Researchers at the National Taiwan Institute of Technology have adopted various AI techniques including fuzzy reasoning, pattern recognition, rule-based reasoning, back-propagation neu- ral network, genetic algorithms and Petri nets for the stamping process planning and design of progressive shearing cut and bending dies [14–16]. However their work lacks an explicit and consistent model to integrate these AI techniques into a compre- hensive design environment.

In this paper, another popular AI technique, blackboard ar- chitecture, is adopted to develop a blackboard-based stamping process planning system. In the last two decades, blackboard ar- chitecture has been successfully used in a wide variety of areas, such as speech recognition, signal processing, engineering de- sign and process planning. Thompson and Lu [17] used a black- board architecture to provide a cooperative decision making en- vironment that is suitable for concurrent product and process design. Srihari et al. [18] developed a real-time CAPP system for printed circuit board (PCB) assembly by integrating multiple KSs, including planning expert and dynamic information pro- cessing modules in the blackboard architecture. Chen et al. [19] developed a concurrent product design evaluation system, using a blackboard architecture to classify knowledge into perse KSs suitable for qualitative and quantitative evaluation, respectively.

In the past few years, blackboard architecture has proven to be suitable for tooling design such as fixture design [20] and in- jection moulding design [21], though this kind of application is still in its infancy stage. Roy and Liao [20] report the preliminary work that investigates the suitability of using a blackboard archi- tecture as a [K1]problem solving model for fixture design. It de-

scribes the creation of various functional KSs for fixture design and their organization in a cooperative problem solving environ- ment. Kwong et al. [21] proposes a blackboard-based system for concurrent process design of injection moulding, which facili- tates the simultaneous considerations of moulding part design, tool design, machine-selection, production scheduling, and cost as early as possible in the conceptual  design stage.  However, we have not found in the literature any attempt to apply the blackboard architecture to stamping process planning for sheet metal parts. It has been mentioned in our earlier work [22] that a blackboard architecture is well suited for constructive prob- lem solving, like process planning of stamping operations, where the problem space is large and knowledge from many different sources must be integrated to achieve a solution. This topic is now to be extensively elaborated in the present paper.

3 Blackboard framework for stamping process planning

Cooperative decision making for knowledge-based stamping process planning involves a variety of KSs such as unfolding knowledge to produce flat pattern, nesting knowledge to produce blank layout, mapping knowledge to transform stamping features into stamping operations, and staging knowledge to sequence the stamping operations. These KSs may be expressed in different representation schemes such as procedures, rules, and objects. This justifies the use of a blackboard framework that can   man-

Fig. 1. Blackboard framework for stamping process planning

age heterogeneous KSs effectively. The KSs interact through the blackboard to develop a solution incrementally.

The proposed blackboard framework consists of three major components: the blackboard data structure, KSs, and a control module (Fig. 1), and was developed using object-oriented expert system shell CLIPS. The different components of the blackboard framework are described as follows.

3.1 Object-oriented blackboard data structure

The blackboard is a globally accessible database, which con- tains the data and partial solutions and is shared by a number of independent KSs. The KSs contribute their partial solutions to the blackboard, which lead to a final solution incrementally. The blackboard is structured as a hierarchy of solution parti- tion levels, which represent different aspects or stages of the solution process. Partial solutions are associated with each level and may be linked to information on other levels using algorith- mic procedures or heuristic rules. Each level contains planning objects that are used to represent the solution space in an object- oriented manner. This leads to the added advantage in knowledge system development because object-oriented approach supports software modularity, reusability, and scalability.