Safetygram-31安全程序-31
Safe Speed Control产品简介与引导说明书
Continued next page...C O N T E N T SSAFE SPEED CONTROLPRODUCTS:Safe Speed ControlF A QsREFERENCE: CatalogsGuard Locking SystemsGK-1, GK-C,EN ISO 13849-1CONTACT INFORMATIONRE FE R EN C EGuard locking systems forprotection of manand machineA supplement to our main catalogcontaining a safety guide book onsafe speed control and solenoid inter-locks. 60 pages.View the PDF catalog: Click here.GK-1 Catalog9th EditionOur main catalog containing technicaldata sheets for all products, a safetytutorial handbook, and a referencesection. 424 pages.View the GK-1 catalog online:. Safe Speed Control European harmonized C-standards, such as DIN EN 12417, define two operating modes in which a machine is generally equipped with: automatic and set-up. When a machine is engaged its production process, or in other words performing its actual purpose, it is said to be running in mode 1 or automatic mode. In this instance all safety guards are closed and secured. Set-up mode or mode 2 is used for adjustment activities, i.e. tool replacement or a size change. During this time the safety guards are opened and the machine is running at significantly reduced speeds. However, manufacturers and operators have expressed the need for additional possibilities of operating speeds for particular set-up and parameter setting tasks. As a result, amendment 1:2006 to DIN EN 12417 allows for “extended manual intervention” known as mode 3. Applica-tions using this mode are taken into consideration in a clause under Appendix I of the new Ma-chine Directive (2006/42/CE: 1.2.5: “Selecting the control and operating modes”). In this situa-tion use of an enabling switch allows an operator to activate a limited number of machine func-tions at speeds higher than mode 2.There are intervention situations however that require the operator to use both hands and for the process to run at speeds near automatic mode 1. “Process observation”, or mode 4, en-ables running the machine with safety guards open and without the continuous actuation of an enabling device. One main difference between mode 3 and mode 4 is that the maximum limits for rotation speeds and feed rates are prescribed in mode 3, whereas in mode 4 the limits are determined by the application needs. Although this fourth mode has been accepted by legisla-tive bodies, it has not yet been addressed in any standards. Nonetheless, the framework for this mode has been defined by the technical committee Machine Building, Production Systems and Steel Construction . S p r i n g 2010Safe Speed Control Products Schmersal offers an ever increasing range of products designed to protect workers from the hazards of dangerous motion: Modular, Programmable Drive Monitoring System PDMS The PDMS is a flexible system for the safe monitoring of drives. This systemmonitors the signals from encoders, resolvers, or proximity switches to detectrotary and linear movement in a safe manner. The system offers a variety ofinput cards in which the drive speed to be monitored can be individually pro-grammed. Multiple and even different input cards can be combined with asafety output module in a housing that can be mounted on a standard DIN rail. More ► Fail-To-Safe Standstill Monitors Standstill monitors provide for the fail-safe registration of machine standstill and control of safety solenoid interlocks on safety guard doors. SSW 301HV / AZR31S1 The SSW301HV is a sensor-free standstill monitor that is directly connected to a three-phase motor and measures the frequency of the inducted voltage up to 690 VAC. On standstill of the motor the enabling paths close. The unit is equipped with light-emitting diodes to indicate the operating conditions and provides three safety outputs and one signaling output. More ►R EF E RE N CE660 WHITE PLAINS ROADI N FO R M A T I O NGK-C Catalog2nd EditionThe Schmersal condensed catalogand overview of product lines. 12pages.View the catalog online: .A New Approach toMachine Safety:EN ISO 13849-1An overview of the new EuropeanSafety Standard that defines Per-formance Levels. 52 Pages.View the guide online:. FWS The FWS range of standstill monitors evaluate the impulses produced by ro-tating movement in inductive proximity switches or incremental shaft encod-ers. When the recorded sequence of impulses falls below the limiting fre-quency, the enabling paths close. FWS range standstill monitors are equipped with Integral System Diagnostics (ISD) for fast and simple recogni-tion of faults. The user receives information regarding the switching condition of the standstill monitor and sensor by means of multi-function LED. A variety of units are available with differ-ent safety and signaling outputs. More ►Fail-to-Safe Timer AZS The AZS fail-safe delay timer provides for the secure measurement of a pre-set time. The enabling signal for the control system is only given when the present time has elapsed which then makes it possible to open a guard de-vice. AZS range fail-safe delay timers are often used where dangerous situa-tion can occur after a machine has been switched off, e.g. because of running-down movements. More ►Solenoid Latching Keyed Interlock Switches AZM161, AZM170, AZM200, AZM415, TK, TZ, and TZK . Schmersal offers a wide variety of solenoid-latching keyed interlock switches to lock machine guards. These switches not only detect the closed safety guard, but provide a lock to prevent the guard from opening until the hazard-ous motion has abated. The switches come in power-to-lock and power-to-unlock versions and many have optional emergency releases to prevent work-ers from being trapped inside the hazardous area. More ► Control Devices TG Door Handle & AZM 200-2568 Pushbuttons and other control devices are often used to signal the system to shut down and unlock the safety guards. The TG door handle and the AZM200-2568 solenoid latching switch have integrated pushbuttons for this purpose. More ►SAFE SPEED CONTROL PRODUCTS Continued from page 1。
安全强化学习综述
安全强化学习综述王雪松 1王荣荣 1程玉虎1摘 要 强化学习(Reinforcement learning, RL)在围棋、视频游戏、导航、推荐系统等领域均取得了巨大成功. 然而, 许多强化学习算法仍然无法直接移植到真实物理环境中. 这是因为在模拟场景下智能体能以不断试错的方式与环境进行交互, 从而学习最优策略. 但考虑到安全因素, 很多现实世界的应用则要求限制智能体的随机探索行为. 因此, 安全问题成为强化学习从模拟到现实的一个重要挑战. 近年来, 许多研究致力于开发安全强化学习(Safe reinforcement learning, SRL)算法, 在确保系统性能的同时满足安全约束. 本文对现有的安全强化学习算法进行全面综述, 将其归为三类: 修改学习过程、修改学习目标、离线强化学习, 并介绍了5大基准测试平台: Safety Gym 、safe-control-gym 、SafeRL-Kit 、D4RL 、NeoRL.最后总结了安全强化学习在自动驾驶、机器人控制、工业过程控制、电力系统优化和医疗健康领域中的应用, 并给出结论与展望.关键词 安全强化学习, 约束马尔科夫决策过程, 学习过程, 学习目标, 离线强化学习引用格式 王雪松, 王荣荣, 程玉虎. 安全强化学习综述. 自动化学报, 2023, 49(9): 1813−1835DOI 10.16383/j.aas.c220631Safe Reinforcement Learning: A SurveyWANG Xue-Song 1 WANG Rong-Rong 1 CHENG Yu-Hu 1Abstract Reinforcement learning (RL) has proved a prominent success in the game of Go, video games, naviga-tion, recommendation systems and other fields. However, a large number of reinforcement learning algorithms can-not be directly transplanted to real physical environment. This is because in the simulation scenario, the agent is able to interact with the environment in a trial-and-error manner to learn the optimal policy. Considering the safety of systems, many real-world applications require the limitation of random exploration behavior of agents. Hence,safety has become an essential factor for reinforcement learning from simulation to reality. In recent years, many re-searches have been devoted to develope safe reinforcement learning (SRL) algorithms that satisfy safety constraints while ensuring system performance. This paper presents a comprehensive survey of existing SRL algorithms, which are divided into three categories: Modification of learning process, modification of learning objective, and offline re-inforcement learning. Furthermore, five experimental platforms are introduced, including Safety Gym, safe-control-gym, SafeRL-Kit, D4RL, and NeoRL. Lastly, the applications of SRL in the fields of autonomous driving, robot control, industrial process control, power system optimization, and healthcare are summarized, and the conclusion and perspective are briefly drawn.Key words Safe reinforcement learning (SRL), constrained Markov decision process (CMDP), learning process,learning objective, offline reinforcement learningCitation Wang Xue-Song, Wang Rong-Rong, Cheng Yu-Hu. Safe reinforcement learning: A survey. Acta Automat-ica Sinica , 2023, 49(9): 1813−1835作为一种重要的机器学习方法, 强化学习 (Re-inforcement learning, RL) 采用了人类和动物学习中 “试错法” 与 “奖惩回报” 的行为心理学机制, 强调智能体在与环境的交互中学习, 利用评价性的反馈信号实现决策的优化[1]. 早期的强化学习主要依赖于人工提取特征, 难以处理复杂高维状态和动作空间下的问题. 近年来, 随着计算机硬件设备性能的提升和神经网络学习算法的发展, 深度学习由于其强大的表征能力和泛化性能受到了众多研究人员的关注[2−3]. 于是, 将深度学习与强化学习相结合就成为了解决复杂环境下感知决策问题的一个可行方案. 2016年, Google 公司的研究团队DeepMind 创新性地将具有感知能力的深度学习与具有决策能收稿日期 2022-08-08 录用日期 2023-01-11Manuscript received August 8, 2022; accepted January 11,2023国家自然科学基金(62176259, 61976215), 江苏省重点研发计划项目(BE2022095)资助Supported by National Natural Science Foundation of China (62176259, 61976215) and Key Research and Development Pro-gram of Jiangsu Province (BE2022095)本文责任编委 黎铭Recommended by Associate Editor LI Ming1. 中国矿业大学信息与控制工程学院 徐州 2211161. School of Information and Control Engineering, China Uni-versity of Mining and Technology, Xuzhou 221116第 49 卷 第 9 期自 动 化 学 报Vol. 49, No. 92023 年 9 月ACTA AUTOMATICA SINICASeptember, 2023力的强化学习相结合, 开发的人工智能机器人Al-phaGo 成功击败了世界围棋冠军李世石[4], 一举掀起了深度强化学习的研究热潮. 目前, 深度强化学习在视频游戏[5]、自动驾驶[6]、机器人控制[7]、电力系统优化[8]、医疗健康[9]等领域均得到了广泛的应用.近年来, 学术界与工业界开始逐步注重深度强化学习如何从理论研究迈向实际应用. 然而, 要实现这一阶段性的跨越还有很多工作需要完成, 其中尤为重要的一项任务就是保证决策的安全性. 安全对于许多应用至关重要, 一旦学习策略失败则可能会引发巨大灾难. 例如, 在医疗健康领域, 微创手术机器人辅助医生完成关于大脑或心脏等关键器官手术时, 必须做到精准无误, 一旦偏离原计划位置, 则将对病人造成致命危害. 再如, 自动驾驶领域, 如果智能驾驶车辆无法规避危险路障信息, 严重的话将造成车毁人亡. 因此, 不仅要关注期望回报最大化,同时也应注重学习的安全性.García 和Fernández [10]于2015年给出了安全强化学习 (Safe reinforcement learning, SRL) 的定义: 考虑安全或风险等概念的强化学习. 具体而言,所谓安全强化学习是指在学习或部署过程中, 在保证合理性能的同时满足一定安全约束的最大化长期回报的强化学习过程. 自2015年起, 基于此研究,学者们提出了大量安全强化学习算法. 为此, 本文对近年来的安全强化学习进行全面综述, 围绕智能体的安全性问题, 从修改学习过程、修改学习目标以及离线强化学习三方面进行总结, 并给出了用于安全强化学习的5大基准测试平台: Safety Gym 、safe-control-gym 、SafeRL-Kit 、D4RL 、NeoRL, 以及安全强化学习在自动驾驶、机器人控制、工业过程控制、电力系统优化以及医疗健康领域的应用.安全强化学习中所涉及的方法、基准测试平台以及应用领域之间的关系如图1所示.本文结构如下: 第1节对安全强化学习问题进行形式化描述; 第2节对近年来的安全强化学习方法进行分类与综述; 第3节介绍5种基准测试平台;第4节总结安全强化学习的实际应用场景; 第5节对未来研究方向进行探讨; 第6节对文章进行总结.1 问题描述M ∪C M =⟨S ,A ,T ,γ,r ⟩C ={c,d }S A T (s ′|s,a )γr :S ×A →R c :S ×A →R d π∗安全强化学习问题通常被定义为一个约束马尔科夫决策过程 (Constrained Markov decision pro-cess, CMDP) [11], 即在标准马尔科夫决策过程 的基础上添加了关于成本函数的约束项 . 表示状态空间集, 表示动作空间集, 表示用于描述动力学模型的状态转移函数, 表示折扣因子, 表示奖励函数; 表示成本函数, 表示安全阈值. 这种情况下, 安全强化学习问题可以表述为在满足安全约束的情况下, 求解使期望回报最大化的最优可行策略J (π)=E τ∼π(∞t =0γtr (s t ,a t ))τ=(s 0,a 0,s 1,a 1,···)τ∼πτπΠc 其中, , 表示一条轨迹, 表示轨迹 根据策略 采样得到, 表示满足安全约束的安全策略集. 值得注意的是, 本文公式所描述的都是单成本约束的形式, 但不失一般性, 这些公式都可以拓展为多成本约束的形式. 对于不同类型的决策任务,安全策略集可以有不同的表达形式.Πc 对于安全性要求严格的决策任务, 例如自动驾驶[12−13]任务, 通常采用硬约束方式, 即在所有的时刻都需要强制满足单步约束. 这种情况下 表示为环境知识人类知识无先验知识拉格朗日法信赖域法策略约束值约束预训练模型图 1 安全强化学习方法、基准测试平台与应用Fig. 1 Methods, benchmarking platforms, and applications of safe reinforcement learning1814自 动 化 学 报49 卷Π其中, 表示可行策略集. 但由于这种约束方式要求过于严格, 因此通常需要借助模型信息加以实现.Πc 在无模型情况下, 软约束方式有着更广泛的应用, 即对折扣累积成本的期望进行约束, 这种情况下 表示为c :S ×A →{0,1}c (s t ,a t )=0c (s t ,a t )=1E τ∼π(∑∞t =0γtc (s t ,a t ))π这种约束方式可以很好地适用于机器人行走[14]、油泵安全控制[15]和电力系统优化[16]等任务, 但对于需要明确定义状态或动作是否安全的任务却难以处理. 为了使软约束方式更好地适用于不同类型的决策任务, 可以将成本函数修改为 ,利用成本函数对当前状态动作对进行安全性判断,若安全, 则 , 否则, , 并且在智能体与环境交互期间遇到不安全的状态动作对时终止当前回合. 这时, 约束项 可以表示 产生不安全状态动作对的概率, 因此经过这样修改后的软约束也被称为机会型约束. 机会型约束由于其良好的任务适应性, 已被成功应用于无模型的自动驾驶[17]和机械臂控制[18]等任务.M =⟨S ,A ,T ,γ,r ⟩π∗=arg max π∈ΠJ (π)B ={(s,a,r,s ′)}π∗另一方面, 离线强化学习[19−20]从一个静态的数据集中学习最优策略, 它避免了与环境的交互过程,可以保障训练过程中的安全性. 因此, 可以将离线强化学习作为安全强化学习的一种特殊形式. 离线强化学习考虑一个标准马尔科夫决策过程 , 它的目标是求解使期望回报最大化的最优可行策略 , 与在线方式不同的是, 智能体在训练过程中不再被允许与环境进行交互, 而是只能从一个静态数据集 中进行学习. 尽管这种方式可以保障训练过程中的安全性, 但分布偏移问题 (目标策略与行为策略分布不同)[19−20]也给求解 的过程带来了困难.因此, 现如今的离线强化学习方法大多关注于如何解决分布偏移问题. 离线强化学习在有先验离线数据集支持的情况下, 借助于其训练过程安全的优势,已被应用于微创手术机器人控制[21]和火力发电机组控制[22]等任务.2 方法分类求解安全强化学习问题的方法有很多, 受Gar-cía 和Fernández [10]启发, 本文从以下三方面进行综述:1) 修改学习过程. 通过约束智能体的探索范围, 采用在线交互反馈机制, 在强化学习的学习或探索过程中阻止其产生危险动作, 从而确保了训练时策略的安全性. 根据是否利用先验知识, 将此类方法划分为三类: 环境知识、人类知识、无先验知识.2) 修改学习目标. 同样采用在线交互反馈机制, 在强化学习的奖励函数或目标函数中引入风险相关因素, 将约束优化问题转化为无约束优化问题,如拉格朗日法、信赖域法.3) 离线强化学习. 仅在静态的离线数据集上训练而不与环境产生交互, 从而完全避免了探索, 但对部署时安全没有任何约束保证, 并未考虑风险相关因素. 因此大多数离线强化学习能实现训练时安全, 但无法做到部署时安全.三类安全强化学习方法的适用条件、优缺点以及应用领域对比如表1所示. 下面对安全强化学习的现有研究成果进行详细综述与总结.2.1 修改学习过程在强化学习领域, 智能体需要通过不断探索来减小外界环境不确定性对自身学习带来的影响. 因此, 鼓励智能体探索一直是强化学习领域非常重要的一个研究方向. 然而, 不加限制的自由探索很有可能使智能体陷入非常危险的境地, 甚至酿成重大安全事故. 为避免强化学习智能体出现意外和不可逆的后果, 有必要在训练或部署的过程中对其进行安全性评估并将其限制在 “安全” 的区域内进行探索, 将此类方法归结为修改学习过程. 根据智能体利用先验知识的类型将此类方法进一步细分为环境知识、人类知识以及无先验知识. 其中环境知识利用系统动力学先验知识实现安全探索; 人类知识借鉴人类经验来引导智能体进行安全探索; 无先验知识没有用到环境知识和人类知识, 而是利用安全约束结构将不安全的行为转换到安全状态空间中.2.1.1 环境知识基于模型的方法因其采样效率高而得以广泛研究. 该类方法利用了环境知识, 需要学习系统动力学模型, 并利用模型生成的轨迹来增强策略学习,其核心思想就是通过协调模型使用和约束策略搜索来提高安全探索的采样效率. 可以使用高斯过程对模型进行不确定性估计, 利用Shielding 修改策略动作从而生成满足约束的安全过滤器, 使用李雅普诺夫函数法或控制障碍函数法来限制智能体的动作选择, 亦或使用已学到的动力学模型预测失败并生成安全策略. 具体方法总结如下.高斯过程. 一种主流的修改学习过程方式是使用高斯过程对具有确定性转移函数和值函数的动力9 期王雪松等: 安全强化学习综述1815学建模, 以便能够估计约束和保证安全学习. Sui等[38]将 “安全” 定义为: 在智能体学习过程中, 选择的动作所收到的期望回报高于一个事先定义的阈值. 由于智能体只能观测到当前状态的安全函数值, 而无法获取相邻状态的信息, 因此需要对安全函数进行假设. 为此, 在假设回报函数满足正则性、Lipschitz 连续以及范数有界等条件的前提下, Sui等[38]利用高斯过程对带参数的回报函数进行建模, 提出一种基于高斯过程的安全探索方法SafeOpt. 在学习过程中, 结合概率生成模型, 通过贝叶斯推理即可求得高斯过程的后验分布, 即回报函数空间的后验.进一步, 利用回报函数置信区间来评估决策的安全性, 得到一个安全的参数区间并约束智能体只在这个安全区间内进行探索. 然而, SafeOpt仅适用于类似多臂老虎机这类的单步、低维决策问题, 很难推广至复杂决策问题. 为此, Turchetta等[39]利用马尔科夫决策过程的可达性, 在SafeOpt的基础上提出SafeMDP安全探索方法, 使其能够解决确定性有限马尔科夫决策过程问题. 在SafeOpt和SafeM-DP中, 回报函数均被视为是先验已知和时不变的,但在很多实际问题中, 回报函数通常是先验未知和时变的. 因此, 该方法并未在考虑安全的同时优化回报函数. 针对上述问题, Wachi等[40]把时间和空间信息融入核函数, 利用时−空高斯过程对带参数的回报函数进行建模, 提出一种新颖的安全探索方法: 时−空SafeMDP (Spatio-temporal SafeMDP, ST-SafeMDP), 能够依概率确保安全性并同时优化回报目标. 尽管上述方法是近似安全的, 但正则性、Lipschitz连续以及范数有界这些较为严格的假设条件限制了SafeOpt、SafeMDP和ST-SafeM-DP在实际中的应用, 而且, 此类方法存在理论保证与计算成本不一致的问题, 在高维空间中很难达到理论上保证的性能.Shielding. Alshiekh等[41]首次提出Shield-ing的概念来确保智能体在学习期间和学习后保持安全. 根据Shielding在强化学习环节中部署的位置, 将其分为两种类型: 前置Shielding和后置Shielding. 前置Shielding是指在训练过程中的每个时间步, Shielding仅向智能体提供安全的动作以供选择. 后置Shielding方式较为常用, 它主要影响智能体与环境的交互过程, 如果当前策略不安全则触发Shielding, 使用一个备用策略来覆盖当前策略以保证安全性. 可以看出, 后置Shielding方法的使用主要涉及两个方面的工作: 1) Shielding触发条件的设计. Zhang等[42]通过一个闭环动力学模型来估计当前策略下智能体未来的状态是否为可恢复状态, 如果不可恢复, 则需要采用备用策略将智能体还原到初始状态后再重新训练. 但如果智能体的状态不能还原, 则此方法就会失效. Jansen等[43]一方面采用形式化验证的方法来计算马尔科夫决策过程安全片段中关键决策的概率, 另一方面根据下一步状态的安全程度来估计决策的置信度. 当关键决策的概率及其置信度均较低时, 则启用备用策略. 但是, 在复杂的强化学习任务中, 从未知的环境中提取出安全片段并不是一件容易的事情. 2) 备用 (安全)策略的设计. Li和Bastani[44]提出了一种基于tube 的鲁棒非线性模型预测控制器并将其作为备用控制器, 其中tube为某策略下智能体多次运行轨迹组成的集合. Bastani[45]进一步将备用策略划分为不变策略和恢复策略, 其中不变策略使智能体在安全平衡点附近运动, 恢复策略使智能体运行到安全平衡点. Shielding根据智能体与安全平衡点的距离来表 1 安全强化学习方法对比Table 1 Comparison of safe reinforcement learning methods方法类别训练时安全部署时安全与环境实时交互优点缺点应用领域修改学习过程环境知识√√√采样效率高需获取环境的动力学模型、实现复杂自动驾驶[12−13, 23]、工业过程控制[24−25]、电力系统优化[26]、医疗健康[21]人类知识√√√加快学习过程人工监督成本高机器人控制[14, 27]、电力系统优化[28]、医疗健康[29]无先验知识√√√无需获取先验知识、可扩展性强收敛性差、训练不稳定自动驾驶[30]、机器人控制[31]、工业过程控制[32]、电力系统优化[33]、医疗健康[34]修改学习目标拉格朗日法×√√思路简单、易于实现拉格朗日乘子选取困难工业过程控制[15]、电力系统优化[16]信赖域法√√√收敛性好、训练稳定近似误差不可忽略、采样效率低机器人控制[35]离线强化学习策略约束√××收敛性好方差大、采样效率低医疗健康[36]值约束√××值函数估计方差小收敛性差工业过程控制[22]预训练模型√××加快学习过程、泛化性强实现复杂工业过程控制[37]1816自 动 化 学 报49 卷决定选用何种类型的备用策略, 从而进一步增强了智能体的安全性. 但是, 在复杂的学习问题中, 很难定义安全平衡点, 往往也无法直观地观测状态到平衡点的距离. 综上所述, 如果环境中不存在可恢复状态, Shielding即便判断出了危险, 也没有适合的备用策略可供使用. 此外, 在复杂的强化学习任务中, 很难提供充足的先验知识来搭建一个全面的Shielding以规避所有的危险.李雅普诺夫法. 李雅普诺夫稳定性理论对于控制理论学科的发展产生了深刻的影响, 是现代控制理论中一个非常重要的组成部分. 该方法已被广泛应用于控制工程中以设计出达到定性目标的控制器, 例如稳定系统或将系统状态维持在所需的工作范围内. 李雅普诺夫函数可以用来解决约束马尔科夫决策过程问题并保证学习过程中的安全性. Per-kins和Barto[46]率先提出了在强化学习中使用李雅普诺夫函数的思路, 通过定性控制技术设计一些基准控制器并使智能体在这些给定的基准控制器间切换, 用于保证智能体的闭环稳定性. 为了规避风险,要求强化学习方法具有从探索动作中安全恢复的能力, 也就是说, 希望智能体能够恢复到安全状态. 众所周知, 这种状态恢复的能力就是控制理论中的渐近稳定性. Berkenkamp等[47]使用李雅普诺夫函数对探索空间进行限制, 让智能体大概率地探索到稳定的策略, 从而能够确保基于模型的强化学习智能体可以在探索过程中被带回到 “吸引区域”. 所谓吸引区域是指: 状态空间的子集, 从该集合中任一状态出发的状态轨迹始终保持在其中并最终收敛到目标状态. 然而, 该方法只有在满足Lipschitz连续性假设条件下才能逐步探索安全状态区域, 这需要事先对具体系统有足够了解, 一般的神经网络可能并不具备Lipschitz连续. 上述方法是基于值函数的,因此将其应用于连续动作问题上仍然具有挑战性.相比之下, Chow等[48]更专注于策略梯度类方法,从原始CMDP安全约束中生成一组状态相关的李雅普诺夫约束, 提出一种基于李雅普诺夫函数的CMDP安全策略优化方法. 主要思路为: 使用深度确定性策略梯度和近端策略优化算法训练神经网络策略, 同时通过将策略参数或动作映射到由线性化李雅普诺夫约束诱导的可行解集上来确保每次策略更新时的约束满意度. 所提方法可扩展性强, 能够与任何同策略或异策略的方法相结合, 可以处理具有连续动作空间的问题, 并在训练和收敛过程中返回安全策略. 通过使用李雅普诺夫函数和Trans-former模型, Jeddi等[49]提出一种新的不确定性感知的安全强化学习算法. 该算法主要思路为: 利用具有理论安全保证的李雅普诺夫函数将基于轨迹的安全约束转换为一组基于状态的局部线性约束; 将安全强化学习模型与基于Transformer的编码器模型相结合, 通过自注意机制为智能体提供处理长时域范围内信息的记忆; 引入一个规避风险的动作选择方案, 通过估计违反约束的概率来识别风险规避的动作, 从而确保动作的安全性. 总而言之, 李雅普诺夫方法的主要特征是将基于轨迹的约束分解为一系列单步状态相关的约束. 因此, 当状态空间无穷大时, 可行性集就具有无穷维约束的特征, 此时直接将这些李雅普诺夫约束(相对于原始的基于轨迹的约束)强加到策略更新优化中实现成本高, 无法应用于真实场景, 而且, 此类方法仅适用于基于模型的强化学习且李雅普诺夫函数通常难以构造.障碍函数法. 障碍函数法是另一种保证控制系统安全的方法. 其基本思想为: 系统状态总是从内点出发, 并始终保持在可行安全域内搜索. 在原先的目标函数中加入障碍函数惩罚项, 相当于在可行安全域边界构筑起一道 “墙”. 当系统状态达到安全边界时, 所构造的障碍函数值就会趋于无穷, 从而避免状态处于安全边界, 而是被 “挡” 在安全域内.为保证强化学习算法在模型信息不确定的情况下的安全性, Cheng等[50]提出了一种将现有的无模型强化学习算法与控制障碍函数 (Control barrier func-tions, CBF) 相结合的框架RL-CBF. 该框架利用高斯过程来模拟系统动力学及其不确定性, 通过使用预先指定的障碍函数来指导策略探索, 提高了学习效率, 实现了非线性控制系统的端到端安全强化学习. 然而, 使用的离散时间CBF公式具有限制性, 因为它只能通过仿射CBF的二次规划进行实时控制综合. 例如, 在避免碰撞的情况下, 仿射CBF 只能编码多面体障碍物. 为了在学习过程中保持安全性, 系统状态必须始终保持在安全集内, 该框架前提假设已得到一个有效安全集, 但实际上学习安全集并非易事, 学习不好则可能出现不安全状态. Yang 等[51]采用障碍函数对系统进行变换, 将原问题转化为无约束优化问题的同时施加状态约束. 为减轻通信负担, 设计了静态和动态两类间歇性策略. 最后,基于actor-critic架构, 提出一种安全的强化学习算法, 采用经验回放技术, 利用历史数据和当前数据来共同学习约束问题的解, 在保证最优性、稳定性和安全性的同时以在线的方式寻求最优安全控制器. Marvi和Kiumarsi[52]提出了一种安全异策略强化学习方法, 以数据驱动的方式学习最优安全策略.该方法将CBF合并进安全最优控制成本目标中形成一个增广值函数, 通过对该增广值函数进行迭代近似并调节权衡因子, 从而实现安全性与最优性的平衡. 但在实际应用中, 权衡因子的选取需要事先9 期王雪松等: 安全强化学习综述1817人工设定, 选择不恰当则可能找不到最优解. 先前的工作集中在一类有限的障碍函数上, 并利用一个辅助神经网来考虑安全层的影响, 这本身就造成了一种近似. 为此, Emam等[53]将一个可微的鲁棒控制障碍函数 (Robust CBF, RCBF) 层合并进基于模型的强化学习框架中. 其中, RCBF可用于非仿射实时控制综合, 而且可以对动力学上的各种扰动进行编码. 同时, 使用高斯过程来学习扰动, 在安全层利用扰动生成模型轨迹. 实验表明, 所提方法能有效指导训练期间的安全探索, 提高样本效率和稳态性能. 障碍函数法能够确保系统安全, 但并未考虑系统的渐进稳定性, 与李雅普诺夫法类似, 在实际应用中障碍函数和权衡参数都需要精心设计与选择.引入惩罚项. 此类方法在原先目标函数的基础上添加惩罚项, 以此修正不安全状态. 由于传统的乐观探索方法可能会使智能体选择不安全的策略,导致违反安全约束, 为此, Bura等[54]提出一种基于模型的乐观−悲观安全强化学习算法 (Optimistic-pessimistic SRL, OPSRL). 该算法在不确定性乐观目标函数的基础上添加悲观约束成本函数惩罚项,对回报目标持乐观态度以便促进探索, 同时对成本函数持悲观态度以确保安全性. 在Media Control 环境下的仿真结果表明, OPSRL在没有违反安全约束的前提下能获得最优性能. 基于模型的方法有可能在安全违规行为发生之前就得以预测, 基于这一动机, Thomas等[55]提出了基于模型的安全策略优化算法 (Safe model-based policy optimization, SMBPO). 该算法通过预测未来几步的轨迹并修改奖励函数来训练安全策略, 对不安全的轨迹进行严厉惩罚, 从而避免不安全状态. 在MuJoCo机器人控制模拟环境下的仿真结果表明, SMBPO能够有效减少连续控制任务的安全违规次数. 但是, 需要有足够大的惩罚和精确的动力学模型才能避免违反安全. Ma等[56]提出了一种基于模型的安全强化学习方法, 称为保守与自适应惩罚 (Conservative and adaptive penalty, CAP). 该方法使用不确定性估计作为保守惩罚函数来避免到达不安全区域, 确保所有的中间策略都是安全的, 并在训练过程中使用环境的真实成本反馈适应性地调整这个惩罚项, 确保零安全违规. 相比于先前的安全强化学习算法, CAP具有高效的采样效率, 同时产生了较少的违规行为.2.1.2 人类知识为了获得更多的经验样本以充分训练深度网络, 有些深度强化学习方法甚至在学习过程中特意加入带有随机性质的探索性学习以增强智能体的探索能力. 一般来说, 这种自主探索仅适用于本质安全的系统或模拟器. 如果在现实世界的一些任务(例如智能交通、自动驾驶) 中直接应用常规的深度强化学习方法, 让智能体进行不受任何安全约束的“试错式” 探索学习, 所做出的决策就有可能使智能体陷入非常危险的境地, 甚至酿成重大安全事故.相较于通过随机探索得到的经验, 人类专家经验具备更强的安全性. 因此, 借鉴人类经验来引导智能体进行探索是一个可行的增强智能体安全性的措施. 常用的方法有中断机制、结构化语言约束、专家指导.中断机制. 此类方法借鉴了人类经验, 当智能体做出危险动作时能及时进行中断. 在将强化学习方法应用于实际问题时, 最理想的状况是智能体任何时候都不会做出危险动作. 由于限制条件太强,只能采取 “人在环中” 的人工介入方式, 即人工盯着智能体, 当出现危险动作时, 出手中断并改为安全的动作. 但是, 让人来持续不断地监督智能体进行训练是不现实的, 因此有必要将人工监督自动化.基于这个出发点, Saunders等[57]利用模仿学习技术来学习人类的干预行为, 提出一种人工干预安全强化学习 (SRL via human intervention, HIRL) 方法. 主要思路为: 首先, 在人工监督阶段, 收集每一个状态−动作对以及与之对应的 “是否实施人工中断” 的二值标签; 然后, 基于人工监督阶段收集的数据, 采用监督学习方式训练一个 “Blocker” 以模仿人类的中断操作. 需要指出的是, 直到 “Blocker”在剩余的训练数据集上表现良好, 人工监督阶段的操作方可停止. 采用4个Atari游戏来测试HIRL 的性能, 结果发现: HIRL的应用场景非常受限, 仅能处理一些较为简单的智能体安全事故且难以保证智能体完全不会做出危险动作; 当环境较为复杂的时候, 甚至需要一年以上的时间来实施人工监督,时间成本高昂. 为降低时间成本, Prakash等[58]将基于模型的方法与HIRL相结合, 提出一种混合安全强化学习框架, 主要包括三个模块: 基于模型的模块、自举模块、无模型模块. 首先, 基于模型的模块由一个动力学模型组成, 用以驱动模型预测控制器来防止危险动作发生; 然后, 自举模块采用由模型预测控制器生成的高质量示例来初始化无模型强化学习方法的策略; 最后, 无模型模块使用基于自举策略梯度的强化学习智能体在 “Blocker” 的监督下继续学习任务. 但是, 作者仅在小规模的4×4格子世界和Island Navigation仿真环境中验证了方法的有效性, 与HIRL一样, 该方法的应用场景仍1818自 动 化 学 报49 卷。
NF3气体
Safetygram-28安全程序-28 Nitrogen Trifluoride (NF3) Safe Handling Practices三氟化氮(NF3)安全操作实践1. General概述三氟化氮（NF3）是有毒、无色、无嗅、不可燃的氧化性压缩气体。
NF3具有在室温条件下相对容易使用和可以用作氟化剂的优点。
由于这些因素，它在许多应用中获得了广泛的认可。
由于它的优点，如高蚀刻率、高选择性、无碳蚀刻和最小限度的残留污染，电子工业把它用在等离子和热清洁应用中。
由于它相对于氟气更容易使用，NF3在高能化学激光器中用作氟源。
在特殊化学品的生产中N F3还用作媒介。
NF3有多种等级，由于Air Products发展的先进的净化技术，还有纯度超过的99.999 %的NF3。
表1列出了NF3的物理和化学性质。
2. Key Considerations关键的考虑Health Effects对健康的影响皮肤接触NF3是没有危险的。
对于眼睛和黏膜，它是一个相对程度较轻的刺激物。
过量吸入NF3会导致血色素转化成高铁血红蛋白。
高铁血红蛋白的形成减少了身体组织可以得到的氧的数量。
这会导致化学黄萎病、头痛、眩晕、虚弱、昏聩和其它伴随着氧气供应减少的表现。
作为高铁血红蛋白症的次要影响，会发生溶血性贫血、脾扩大和肝、肾、心肌的病变，这些影响是不可逆转的。
在NF3暴露停止后，高铁血红蛋白转化回血红素。
高铁血红蛋白症会在几个小时后自动消失，而溶血性贫血需要几个星期才能解除。
Toxicological Properties毒物学性质Inhalation吸入L C50 = 6,700 ppm (1 hour小时) rat鼠对于测试的所有物种，急性暴露于高浓度NF3的直接影响是大范围高铁血红蛋白形成和并发的组织缺氧。
随后的通常是会导致肝、肾、脾和有时候的心脏病变的溶血性贫血。
在已经吸入了NF3的狗和猴子身上还观察到了呕吐。
老鼠通过吸入暴露在100 ppm NF3中连续19周，每周5天，每天7小时，导致了肝和肾的轻微到中度的病变。
安全管理器-过程解决方案安全管理器说明书
Process SolutionsSafety Manager combines Honeywell’s proven Quadruple Modular Redundancy (QMR®) 2oo4D technology with extensive process safety management expertise in integrating process safety data, applications, system diagnostics and critical control strategies.Honeywell’s IEC 61511 and IEC 61508 SIL 3 TÜV certified solution provides the optimal level of safety and process integration while still maintaining functional safety separation as mandated by those standards. Through Experion operational integration, all systems are unified into one operationally integrated architecture, providing a unique opportunity to improve safety, process availability and efficiency.Experion provides unprecedented connectivity through all levels of process and business operations to optimize work processes, improve routine maintenance efficiencies, enhance safety management and release personnel from manual processes.Benefits∙Safe and Secure– Safety Manager is designed to be securely integrated into customer systems and has passed very rigorous security testing as defined by ISA Security Compliance Institute (ISCI).Safety Manager was the first safety system to achieve Embedded Device Security Assurance (EDSA) certification. ISCI developed this certification within the framework of the ISA Industrial Automation and Control Systems security standards (ISA 99). Because of the built in protection mechanisms, the Experion Safety Manager is protected from cyber attacks and disruption of service. ∙High Availability Architecture–Honeywell’s field-proven QMR 2oo4D architecture provides the highest availability with a safe architecture. Applying QMR technology allows uninterrupted process operation in the event of any system degradation or on-process modification without jeopardizing the SIL 3 level. The optional Safety Manager A.R.T. (Advanced Redundancy Technique) provides additional benefits for locations where timely maintenance is not available.∙Easy and Intuitive Engineering and Modifications– Safety Builder, an intuitive and comprehensive configuration tool, provides plant-wide management of safety-critical databases and application programming for easy network design. TÜV-approved, menu-driven online modifications prevent errors while maintaining and optimizing the safety application.∙Defense-in-Depth– SafeNet and remote distributed Safety Manager provide the ability to design defense-in-depth safety strategies that maximize safety and security while minimizing risk and scope-of-loss concerns.∙Safety Networking - The networking capabilities of Safety Manager are unsurpassed. Up to 1024 redundant nodes can be included in one safety network, acting as one integrated safety solution. The SIL 4 certified SafeNet communication protocol guarantees fast and safe communication over any media and distance. The remote management capabilities support centralized management of all connected safety systems.Honeywell’s Safety Manager, part of the Experion® Process Knowledge System (PKS), enhances the safety, reliability and efficiency of critical processes. Experion® PKS – The Knowledge to Make it Possible.∙SafeNet Flexibility - SafeNet can run over any network, such as a dedicated separated safety network as well as the Honeywell Fault Tolerant Ethernet (FTE) network infrastructure. SafeNet is the only SIL 4 certifiedcommunication protocol available in process networks today.∙Self-Learning – Replacing any module, including the safety processor, is possible when the plant is in operation, and data and programs are automatically copied from the running processor. There is no manual loading required, which simplifies handling and avoids problems. The total system will continue to meet the stringent SIL 3 requirements.∙High Performance – Safety Manager has been optimized to manage large applications with over 1,000 I/O as well as high-speed applications with fast processing requirements of cycle times well below 100 milliseconds.∙Universal Safety I/O – Safety Manager Universal Safety I/O enables maximum architectural flexibility and lowest cost ofownership when safety is required at distributed locations. It has the unique feature that each channel can be configured individually to a different I/O type. Every Universal Safety I/O module has a capacity of 32 freely configurable channels, enabling savings on both installation and operational costs. By using soft-marshalling, the Universal Safety I/O module can be mounted close to the process unit, eliminating the need for marshalling panels, homerun cables and reducing oreliminating field auxiliary rooms. This approach is ideally suited to highly distributed applications such as oil and gas upstream applications, and reduces cost while increasing availability and efficiency. This reduces overall capital expenditure, as well as maintenance costs.∙Localized Safeguarding - With Universal Safety Logic Solver,the safety application can be distributed into the field close to the process unit while maintaining a transparent overview of the overall safety application. The unique feature of this Universal Safety IO module is the fact that besides being an IO module to Safety Manager, it can execute the safetyapplication locally. Safeguarding the process even in the event communications to the Safety Manager are interrupted.∙Standardized Solutions - Universal Channel Technology enables Universal Cabinet designs to be standardized,significantly reducing engineering cost and schedule when applied broadly across a project.∙Advanced Experion Integration – Supports Safety Manager integration in Experion, providing an integrated safety and control solution. It enables, for example, transmitter data sharing between the CEE (Control Execution Environment)controllers and Safety Manager, via direct peer to peer communication, to save installed and operational costs. Peer to peer communication further allows for alarm suppression,automatic bypassing and interlocks between shutdown and control functions as well as “soft landing” in case of process upset. It also provides easy operator access and full Console Station support. As p art of the “enter data only once”philosophy, the Experion-related properties are configured from the Safety Builder tool simplifying maintenance and reducing total cost of ownership.∙Built on QMR Technology – Safety Manager is based on the unique and field-proven QMR diagnostic-based technology with 2oo4D architecture. QMR enhances system flexibility,increases diagnostic messaging capabilities and improves system fault tolerance for critical applications. It enables the handling of multiple system faults within Experion Safety Manager, matching the needs of critical control applications.In addition, Safety Manager provides the basis for integrating SIL-rated field sensors and valve actuators, ensuring that safety functions are well established to protect complex and hazardous processes. It integrates SIL 1-3 safety transmitters (such as Honeywell ST3000 and STT250) or safety valve positioners for improved safety and field asset management.∙Optimized field maintenance - Without the need for extra infrastructure or engineering, HART devices are integrated within Honeywell’s Field Device Manager. This provides all required data for field asset management. To prevent inadvertent device changes, the safety manager prevents FDM from writing parameter changes unless the device safety lock has been disabled from Safety Builder.Compliance to Safety StandardsA major requirement for compliance with IEC 61511 and IEC 61508 is the availability of a change history of applications. With Safety Builder, change history is efficiently tracked with the Safety Audit Tracker through an automatically enabled audit trail. Difficult procedures or extensive loggings are not required. The Safety Audit Tracker, together with the automated embedded Application Verification mechanism, is all that is required.Safety Manager complies with the following international standards:∙For burner management: NFPA 85, 86, EN50156∙For emergency shutdown and other critical applications: IEC 61508, IEC61511, ISA S84.01, DIN V 19250, UL,FM, ATEX∙For fire and gas: EN54-2, NFPA 72, Lloyd’s Register and offshore installations ABSWith all SIL 3 safety hardware and software compliance tools, Safety Manager provides excellent protection for safety applications across multiple industries throughout the entire life of an installation. Safety Manager provides the basis for critical control and safety unification, reducing risks and installed costs, and improving safety while increasing uptime.Optimized Engineering EnvironmentSafety Builder software improves engineering and design efficiency. With simple drag and drop functionality, a complete and complex network can be designed within minutes without programming, saving valuable engineering and testing time. The complete network design is available on a one-page view without requiring additional documentation.An integrated editor facilitates fast and effective application design, allowing clear and distinct views of all logic with full compliance to IEC 61131 standards. Logic inputs, outputs and symbols are placed with drag and drop functionality from the toolbar and are easily configurable. Through the Safety Manager simulation mode any application can be loaded and tested on a minimum size system, a tool that facilitates easy application design and testing. The simulation mode also allows execution of online modifications and testing of all communication interfaces.In absence of a Safety Manager system the Honeywell’s UniSim® simulation environment for Safety Manager supports offline simulation as well. It can help in the early implementation phase of a project or as part of a plant-wide system simulation. It supports step by step simulation, freezing the application and building snapshots.Optimal Process AvailabilityApplying QMR technology to Safety Manager delivers unlimited runtime for single channel operation. This increases process availability, allowing uninterrupted process operation in the event of any system degradation. Without incurring any process downtime, the system can be kept up to date with the latest system software as well as application changes or additions through a four-step online system modification procedure The on-process modification to the application can be carried out remotely without physical presence to the system.I/O faults are detected and isolated on a per-channel basis and immediately reported to the appropriate level. This minimizes the time to repair and further increases system robustness.Integrated Operation and MaintenanceSafety Manager unifies critical safety process data with process control information, providing single-window access for operation and maintenance. When connected to the Honeywell FTE network through TÜV SIL 3 approved Universal Safety Interfaces, multiple Safety Managers can be unified into one safety system architecture.Safety Manager integration delivers fast, safe and reliable data exchange with Experion, enhancing operator and maintenance performance. In addition, Safety Manager extends the system proof test interval with inherent extensive system self-testing and diagnostic capability, reducing operational and maintenance costs. Integrated sequence of events (SOE) functionality for all process and safety-related activities supports analysis at a glance.Safeguards are built into Safety Manager to eliminate the possibility of systematic failures caused by errors made duringthe design, planning, construction, operation and decommissioning of the system. A systematic failure in thedesign of a common tool can result in an unsafe reaction of both the safety and control systems.Safety through SeparationSafety and control systems must be integrated to allow for smooth and safe plant operation, while still maintaining a safe separation where appropriate.∙Secure Separated Databases - Within H oneywell’s unique solution, separate databases store the safety and control strategies, and separate software modules are available for safety and control through dedicated tools such as Safety Builder and Control Builder. Maintaining separate tools with separate databases prevents unauthorized changes or corruptions, decreases safety risks and prevents common cause failures.∙Managed and Protected Database Environment - A unique, secure login scheme protects Safety Manager from off- and on-process changes. This login scheme uses a dedicated protection mechanism with several access levels for the engineering application, loading of the application in the controller and forcing points in Safety Manager. A user expiration mechanism downgrades the access level after auser-defined period of time elapses to protect the application from accidental or unauthorized changes when Safety Builder is unmanned over a specified period.∙Dedicated Software and Hardware - Using dedicated and specifically developed hardware and software in accordance with the IEC61508 safety standard reduces the risk of a common cause failure. Using dedicated hardware and software for both safety and control protects the safety system from any defects in control-related operations. In addition, the safety and control strategies are developed by different groups using dedicated methods.Conversely, using the same hardware or software for both safety and control increases the possibility of systematic controller failures, including those that result from design errors. A clear separation reduces the effort for testing and designing safety systems.∙Secure Environment - It is crucial that critical control and SIS are protected from intentional or accidental cyber threats. In general, functional security in combination with functional safety is critical to assessing the overall integrity of a SIS. Safety Manager architecture is secure by design and is certified to the Embedded Device Security Assurance program as defined by the ISA Security Compliance Institute. Adherence to this standard provides assurance of safety, security and robustness, meeting stringent industry best practices and performance benchmarks.In addition, Safety Manager is protected from outside threats by an embedded certified hardware firewall. This firewall isolates the safety application during runtime execution from external devices so they can never jeopardize the safety or availability of the application. With this firewall and the use of a SIL 4 certified proprietary protocol between safety managers, the data integrity between control and safety is protected and guaranteed.∙Safety Inside - Using dedicated firmware for safety and control ensures that safety is embedded into the system—no additional programming is needed to establish the required safety level. Strategies with a common platform for safety and control require that safety be built into the application. This customized safety level is a manual process and requires fundamental knowledge of the safety system to establish safety functions without jeopardizing the integrity of the application.Honeywell’s integrated control and safety solution is driven by the separation principle—hardware and software diversification, integrated operator interface, integrated data processing and analysis, and integrated alarm management.For More InformationTo l earn more about Honeywell’s Safety manager, visit our website or contact your Honeywell account manager. Honeywell Process Solutions Honeywell1250 West Sam Houston Parkway South Houston, TX 77042Honeywell House, Arlington Business Park Bracknell, Berkshire, England RG12 1EB UK Shanghai City Centre, 100 Junyi Road Shanghai, China 20051 The operational integration provided with Experion and Safety Manager allows plant personnel to have a seamless interface to the process while maintaining safe separation. This allows for a wide range of applications to be monitored plant-wide from any operator console. A complete overview of all information needed from the operator’s point of view is available through Experion Station or Experion Console Station. This communication architecture, supplied by Honeywell, delivers a scalable solution, from small control and safety networks to huge plant architectures with over 100,000 monitored I/O points. Interoperability of Safety Manager with the SafeNet protocol extends the functionality of one Safety Manager and allows for plant-wide implementation, binding the separate functionalities into one safety application with different protection layers.Engineering ExcellenceHoneywell’s Global Safety Discipline program enables consistent project execution excellence across Honeywell engineering locations. TÜV certified procedures and resources guarantee a global and transparent safety project execution by using certified standard builds, including templates, guidelines solution libraries, checklists, methodologies and tools.Safety Manager HMIWeb Solution Pack shapes and faceplates provide all projects with a highly flexible and functional library, enabling maximum advantage of the principles of safe and effective operations as described by the Abnormal Situation Management (ASM) Consortium.Honeywell Safety ServicesHoneywell’s offerings go beyond supplying hardware and software, establishing a unique safety knowledge community located in our expertise centers around the world in North America, Europe, South Africa, Asia and Australia.Over 200 certified safety engineers employed in these centers offer a wide range of consulting, project and lifecycle support services, including:∙Safety system audits∙Process hazard and risk assessment ∙SIL classification∙IEC61508 and IEC61511 CFSE training ∙Safety requirement specification development ∙FEED studies with customers to jointly develop their requirements∙IEC61508, IEC61511 and ISA S84 compliant solutions development∙Safety Instrumented Systems implementation ∙Live, hot cutover implementation and execution of revamp projects∙Installation and commissioning ∙SIL verification ∙SIL validation ∙Periodic proof-testing ∙System maintenance∙Solution Enhancement Support Program (SESP)∙Parts managementPN-12-25-ENG March 2013© 2012 Honeywell International Inc.Experion®, QMR® and UniSim® are registered trademarks of Honeywell International Inc.。
压缩气体钢瓶操作、储存和使用
Safetygram-10安全程序-10 Handling, Storage, and Use of Compressed Gas Cylinders压缩气体钢瓶操作、储存和使用General概述As a member of the chemical industry, Air Products is committed to the tenets of Responsible Care to help guide our performance with respect to environmental, health and safety performance, and the distribution and use of our products.作为化学工业的一员，Air Products忠于责任心的原则，该原则指导着我们在环境、健康、安全以及我们的产品的销售和使用方面的表现。
As our customer, you need to share in the responsibility for safe handling, storage, and use of our products. Follow these seven general safety recommendations:作为我们的顾客，你需要分担安全地操作、储存和使用我们的产品的责任。
请遵守这七项关于安全的建议：1. Know and understand the properties, uses, and safety precautions before using any gas or gas mixture. Consult the Air Products Material Safety Data Sheet (MSDS) and Safetygrams for safety information on the gases and equipment you will be using.在使用任何气体或气体混合物之前，要知道和理解性质、用途和安全防范。
材料安全数据表（MSDS）紧急联系电话
材料安全数据表(MSDS)第1部分产品概述产品名称：三氟化硼Boron Trifluoride化学名称：三氟化硼分子式：BF3代名词：三氟化硼三氟硼烷第2部分主要组成与性状BF3纯度> 99%CAS 号码: 7637-07-2暴露极限:OSHA :PEL=1ppm(ceiling) ACGIH:TWA/TLV=1ppm(ceiling) NIOSH：IDLH=25ppm第3部分危害概述紧急情况综述三氟化硼是一种有毒、腐蚀性、不可燃的气体，它与空气接触形成白色浓雾。
雾的密度与空气中的湿度大小有关。
它的密度比空气大，所以聚积在较低区域。
与水反应产生热并形成酸性的三氟化硼水合物。
它有令人厌恶的气味，通常在较低浓度下能被检测到。
当进入未知浓度区域或超过暴露极限区域时需配戴自给式呼吸器(SCBA)。
在大量泄漏时需穿戴全身防护服。
紧急联系电话0532－388 9090急性潜在健康影响暴露途径：眼接触：能引起严重的损伤（刺激和腐蚀）可能导致失明。
摄入：不适用吸入：三氟化硼腐蚀并刺激呼吸道及粘膜。
影响中枢神经系统，可能造成深度肺炎（化学性肺炎），出血（肺动脉出血），肺水肿和系统性副作用并可致命。
症状可能滞后出现。
皮肤接触：高浓度的三氟化硼引起脱水性灼伤，类似于酸灼伤。
其潜在的氟化氢形成可引起其它的组织损坏和系统性副作用而致命。
灼伤可能不会立即引起疼痛和被看到。
MSDS 三氟化硼1/6多次暴露的潜在健康影响：进入路径：吸入，皮肤或眼接触。
损害器官：眼睛、皮肤、呼吸道、肺、肾、肝、心脏、牙和骨骼。
症状：系统低水平的吸收三氟化硼会造成慢性支气管炎、哮喘和在骨骼、牙齿上有异常的氟化物积累（氟中毒）。
过份暴露造成的病状恶化：可加重哮喘、肺气肿或其他呼吸科疾病。
致癌性：三氟化硼未被 NTP、OSHA 及 IARC列为致癌物或潜在的致癌物。
第4部分急救措施任何暴露在三氟化硼中的情况都要立即治疗。
症状可能滞后出现。
眼接触：翻开眼睑立即用水冲洗直至有葡萄糖酸钙溶液可用。
CGA DISS
Safetygram-31安全程序-31 Cylinder Valve Outlet Connections钢瓶阀门排气口接头Introduction引言The Connections Standards Committee of the Compressed Gas Association (CGA) is responsible for assigning standard connections for specific gases and establishing detailed dimensions for the manufacture of these connections. The main purpose for establishing such standards is to prevent interconnection with non-compatible gases and to provide continuity among manufacturers. Assigned connections also prevent interconnection of the same gas at incompatible pressures.压缩气体协会（CGA）的接头标准委员会负责为特殊气体指定标准接头和为这些接头的制造确定详细的尺寸。
建立这些标准的主要目的是防止不相容气体的相互连接和在制造商之间提供连续性。
指定接头还防止了在不相容压力下同样气体的交叉连接。
There are four basic groups of valve outlet connections: (1) connections for general, industrial compressed gas service; (2) connections for self-contained breathing apparatus (SCBA) service; (3) connections for ultra-high-integrity service; and (4) pin-indexed connections for medical gas service. Although this Safetygram addresses connections only for industrial compressed gas service and ultra-high-integrity gas service, much of this information also applies to the other two groups.有四组基本的阀门排气口接头：(1) 用于普通的工业气体设施的接头；(2)用于自给式呼吸器（SCBA）装置的接头；(3)用于超高完整性装置的接头；和(4)用于医疗气体设施的指度针接头。
物流专用术语中英文翻译
物流专用术语物流基本概念术语1. 物品article2. 物流logistics3. 物流活动logistics activity4. 物流作业logistics operation5. 物流模数logistics modulus6. 物流技术logistics technology7. 物流成本logistics cost8. 物流管理logistics management9. 物流中心logistics center10. 物流网络logistics network11. 物流信息logistics information12. 物流企业logistics enterprise13. 物流单证logistics documents14. 物流联盟logistics alliance15. 供应物流supply logistics16. 生产物流production logistics17. 销售物流distribution logistics18. 回收物流returned logistics19. 废弃物物流waste material logistics20. 绿色物流environmental logistics21. 企业物流internal logistics22. 社会物流external logistics23. 军事物流military logistics24. 国际物流international logistics25. 第三方物流third-part logistics (TPL)26. 定制物流customized logistics27. 虚拟物流virtual logistics28. 增值物流服务value-added logistics service29. 供应链supply chain30. 条码bar code31. 电子数据交换electronic data interchange (EDI)32. 有形消耗tangible loss33. 无形消耗intangible loss物流作业术语1. 运输transportation2. 联合运输combined transport3. 直达运输through transport4. 中转运输transfer transport5. 甩挂运输drop and pull transport6. 集装运输containerized transport7. 集装箱运输container transport8. 门到门door-to-door9. 整箱货full container load (FCL)10. 拼箱货less than container load (LCL11. 储存storing12. 保管storage13. 物品储存article reserves14. 库存inventory15. 经常库存cycle stock16. 安全库存safety stick17. 库存周期inventory cycle time18. 前置期(或提前期) lead time19. 订货处理周期order cycle time20. 货垛goods stack21. 堆码stacking22. 搬运handing/carrying23. 装卸loading and unloading24. 单元装卸unit loading and unloading25. 包装package/packaging26. 销售包装sales package27. 定牌包装packing of nominated brand28. 中性包装neutral packing29. 运输包装transport package30. 托盘包装palletizing31. 集装化containerization32. 散装化containerization33. 直接换装cross docking34. 配送distribution35. 共同配送joint distribution36. 配送中心distribution center37. 分拣sorting38. 拣选order picking39. 集货goods collection40. 组配assembly41. 流通加工distribution processing42. 冷链cold chain43. 检验inspection物流技术装备及设施术语1. 仓库warehouse2. 库房storehouse3. 自动化仓库automatic warehouse4.4. 立体仓库stereoscopic warehouse5. 虚拟仓库virtual warehouse6. 保税仓库boned warehouse7. 出口监管仓库export supervised warehouse8. 海关监管货物cargo under customer’s supervision9. 冷藏区chill space10. 冷冻区freeze space11. 控湿储存区humidity controlled space12. 温度可控区temperature controlled space13. 收货区receiving space14. 发货区shipping space15. 料棚goods shed16. 货场goods yard17. 货架goods shelf18. 托盘pallet19. 叉车fork lift truck20. 输送机conveyor21. 自动导引车automatic guided vehicle (AGV)22. 箱式车box car23. 集装箱container24. 换算箱twenty-feet equivalent unit (TEU)25. 特种货物集装箱specific cargo container26. 全集装箱船full container ship27. 铁路集装箱场railway container yard28. 公路集装箱中转站inland container depot29. 集装箱货运站container freight station (CFS)30. 集装箱码头container terminal31. 国际铁路联运international through railway transport32. 国际多式联运international multimodal transport33. 大陆桥运输land bridge transport34. 班轮运输liner transport35. 租船运输shipping by chartering36. 船务代理shipping agency37. 国际货运代理international freight forwarding agent38. 理货tally39. 国际货物运输保险international transportation cargo insurance40. 报关customs declaration41. 报关行customs broker42. 进出口商品检验commodity inspection 奖罚物流管理术语1. 物流战略logistics strategy2. 物流战略管理logistics strategy management3. 仓库管理warehouse management4. 仓库布局warehouse layout5. 库存控制inventory control6. 经济订货批量economic order quantity (EOQ)7. 定量订货方式fixed-quantity system (FQS)8. 定期订货方式fixed-quantity system (FIS)9. ABC分类管理ABC classification10. 电子订货系统Electronic order system (EOS)11. 准时制just in time (JIT)12. 准时制物流just-in-time logistics13. 零库存技术zero-inventory logistics14. 物流成本管理logistics cost control15. 物料需要计划material requirements planning (MRP)16. 制造资源计划manufacturing resource planning (MRP II)17. 配送需要计划distribution requirements planning (DRP)18. 配送资源计划distribution resource planning (DRP II)19. 物流资源计划logistics resource planning (LRP)20. 企业资源计划enterprise resource planning (ERP)21. 供应链管理supply chain management (SCM)22. 快速反映Quick response (QR)23. 有效客户反映efficient customer response(ECR)24. 连续库存补充计划continuous replenishment program (CRP)25. 计算机付诸订货系统computer assisted ordering (CAO)26. 供应商管理库存vendor managed inventory (VMI)27. 业务外包outsourcing度量单位汇总克 Gram g、公斤 Kilogram Kg、公担 Quintal q、公吨 Metric ton m、t、长吨 Long ton l、t、短吨 Short ton Sh、t、英担 Hundredweight Cwt、美担 Hundredweight cwt、磅 Pound lb、两(常衡) Ounce oz、两(金衡) Ounce oz、t司马担 Picul米 Metre m、公里 Kilometre Km、厘米 Centimetre cm、毫米 Milimerte mm、码 Yard Yd、英尺 Foot Ft、英寸 Inch In、平方米 Square metre Sq、m、平方英尺 Square foot Sq、ft、平方码 Square yard Sq、yd、立方米 Cubic metre Cu、m、立方英尺 Cubic foot Cu、ft、升 Litre l、毫升 Millilitre Ml、加仑 Gallon Gal、蒲式耳 Bushel Bu、克拉 Carat Car、马力 Horse Power h、p、千瓦 Kilowatt Kw、公吨度 Metric ton unit m、t、u、表 6 :常见英文单位Bag 袋 bale 包 bottle 瓶Box 箱 carton/ctn 纸盒箱 Case(c/s) 箱、合、套coil 圈 Container 集装箱 Crate 板条箱Dozen 打 gross 罗(12打) drum 桶In bulk 散装 lot 批 package 件pallet 托盘 Pieces/pcs 件 Ream 令roll 卷 set 套、台、坐 Sheet 张、件strand 股 unit 辆、台、单位 Vial 药瓶专有名词汇总出口信贷export credit出口津贴export subsidy商品倾销dumping外汇倾销exchange dumping优惠关税special preferences保税仓库bonded warehouse贸易顺差favorable balance of trade贸易逆差unfavorable balance of trade进口配额制import quotas自由贸易区free trade zone对外贸易值value of foreign trade国际贸易值value of international trade普遍优惠制generalized system of preferences-GSP 最惠国待遇most-favored nation treatment-MFNT 价格术语trade term (price term)运费freight单价price码头费wharfage总值total value卸货费landing charges金额amount关税customs duty净价net price印花税stamp duty含佣价price including commission港口税port dues回佣return commission 、装运港port of shipment折扣discount, allowance卸货港port of discharge批发价wholesale price目的港port of destination零售价retail price进口许口证import licence现货价格spot price出口许口证export licence期货价格forward price现行价格(时价)current price prevailing price国际市场价格world (International)Market price离岸价(船上交货价)FOB-free on board成本加运费价(离岸加运费价) C&F-cost and freight到岸价(成本加运费、保险费价)CIF-cost,insurance and freight交货delivery轮船steamship(缩写S、S)装运、装船shipment租船charter (the chartered ship)交货时间time of delivery定程租船voyage charter装运期限time of shipment定期租船time charter托运人(一般指出口商)shipper, consignor收货人consignee班轮regular shipping liner驳船lighter舱位shipping space油轮tanker报关clearance of goods陆运收据cargo receipt提货to take delivery of goods空运提单airway bill正本提单original B\\L选择港(任意港)optional port选港费optional charges选港费由买方负担optional charges to be borne by the Buyers 或optional charges for Buyers’ account一月份装船shipment during January 或January shipment一月底装船shipment not later than Jan、31st、或shipment on orbefore Jan、31st、一/二月份装船shipment during Jan、/Feb、或Jan、/Feb、shipment 在、、、、、、(时间)分两批装船shipment during、、、、in two lots 在、、、、、、(时间)平均分两批装船shipment during、、、、in two equal lots 分三个月装运in three monthly shipments分三个月,每月平均装运in three equal monthly shipments立即装运immediate shipments即期装运prompt shipments收到信用证后30天内装运shipments within 30 days after receipt of L/C 允许分批装船partial shipment not allowed partial shipment not permitted partial shipment not unacceptable订单indent订货;订购book; booking电复cable reply实盘firm offer递盘bid; bidding递实盘bid firm还盘counter offer发盘(发价) offer发实盘offer firm询盘(询价) inquiry;enquiry指示性价格price indication速复reply immediately参考价reference price习惯做法usual practice交易磋商business negotiation不受约束without engagement业务洽谈business discussion限**复subject to reply **限* *复到subject to reply reaching here **有效期限time of validity购货合同purchase contract销售合同sales contract购货确认书purchase confirmation销售确认书sales confirmation一般交易条件general terms and conditions以未售出为准subject to prior sale需经卖方确认subject to seller’s confirmation需经我方最后确认subject to our final confirmationINT (拍卖auction)寄售consignment招标invitation of tender投标submission of tender一般代理人agent总代理人general agent代理协议agency agreement累计佣金accumulative commission补偿贸易compensation trade (或抵偿贸易)compensating/compensatorytrade(又叫:往返贸易) counter trade来料加工processing on giving materials来料装配assembling on provided parts独家经营/专营权exclusive right独家经营/包销/代理协议exclusivity agreement独家代理sole agency; sole agent; exclusive agency;exclusive agent轮船steamship(缩写S、S)装运、装船shipment租船charter(the chartered shep)交货时间time of delivery定程租船voyage charter;装运期限time of shipment定期租船time charter托运人(一般指出口商)shipper,consignor收货人consignee班轮regular shipping liner驳船lighter舱位shipping space油轮tanker报关clearance of goods陆运收据cargo receipt提货to take delivery of goods空运提单airway bill正本提单original B\L选择港(任意港)optional port选港费optional charges选港费由买方负担optional charges to be borne by the Buyers或optional charges for Buyers' account一月份装船shipment during January 或January shipment一月底装船shipment not later than Jan、31st、或shipment on or before Jan、31st、一/二月份装船shipment during Jan、/Feb、或Jan、/Feb、shipment在……(时间)分两批装船shipment during、、、、in two lots在…(时间)平均分两批装船shipment during、、、、in two equal lots分三个月装运in three monthly shipments分三个月,每月平均装运in three equal monthly shipments立即装运immediate shipments即期装运prompt shipments收到信用证后30天内装运shipments within 30 days after receipt of L/C允许分批装船partial shipment not allowed partial shipment not permitted partial shipment not unacceptable运费freight单价price码头费wharfage总值total value卸货费landing charges金额amount关税customs duty净价net price印花税stamp duty含佣价price including commission港口税portdues回佣return commission装运港portof shipment折扣discount,allowance卸货港port of discharge批发价wholesale price目的港portof destination零售价retail price进口许口证inportlicence现货价格spot price出口许口证exportlicence期货价格forward price现行价格(时价)current price国际市场价格world (International)Marketprice离岸价(船上交货价)FOB-free on board成本加运费价(离岸加运费价) C&F-cost and freight到岸价(成本加运费、保险费价)CIF-cost,insurance and freight。
- 1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。
Safetygram-31安全程序-31 Cylinder Valve Outlet Connections钢瓶阀门排气口接头Introduction引言The Connections Standards Committee of the Compressed Gas Association (CGA) is responsible for assigning standard connections for specific gases and establishing detailed dimensions for the manufacture of these connections. The main purpose for establishing such standards is to prevent interconnection with non-compatible gases and to provide continuity among manufacturers. Assigned connections also prevent interconnection of th e same gas at incompatible pressures.压缩气体协会（CGA）的接头标准委员会负责为特殊气体指定标准接头和为这些接头的制造确定详细的尺寸。
建立这些标准的主要目的是防止不相容气体的相互连接和在制造商之间提供连续性。
指定接头还防止了在不相容压力下同样气体的交叉连接。
There are four basic groups of valve outlet connections: (1) connections for general, industrial compressed gas service; (2) connections for self-contained breathing apparatus (SCBA) service; (3) connections for ultra-high-integrity service; and (4) pin-indexed connections for medical gas service. Although this Safetygram addresses connections only for industrial compressed gas service and ultra-high-integrity gas service, much of this information also applies to the other two groups.有四组基本的阀门排气口接头：(1) 用于普通的工业气体设施的接头；(2)用于自给式呼吸器（SCBA）装置的接头；(3)用于超高完整性装置的接头；和(4)用于医疗气体设施的指度针接头。
尽管本安全程序只说明用于工业压缩气体设施和超高纯度气体设施的接头，这些资料中的许多也适用于其它两组接头。
In North America, outlet connections are usually designated by a three-digit number preceded by the letters CGA, the acronym for the Compressed Gas Association, for example, CGA 350. Sometimes an ultra-high-integrity connection is preceded by a ―DISS‖ designation rather than the more common CGA designation. DISS is the acronym for Diameter Index Safety System. 在北美，排气口接头通常由在字母CGA之后的三位数字指定名称，例如CGA 350。
CGA是压缩气体协会的只取首字母的缩写。
有时候超高完整性接头的指定名称的前面是“DISS”，而不是更常用的CGA。
DISS是直径指数安全系统的只取首字母的缩写。
Typical Connections典型的接头A typical connection consists of three or four parts depending on whether it is a bullet-nose or a gasketed connection. A bullet-nose connection consists of a valve outlet, a nut, and a nipple. The nut is placed on the nipple so the shoulder of the nipple rests against the pushing surface of the nut. The nut has straight threads that engage the mating threads on the valve outlet and pull the nipple against the sealing surface of the valve outlet. No sealing takes place at the threads. The gas-tight seal takes place between the nipple and the valve outlet seat at a verysmall contact circle where they touch. The success of this connection depends on the surface condition of both the nipple and the valve outlet sealing area at the point of contact.根据它是弹头接头还是衬垫接头，一个典型的接头由三个或四个零件组成。
一个弹头接头由一个阀门排气口、一个螺母和一个管接口组成。
螺母放在管接口的上面，因而管接口靠在螺母的螺纹正面。
螺母有直螺纹，直螺纹咬合阀门排气口上的配合螺纹，拉动管接口靠在阀门排气口的密封面上。
在螺纹上不形成密封。
不漏气的密封形成在管接口和阀门排气口座之间，在它们接触的一个非常小的接触圆周处。
这种接头的成功依赖于管接口和阀门排气口密封在接触点的表面状况。
Some bullet-nose connections have a soft tipped nipple or an O-ring on the nipple to improve the seal. This allows the connection to be made without a wrench, using a nut equipped with a hand wheel. These connections are called hand-tight connections.一些弹头接头具有一个软头管接口或在管接口上有一个O形圈，以改善密封。
这使得不用扳手，而是使用带手轮的螺母就可以进行连接。
这些接头称为手紧接头。
A gasketed connection has four parts: the valve outlet, a nipple, a nut, and a gasket (or washer).The nipple of the gasketed connection is not bullet-shaped as in the bullet-nose connections, but is flat to compress the gasket. The valve outlet sealing area is also flat. The nut fits on the nipple so the shoulder of the nipple rests on the pushing surface of the nut. The washer is placed on the flat surface of the nipple inside the nut. The straight threads of the nut engage with the mating threads of the valve outlet and are tightened to compress the gasket between the two sealing surfaces.一个衬垫接头有四个零件：一个阀门排气口、一个管接口、一个螺母和一个衬垫（或垫圈）。
衬垫接头的管接口不是象弹头接头那样的子弹形，而是平的，以便压缩衬垫。
阀门排气口密封区域也是平的。
螺母配合在管接口上，因而管接口靠在螺母的螺纹正面。
垫圈放在螺母内的管接口的平面上。
螺母的直螺纹同阀门排气口的配合螺纹咬合在一起并上紧，来压缩在两个密封表面之间的衬垫。
Standard industrial valve outlet connections are designed to provide a ―bubble-tigh t‖ connection, which refers to leak-checking the connection with a soap solution or by immersing it in water. Bubbles indicate a leak. Leak detection solutions vary in their ability to detect leaks. However, leakage rates for standard connections range from 1 x 10-3 to 1 x 10-5cc of helium per second. Slower leak rates can be obtained with these connections depending on surface finishes, gasket materials, and mechanical condition. However, achieving slower leak rates can be difficult and inconsistent, one of the primary reasons why the ultra-high-integrity outlet connections were developed.标准工业阀门排气口接头设计得可以提供一种“不冒泡”接头。