【正文】 vity of PCB assembly. Only few investigations are for PCB assembly process control (Chandraker et al, 1990。 Randhawa et al, 1986。 Cecil et al, 1994). These investigations are still short of the following capabilities: (1) to be applied in overall processes。 (2) to learn experience from successful operations。 (3) to infer theprocesses with environmental variables.2. Research ObjectivesThe objective of this research is to design and to implement an intelligent realtime decision support system for PCB assembly process control. The expected functions of the system include abilities (1) to identify process performance by monitoring process variables, (2) to decide necessary adjustments, and (3) to make suggestions by using the ES and the ANNs for different steps of PCB assembly. The system will have a generalized architecture that allows users to create customized systems for different steps such as adhesive dispensing, cleaning, etc. The inference process is expected to be accurate and fast enough to response optimal or nearoptimal values of process input variables.3. Proposed MethodologiesThis proposed realtime decision support system for PCB assembly process control will be able to analyze the output data of a process and recognize the data if there is any process output that exceeds desired specification or any pattern that shows the process is going to be out of control. If the process is not declared out of control, the process input would be used for the production process. If there is any signal that shows a situation of out of control, the output data will be sent to the inference engine to detect possible reasons or faults. After the reasons or faults have been detected, remendations and explanations will be provided to the users. Finally, users can adjust the process input based on the remendations. The process control flow is shown in Figure 1 Statistical process control (SPC)Statistical process control (SPC) is a proven statistical technique for any process to detect the signals that show the situations of out of control or the patterns of potential problems. It consists of systematic collections of data related to a process and graphical summaries of the data for analysts39。 visibility. The main purpose of the SPC is for an analysis and improvement of any process (ReVelle et al, 1992). Among the techniques of SPC, the controlcharts are easily for operators to determine if performance of a process is maintaining an acceptable level of quality over time. When an out of control signal is shown by control charts, a search for possible fault causes of variation begins (Dietz et al, 1989。 Feigenbaum, 1991). Unfortunately, use of the classical SPC technique is usually cumbersome because it requires a lot of data manipulation, control chart construction, and expert judgment (Jacobs, 1993). Therefore, it is necessary to propose a technique for automatically analyzing control charts arid the ANN approach will be applied in this research due to its ability in pattern recognition. Knowledgebased expert systemESs have been successfully appliled in most decision making processes by simulating human behavior. They perform reasoning by using preestablished rules in a welldefined domain. The rules bine knowledge of domain experts and facts of the welldefined domain. The knowledge database contains the rules used in the inference process. Over the past decades, the ESs have been applied in many fields such as diagnostic, process control, scheduling, and quality control. An important advantage of the ES is that the users can easily understand results because reasoning can be explained. The other advantage is the ease in modifying the knowledge base. However, ESs cannot solve problems unless they have specific rules. This knowledge acquisition bottleneck: may arise when the human experts are too busy or difficult to deal with, or when written materials may be difficult to obtain. Another difficulty of ESs is the validation and the verification of the systems when the systems bee large. Other limitation is that the systems can not learn experience automatically. Artificial neural networks (ANNs)An ANN is a model that mumics activities of human brains. It consists of many simple neurons operating in parallel based on the simplistic mathematical representation of what we think real neurons look like. Application fiellds of the ANN are those areas that rules are unkno,wn or inplete. After training, the ANN can learn from standard patterns (historical data) and store the optimal weights for future inputs. This approach has the advantage of easy update for the system with updated data. Thus, it eliminates the burden of rule reconstruction and programming modification of the system. Another advantage is its execution speed after the networks are trained especially in parallel puters.The major limitation of the ANN is that its output is just sets of numbers that has no obvious meaning to humans. An explanation system has to be provided for easy understanding to humans. Another limitation is its training time. When the number of layers increases, the training time increases significantly (Medsker, 1994). Hybrid intelligent systemAs mentioned above, both artificial intelligent technologies represent various aspects of human intelligence. A hybrid artificial intelligent system that bines two technologies may enhance performance of the system. Such an intelligent system can have advantages of both methodologies and get rid of their limitations.The models of theses hybrid