Earthquake Reaearch in China  2017, Vol. 31 Issue (1): 1-11
Review and Prospects of the Global Earthquake Model Project (GEM)
Li Changlong1,2, Wu Jian1,2, Xu Weijin1,2, Gao Mengtan1,2     
1 Institute of Geophysics, China Earthquake Administration, Beijing 100081, China;
2 China Earthquake Risk and Insurance Laboratory, Beijing 100081, China
Abstract: This paper reviewed the main target, functions, tool (OpenQuake software) and research achievements of the Global Earthquake Model (GEM) Foundation, and made a simple prospect on the development and application of projects of GEM in the future. Learning from GEM and OpenQuake is helpful to improve the seismic hazard model of China and enhance the scientificity of the seismic hazard assessment for metropolitans and major engineering facilities near major seismogenic structures.
Key words: Global Earthquake Model (GEM)     Probabilistic seismic hazard assessment (PSHA)     Seismic risk assessment     OpenQuake virtual machine    

INTRODUCTION

Strong earthquake ground motion can cause destruction to buildings and lead to earthquake disasters. It is of great help to preventing earthquake disasters by providing appropriate seismic fortification parameters for structural design. Therefore, seismic hazard and risk assessment has important practical and social value.

Since Probabilistic Seismic Hazard Analysis (PSHA) was proposed by Cornell (1968), PSHA has been carried out on a global scale, and most parts of the world have formed preliminary conclusions of seismic hazard (Giardini et al., 1999). With the development of modern society, the concept of Seismic Risk is being used more and more frequently. Seismic hazards, vulnerability and exposure of assets are three factors that influence earthquake risk. Therefore, as compared with seismic hazard, the concept of seismic risk better reflects the relationship between earthquakes and society, economic activities and human life.

In western developed countries, PSHA results with high confidence and building structures reinforced by high-standard criteria have been able to reduce the earthquake disaster to a low level. However, in less developed countries and regions, due to the poor seismic hazard research and the weak seismic performance of building facilities, the tragedy of the "catastrophe caused by common earthquakes" still occurs. A typical example is the M7.3 earthquake in 2010 in the Haitian capital Port-au-Prince, resulting in more than 300, 000 casualties. Therefore, there is an urgent need to promote seismic hazard and risk assessment on a global scale. Under this background, the Global Earthquake Model (GEM) Foundation came into being. It is committed to addressing the global seismic risk assessment issues, and to providing scientific support for global social development.

In recent years, the socio-economic development of some developing countries has been fast, with rapid population growth and high concentration in large cities. Many of these countries have had serious earthquake disasters, such as China and some Southeast Asian countries. These countries and regions are weak in seismic fortification and have high vulnerability. A comprehensive assessment of the earthquake risk in these areas is of great importance and urgency. Therefore, using the platform provided by GEM to carry out seismic risk assessment is a feasible method, and the first step to do it is to understand the status quo of GEM research in recent years and to grasp the earthquake risk assessment tools developed by it.

This paper reviews the objectives, main functions, tools (OpenQuake software) and research achievements of GEM, and forecast the developing trend of GEM and OpenQuake in the future.

1 INTRODUCTION OF GEM

Founded in 2009, GEM is a global association of seismic hazard analysis and seismic risk assessment, headquartered in Pavia, Italy. Kate Stillwell is currently the head of the GEM Science Department. Since its foundation, GEM has worked with global scientists on global seismic risk assessment (Danciu et al., 2010; Crowley et al., 2010). GEM has four major work areas (GEM, 2013) :

(1) Develop and support high-performance seismic risk assessment tools.

(2) Collect and update seismic hazard data (data, models, methods and criterions).

(3) Develop and apply collected seismic risk data in various projects.

(4) Technological transformation and capacity upgrading.

Compared with the previous seismic hazard analysis, GEM has made two major advances. First, GEM developed the powerful seismic hazard and risk assessment software, OpenQuake, to describe more complex seismic hazard models for a wide variety of seismic tectonics worldwide. Second, GEM allows the establishment of a complete and accurate model of vulnerability and the scale of the affected to calculate scientifically the seismic risk, thus, providing a convenient method for seismic risk assessment and promoting the development of seismo-sociology.

GEM provides a good platform for cooperation, communication, learning, and resource sharing for the

world's seismologists. As of 2014, earthquake experts from more than 100 countries on six continents have been involved in GEM's work. GEM has completed PSHA research work in the western Eurasia Continent (Europe, West and Central Asia) (Danciu et al., 2015). At the same time, OpenQuake software codes continue to be updated by the world's scientists, and become increasingly powerful. Advances in OpenQuake software and GEM research are detailed as below.

2 OPENQUAKE, THE SOFTWARE FOR SEISMIC HAZARD AND RISK ASSESSMENT

In order to meet the high accuracy, high reliability and high adaptability of earthquake risk assessment, GEM has developed OpenQuake as its main product and tool. It is an open-source software for seismic hazard and risk assessment. A complete set of OpenQuake installation files, user manuals and source codes can be downloaded from GEM's official website at http://www.globalquakemodel.org/. It runs on Python and Java language (Silva et al., 2012) and operates on Linux virtual machines. A variety of seismic hazard and risk assessment models, (Stewart et al., 2012a; Pagani et al., 2015), sophisticated algorithms and processes (Silva et al., 2012) and user-friendly user interfaces (Weatherill, 2012) enable it to perform many seismic risk assessment functions.

2.1 Main Functions of OpenQuake

There are three categories of seismic risk calculations that can be performed by OpenQuake (Silva et al., 2014a):

(1) Scenario-based Earthquake Risk Calculation that calculates the total loss of various assets (e.g. buildings, population) caused by a single earthquake (deterministic scenario).

(2) Event-based risk calculation that calculates the probability of a loss caused by a single random earthquake event exceeding a certain level over a certain period of time.

(3) PSHA based seismic risk calculation for calculating the probability of single asset loss according to the seismic hazard curve obtained by PSHA.

The output of OpenQuake includes (Silva et al., 2012) :

(1) Hazard curve: The relationship between the values of ground motion parameters and the probability of exceedance in a given period.

(2) Hazard map: The map describing the distribution of ground motion values with a particular probability of exceedance over a given time period.

(3) Random earthquake event set: A set of seismicity (occurring within a given time period) predicted by a Poisson seismicity model.

(4) Ground motion field: Each ground motion field describes the distribution of ground motion parameters calculated from a set of earthquake rupture and ground motion prediction equation (GMPE).

(5) Disaggregation of hazard parameters: The contributions of seismic sources with different magnitudes and locations to seismic hazard are given.

(6) Loss curve: The relationship between loss and the probability of exceedance in a given period of time. It can describe the loss of individual assets or the total loss of various assets.

(7) Loss map: A map of loss distribution describing a specific probability of exceedance in a given time period.

(8) Average loss map: A map describing the average loss over a given period of time (e.g. annual average loss).

(9) Map of collapsed building: A map describing the distribution of percentage and number of buildings collapsed.

(10) Loss statistics: The loss of various statistical indicators (such as the average loss, the standard deviation of the loss and the average loss distribution.).

(11) Revenue-cost ratio: The ratio of annual average earthquake losses before and after mitigation measures.

OpenQuake's functions depend on its model and its operational processes. In OpenQuake, the PSHA models and algorithms are part of the seismic risk assessment model and are the basis of earthquake risk assessment.

2.2 Openquake's Seismic Hazard and Risk Assessment Model 2.2.1 OpenQuake's PSHA Model

OpenQuake's PSHA model includes the source model and GMPE model. It supports the creation of area and fault models. The location and shape of the source are determined by the coordinates of the input source boundary points. The seismicity of the source can be determined by setting the a and b values in the G-R relationship (Gutenberg et al., 1944) or the annual occurrence rate of each magnitude interval.

OpenQuake includes a variety of GMPE models (Stewart et al., 2012a; Akkar et al., 2012). The range of use includes all regions of the world, a variety of source types and a variety of distance attenuation parameters. The users can select the appropriate GMPE model as needed. However, the current version of OpenQuake does not support users editing and inputting the GMPE model themselves (Crowley et al., 2012). In addition, the site condition model (Stewart et al., 2012b) is also considered in OpenQuake.

2.2.2 OpenQuake' Seismic Risk Assessment Model

OpenQuake' seismic risk assessment model includes seismic hazard, vulnerability, exposure, and ground-motion field model. The seismic hazard model is derived from the PSHA model. The vulnerability model describes the loss rate distribution under a given intensity. The exposure model includes type, location and taxonomy of assets. The ground motion field model represents a random variation of GMPE (including for the same earthquake and between different earthquakes).

2.3 OpenQuake's Algorithm of Seismic Hazard and Risk Assessment 2.3.1 OpenQuake's Algorithm of PSHA

OpenQuake's PSHA algorithm is divided into three parts (Monelli et al., 2012) :

(1) Logic tree processor (LTP)

LTP inputs the PSHA input model and generates the source model. LTP uses the initial source model information, and takes into account the source logic tree, which is a different model scheme designed due to the uncertainty of the source model's knowledge, to generate the source model. After that, LTP generates a ground motion model.

(2) Earthquake rupture forecast (ERF) calculator

The generated source model is then called the ERF calculator. The calculator calculates the rate of occurrence of each earthquake rupture in a particular time frame for the source model.

(3) Classic PSHA calculator

Classic PSHA calculator uses ERF and ground motion models to calculate the hazard profile for each site.

Fig. 1 Flow chart of PSHA by OpenQuake (Crowley et al., 2011)
2.3.2 OpenQuake's Algorithm of Seismic Risk Assessment

OpenQuake' seismic risk assessment algorithm is divided into four types (Silva et al., 2012) :

(1) Scenario risk algorithm

Scenario risk algorithm calculates the deterministic loss of a single earthquake. The algorithm inputs an earthquake rupture and a GMPE to calculate the ground motion field. Other inputs include the vulnerability model and exposure model. The outputs are loss maps and loss statistics.

(2) Event-based probabilistic risk algorithm

This algorithm is based on the probability of loss and the losses calculated from probabilistic seismic hazard calculations. The loss is calculated by setting a single seismic event, and the time of occurrence of the earthquake is random. The loss curves and loss maps can be calculated for different return periods, and the damage distribution of each asset and total assets can be output.

(3) PSHA based risk algorithm

This algorithm uses the risk curve of probabilistic seismic hazard to calculate the loss exceedance curve of each asset. The probability of a single asset losing more than a certain percentage at a certain location is obtained, and the probability loss map is output. Unlike other algorithms, this algorithm cannot output the total loss curve because the residuals and the vulnerability of the ground motion field are not taken into account.

(4) Ratio of profit/cost algorithm

This algorithm is used to calculate whether the improvement measures of an existing building bring benefits from an economic point of view. It provides the expected loss estimate before and after the measures, and can calculate the ratio between the economic benefits and the cost after the measures to determine the transformation plan of the building.

All four kinds of algorithms are made up of several calculation modules. The calculation area, algorithm type, input path, and output type also need to be defined in the calculation. The four algorithms require the input of the vulnerability and the exposure model, and the calculation flow charts have a similar form, as shown in Fig. 2.

Fig. 2 Flow chart of seismic risk assessment by OpenQuake
3 GEM'S DEVELOPMENT IN RECENT YEARS

In recent years, GEM has made significant progress with the use of OpenQuake, including PSHA, global seismic databases, and building taxonomy around the world, covering all aspects of seismic risk assessment.

3.1 PSHA in Europe

From 2009 to 2013, the 2013 European Seismic Hazard Model (Giardini et al., 2013; 2014; Woessner et al., 2015) was established with the Seismic Hazard Harmonization in Europe (SHARE) Project. The model crosses national borders and integrates seismic hazard data from different European countries such as Switzerland (Schwarz, 2015), Portugal (Silva et al., 2014b), Belgium (Kris et al., 2014) and Turkey.

SHARE has six aspects of improvements compared to the previous PSHA work in Europe. (1) A unified PSHA database was developed and updated. (2) With multi-party cooperation, the results of hundreds of European experts are used. (3) Multi-disciplinary work in earthquake engineering is included. (4) Research results are directly used by the relevant European architectural design criterion. (5) The uncertainties of the model parameters and hazard results are considered. (6) All data, results and methods are fully open to public. SHARE derives a map of peak ground acceleration (PGA) with 10% probabilities of exceedance in 50 years for Europe as shown in Fig. 3.

Fig. 3 PGA map with 10% probability of exceedance in 50 years in Europe (Woessner et al., 2015)
3.2 Seismic Hazard and Risk Assessment in Central Asia

Bindi et al. (2012) conducted a PSHA study in Central Asia. In this study, the area source model is used. The seismicity model only considers shallow earthquakes with focal depth less than 50 km. Seismic hazard parameters use macro-seismic intensity. The study used OpenQuake to produce an intensity distribution with 10% probability of exceedance in 50 years in Central Asia (Fig. 4). The results show that the intensity of Ⅷ in the region with the highest seismic hazard in Central Asia is less than 475 years in the vicinity of the South Tienshan area.

Fig. 4 Intensity map with 10% probability of exceedance in 50 years in Central Asia (Bindi et al., 2015)

Pittore et al. (2011) studied the seismic risk in Central Asia. The study, which combines the distribution of seismic hazard and population distribution in Central Asia, assesses the risk of casualties in the region over the next 50 years and draws an estimate of 10% probability of exceedance of the population's casualty over the next 50 years in Central Asia (Fig. 5). Based on the PSHA study of Central Asia, Wieland et al. (2015) studied the exposure model in this region. The study unifies the exposure models of the five Central Asian countries and builds the building model according to GEM's building taxonomy. The results highlight the multi-scale and time-dependent dynamics of the exposure model.

Fig. 5 Expected casualties map with 10% probability of exceedance in 50 years in Central Asia (Pittore et al., 2011)
3.3 Other Research Results

In other parts of the world, GEM research is also widely carried out. Benito et al. (2012) conducted a PSHA study in Central America. The study introduced a new source region partitioning and GMPE applied to various tectonic types in Central America, and derived PGA and spectral acceleration (Sa) distributions for 500 and 2500 years in the region. Danciu et al. (2015) conducted a PSHA study of the Middle East, weighted the average of the fault source model and area source model, and produced a 10% earthquake risk map of the region over 50 years. In addition, seismic hazard and risk assessment in Sub-Saharan Africa (Ayele et al., 2013) is ongoing.

In the GEM database construction, Albini et al. (2013) established a global historical earthquake record of 1000-1903. GEM, in collaboration with the International Seismological Center (ISC), established the Global Earthquake Catalog for 1900-2009 (Michael, 2014). Pagani et al. (2015) established a seismic hazard database. Yepes-Estrada et al. (2014) developed a vulnerability database. Christophersen et al. (2015) established a new tectonic fault database. In addition, the OpenQuake program updates and function extensions have been in progress, and the latest development has been published in the GEM official website.

4 APPLICATION PROSPECTS OF GEM AND OPENQUAKE IN CHINA

In China, seismogenic faults are widely-distributed, with various types and complex forms. At present, the PSHA work in China considers all seismogenic structures of the whole country as the area source model (Zhou Bengang et al., 2013). Using the OpenQuake algorithm, we can further develop the seismic hazard model of China, and more scientifically and rationally consider the three-dimensional fault structure and source characteristics to improve the scientific nature of seismic hazard analysis in the near-field region. Using the seismic risk assessment algorithm of OpenQuake, we can further improve the existing earthquake risk assessment model in China.

In accordance with GEM's work progress, PSHA work has been done using OpenQuake in Europe, West Asia and Central Asia. If OpenQuake can be widely used in China, it can not only expand GEM's global model, but also help China to carry out earthquake disaster reduction work.

In recent years, with the rapid urbanization advance, China's population and social wealth have a strong trend to concentrate in the city cluster. However, the distribution of some city clusters, such as Beijing-Tianjin-Hebei, North China Plain, Hohhot-Baotou, Ningxia Plain, Shanxi and Guanzhong, coincides highly with the high seismic hazard region. In South China, dozens of dams in the Qinghai-Tibetan Plateau and the western Sichuan Plateau are located in high seismic hazard areas, posing a threat to densely populated people living in the middle and lower reaches of the Yangtze River. Therefore, the use of more scientific and reasonable models to carry out seismic risk assessment in large cities and major engineering facilities is of great significance.

GEM provides theoretical support and powerful tools to help the government in seismic risk assessment and disaster scenarios establishment. Using various earthquake risk models in OpenQuake, it is possible to finely set various parameters of an earthquake disaster, and to construct a precise, reliable, concrete and intuitive earthquake disaster scenario, and to provide information for planning and disaster preparedness of the government.

Earthquakes may cause serious economic and property losses, thus the earthquake insurance industry came into being. At present in the United States and other developed countries, earthquake insurance has achieved commercialization and market-oriented management, and its business volume is growing rapidly. At present, the earthquake insurance industry has begun to appear in China. In cooperation with GEM, the establishment of China's earthquake risk model will also contribute to the earthquake insurance business.

5 SUMMARY

In this paper, the business scope of the GEM Foundation, the development tool OpenQuake, its recent research progress and future development prospects are briefly introduced. Seismic hazard analysis in developing countries requires innovative models and ideas. Taking advanced international experience as a reference will contribute to the progress of seismic hazard and risk assessment.

This paper has been published in Chinese in the Journal of Technology for Earthquake Disaster Prevention, Volume 11, Number 3, 2016.

REFERENCES
Akkar S. , Douglas J. , Di Alessandro C. , et al. Defining a Consistent Strategy to Model Ground-motion Parameters for the GEM-PEER Global GMPEs Project[C]. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012. Paper No. 2743.
Albini P. , Musson R. M. W. , Gomez Capera A. A. , et al. Global Historical Earthquake Archive and Catalogue (1000-1903)[R]. GEM Technical Report 2013.
Ayele A., Midzi V., Ateba B., et al. Earthquake hazard and risk in Sub-Saharan Africa:current status of the Global Earthquake Model (GEM) initiative in the region[J]. Geophysical Research Abstracts, 2013, 15: 1181.
Benito M.B., Lindholm C., Camacho E., et al. A new evaluation of seismic hazard for the Central America region[J]. Bulletin of the Seismological Society of America, 2012, 102(2): 504–523. DOI:10.1785/0120110015.
Bindi D., Abdrakhmatov K., Parolai S., et al. Seismic hazard assessment in Central Asia:Outcomes from a site approach[J]. Soil Dynamics and Earthquake Engineering, 2012, 37: 84–91. DOI:10.1016/j.soildyn.2012.01.016.
Christophersen A., Litchfield N., Berryman K., et al. Development of the Global Earthquake Model's neotectonic fault database[J]. Natural Hazards, 2015, 79(1): 111–135. DOI:10.1007/s11069-015-1831-6.
Cornell C.A. Engineering seismic risk analysis[J]. Bulletin of the Seismological Society of America, 1968, 58(5): 1583–1606.
Crowley H. , Colombi M. , Crempien J. , et al. GEM1 Seismic Risk Report: Part 1, GEM Technical Report 2010-5[R]. GEM Foundation, Pavia, Italy, 2010.
Crowley H. , Monelli D. , Pagani M, et al. OpenQuake Book Version 0. 1[R]. GEM Foundation, 2011.
Crowley H. , Monelli D. , Pagani M, et al. OpenQuake User's Manual Version 1. 1[R]. GEM Foundation, 2012.
Danciu L., Giardini D. Global Seismic Hazard Assessment Program-GSHAP legacy[J]. Annals of Geophysics, 2015, 58(1): S0109.
Danciu L. , Giardini D. , Sesetyan K. Seismic Hazard Assessment in the Middle East Region[R]. Earthquake Model of the Middle East Region Project, 2015.
Danciu L. , Monelli D. , Pagani M. , et al. GEM1 Hazard: Review of PSHA Software, GEM Technical Report 2010-2[R]. GEM Foundation, Pavia, Italy, 2010.
GEM. GEM Vision: Celebrating achievements and looking forward, 2013.
Giardini D., Gruenthal G., Shedlock K.M., et al. The GSHAP global seismic hazard map[J]. Annals of Geophysics, 1999, 42(6): 1225–1230.
Giardini D., Woessner J., Danciu L., et al. Mapping Europe's seismic hazard[J]. EOS, Transactions, American Geophysical Union, 2014, 95(29): 261–268.
Giardini D. , Woessner J. , Danciu L. , et al. Seismic Hazard Harmonization in Europe (SHARE): Online Data Resource, 2013.
Gutenberg B., Richter C.F. Frequency of earthquakes in California[J]. Bulletin of the Seismological Society of America, 1944, 34(4): 185–188.
Kris V. , Bart V. , Koen V. , et al. Development of Seismic Hazard Maps for Belgium[R]. Workshop: Results of the European Project, 2014.
Michael A.J. How complete is the ISC-GEM global earthquake catalog?[J]. Bulletin of the Seismological Society of America, 2014, 104(4): 1829–1837. DOI:10.1785/0120130227.
Monelli D. , Pagani M. , Weatherill G. , et al. The Hazard Component of OpenQuake: the Calculation Engine of the Global Earthquake Model[C]. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012. Paper No. 4180.
Pagani M., Garcia J., Monelli D., et al. A summary of hazard datasets and guidelines supported by the Global Earthquake Model during the first implementation phase[J]. Annals of Geophysics, 2015, 58(1): S0108.
Pittore M. Seismic Hazard and Risk in Central Asia[R]. Deutsches GeoForschungs Zentrum GFZ, 2011.
Schwarz C. Towards a Swiss National Earthquake Risk Model: Sensitivity and Gap Analysis[R]. TU Delft, Delft University of Technology, 2015.
Silva V. , Crowley H, Yepes C. , et al. Presentation of the OpenQuake-engine, an Open Source Software for Seismic Hazard and Risk Assessment[C]. Proceedings of the 10th US National Conference on Earthquake Engineering, Anchorage, Alaska, 2014a.
Silva V. , Crowley H. , Pagani M. , et al. Development and Application of OpenQuake, an Open Source Software for Seismic Risk Assessment [C]. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012. Paper No. 4923.
Silva V. , Crowley H. , Varum H. , et al. Seismic Hazard and Risk Assessment of Portugal[C]. Second European Conference on Earthquake Engineering and Seismology, Istanbul, 2014b.
Stewart J. P. , Douglas J. , Di Alessandro C. , et al. Selection of a Global Set of GMPEs for the GEM-PEER Global GMPEs Project [C]. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012a. Paper No. 2320.
Stewart J. P. , Seyhan E. , Boore D. M. , et al. Site Effects in Parametric Ground Motion Models for the GEM-PEER Global GMPEs Project[C]. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012b.
Weatherill G. A. , Pagani M. , Monelli D. The Hazard Component of the GEM Modeller's Toolkit: A Framework for the Preparation and Analysis of Probabilistic Seismic Hazard (PSHA) Input Tools[C]. Selection of a global set of GMPEs for the GEM-PEER Global GMPEs Project. Proceedings of the 15th World Conference on Earthquake Engineering, Lisbon, 2012. Paper No. 231.
Wieland M., Pittore M., Parolai S., et al. Towards a cross-border exposure model for the Earthquake Model Central Asia[J]. Annals of Geophysics, 2015, 58(1): S0106.
Woessner J. , Laurentiu D. , Giardini D. , et al. The 2013 European seismic hazard model: key components and results[R]. Bulletin of Earthquake Engineering, 2015, 13(12): 3553-3596.
Yepes-Estrada C. , Silva V. , Crowley H. GEM Vulnerability Database for the OpenQuake-platform[C]. Second European Conference on Earthquake Engineering and Seismology, Istanbul, 2014.
Zhou Bengang, Chen Guoxing, Gao Zhanwu, et al. The technical highlights in identifying the potential seismic sources for the update of national seismic zoning map of China[J]. Technology for Earthquake Disaster Prevention, 2013, 8(2): 113–124.
全球地震模型(GEM)研究进展综述及应用前景展望
李昌珑1,2, 吴健1,2, 徐伟进1,2, 高孟潭1,2     
1 中国地震局地球物理研究所, 北京 100081;
2 中国地震风险与保险实验室, 北京 100081
摘要:本文综述了全球地震模型(GEM)组织的工作目标、主要职能、主要工具(OpenQuake软件)、主要研究成果, 并对GEM今后的发展和应用趋势进行了展望。借鉴GEM和OpenQuake, 可以进一步完善我国的地震危险性分析模型, 提升大型活动断裂带附近大城市和重大工程设施地震危险性分析的科学性。
关键词球地震模型(GEM)    概率地震危险性分析(PSHA)    地震风险评估    OpenQuake虚拟机