design is in place, a stepwise decomposition of the overall design into a hierarchy of manufacturing components is performed. The manufacturing process then starts with the individual basic building blocks such as steel plates and pipes. These building blocks are put together into ever more complex structures and finally assembled in the dock to form the finished ship.
Thus a very useful thing to know as soon as possible after design time are the manufacturing consequences of design decisions. This includes issues such as whether the intermediate structures can actually be built by the available production facilities, the implications on the use of material and whether or not the production can be efficiently scheduled [7].
Fig.2. shows schematically how a redesign decision at a point in time during construction implies future costs, only some of which are known at the time. Thus a decision support tool is required to give better estimates of the implied costs as early as possible in the process.
Simulation, both of the feasibility of the manufacturing tasks and the efficiency with which these tasks can be performed using the available equipment, is a very compute-intense application of simulation and optimization. In the next section, we describe how a decision support tool can be designed and implemented as a parallel application by modeling the main actors in the process as agents.

Fig.2. Economic consequences of design decisions. A design decision implies a future commitment of economic resources which is only partially known at design time.
6 Multi-Agent Systems
The notion of a software agent, a sort of autonomous, dynamic generalization of an object (in the sense of Object Orientation) is probably unfamiliar to the typical HPCN reader in the area of scientific computation. An agent possesses its own beliefs, desires and intentions and is able to reason about and act on its perception of other agents and the environment.
A multi-agent system is a collection of agents which try to cooperate to solve some problem, typically in the areas of control and optimization. A good example is the process of learning to drive a car in traffic. Each driver is an autonomous agent which observes and reasons about the intentions of other drivers. Agents are in fact a very useful tool for modeling a wide range of dynamical processes in the real world, such as the motion of protein molecules [8] or multi-link robots [9]. For other applications, see [4].
One of the interesting properties of multi-agent systems is the way global behavior of the system emerges from the individual interactions of the agents [10]. The notion of emergence can be thought of as generalizing the concept of evolution in dynamical systems.
Examples of agents present in the system are the assembly network generator agent which encapsulates knowledge about shipbuilding production methods for planning assembly sequences, the robot motion verification agent, which is a simulator capable of generating collision-free trajectories for robots carrying out their tasks, the quantity surveyor agent which possesses knowledge about various costs involved in the manufacturing process and the scheduling agent which designs a schedule for performing the manufacturing tasks using the production resources available.
7 Parallel Implementation
The decision support tool which implements all these agents is a piece of Object- Oriented software targeted at a multi-processor system, in this case, a network of Silicon Graphics workstations in the Design Department at Odense Steel Shipyard. Rather than hand-code all the communication between agents and meta-code for load balancing the parallel application, abstract interaction mechanisms were developed. These mechanisms are based on a task distribution agent being present on each processor. The society of task distribution agents is responsible for all aspects of communication and migration of tasks in the system.
The overall agent system runs on top of PVM and achieves good speedup and load balancing. To give some idea of the size of the shipbuilding application, it takes 7 hours to evaluate a single design on 25 SGI workstations.

From：Applied Parallel Computing Large Scale Scientific and Industrial Problems Lecture Notes in Computer Science, 1998, Volume 1541/1998, 476-482, DOI: 10.1007/BFb0095371 .
汉字翻译：
船舶设计优化
这一贡献致力于开拓类比当代先进制造工厂和一个异构并行计算机,构建了一个HPCN决议支援工具给船舶设计师。这个应用程序是一个HPCN一个原因是造船规模——一个巨大集装箱船经过装配了大约150万原子部件在生产过程中。该决议支持工具作用是快速评定修改设计造成制造后果。这个应用程序可描述为一个实现在PVM上运行分布式多智能体。
1.制造业与HPCN相同性
制造复杂产品有许多相同之处，如船舶、飞机、汽车和执行并行程式。制造了一艘船按照生产计划展开,必须确保全部部件在适当时间、适当地点结合在一起。类似计算机应用应确保适宜数据可在适当处理器下运行。
这并不奇怪，所以，制造业困扰是不确定性和在多处理器硬件下执行并行程序完全不一样。这已经引发了一部分研究人员在生产工程中在管理复杂性和不可预测性主要领域寻求灵感地方。很多新范例,如Holonic制造系统和分形工厂出现[1,2]包含着思想,而让人回顾起了多-代理领域系统[3、4]。
生产任务,就像是在执行操作数据,在其职权范围内策划、调度和控制。同时,复杂产品都聚集在身体上分布式讨论会或生产设施,所以组件类必须被流放分给他们。这好比通信之间数据在一个平行处理器计算机上,它也明确了车间和处理器之间相同性。
本文其余部分汇报，企图利用这个理论，建立一个并行应用程序优化船舶设计对于生产中出现问题。
2. 欧登塞钢船厂造船
欧登塞钢船厂位于Munkebo镇Funen岛上。它被公认为是世界上其中一个最当代化，高度自动化船厂。
它专门建设超大型油轮（超级油轮）和超大型集装箱船。企业厂区世界上第一个建造双壳超级油轮，现在正在为马士基航运企业建造船舶15份订单是有史以来建造最大集装箱。这些集装箱船约340米长，在28节最高速度下能够携带12名船员与约7000箱集装箱。
欧登塞钢船厂相比传统船厂建造更像是一个船厂。这艘船设计能够把制造模块分解为组装和加工,在许多车间投入生产,比如,切割、焊接和表面处理。不论什么时候,都有3个相同船只在建, 约每100天建成一艘新船。
这个企业能在世界上竞争非常激烈船舶业生存，是因为广泛应用信息技术和机械,现在有40台机器在这个企业从事各种生产活动。企业里研究任务,约有十个工业博士学生组成;他们来自丹麦不一样注册工科院校。
3.未来制造系统
信息科技渗透到我们生活中也将影响制造业。比如，互联网有望在商品交易中占主导地位。这意味着客户能够直接用数字与制造商联络。
直接与客户接触数字将使他们能够参加在设计过程中，使他们得到了他们产品有一定影响。考虑到用户需求增加了灵活性，在生产过程中是不可预测元素，尤其是在生产全球化趋势下。智能机器人系统，如AMROSE，数字化CAD模型上信息主要起源是工件和工作单元[5,6]。这些信息提供机械用来构建任务执行，防止轨迹碰撞，因为造船工艺精度高，能够进行修正偏差小，实际世界使用一个虚拟非常简单传感器系统。求解约束运动微分方程所产生数值运动轨迹模型,机器移动，在人为设计引发刀心达成目标和疏去除障碍，如工件一块或者本身一部分。最终,有限制哪些是能够得到一个机器人做了,所以实际制造将推行人类与机电合作代理。最终，机器能做是有不足，所以实际制造业将经过人与机电一体化合作完成。
大多数工业产品，如在图1所表示风房组成部分，在各种CAD系统下电子化设计。

图.1 显示了一个风车房CAD模型。该模型，用本特利MicroStation制作，包含工件和任务曲线几何结构。
4.当今制造系统
与上述情况相比，今天现实执行大规模生产是依照传统生产工程哲学观念1前从亨利福特引进。一个经典生产线作为一个序列电脑程序具备相同结构,使整个生产过程是由需求驱动。这种刚性反应在自上而下规划和控制系统用于制造业，这两个都是很适适用于复杂性和不可预测性。
实际上,生产环境一直具备不可预测性。今天制造系统是建立在理想化模型上，它具备没有考虑到不可预知性，不过处理使用了复杂、昂贵物流和缓冲系统。
制造商也开始意识到，对自上而下串行方法是人工异化结果之一。比如，汽车制造商已经试验了对一些特定汽车进行人工团体负责而不是在生产线上重复操作。这种模型反应了并行生产过程比流水线生产愈加好。
5.一个船舶设计优化决议支持工具
大型船只，以及飞机，一向是建造最复杂。集装箱船由约150万原子构件组装成一个个原来越复杂层次结构组成。所以，任何制造过程支持工具，能够预期将成为一个大型HPCN应用。
船舶设计要同时考虑功效和施工方便，诸如经济，安全，保险问题，维修，甚至拆卸问题。一旦功效设计到位，逐步分解整体设计成制造组件层次结构进行制造。制造过程，开始个人基本构建模块，如钢板和钢管。这些构建块组合成愈加复杂结构，最终在船坞组装成成品船。
在设计时间后尽快知道设计制造决定结果非常有用。这包含诸如是否中间结构是否能够由现有生产设施建造，在材料利用方面是否有影响和是否可有效地安排生产等问题[7]。
图2示意图显示了在一个时间点上重新设计施工决定预期花费费用，其中只有一些是已知。所以，一个决议支持工具，在生产过程中必须给隐含成本提供愈加好估算。
用现有生产设备能否完成生产，仿真生产任务可行性和效率，是一个计算非常密集型应用模拟与优化。在下一节中，我们描述了一个决议支持工具，能够设计和用一个并行程序去建模，经过模拟在这个过程中主要角色程序。

已知成本
未知成本
总成本

图 2 设计决议经济影响。一个设计决定隐含着未来负担经济支出，它是设计过程中一部分。
6．多智能体系统
一个软件代理，自主，一个对象（在面向对象意义上）概念是动态一个推广，可能不熟悉经典HPCN在科学计算领域读者。代理人拥有自己信念，愿望和意图，并能采取行动原因和其关于其余代理和环境看法。
多智能体系统是一个试图协调处理一些问题,尤其是地域控制和优化代理集合。一个很好例子就是在交通中学习开车过程开车。每个司机都是独立代理来观察和感受其余司机意图。代理在现实世界中其实是一个很有用为动力学过程建模工具, 如蛋白质分子运动[8]多连杆机器人[9]。对于其它应用，见[4]。
对于多智能体系统有趣特征之一是该系统全球化出现是因为个体相互作用 [10]。这个概念出现能够被看作是动力系统概念进化。
关于系统中现在代理，封装了关于造船生产装配序列规划方法知识，验证代理机器人运动，这是一个能生成无碰撞轨迹模拟器提供给机器人执行任务，工料测量师代理它拥关于于在制造过程中包括各项费用知识和日程安排代理即设计用于执行任务使用生产制造资源时间表。
7．并行实施
决议支持工具，实现了全部这些代理是一个在多处理器系统有针对性面向对象软件在这种情况下，在欧登塞钢船厂有一个硅谷制图工作站网络。不是全部手写通信代码在代理和源代码之间为负载平衡并行应用程序，抽象互动机制被开发。这些机制是基于对被代理人任务分配在每个处理器中。该协会任务分配机构在系统中负责各方面通讯和任务迁移。
代理系统整体运行于PVM顶级，取得了良好加速比和负载平衡。为了融入一些尺度造船应用新想法，在25 SGI工作站花费了7个小时计算一个单一设计。