Knowledge-Based Engineering Systems Research in Progress

ion level may be viewed as a model 2. P.A. Fishwick, Hierarchical Reasoning: for reasoning. The interface rules that let Simulating Complex Processes Over Multhe user bridge abstraction levels are tiple Levels ofAbstraction, technical report PaulA. Fishwick representative of what is learned when and PhD proposal, University of PennsylvaDepartment of Computer and traversing levels. Thus, Hires can be used nia, Philadelphia, 1985. Information Science as a basis for computer-aided instruction 3PA SishcHres: Hiera sonUniversity ofPennsylvania ing System Version 1.0/User's Manual, because it can teach the user about the technical report, University of Pennsylvania, Moore School/D2 process being simulated. Philadelphia, 1985. Philadelphia, PA 19104 Merging Statistical and Knowledge-Based Techniques for Process Diagnosis Promptly locating process problems the functional test for a particular type of System inputs. The following data during the manufacturing process is a comcircuit board. Using this information, samsources are available as inputs to IAS: plex and important task: complex because pled data from the shop floor control sysinformation on when and where each operit requires a thorough knowledge of the tem, and knowledge ofthe manufacturing ation was performed on each circuit board, particular manufacturing process; imporprocess, the system might determine that test and repair information for each circuit tant because especially on continuous a malfunction in the solder application was board, and circuit board design informain-line manufacturing lines where product causing the downward trend. Shop persontion, such as what components are located volumes and velocities are high nel would then be alerted to the problem. on a board and who each component's uncaught process problems can quickly manufacturer is. result in many defective products. Other knowledge-based process control The system uses these inputs to diagnose systems, such as PDS3 and Meld,4 tend to ' . ' ~~~~~~process problems such as: rely on direct acquisition of process data System strategy. The Intelligent Analyfrom sensors (such as the temperature of * a particular manufacturer's composis System detects manufacturing process nents have high defect levels; the solder bath) Because most measure... faults through a hybrid combination of ments available to it are product-related, a particular automatic insertion knowledge-based and probabilistic IAS statistically filters the data. To illusmachine is clipping the leads on a parmethods. The first IAS implementation ticular component too short; and will be to diagnose process faults in circuit trate, test measurement data rates on the * the testing software has a bug and is order of two million observations per day board manufacturing. The work will focus o o t indicting the wrong component. are not uncommon on high-volume circuit on diagnosis of the process, not the prodboard lines. uct as in the Dart' and Sophie2 systems, Detecting process trends. Using empirfor example. ical Bayesian techniques, the Real-Time Specifically, IAS does not attempt to Circuit board manufacture. Typically, Analysis System (RTAS)5 abstracts test diagnose what caused an individual circuit circuit board manufacturing includes four and repair data to construct estimates of board to fail. Rather, the system detects major steps: (1) automated insertion of first-pass yield distributions for the circuit shifts in process parameters and locates the components; (2) hand/robot assembly of boards as well as the defect levels for manufacturing step that introduced the components; (3) solder; and (4) one or individual components on the circuit process fault. more stages of test and (if the test fails) boards. When RTAS detects a sudden or The system might, for example, detect a repair. Each major step may have one or gradual trend in either the yield of a pardownward trend in the first pass yield of more operations. ticular board type or a particular compo-