UML models may be automatically transformed to other representations (e.g. Java) by means of QVT-like transformation languages. UML is extensible, with two mechanisms for customization: profiles and stereotypes.
History of object-oriented methods and notation.
UML has been evolving since the second half of the 1990s and has its roots in the object-oriented methods developed in the late 1980s and early 1990s. The timeline (see image) shows the highlight of the history of object-oriented modeling methods and notation.
In 1996, Rational concluded that the abundance of modeling languages was slowing the adoption of object technology, so repositioning the work on a unified method, they tasked the "Three Amigos" with the development of a non-proprietary Unified Modeling Language. Representatives of competing object technology companies were consulted during OOPSLA '96; they chose boxes for representing classes rather than the cloud symbols that were used in Booch's notation.
Under the technical leadership of the "Three Amigos", an international consortium called the UML Partners was organized in 1996 to complete the Unified Modeling Language (UML) specification, and propose it as a response to the OMG RFP. The UML Partners' UML 1.0 specification draft was proposed to the OMG in January 1997. During the same month the UML Partners formed a Semantics Task Force, chaired by Cris Kobryn and administered by Ed Eykholt, to finalize the semantics of the specification and integrate it with other standardization efforts. The result of this work, UML 1.1, was submitted to the OMG in August 1997 and adopted by the OMG in November 1997.
As a modeling notation, the influence of the OMT notation dominates (e. g., using rectangles for classes and objects). Though the Booch "cloud" notation was dropped, the Booch capability to specify lower-level design detail was embraced. The use case notation from Objectory and the component notation from Booch were integrated with the rest of the notation, but the semantic integration was relatively weak in UML 1.1, and was not really fixed until the UML 2.0 major revision.
Concepts from many other OO methods were also loosely integrated with UML with the intent that UML would support all OO methods. Many others also contributed, with their approaches flavouring the many models of the day, including: Tony Wasserman and Peter Pircher with the "Object-Oriented Structured Design (OOSD)" notation (not a method), Ray Buhr's "Systems Design with Ada", Archie Bowen's use case and timing analysis, Paul Ward's data analysis and David Harel's "Statecharts"; as the group tried to ensure broad coverage in the real-time systems domain. As a result, UML is useful in a variety of engineering problems, from single process, single user applications to concurrent, distributed systems, making UML rich but also large.
The Unified Modeling Language is an international standard:
ISO/IEC 19501:2005 Information technology – Open Distributed Processing – Unified Modeling Language (UML) Version 1.4.2
UML has matured significantly since UML 1.1. Several minor revisions (UML 1.3, 1.4, and 1.5) fixed shortcomings and bugs with the first version of UML, followed by the UML 2.0 major revision that was adopted by the OMG in 2005.
Although UML 2.1 was never released as a formal specification, versions 2.1.1 and 2.1.2 appeared in 2007, followed by UML 2.2 in February 2009. UML 2.3 was formally released in May 2010. UML 2.4.1 was formally released in August 2011. UML 2.5 was released in October 2012 as an "In process" version and has yet to become formally released.
There are four parts to the UML 2.x specification:
The Superstructure that defines the notation and semantics for diagrams and their model elements
The Infrastructure that defines the core metamodel on which the Superstructure is based
The UML Diagram Interchange that defines how UML 2 diagram layouts are exchanged
The current versions of these standards follow: UML Superstructure version 2.4.1, UML Infrastructure version 2.4.1, OCL version 2.3.1, and UML Diagram Interchange version 1.0.
Although many UML tools support some of the new features of UML 2.x, the OMG provides no test suite to objectively test compliance with its specifications.
Software development methods
UML is not a development method by itself; however, it was designed to be compatible with the leading object-oriented software development methods of its time (for example OMT, Booch method, Objectory). Since UML has evolved, some of these methods have been recast to take advantage of the new notations (for example OMT), and new methods have been created based on UML, such as IBM Rational Unified Process (RUP). Others include Abstraction Method and Dynamic Systems Development Method.
It is important to distinguish between the UML model and the set of diagrams of a system. A diagram is a partial graphic representation of a system's model. The model also contains documentation that drives the model elements and diagrams (such as written use cases).
UML diagrams represent two different views of a system model:
UML 2.2 has 14 types of diagrams divided into two categories. Seven diagram types represent structural information, and the other seven represent general types of behavior, including four that represent different aspects of interactions. These diagrams can be categorized hierarchically as shown in the following class diagram:
UML does not restrict UML element types to a certain diagram type. In general, every UML element may appear on almost all types of diagrams; this flexibility has been partially restricted in UML 2.0. UML profiles may define additional diagram types or extend existing diagrams with additional notations.
In keeping with the tradition of engineering drawings, a comment or note explaining usage, constraint, or intent is allowed in a UML diagram.
Structure diagrams emphasize the things that must be present in the system being modeled. Since structure diagrams represent the structure, they are used extensively in documenting the software architecture of software systems.
Class diagram: describes the structure of a system by showing the system's classes, their attributes, and the relationships among the classes.
Component diagram: describes how a software system is split up into components and shows the dependencies among these components.
Deployment diagram: describes the hardware used in system implementations and the execution environments and artifacts deployed on the hardware.
Object diagram: shows a complete or partial view of the structure of an example modeled system at a specific time.
Package diagram: describes how a system is split up into logical groupings by showing the dependencies among these groupings.
Profile diagram: operates at the metamodel level to show stereotypes as classes with the <<stereotype>> stereotype, and profiles as packages with the <<profile>> stereotype. The extension relation (solid line with closed, filled arrowhead) indicates what metamodel element a given stereotype is extending.
Behavior diagrams emphasize what must happen in the system being modeled. Since behavior diagrams illustrate the behavior of a system, they are used extensively to describe the functionality of software systems.
Activity diagram: describes the business and operational step-by-step workflows of components in a system. An activity diagram shows the overall flow of control.
Interaction diagrams, a subset of behavior diagrams, emphasize the flow of control and data among the things in the system being modeled:
Communication diagram: shows the interactions between objects or parts in terms of sequenced messages. They represent a combination of information taken from Class, Sequence, and Use Case Diagrams describing both the static structure and dynamic behavior of a system.
The Object Management Group (OMG) has developed a metamodeling architecture to define the Unified Modeling Language (UML), called the Meta-Object Facility (MOF). The Meta-Object Facility is designed as a four-layered architecture, as shown in the image at right. It provides a meta-meta model at the top layer, called the M3 layer. This M3-model is the language used by Meta-Object Facility to build metamodels, called M2-models.
The most prominent example of a Layer 2 Meta-Object Facility model is the UML metamodel, the model that describes the UML itself. These M2-models describe elements of the M1-layer, and thus M1-models. These would be, for example, models written in UML. The last layer is the M0-layer or data layer. It is used to describe runtime instance of the system.
Beyond the M3-model, the Meta-Object Facility describes the means to create and manipulate models and metamodels by defining Common Object Request Broker Architecture (CORBA) interfaces that describe those operations. Because of the similarities between the Meta-Object Facility M0-model and UML structure models, Meta-Object Facility metamodels are usually modeled as UML class diagrams. A supporting standard of the Meta-Object Facility is XMI, which defines an XML-based exchange format for models on the M3-, M2-, or M1-Layer.
Although UML is a widely recognized and used modeling standard, it is frequently criticized for the following:
Bertrand Meyer, in a satirical essay framed as a student's request for a grade change, apparently criticized UML as of 1997 for being unrelated to object-oriented software development; a disclaimer was added later pointing out that his company nevertheless supports UML.Ivar Jacobson, a co-architect of UML, said that objections to UML 2.0's size were valid enough to consider the application of intelligent agents to the problem. It contains many diagrams and constructs that are redundant or infrequently used.
Problems in learning and adopting
The problems cited in this section make learning and adopting UML problematic, especially when required of engineers lacking the prerequisite skills. In practice, people often draw diagrams with the symbols provided by their CASE tool, but without the meanings those symbols are intended to provide. Simple user narratives e.g. "what I do at work ..." have shown to be much simpler to record and more immediately useful.
The standards have been cited as being ambiguous and inconsistent. The UML 2.0 standard still suffers many issues.
Capabilities of UML and implementation language mismatch
Typical of other notational systems, UML is able to represent some systems more concisely or efficiently than others. Thus a developer gravitates toward solutions that reside at the intersection of the capabilities of UML and the implementation language. This problem is particularly pronounced if the implementation language does not adhere to orthodox object-oriented doctrine, since the intersection set between UML and implementation language may be that much smaller or equal in size.
Dysfunctional interchange format
While the XMI (XML Metadata Interchange) standard is designed to facilitate the interchange of UML models, it has been largely ineffective in the practical interchange of UML 2.x models. This interoperability ineffectiveness is attributable to several reasons. Firstly, XMI 2.x is large and complex in its own right, since it purports to address a technical problem more ambitious than exchanging UML 2.x models. In particular, it attempts to provide a mechanism for facilitating the exchange of any arbitrary modeling language defined by the OMG's Meta-Object Facility (MOF). Secondly, the UML 2.x Diagram Interchange specification lacks sufficient detail to facilitate reliable interchange of UML 2.x notations between modeling tools. Since UML is a visual modeling language, this shortcoming is substantial for modelers who don't want to redraw their diagrams. The Diagram Definition OMG project is another alternative.
As with database Chen, Bachman, and ISO ER diagrams, class models are specified to use "look-across" cardinalities, even though several authors (Merise, Elmasri & Navathe  amongst others ) prefer same-side or "look-here" for roles and both minimum and maximum cardinalities. Recent researchers (Feinerer, Dullea et. alia ) have shown that the "look-across" technique used by UML and ER diagrams is less effective and less coherent when applied to n-ary relationships of order >2.
In Feinerer it says "Problems arise if we operate under the look-across semantics as used for UML associations. Hartmann  investigates this situation and shows how and why different transformations fail." (Although the "reduction" mentioned is spurious as the two diagrams 3.4 and 3.5 are in fact the same) and also "As we will see on the next few pages, the look-across interpretation introduces several difficulties which prevent the extension of simple mechanisms from binary to n-ary associations."