Clemens Schwenke

Automated planning and creation of simulation experiments with a domain specific ontology for semiconductor manufacturing AMHS

To successfully manufacture logic and power semiconductors in existing high mix semiconductor factories, fast ramp up phases and frequent product changes are necessary. Especially for power semiconductor production, new manufacturing and automation concepts are required, e.g. regarding the use of other substrates than silicon wafers. To allow a judgment on how an existing automated material handling system (AMHS) can cope with the new challenges or which alterations are required, a material flow simulation is essential. However, the planning and creation of such simulation experiments is difficult because of the systems complexity, the large amount of boundary conditions and the effort of manually modifying and testing many different variants, partially using currently unforeseen automation concepts. In order to assist in this process, the authors suggest a method for rapidly creating valid simulation experiments using an ontology that allows for the reuse of previous experiments and system experts knowledge.

Simulation and analysis of buying behavior in supermarkets

Future supermarkets will provide instrumented shopping carts, shelves and products so that large amounts of event data of customers actions inside the store will be recorded. Special data mining algorithms will be necessary for analyzing this data effectively. At present, respective event data is not available. In order to create test data sets for developing, implementing and testing respective data mining algorithms presently, a simulation of buying behavior in supermarkets is currently being developed by the authors. It will be used to investigate tailored data mining methods.

Event-based recognition and source identification of transient tailbacks in manufacturing plants

Automated material handling systems in complex manufacturing plants oftentimes exhibit, possibly transient, tailback phenomena, which reduce a production lines throughput. In order to identify origins and causes of observed tailbacks, historic event log data of loads passing certain waypoints has to be inspected. This paper introduces an approach to automatically carry out transient tailback recognition and cause identification. The approach is based on analysis of holding times and capacities of transport segments. As a result, complete lists of tailbacks and affected segments are provided. Each tailback consists of a reconstructed queue of loads waiting for preceding loads. For each tailback an initial cause event is determined. Additionally, identified tailbacks can be ranked by length or by impact on the transport delays. The developed demonstrator frees the user from time consuming visual inspection of log files by providing clearly represented complete tailback information instead.

Analysis and simulation of sales receipt data in supermarkets

For the prediction of product sales, simulations of interactions of different individual customer behaviors in supermarkets are an appropriate approach. In order to parameterize the simulation close to reality, analysis of shopping baskets based on real sales receipt data is one of the key steps. The combination of the data analysis and simulation is in the scope of this paper Analysis of real sales receipt data was done by determining so-called customer prototypes, which are representatives of different customer classes. In order to make the simulation insensitive against short transient changes, longitudinal analysis and clustering was carried out, additionally to the common cross section analysis. Longitudinal analysis was necessary to calculate similarities between shopping basket clusters of different points in time. The resulting stabile customer prototypes then were used to parameterize a supermarket simulation. As result, this paper shows the analysis of consumer data as well as the parameterizing and validation of a supermarket simulation. The combining of data mining with simulation results in a better accuracy of analysis and the ability to evaluate influencing factors that can not be extracted directly through data mining.

Modeling and wafer defect analysis in semiconductor automated material handling systems

Modeling and analysis of automated material handling systems in semiconductor manufacturing is difficult because of its complexity, the large amount of data originating from different sources as well as the often incomplete monitoring of transport processes. This article proposes an automated method and tool for building high level models of such systems based on transport logs, routing information and static system data. On the basis of this model, a method for tracing and correlating lot movements is presented and used to support system experts in locating fab areas that most likely caused defects measured on wafers, e.g. due to temporarily contaminated clean room air. In addition, several methods to analyze the transport systems performance, such as the determination of lot detours or causes for a potentially critical load of certain system parts are discussed.

Analysis of maintenance histories of industrial equipment with frequent maintenance demand

In order to guarantee high operational availability, the modular industrial equipment requires frequent maintenance, which oftentimes is carried out by the manufacturer. Reports about service technicians activities are stored in maintenance histories. Manufacturers of such equipment would benefit significantly from analysis of recorded maintenance and fault histories for planning of maintenance activities, offering scalable service contracts and finding reasons for product faults. This paper introduces a methodology that supports the interpretation of the maintenance histories to allow the manufacturers the analysis and optimization of maintenance operations. The methodology interprets the maintenance histories as sequences of events, containing meaningful patterns. Tailored data mining algorithms are applied, that provide causality details going beyond the results of standard techniques. The paper uses the example of maintenance reports of gas analytic equipment.

Automated planning, execution and evaluation of simulation experiments of semiconductor AMHS

Increasing variety and complexity of products in existing semiconductor factories cause an increased amount of production steps. Accordingly, this leads to a significant increase of non value added transportation processes. Therefore, transport and storage durations shall be minimized by optimal alteration of the given automated material handling system (AMHS). This can be achieved by simulation and analysis of possible AMHS alterations. However, this is a difficult task because of the systems complexity, the large amount of data and the high effort of manually modifying and testing many different AMHS alterations. In order to assist the system experts in executing these tasks, the authors suggest a method for automatic planning, execution and comparison of simulation experiments, including the automatic alteration of the transportation systems layout by introducing additional AMHS segments as shortcuts. The approach is feasible for existing simulation models as well as for generating simulations from the factorys core data.

Event Based Identification and Prediction of Congestions in Manufacturing Plants

In modern manufacturing plants, most prominently in semiconductor manufacturing, complex interwoven automated material handling systems are used to transport lots between productive service stations. In busy areas, these systems often show transient congestion phenomena that have a negative impact on the throughput of the factory. Because of the systems complexity and the large amount of data originating from several thousand transports a day, these congestions are difficult to detect and to trace or even to predict. In this paper, the authors present an approach to detect congestions in such systems using an event-based model building and analysis approach. Once such effects have been identified, an algorithm is proposed to derive congestion prognosis rules which can then be used to implement effective congestion prevention mechanisms.

Simulation based forecast of supermarket sales

Analysis and forecasting of the sales of products is extraordinary important for managers of supermarkets. In special, the forecasting of special bargain offers is of interest. This paper approaches this task by low level simulation of customers and products of a supermarket. The simulation is parameterized by results of analysis of real sales receipt data to achieve the results close to reality. The analysis includes building of probabilistic models based on clustering and time series analysis, as well as simulation, which is used to forecast time series of product sales. Along with results of analysis, the proposed simulation approach also incorporates results of empiric analysis of customer psychology. Further, the user can set other parameters like composition of customer types, prices of products, assortment of bargain offers, configuration of shelves and product locations, which allows investigation of different influences and specialized simulation of selectable products of interest.