This paper describes a constraint-based knowledge compiler for parametric design problems in mechanical engineering. The purpose of our research is to realize tools which enable designers to build knowledge-based systems for parametric design in mechanical engineering simply by using a knowledge compilation approach. The knowledge compilation approach seeks to automate 1) the process of producing knowledge-based design systems from higher level design specifications, and 2) the process of restructuring existing software systems to produce new systems that exhibit a) an increase in efficiency or usability,b) a change in representation level, and c) a reduction of reasoning. Consequently, we realize knowledge compilation that adopts the advantages of the constraint-based formalization of parametric design systems; we define this technique as constraint-based knowledge compilation. Constraint-based knowledge compilation automates the process of producing knowledge-based design systems for each design object, from input design specifications, by regarding 1) a design problem in mechanical engineering as parametric design, 2) an explicit declarative knowledge of this design as a constraint, and 3) the design process as a constraint satisfaction problem. The advantage of this technique is that it is easy to modify and reuse this knowledge in order to realize problem-solving mechanisms (i.e. constraint satisfaction) for other design tasks. We have developed MECHANICOT, a tool which implements our constraint-based knowledge compilation technique for building knowledge-based systems for parametric design in mechanical engineering. MECHANICOT generates a design plan that embeds the constraint-incorporated problem-solving mechanisms realized using problem-solving primitives and that solves the parametric design problem in mechanical engineering. We have applied the constraint-based knowledge compilation technique to the following functional machine units : a gear unit and a main spindle head of a lathe. The results show that MECHANICOT can compile the main spindle head problem constraints derived from the input design specification and generate a design plan more efficiently than engineers employing conventional means. The significance of this work lies in demonstrating an improvement of the development of knowledge-based design systems and an enhancement of the productivity of designers and end users.

Traditional approaches to learning from classified examples for knowledge acquisition may be divided into two categories: artificial intelligence approaches (e.g., ID3) and pattern recognition approaches (e.g., neural network). Most artificial intelligence approaches are designed to treat symbolic information, while they are weak in treating noisy quantitative data. Conversely, most pattern recognition approaches are designed to treat noisy quantitative data, while they are weak in treating symbolic information. Both approaches have strengths and weaknesses based on principles used. This paper presents a new approach to knowledge acquisition systems located between artificial intelligence approaches and pattern recognition approaches. Our knowledge acquisition system is based on the Cartesian space model (CSM) which is a mathematical model to treat symbolic data where each sample is described not only by quantitative features but also qualitative and structural features. Our system is composed of six subsystems: pattern categorizer, event generator, feature selector, production rule generator, inference engine, and graphic subsystem. If we have no class concepts before hand, the pattern categorizer can generate the class concepts automatically. The pattern categorizer uses a hierarchical conceptual clustering based on the generalized Minkowski metrics on the CSM. Our classification method uses a region oriented approach, and the approach is realized by the event generator in our system. The event generator produces the regions called events which describe each pattern class. If a test sample is included only in events for a pattern class, the sample is decided to come from the class. An event is reduced to a rectangular form when each sample is based only on quantitative information. In our method, some test samples are rejected to assign class name, when the samples are not included in any event. In this case, our system can suggest the nearest pattern class by using functions similar to fuzzy membership functions. We compare our knowledge acquisition system to the ID3 symbolic learning system and the backpropagation neural network learning system based on some simple examples. Our knowledge acquisition system is useful as a tool to support knowledge engineers.

Consider the situation that an analogical reasoning model is investigated considering the practical application, it is necessary to consider the effective knowledge retrieval to solve given problems. In most of the analogical reasoning models which have been proposed, knowledge is retrieved with some criteria from knowledge base, and problems are solved by using of the retrieved and/or general knowledge. However, these models were using the search and combination of all the elements in the knowledge base, impricitly. So, the knowledge retrieval often becomes explosively complex. In our method at first one problem is divided into several sub-problems, then they are analized individually with appropriate knowledge base and the results are harmonized to solve. As the sub-problems can refer several different knowledge base, the combination of the targets will not be complicated. Thus it is not necessary to deal with the large targets all at once and the retrieval time is reduced. In this paper, we propose analogical reasoning model based on knowledge classification. Further-more, we applied the pilot system to the domain of mathematical problems such as equation problems. By using this system, we compared the time during problem solving with our model to the time without ours, and show the efficiency of our model.

We have already proposed a new type of binary coded GA, called the forking GA (fGA), as a kind of multi-population GAs and showed that the searching power of the fGA is superior to the standard GA. The feature of the fGA is that each population takes a different role in optimization. That is, each population is responsible for searching in a non-overlapping sub-area of the search space. In this paper, the extended forking GA for order representation, called the o-fGA, is proposed. The results of experiments for the traveling salesman problems (TSP) and flowshop scheduling problems show that the approach of fGA is also effective for the order representation.

Parallel software consists of many asynchronous processes. It is important for parallel software to behave on valid timing conditions. As timing constraints and state transitions can be represented by temporal logic, logic is useful for computer aided design for parallel software. But as the problem of deciding whether a given real-time temporal logic formula is finitely-satisfiable isΣ^1_1-complete, it is impossible to generate system specification from real-time temporal logic. Specification method is more practical than generation method. In this paper, I propose specification and verification method for parallel software as follows. (1) Hierarchical specification method: Hierarchical specification method consists of system configuration specification and process specification. System configuration specification represents processes configuration by process algebra. Process specification represents internal process behavior by timed automaton. System specification is automatically generated from process algebra and timed automaton. (2) Formal timing verification method: Formal timing verification method is based on real-time model-checking. System specification is interpreted as timed Kripke structure, and verification algorithm consists of labeling algorithm and timing simulation. Timing verification cost is redued compared with region graph method. Based on (1) and (2), we can specify and verify parallel software, which includes timing constraints and infinite computation, asynchronous behavior. I show this method effective by automobile control system specification and verification.

Functional modelling techniques are recently used to include the intensions of system designers into models of a system. Functions are higher level than behaviour and they are sometimes given different meanings depending on a system situation. Therefore, it is important to be able to derive behaviour from a functional model. This paper describes techniques to represent goals-functions-structure and to derive system behaviour from a functional model through a structure model, where the Multilevel Flow Modelling (MFM) and the Hybrid Phenomena Theory (HPT) are effectively combined. The MFM is a methodology to model an engineering system from the standpoint of means and goals. It has been applied to diagnostic, planning, and man-machine interface design problems. The HPT is a method to model the relations between structure and behaviour. One useful application of the HPT is to derive mathematical equations describing system's behaviour from structural information. The MFM is extended to be able to represent systematically abstracted information of structure of a system. The HPT is applied to derive the behaviour of a system from its structure model. A transformation mechanism from a MFM model to its corresponding HPT representation is developed to bridge the MFM and HPT. A simple example to model a central heating system and to derive its behaviour demonstrates the proposed techniques.

Inductive Logic Programming (ILP), a variation of inductive inference, has been gaining attention as a way to synthesize logic programs from examples. One of the important tasks of ILP is to search the hypothesis space for definite clauses which explain examples under the background knowledge. Many researchers have studied methods which perform the task, and developed the method which employs refinement operators to generate definite clauses, the one which calculates a least general generalization (lgg) of examples relative to the background knowledge which consists of ground atoms, etc. In this paper we describe computing-lggs-first strategy, that is to say, to search the hypothesis space, we first calculate possible lggs of atoms provable from the background knowledge, then combine them to form clauses. It is more efficient than the method employing refinement operators, and can use non-ground atoms as the background knowledge unlike the method calculating an lgg relative to the background knowledge. Practically, we have to employ the strategy with some restrictions. We have been developing an inductive inference system KISS. It calculates lggs of atoms in the background knowledge, instead of calculating lggs of provable atoms. Due to this simple strategy, it becomes possible to find various clauses such as those which are necessary to synthesize logic programs and those which represent regularities of predicates.

It has long been recognized that abstraction can be used to reduce a problem-solving search. Since combinatorial optimization problems have extraordinary large search spaces, it is meaning-full to apply abstraction those problems. But it has never been shown that abstraction is useful to solve them. Inductive Algorithms that is a new problem-solving paradigm based on abstraction efficiently solved the knapsack problem that is a typical example of the combinatorial optimization problem. This article provides experimental results of applying Inductive Algorithms to the traveling salesman problem (for short, TSP). In particular, it is shown that this algorithm is applicable to the large scale TSP, consisting of 532 cities.

Abduction has been recognized as important human reasoning and has been applied in various fields (See for example[Eshghi 89]). It is also related with logic programming, especially with negation as failure [Eshghi 89, Kakas 90a, Kakas 90b]. Eshghi and Kowalski [Eshghi 89] introduce abduction to handle negation as failure and Kakas and Mancarella [Kakas 90a, Kakas 90b] extend [Eshghi 89] to include any arbitrary abducible predicate. We have already proposed a correct bottom-up procedure to compute abduction [Satoh 91]. This procedure computes generalized stable models [Kakas 90a] for computing abduction. However, this procedure is not suitable for a query evaluation. Although [Kakas 90b] proposes a query evaluation method by extending the procedure in [Eshghi 89], there is a problem of incorrectness in the procedure in [Eshghi 89] and the problem is inherited to the procedure in [Kakas 90b]. Moreover, their procedures can only handle a limited class of integrity constraints. Our proposed procedure in this paper is correct for any consistent abductive framework proposed in [Kakas 90a]. In other words, if the procedure succeeds, there is a set of hypotheses which satisfies a query, and if the procedure finitely fails, there is no such set. We can guarantee the above correctness since we adopt a forward evaluation of rules and check consistency of "implicit deletion" [Sadri 88]. Thanks to the forward evaluation of rules, we can also handle any form of integrity constraints.

In this paper, we describe a basic framework of the narrative generation system for supporting human creative tasks. The narrative generation process by computer is divided into the conceptual representation level and the surface language generation level, and we deal with only the former level here. The conceptual representation is divided into three aspects; story, plot, and construction. While the story is an events sequence that was arranged according to a temporal order, the plot is an events sequence that was reorganized by an order which each event is introduced into a narrative. These three levels in a narrative are constructed as tree structures. Terminal nodes in the tree structures are events and all nodes other than them are relations that connect subordinate nodes. Narrative generation is performed by expanding or transforming a tree structure. In the story and construction generation, the system enlarges each tree by expanding events or partial trees using appropriate relations, and in the plot generation, a story tree is transformed into a plot tree through the connection relations among nodes in it are rearranged. We call narrative techniques the procedures to expand a tree through applying relations to nodes or to transform a tree using actors" viewpoints or plot patterns. On the other hand, we call narrative strategies the rules to decide a current executable narrative technique and the node to which it is applyed according to narrative parameters that define the features of a narrative to be generated through narrative generation process. The system generates a narrative by executing appropriate narrative techniques under the control of narrative strategies based on a set of events and narrative parameters were given by user. This narrative generation mechanism has some remarkable characteristics. First, the system can flexibly generate a variety of narratives from one input. Next, the system has an ability that integrates a variety of theories or knowledge representations and that extends the system itself. These advantages are relate to clear separation among narrative techniques, narrative strategies, and knowledge base. Lastly, by above reason, the system has potentiality that can use for various purposes. We can change or add each modules in it to apply to specific areas.