{"id":3435,"date":"2021-03-04T08:00:00","date_gmt":"2021-03-04T07:00:00","guid":{"rendered":"http:\/\/www.se-trends.de\/?p=3435"},"modified":"2021-03-03T15:44:32","modified_gmt":"2021-03-03T14:44:32","slug":"make-requirements-machine-readable","status":"publish","type":"post","link":"https:\/\/www.se-trends.de\/en\/anforderungen-maschinenlesbar-machen\/","title":{"rendered":"Making requirements machine-readable with AI"},"content":{"rendered":"

Requirements management is a key discipline in product development, for example, requirements are an enabler for communication. Customers and suppliers use requirements to capture a common understanding of what needs to be done. But requirements are not machine-readable.<\/p>\n\n\n\n

There are powerful tools that make it easier for us to deal with requirements. But for these tools, the actual content is a \"black box\". The important part of the requirement is hidden in a text attribute that has to be read by people again and again... and again... and again...<\/p>\n\n\n\n\n\n\n\n

From documents to requirement elements<\/h2>\n\n\n\n

With the advent of computers, engineers began to look for ways to increase efficiency. And they found what they were looking for: Word and Excel, the most popular tools for requirements engineering. Whether these are really better than paper is another question...<\/p>\n\n\n\n

Unbelievable: Word and Excel are still the most popular tools for recording requirements<\/p><\/blockquote>\n\n\n\n

For those who can afford it, there are more powerful tools. Telelogic pioneered requirements engineering tools in the 1990s with DOORS (since acquired by IBM). DOORS is still around, but other popular requirements engineering tools include Polarion (Siemens), Jama Connect (Jama Software), Codebeamer (Intland) and Integrity (PTC).<\/p>\n\n\n\n

All these tools work in a similar way: They allow the definition of data types: These are entities with properties. For example, a data type \"User Need\" can have properties for \"Name\", \"Description\" and \"Priority\", in addition to a unique ID generated by the system.<\/p>\n\n\n\n

Elements can be created by the Traceability<\/a> can be linked with each other. This supports activities such as coverage or impact analysis and supports change management.<\/p>\n\n\n\n

A requirement is a black box<\/h2>\n\n\n\n

However, all these tools have one thing in common: the actual content is almost always located in an attribute as formatted text. In DOORS this is the \"Object Text\", in Jama it is the \"Description\". Users can write anything they want in these fields.<\/p>\n\n\n\n

Engineers like this because they are not restricted in what they write. They can even insert images. But the downside is that engineers often have to look at the actual content. The tools help engineers decide when to look at the content, but not why. For the tool, the requirement is a black box.<\/p>\n\n\n\n

A simple example: The stakeholder specifies a timeout of 10 minutes for a web system. A test engineer reads the requirement and writes a test case. The tester should, for example, wait a little less than 10 minutes and ensure that the user is still logged in.<\/p>\n\n\n\n

If the stakeholder changes the timeout to 20 minutes, then the test may be masking a latent problem: the test will still pass, but it would not test the changed requirement! A typical requirements management tool would mark the test as \"suspect\", but a human has to look at it.<\/p>\n\n\n\n

Is MBSE the answer?<\/h2>\n\n\n\n

Model Based Systems Engineering (MBSE<\/a>) uses a formal modeling notation to capture requirements. Depending on the notation, MBSE can actually solve this problem.<\/p>\n\n\n\n

Notations such as SysML create transparency, but are so difficult to use that many stakeholders reject them and only a few stakeholders actually master them.<\/p><\/blockquote>\n\n\n\n

There is only one catch: the notations used are difficult. The most popular notation is SysML<\/a>which consists of dozens of diagram types and hundreds of element types.<\/p>\n\n\n\n

\"SysML