Knowledge base

From WikID

A knowledge base (or knowledgebase; abbreviated KB) is a special kind of database for knowledge management. It provides the means for the computerized collection, organization, and retrieval of knowledge (Wikipedia [1], copied May 26, 2008)

In product design, knowledge bases are typically employed to enable knowledge utilization by the end user of the product, to support the use of the product, or to enable knowledge utilization by the designers of the product, to support the design process.

Computer support of knowledge utilization by the user typically comes in the form of embedded software, where the product is equipped with its own ‘computer’, usually a microchip.

Knowledge-utilization support in the design process typically comes in the form of knowledge-based engineering (KBE) systems, a specific type of knowledge-based systems (KBSs). KBS technology has mainly developed from a branch of artificial intelligence (AI) called rule-based systems. In rule-based systems, the knowledge of interest is compiled by human experts into a system of logical rules and equations and stored in a knowledge base. The rules are a formal representation of the knowledge concerning the given area or domain. The representation typically uses symbolic (i.e., non-numeric) terms defined in a language based on a predefined syntax with semantics to express meaning. In most current KBSs, the domain expertise is captured in the form of so-called production rules, relationships and facts. They are generally formulated as

IF (condition) THEN (draw conclusion or take action)

For instance,

IF material(part_A OR part_B) = thermoplastic AND no_extra_part_allowed = TRUE THEN connection_type = snap_fit

is a production rule for the selection of a connection. A more complex version could prompt the user to choose between alternatives that do not require a separate fastener, e.g., snap fit or ultrasonic welding. Additional rules could guide the user through the process of dimensioning a snap finger based on parameters such as the insertion force allowed and the insertion distance. Further possible add-ons are the automatic creation of geometry that can be imported into a CAD application, or even complete integration into a CAD system. In the knowledge base, production rules are usually formulated in a more concise way than the above IF-THEN statement, but typically still in an English-like ‘language’. To cover more complex relationships, sometimes object-oriented representations called ‘frames’, ‘graphs’ and ‘networks’ are used. For the end user of the system, the difference is not clearly noticeable. Two other alternatives for traditional production rules that do make a difference to the end user are KBs based on the AI concepts of fuzzy logic (or fuzzy sets) and those based on neural networks. The typical characteristic of fuzzy KBs, is that they are able to deal with vague descriptions, while neural networks, enable a KB to become self-learning: the system can be trained by feeding it with examples of input and the desired output, instead of feeding it with formal rules. This is useful in case it is known that a relationship exists, but it is not possible to express it in a formula or a logical statement. Currently, however, most of the work involved in preparing a knowledge based is still done by human experts or knowledge authors. KBs that create part of their knowledge output automatically based on fuzzy logic and/or neural networks are not widespread in design and engineering.

Creating knowledge bases

The knowledge in the KB itself is usually created and processed via a workflow that involves knowledge acquisition or knowledge capturing, knowledge structuring, knowledge formalization and knowledge storage. The extraction of the information from the knowledge base is carried out by a part of the system known as inference engine, which is generally separate from the KB itself. Inference normally involves searching the knowledge base for the appropriate rules that apply in order to reach a given objective, e.g., to optimize the parameters of a design. In most cases, the KBE system will be rule-based and it has to deal with multiple rules that all apply to the problem at hand. The most important task of the inference engine is to identify these rules and to put them in the right order for processing, or to combine them into more complex (e.g. nested) rules. It is probably true to say that outside of the field of engineering, KBSs are mostly used by large companies because these systems are custom-built and therefore expensive. Typical applications are routine commercial operations – such as order processing, and diagnostic systems – such as problem solving in a car workshop. Another widespread application of KBSs in troubleshooting can be found on the websites that many companies have opened for customer support, for instance Apple and Microsoft. Engineering is a field where people are used to formulating problems in terms of equations and logic; therefore it is not surprising that – when it comes to application in smaller companies – knowledge-based engineering (KBE) is ahead of most KBS application areas. A KBE system represents product knowledge and the skills of an individual (the knowledge author) within an engineering domain; it incorporates them and makes them available within a computerized application. As a sub-domain of KBSs dedicated to engineering and design applications, KBE has grown out of CAD/CAE technology. We distinguish custom-built KBE systems in the tradition of KBSs , KBE systems in close connection to CAD and stand-alone KBE systems that do not intensively interface with CAD systems. In the latter category are KBE tools for problem solving and idea finding such as TRIZ and knowledge-based ‘design-for-X’. The other two categories are discussed below.

Company-specific customized KBE systems

Customized KBE systems specifically tailored for a specific context in a specific company. These systems are not very widespread but they have a long history of successes particularly in the aerospace and automotive industries. The introduction of customized KBE is considered to have resulted in substantial reductions in development times. The systems do not necessarily have a high level of technological sophistication. Typically, a product model is held in a rule-based format such that properties of a product can be linked together with constraints. Design rules are stored in a knowledge base from which they are retrieved by an inference engine. Current KBE systems are problem solvers based on declarative models with no concept of a design process. They have not been devised for capturing the designer’s reasoning and are often built in an ad hoc fashion, which makes maintenance of the system difficult. Many KBE systems focus on geometric knowledge, i.e., linking the dimensional relationships of a product. Outputs include complete detail design drawings, costs, bills of material and manufacturing process information. Typically, they rely on some kind of macro language that is already included in conventional CAD/CAE software, but not intended for the average CAD user. Knowledge-based engineering is especially beneficial for companies with one or more of the following characteristics: • Products are based on design-to-order or engineering-to-order: each product is slightly different from the other; the differences depend on the assignment from the customer; • The design of the products includes repetitive design tasks; • The products can be defined by a set of rules and equations. An example [2] is the KBE application for headlamp design implemented at Jaguar Cars, where repetitive design tasks were captured into rules and equations. A car designer who makes a style rendering initially determines the form of the headlamps. Before KBE implementation, this rendering was entered into a CAD system and the CAD file was sent to the headlamp supplier who usually came up with modifications. These modifications on their turn would lead to modifications that the designer had to apply to the bonnet and the wing of the car body, which again made it necessary to reconsider the headlamp design, and so on. This tedious trial-and-error process could take up to three months. As early as 1989, the car company and the supplier decided to work on a KBE system that incorporates all the headlamp knowledge concerning practical design and legislation, and which enables the designer to play with various styling options in the certainty that the design of the front of the car body and the headlamps will be compatible. What once took months to accomplish now takes minutes.

User-customizable KBE in combination with CAD

Software that more explicitly profiles itself as KBE software is typically integrated with CAD systems. These off-the-shelf solutions are typically based on the same macro languages as the customized KBE applications, but they provide a more easily accessible user interface to everyday CAD users, making it possible to define rules and equations by using a graphical interface. Key players in the KBE world are Genworks, Solid Works in combination with DriveWorks or RuleStream, and CATIA (Product Knowledge Template)